Citation
Streamflow droughts in major watershed regions of the conteminous United States

Material Information

Title:
Streamflow droughts in major watershed regions of the conteminous United States investigating spatio-temporal characteristics and climatic connections
Added title page title:
Investigating spatio-temporal characteristics and climatic connections
Creator:
Poshtiri, Maryam Pournasiri ( author )
Language:
English
Physical Description:
1 electronic file (132 pages) : ;

Thesis/Dissertation Information

Degree:
Doctorate ( Doctor of philosophy)
Degree Grantor:
University of Colorado Denver
Degree Divisions:
Department of Civil Engineering, CU Denver
Degree Disciplines:
Civil engineering
Committee Chair:
Mays, David
Committee Members:
Pal, Indrani
Rajagopalan, Balaji
Livneh, Ben
Towler, Erin

Subjects

Subjects / Keywords:
Streamflow -- United States ( lcsh )
Droughts -- United States ( lcsh )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Review:
This PhD thesis evaluates historic patterns and spatial-temporal variability of streamflow drought indicators and their hydro-meteorological characteristics across the conterminous U.S. (CONUS). Streamflow drought indicators are developed using known approaches, such as low flow magnitude characteristics and deficit characteristics, based on observed daily streamflow records for 603 locations. A comparison of space-time variability patterns of different streamflow drought indicators show the necessity of evaluating the multi-feature characteristics of this natural hazard that has a different onset time and spatial extent from meteorological droughts in a given region of interest. A general increase in drying tendency of streamflow droughts is presented across the CONUS, corresponding with regionally consistent trends in climatic conditions on an annual and seasonal scales. Specifically, hotspot locations of negative trends are found in the northern Great Lakes region, Pacific Northwest, Southeast (Atlantic plains) and Southwestern U.S. Regional differences in the low flow trends are also notable. A reversal of trends in streamflow drought indicators is also seen in the Northeast and Pacific Northwest since 1980s. Patterns in return periods and corresponding return values of low flows are also reported, suggesting changing risk conditions that are important for water resources decision-making. Annual total number of days with zero flows and annual total number of days with flows going under a specific threshold were helpful to identify severely intermittent streams, which are found in southwestern (California, lower Colorado), western (Rio Grande, Texas-Gulf), some of the central locations of the U.S. (such as Arkansas White red and Ohio), and in some parts of northern U.S. (upper Missouri). The spatial clusters of locations were identified and presented, indicating homogeneous regions for streamflow drought indicators with similar variability. Further analyses indicated how these clusters of locations connect to the different regional and large-scale climate. There are strong associations between the streamflow drought indicators in the homogeneous regions and historic variations in climate. In general, driest river conditions co-occur with dry and warm regional climatic conditions in an annual scale displaying a significant link with both Pacific and northern Atlantic Ocean warm pools.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: Adobe Reader.
Restriction:
Embargo ended 12/11/2019
Statement of Responsibility:
by Maryam Puurnasiri Poshtiri.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
on10287 ( NOTIS )
1028749425 ( OCLC )
on1028749425
Classification:
LD1193.E53 2017d P67 ( lcc )

Downloads

This item has the following downloads:


Full Text
20
Most drought impacts are not simply caused by lack of precipitation, but it is mainly caused by a lack of available water resources at the specific time such as seasonal anomalies of soil moisture, groundwater, or streamflow [Laaha el al, 2016], Streamflow droughts directly impact the quantity, quality, and reliability of streamflows at various temporal and spatial scales, and consequently affect a range of different water use sectors [Laaha el al., 2016], It determines the adequacy of flow of a river to supply and maintain enough water for disposal of liquid wastes, municipal or industrial supplies, supplemental irrigation, and ecological demand such as suitable conditions for fish [Riggs, 1972], Stream low flow influence physical habitat, ecological processes, and biodiversity within and among stream networks [Falke el al, 2011; Rolls el al, 2012; Jaeger el al, 2014; Satin el al.,2016], Thermal plant cooling in the U.S., supported by surface water withdrawals, has already been affected in areas where surface water supplies are diminishing, such as the southern United States [ Van Vliet el al., 2012; Wilbanks el al., 2012], Persistent low flows during warm seasons can limit the ability of utilities to discharge heated water to streams from once-through cooled power systems due to regulatory requirements and concerns about how the release of warmer water into rivers and streams affects ecosystems and biodiversity [Laaha etal, 2016], That may also require wastewater facilities to increase treatment to meet stream water quality standards [EPA, 2011], Streamflow droughts decrease reliability of streamflows for navigation, which consequently increase shipping costs and result in commodity price volatility [Melillo et al., 2014; Laaha et al., 2016], Persistent dry condition in rivers can alter flow continuity in time and space [Bogan etal, 2015; Dairy etal, 2016] and consequently streams may become more intermittent [Doll andSchmied, 2012; Jaeger et al, 2014; Reynolds et al, 2015; Datry et al, 2016], Social societal/cross boundary conflicts during low-flow periods have also been


Souris Red
Missour
♦Great Basin
rkansas White Red
Data Length (Year)
-zi
ZZ - ot'
Ot- zz oe-ot; m -oe
Figure 2.1: Water resources regions (pink boundaries) comprising of the major river basins in the CONUS, green bubbles indicate locations of HCDN-2009 stream gauges. The size of the bubbles is proportional to the data length in years.
to
00


I
r ^ Daily streamflow Data L. A
|
Low flow magnitude characteristics
Minimum n-day mean flow

Percentile (Qx) from the flow duration curve
I
\
Deficit characteristics
j
-
r >
Threshold level
i j
.
t >
Deficit indicators
J
1
Streamflow drought event characteristics
Streamflow drought event indicators
V_
Figure 2.2: The flowchart of Streamflow drought indicators calculation.


Figure 2.3: Illustration of the deficit characteristics and streamflow drought event features as defined with the threshold level method. Duration (di), Deficit volume or severity (Vi), Minimum deficit volume (Vmin,i) , Time of occurrence, Mutually dependent droughts (di, Vi; di+i ,Vi+i; ti < tc) and Minor droughts.


(A)
0.15
Log(absolute(trend value)x10A6)
Log(absolute(trend value)x10A6)
Log(absolute(trend value)x10A6)
50-
45-
40-
35-
30-
-100
long
-80
25-
-120

(9v0i.x(9n|BA puaj))9)n|osqe)6o~| (9vlHx(9nleA pu9Ji)9jn|osqe)6o_|


Trend Value
(B)
--------1-----------------1-----------------r
start yr-1960(81) start yr-1970(162) start yr-1980(248) start yr-1990(347) start yr-2000(487) start yr-2012(570)
Decade
Figure 3.1: (A) Trends of annual 7-days minimum flow (q7). Color bubbles indicate location of the stations, and sign and significance of trend estimates (in cubic feet per second (cfs) per day per year). The size of the bubble is proportional to the magnitude of trend value. The probability density functions of q7 trends are determined pooling together all the stream gauges located in the western, central, and eastern U.S. (top three panels), as well as northern and southern U.S. (right two panels); (B) Box plot showing distributions of trend values (in cubic feet per second (cfs) per day per year) over all locations having at least 30-years of data length, pooling together all the stations across CONUS, for different end year (decadal shift in trends). Number of stations considered for each decadal analysis is stated in the brackets. For example, only 81 stations are included for the analysis when 1960 is the end year, while 570 stations are included for 2012. Also shown in color bar plot is fraction of total number of stations showing positive or negative trends in the box plot.


56
Figure 3.2: Trends of seasonal q7. Color bubbles indicate location of the stations, sign and significance of trend estimates (in cubic feet per second (cfs) per day per year). The size of the bubble is proportional to the magnitude of trend value.
Figure 3.3 shows temporal shifts in trends, like Figure 3.1-B, but for each station. This figure suggests a reversal in the sign of trends over some regions in the CONUS, specifically, the Pacific Northwest (shift after 1980) and the Northeast (shift after 1970), which is not visible in Figure 3.1-B when all the stations are pooled together. To confirm this shift before and after 1980s, Figure 3.4 shows trends in annual q7 magnitudes, considering start year - 1979 and 1980 - 2012. This figure displays a reversal in the sign of trends over most regions in the CONUS after 1979, mostly shifts from positive significant trend to negative significant trend. Overall, this figure confirms the reduced wetting tendency in low flows in the recent decades. Figure 3.5 shows boxplots specific to each river basin (river basins information is in Table 3.2), pooling together all the stations’ q7 data, before and after 1980. Majority of the basins under study show drying trends (shift from red color to blue color, before and after 1980),


58
Figure 3.3: Trends of annual q7 where the end years are changing decade by decade. Color bubbles indicate location of the stations, sign and significance of trend estimates (in cubic feet per second (cfs) per day per year). The size of the bubble is proportional to the magnitude of trend value.


59
45
40
TO
35
30 25
-120 -100 -80
long
(A) Start year - 1979
(B) 1980-2012
Figure 3.4: Trends of annual q7: (A) start year to 1979, (B) 1980 to 2012. Color bubbles indicate location of the stations, sign and significance of trend estimates (in cubic feet per second (cfs) per day per year). The size of the bubble is proportional to the magnitude of trend value.


Trend Value
-0.6 -0.4 -0.2 0.0 02 0.4
Trend Value
0.0 0.1 0.2 0.3
Trend Value
-0.05 0.00 0.05 0.10
start yr-1980(30) 1980-2612(52)
Decade
Trend Value
-0.2 -0.1 0.0 0.1 0.2 0.3
Trend Value
-0.1 0.0 0.1 0.2
o
Trend Value
-1.5 -1.0 -0.5 0.0 0.5 1.0
Trend Value
-1.0 -0.5 0.0 0.5 1.0


62
start yr-1980(2) 1980-2012(27)
Decade
3 P
03 °
10
start yr-1980(9) 1980-2012(20)
Decade
start yr-l980(12) 1980-2012(35)
Decade
start yr-1980(6) 1980-2612(12)
Decade
start yr-l980(13) 1980-2612(24)
Decade
Figure 3.5: Box plots of annual q7 trend magnitudes for major river basins, before and after 1980. Most of the basins show drying trends (red color before 1980 to blue color after 1980), where median trend value decreased, only except basin 9. Number of stations included in the analyses are indicated in the brackets. Stations having at least 30-y ears of data availability were included.


350
63
Return Period! Year)
Return Period! Year)
02
Return Period!Year)
04
Return Period! Year)
06


64
Figure 3.6: Return level plots of q7 for different river. Different colors indicate different end year of analysis. Bigger graph shows return periods of q7 where 2-yr, 5-yr and 10-yr return levels are marked by dotted lines, and the graph in the inset shows the changes of 2q7, 5q7, and 10q7 values over the decades.


65
Trends in Deficit Indicators
The map of monotonic trends in CDO is shown in Figure 3.7. It is expected that the trends in CDO would be similar to those in q7 and q30, but opposite in sign, which is found in Figure 3.7. In a location where low flows are declining, the occurrence of dry days are increasing. For example: Northern Great Lakes region, Pacific Northwest, the Southeast (Atlantic plains) and parts of Southwestern U.S. (western side of upper and lower Colorado River basins) have positive significant trends in CDO, which makes these regions “hotspots” where a focus on adaptive management measures are most needed. Thus, the results in Figure 3.7 are consistent with those from Figure 3.1 (and so is also found for q30, Figure A2 in Appendix).
Seasonal trends in CDO corresponding to the 10q7 threshold are shown in Figure 3.8. Consistent with q7 in Figure 3.2, summer and autumn (JJA and SON) show similar outcomes as annual but Pacific Northwest shows positive trends in the spring (MAM).
In order to confirm the decadal shifts in the sign of trends, Figure 3.7-B displays how the boxplot of the CDO trend values shifted in the previous decades. The median trend values, although remained negative, increased in the last four decades in the overall CONUS. This confirms that the natural flows of the major rivers overall are becoming drier, most notably since 1980. This reduced wetting pattern is also confirmed in the increasing number of stations having positive trends in CDO in Figure 3.7-B color bar plot.


(A)
Significance
insig
+msig
-I'J
• -sig
-100
long
Log(absolute(trend value)x10A3)
Log(absolute(trend value)x10A3)
Log(absolute(trend value)x10A3)
40
+â– <
ra
35-
30-
45-
25-
^120
Trend Value • 0 • 2
• 4
• 6
• 8
On
On
(Cv(Hx(»nlBA puaj})ajn|osqe)6o-| (CvOlx(3nlBA puaj|)a}n|osqe)6on


(B)
start yr-1960(81) start yr-1970(162) start yr-1980(248) start yr-1990(347) start yr-2000(487) start yr-2012(570)
Decade
Figure 3.7: (A) Trends of annual CDO (total number of days below 10q7). Color bubbles indicate location of the stations, and sign and significance of trend estimates (in days per year). The size of the bubble is proportional to the magnitude of trend value. The probability density functions are determined pooling together all the stream gauges located in the western, central, and eastern U.S. (top three panels), as well as northern and southern U.S. (right two panels); (B) Box plot showing distributions of trend values (days per year) over all locations having at least 30-years of data length, pooling together all the stations across the CONUS, for different end year (decadal shift in trends). Number of stations considered for each decadal analysis is stated in the brackets. For example, only 81 stations are included for the analysis when 1960 is the end year, while 570 stations are included for 2012. Also shown in color bar plot is fraction of total number of stations showing positive or negative trends in the box plot.
On
^1


68
DJF
*.8 1/ •
f.

• ^

• • * • • •# • • •• • • 0 \ • ■t. - • \ • #1 * Vt • • ». T
u*
**-_. f . V, -
:z |
r I
.
•'» •*'U • J-#
MAM
f? *>
s% - *
if . I* > '
;*V
•'-N J -' Uk>v*'
V A
. ifiL f.' / 71."“ •*
-120 -100 long SON

c » 8 *.
NT . * .T •« •••••) •
-/• * • >-# V •• • • V « 0 • r. . .1 • * •«t* . *
• . «• . •* • vS? »****•
.4^.
. *?
*
V
• •iraig • 10
• -m • is
• ">*J • 2 0
long
Figure 3.8: Trends of seasonal CDO (total number of days below 10q7). Color bubbles indicate location of the stations, and sign and significance of trend estimates (in days per year). The size of the bubble is proportional to the magnitude of trend value.
Patterns in annual SDS and CDV are also found to be similar to annual CDO in Figure 3.9 and 3.10, respectively. Patterns in SDS, in Figure 3.9, indicate longer deficit periods in the Southeast, northern Great Lakes, and the western U.S. Therefore, annual persistence of dry days, indicated by SDS, and cumulative flow deficit also confirm changes found for the low flow magnitudes and frequency of dry days. Thus, multiple indicators of hydrological droughts considered in this study consistently depict a reduced wetting tendency and often drying
tendency of the major watershed regions comprising of the key rivers in the mainland U.S.


69
(A)
10q30
(B)
40
q30
Figure 3.9: (A) Trends of annual SDS (total number of consecutive dry days below 10q7 and 10q30). Color bubbles indicate location of the stations, and sign and significance of trend estimates (in days per year). The size of the bubble is proportional to the magnitude of trend value; (B) Proportions of total number of stations showing different types of trends in annual persistence of dry days corresponding to 10q7 and 10q30.


70
start yr-1960(81) start yr-1970(162) start yr-1980(248) start yr-1990(347) start yr-2000(487) start yr-2012(570)
Decade
Figure 3.10: (A) Trends of annual CDV. Color bubbles indicate location of the stations, and sign and significance of trend estimates (in cfs per year). The size of the bubble is proportional to the magnitude of trend value. (B) Box plot showing distributions of trend values over all the locations having at least 30-years of data length, pooling together all the stations across CONUS, for different end year (decadal shift in trends). Number of stations considered for each decadal analysis is stated in the brackets. For example, only 81 stations are included for the analysis when 1960 is the end year, while 570 stations are included for 2012. Also shown in color bar plot is fraction of total number of stations showing positive or negative trends in the box plot.


71
Summary and Discussion
This national scale study on streamflow drought indicators is based on natural daily flow records from 603 USGS stream gauge stations that are minimally influenced by upstream water uses, diversions, impoundments, or land-use changes. A non-parametric trend analysis of different indicators yields a general reduced wetting trend in the U.S., as opposed to increase in average U.S. precipitation since 1900 [Melillo et al., 2014], increase in the wet days [Palet al., 2013] and increase in water availability within the U.S. [McCabe and Wolock, 2002], Furthermore, patterns in streamflow drought indicators along the latitudes and longitudes differ, specifically, more drying patterns are observed over the southern U.S., which corresponds to previous precipitation studies [Milly et al., 2005],
The significant spatial clustering of similar sign trends in streamflow drought indicators shown in this study suggests systematic causes. Increases in the dry conditions could be driven by different environmental changes [DeFries and Eshleman, 2004], Generally, the reductions in precipitation with increased evaporation along with higher temperatures can trigger the changes in streamflow drought indicators [Sheffieldand Wood, 2008], The streamflow drought indicators in this study are associated with the regionally consistent trends in climatic conditions, as synthesized in the NCA. For example, increases in the 7-day minimum flow magnitudes and decreases in the frequency of dry days in the Midwest and Northeast are associated with average increases in the annual precipitation and runoff patterns in that region [Melillo et al., 2014], Further, runoff was shown to increase in the Mississippi Basin and the Northeast, but a declining trend was documented in annual runoff in the Colorado River Basin [Melillo et al., 2014], Heavy downpours were also reported to be increasing over the last three to five decades in the CONUS, with the largest increases in the Midwest and Northeast, and


72
upper Great Plains [Melillo et al., 2014], Although attribution of trends of streamflow drought indicators cannot simply be explained without considering decadal to multi-decadal variability [Novotny and Stefan, 2007], there is significant correspondence between the changes in annual precipitation, runoff, and surface soil moisture [Melillo et al., 2014], and streamflow drought indicators reported in this study. Furthermore, this study finds an increase in streamflow dry days (CDO), streamflow dry spells (SDS), and cumulative streamflow deficits (CDV), and a decrease in low flow magnitudes (q7 and q30) in the Southwest, where according to NCA, surface soil moisture has exhibited drying trends [Melillo et al., 2014], Recent soil moisture trends also show moistening in the Northeast and northern Great Plains and Midwest [Melillo et al., 2014],
Some regions went through a complete shift in trends before and after 1980, for example, the Northeast and Pacific Northwest. A comparison between the probability density functions of magnitude and sign of trends for every basin in this study, before 1980 and after 1980, yield reduced wetting and often drying patterns for most basins. Yulianti and Bum (1998) noticed opposite streamflow trends in Canada before and after 1970 [Yulianti and Burn, 1998], Milly et al. (2005) selected 1970 as a turning point year because of noticeable changes in global-mean measures of hydro-climate around this time [Milly et al., 2005],
Although climate is an important driver of patterns in streamflow drought indicators, it would be inappropriate to attribute these hydrologic trends to climate change alone, without consideration of how land use and water management may have influenced the trends [DeFries andEshleman, 2004; Schilling et al., 2008; Tomer and Schilling, 2009]. Given that the HCDN is not fully free of changes in water and land management, these factors could also attribute to the changes in streamflow drought indicators. For example: in the northern midwest, increasing


73
low flows may be attributable to increased cropping [Schilling et al., 2008; Tomer and Schilling, 2009] and/or increased irrigation because agricultural changes could be associated with ecohydrologic shifts [Schilling et al., 2008; Tomer and Schilling, 2009] and irrigation may alter the atmospheric transport of water vapor via partitioning of surface energy fluxes [Huber et al., 2014], In the Southeast, Pacific Northwest, and Rocky Mountains, the creation of forest plantations may have contributed to decreasing late summer low flows because higher leaf-area index may result in lower streamflow and higher evapotranspiration [Stohlgren et al., 1998; VanShaar et al., 2002], Similarly, in the northeast, increasing low flows may also be the result of forest succession.
Finally, there are important implications of this work for the different regions. The combination of longer low flow periods with rising stream temperatures, especially during summer, may affect the efficiency of thermoelectric power production, make electric power plant cooling water withdrawals unreliable [Van Vliet et al., 2012], may degrade habitats and favor invasive, non-native species, and thus affecting aquatic and riparian ecosystems [Falke et al., 2011], It also limits the ability of utilities to discharge heated water to streams from once-through cooled power systems due to regulatory requirements and concerns about how the release of warmer water into rivers and streams affects ecosystems and biodiversity. The longer low flow periods are observed in the Southeast, northern Great Lakes, and the western U.S.; thus, if stream temperatures increase in these regions, it could lead to potential threats for electric power plants. In addition, a combination of a decrease in magnitude and an increase in the frequency of low flow results in poor water quality conditions in a river, with negative implications for aquatic life. This brings another concern for the Southeast, northern Great Lakes, and the western U.S. locations.


74
Furthermore, water sector withdrawals and uses vary significantly by region and correlate with climate [Melillo et al., 2014], The Southwest, Great Plains, and Southeast are particularly vulnerable to changes in water supply and demand. Irrigation is dominant along the Mississippi Valley, in Florida, and in southeastern Texas, where dry hydrological extremes are trending to be drier (Figure 3.1). Groundwater withdrawals are especially intense in parts of the Southwest, Southeast, Northwest, and Great Plains, the Mississippi Valley, Florida and South Georgia, and near the Great Lakes. Drier conditions in these regions could drive greater groundwater usage.
Overall, this research presents new findings on the regional changes and decadal shifts in patterns in streamflow drought indicators in major watershed regions in the CONUS. This work provides additional evidence that water resources could possibly being systematically impacted by climate and environmental changes. Therefore, water resources managers, engineers and planners may face changes in the risks and impacts, as well as opportunities, which may not be properly identified and incorporated in the existing practices. Water management should thus rely less on historical practices and responses and more on robust, risk-based, and adaptive decision approaches [Milly et al., 2015],


75
CHAPTER IV
SPATIO-TEMPORAL VARIABILITY OF STREAMFLOW DROUGHT INDICATORS ACROSS CONTERMINOUS U.S.
Abstract
This study presents spatial-temporal variability of streamflow drought indicators across the conterminous United States (CONUS) spanning the period from 1963-2012, and its connection to regional and large scale climate anomalies. This study uses Partitioning Around Medoids (PAM) algorithm along with F-Madogram, a non-parametric clustering method, to systematically identify the homogeneous regions for streamflow drought indicators. There are strong associations between the streamflow drought indicators and historic variations in regional climate indices, including PMDI, a known meteorological drought indicator. In general, driest river conditions co-occur with dry and warm regional climatic conditions. Hence, in an annual scale, streamflow drought events show coincidence with known historic meteorological drought events. Furthermore, river dry days’ indicators also display a significant link with the combination of Pacific and Atlantic Ocean warm pools.
Introduction
The significant spatial patterns of similar sign trends in streamflow drought indicators (Chapter III) echo the local climatic changes and soil moisture trends documented in the recent National Climate Assessment (NC A)\Pournasiri Poshtiri and Pal, 2016], In the last four decades, low flow magnitudes have shown negative patterns in most river basins within the conterminous U.S. (CONUS), particularly, in the northern Great Lakes region, Pacific Northwest, southeast (Atlantic plains) and southwestern U.S. [Pournasiri Poshtiri and Pal, 2016], indicating a gradual increase in the drying tendency of rivers across the CONUS.


76
Spatial and temporal configurations of stream low flow influence physical habitat, ecological processes, and biodiversity within and among stream networks [Rolls et al., 2012; Jaeger et al., 2014], Tests of regional homogeneity form an integral part of river flow regionalization procedures, such as homogeneous low flow regions, will extract relevant information hidden in the complex spatial-temporal low flow data, and be used to transfer those statistics from gaged to ungaged sites [Ries, 2007], Numerous researchers have used and documented statistical approaches to regionalize drought indices, predominantly in the context of climate [e.g., Fovell, 1997; Lund and Li, 2009], or annual streamflows [McCabe and Wolock, 2014], Despite the recognition that the low flows have widespread hydrological, economic, and social implications, to our knowledge, no study so far has presented spatio-temporal characteristics of streamflow drought features for U.S. rivers.
Persistent dry condition in rivers can alter flow continuity in time and space [Bogan et al., 2015; Datry et al., 2016], As a result, under climate change and landuse modifications there can be a growing tendency of potential shift in perennial to intermittent streams [Doll and Schmied, 2012; Jaeger et al., 2014; Reynolds et al., 2015; Datry et al., 2016] with longterm effects on biodiversity and ecosystem functions [Bogan et al., 2015; Datry et al., 2016; Vander Vorste et al., 2016; Leigh and Datry, 2017], In fact, many of the CONUS rivers are already experiencing increasing intermittency [Jaeger et al., 2014; Reynolds et al., 2015], However, headwater streams are still poorly characterized in the U.S. intermittency [Jaeger et al., 2014; Reynolds etal., 2015],
Streamflow drought is part of local hydrology. During dry weather, runoff is usually low, but potential evapotranspiration may be high due to increased radiation, wind speed, or vapor pressure deficits. This can lead to increased actual evapotranspiration, resulting in an extra loss


77
of water from the open water bodies, and consequently, sustained streamflow deficit conditions \Menzel and Burger, 2002; Whitfield et al., 2003; Van Loon and Laaha, 2015; Wang et al., 2015; Huang et al. ,2016], On the other hand, spatiotemporal variability of streamflow drought characteristics can be associated with large-scale atmospheric or oceanic patterns (e.g., El Nino-Southern Oscillation (ENSO), North Atlantic Oscillation (NAO)) [Lins and Slack, 1999; Van Loon and Van Lanen, 2012; Van Loon, 2015], Occurrence of many extreme droughts and megadroughts over most of the CONUS have been linked with cool SSTA (e.g., La Nina phase) in the middle and eastern tropical Pacific (ENSO region) or the warm SSTA in northern Atlantic Ocean [Palmer andBrankovic, 1989; McCabe et al., 2004; Cook et al., 2010; Dai, 2013; Peterson etal., 2013], However, no study so far has looked into how regional and large-scale climate directly links with regional river drying or streamflow drought patterns.
The main objective of this study are to (1) use a spatial clustering approach for important streamflow drought indicators and identify homogeneous groups of locations having similar variability, (2) characterize intermittency of streams in the headwater locations across the CONUS, and (3) present streamflow drought connections with regional and large-scale climate. This study identifies region-specific climate indices those associate with streamflow drought occurrence and magnitudes within the CONUS such that water managers are able to forecast streamflow drought magnitudes once climate model forecasts become available some time in advance.
Data
Streamflow Drought Indicators
For this study, annual minimum 7-day mean flow (q7) and annual cumulative stream dry days occurrence (CDO) are selected from the dataset developed in Chapter II. q7 is most


78
popularly used regulatory tool by the USGS and the US-EPA for low flow frequency analysis and water quality studies library [Pournasiri Poshtiri et al., 2016], CDO is defined as the annual total number of days falling below 10-year q7 low flow.
Climate Data
Regional climate data (Precipitation and Temperature data): daily temperature (maximum and minimum values) and precipitation data are selected from United States Historical Climatology Network (USHCN) [Menne et al., 2010], The USHCN stations represent the long term high quality daily data from 1218 observing stations across the contiguous U.S. [Menne etal., 2010],
Palmer Modified Drought Index (PMDI): Gridded instrumental Palmer Modified Drought Index (PMDI) data is a small variant of the original PDSI that enables real time drought monitoring. PMDI accounts for evapotranspiration, soil moisture conditions, and precipitation. This study uses 1 degree gridded monthly data across the CONUS covering the period of 1895 to 2004 [Heim et al., 2007],
Large scale climate data: To study large-scale climatological patterns connected to streamflow droughts, Extended Reconstructed Sea Surface Temperature anomalies is slecelted from NOAA NCDC ERSST version3b, obtained in ready to analyze format from the IRI Data Library (http://iridl.ldeo.columbia.edu).
Method (Spatial Clustering)
Streamflow drought indicators are clustered using Partitioning Around Medoids (PAM) algorithm based on F-madogram [Bernard et al., 2013], PAM algorithm generates clusters around the representative stations called medoids. In this study, a medoid is a station where
CDO time series mimics the statistical distribution of the data within the other stations in the


79
cluster, and consequently, each data point keeps its properties [.Kaufman andRousseeuw, 2009; Bernard et al., 2013], A nonparametric approach called F-Madogram is used for measuring pairwise dependence among distributions (or time series or variables) and can be used as distance matrix in the PAM clustering algorithm [.Bernard et al., 2013], For a sample of variables (M(t)1, M(t)1)1 from two locations i and j at T different time units, madogram estimator dij is defined as:
T
d‘> = fZ ■ /?'(M'W) 1 (4 1)
t=l
Where Ft is the empirical distribution function, defined as [ Van der Vaart, 2000]:
number of elements in the sample < u , .
F,(u) =----------------------f---------------------- (42)
In this way, variable M1 is applied to its own distribution Fj(u) = P(Mj < it), and hence the entire function returns the empirical cumulative distribution, a proportion of the number of data points less than or equal to value u [.Bernard et al., 2013],
PAM moves around A number of medoids and try to minimize total intra-cluster distance. Medoids form the center of the cluster representing the valid annual q7 (or CDO) at each step of the algorithm.
To select the optimal number of clusters (K) and to distinguish the well classified station, silhouette coefficient [Zelenhasic and Salvai, 1987] can be used, which compares the tightness and separation of clusters. The silhouette coefficient (SC) for a station i is defined as:
SCj(K) = 1 — -t— (4.3)
ai,-K
Where diK is intracluster distance between medoid K and station i and di, -k is the smallest distance between station i and all the other medoids but K. SCi(K) has a value between -1 and


80
+1, where SCi(K) ~ +1 indicates that the intra-cluster distance is much smaller than the intercluster distance and the station i is well-classified; if SCi(K) ~ 0 then it is viewed as non-informative.
After running the PAM algorithm for a given number of clusters K, SC is calculated for each station. The average of all SCs helps evaluating the quality of a partitioning into K clusters. For a more detailed theoretical description of both PAM and FM algorithm applied here, the readers are kindly referred to [Cooley etal., 2006], \Naveau etal., 2009], and \Lehner et al., 2006],
Finally, 95th percentile significance levels of SCs for a given K is determined by shuffling the q7 (and CDO) data at each station, sampling that randomly, and running the PAM algorithm on it. This process breaks down any spatial and temporal dependence in the data. This algorithm is repeated 50 times. The 95th quantile values from 50 average SCs from randomized experiments are considered as significance levels, above which a station is marked as significant.[Bernardet al., 2013],
Results
This section presents the main findings and results of the study as follows: Section 4.4.1 presents spatio-temporal variability lowflow indicators (i.e., q7) and deficit indicators (i.e., CDO). Section 4.4.2 presents the river intermittency behavior within CONUS. Section 4.4.3 presents the association between river intermittency indicators (i.e., CDO) and climate
anomalies.


81
Spatial Variability of Streamflow Drought Indicators
Spatial Variability of Low Flow Indicators (i.e.. q7)
Figure 4.1-A summarizes SC distributions for q7 with 50 years data, where is varying from 3 to 25. The red line represents average SCs among all stations for a given K, the solid black lines correspond to the median values, and dotted blue lines displays 95th percentile values of SCs. Maximizing the mean SC and minimizing the number of negative SC values helps to select a K = 12. Figure 4.1-B shows the spatial clustering pattern of q7 for K = 12, which indicates that the low flow distribution in one cluster is different from the low flows in another cluster, as the spatial relationships between neighboring stations i.e. the bivariate distributions are stronger within individual cluster. Figure 4.2 shows the time series of q7 at stations corresponding to each cluster. The spatial relationships between clusters are variable due to the timing of occurrence of annual low flows. Figure 4.3 indicates a range of occurrence timings, often within a given cluster itself, where the spread is lower for the northeastern clusters (1, 2, 3), cluster 8 and 9, and Pacific northwest cluster (12), where low flows occur around September on an average, but the same for cluster 10 has the largest spread with a median timing of q7 occurrence in the fall.
It is also checked if the spatial patterns in Figure 4.1-B remain consistent for the variable data length, as we considered a tradeoff between the number of stations included in the analysis and their time lengths. Figures 4.4 and 4.5 show SC and spatial clustering patterns for the different lengths of low flow time series. An overall general consistency is noticed between Figure 4.4 and Figure 4.1-B, despite the differences in the data length and number of stations and often differing K (varied between 10 and 12).


82
(A)
50
45
40
35
30
25
(B)
• 12
-120
-100
-80
Figure 4.1: (A) The distribution of silhouette coefficients (SC) for annual q7 across stations within U.S. for 50 years data where the average SC is represented by the red line, the solid black lines correspond to the SC median values, and the dotted blue line corresponds to the 95 percentile levels; (B) Spatial map for clusters where medoids are represented by triangles and stations are attached to their medoids by gray lines, gray circles correspond to nonsignificant locations.


|enuuv
Cluster 1 Cluster 2
1970 1980 1990 2000 2010 1970 1980 1990 2000 2010
Cluster 3
1970 1980 1990 2000 2010
Cluster 5

....
1970 1980 1990 2000 2010
Cluster 6
Cluster 4
1970 1980 1990 2000 2010
1970
1980
1990
2000
2010
Cluster 7
1970 1980 1990 2000 2010
Cluster 8
1970
1980
1990
2000
2010
Cluster 9 Cluster 10
A 5.0- A AM
2.5
/ \; y A J A . /y o.o-
-2.5-
1970 1980 1990 2000 2010 1970 1980 1990 2000 2010
1970 1980 1990 2000 2010
1970 1980 1990 2000 2010
Year
Figure 4.2: Spaghetti plots (standardized) for all the time series of annual q7 for different clusters (1963-2012). Medoids are indicated as solid black line and gray colors indicate all the other stations.
OO


Month
Feb
Figure 4.3: Distribution of the occurrence time of q7 over all the locations within each cluster. Black line in the middle of the box is the median.


85
(A) 557 stations with 30 years data (1983-2012)
(B) 466 stations with 40 years data (1973-2012)
(C) 280 stations with 60 years data (1953-2012)
(D) 196 stations with 70 years data (1943-2012)
Figure 4.4: The distributions of SC for annual q7 across the U.S. with different data lengths where the average SC is represented by the red line, the solid black lines correspond to the SC median values. The dotted blue line corresponds to the 95 percentile level SC.


86
(A) 557 stations with 30 years data (1983-2012)
\ . Mf'
f-t. t . • «• fc. •. <■. • % « * 1 ♦£*4
ip • l • 1 4 fl ' a • . * tm'
V.*
â– if
5
(B) 466 stations with 40 years data (1973-2012)
d-.t /
*• •. • •• 5 . •

Cluster
• 1
• 2
• 3
• *
• 5
• e
• 7
• 8
• 9
• 10 • 11 • 12
(C) 280 stations with 60 years data (1953-2012)
’ $K
> ■ •
* % *
(D) 196 stations with 70 years data (1943-2012)
«r:
4 ;
* r^T-.
• •* l
» r • .*ci*
• I i» * IK
# • * •
!•»
Figure 4.5: Spatial clustering maps for annual q7 for K = 12 and different data length where medoids are represented by triangles and stations are attached to their medoids by gray lines, gray circles correspond to nonsignificant locations.
Spatial Variability of Deficit Indicators (CDO)
Figure 4.6 summarizes SC distributions for different data length of CDO where K = 2 to 25 are checked. Similar to q7, 50-years data is selected in the end for further analyses because this data length provides a good tradeoff between the number of years used in the study and corresponding number of stations across the CONUS providing a fair spatial distribution for clustering analysis. Nonetheless, we find a nice consistency between spatial clusters corresponding to the data having different lengths (Figure 4.7). Figure 8-A shows an enlarged version of the distribution of silhouette coefficients for 50 years’ data (1963-2012 and 393 stations) (also in Figure 4.6) where the red line represents average SC values among all stations


87
for each given K and the solid black lines correspond to the median values. K is selected based on where the SC has the highest average value and the lowest number of stations having negative SC values. Such condition yields for a K=3, where intra-cluster heterogeneity (i.e. higher intra-cluster distance) is also maintained, which is a condition to be checked in addition to average SC, as suggested by previous researchers [e.g., Bernardet al., 2013], As the second best K, this study also presents the spatial clustering pattern corresponding to K=12 to compare results with K=3. Figure 4.8-A (and Figure 4.6) displays 95th percentile values of SCs from 50 average samples as dotted blue lines. The stations having SCs below that threshold are considered as non-significant. Figure 8-B corresponds to the spatial clustering pattern for K=3 and 50 years data, while Figure 4.9 displays that for the other data length. Non-significant stations are displayed by grey colors and the medoids are displayed by triangles. Figure 8-B indicates that K=3 divides the CONUS into three large regions (east, midwest, and west), while K=12 in Figure 9-C creates smaller regions (6 regions in total in the east, 4 regions in the midwest, and 2 in the west). A cross-correlation analysis between CDO time series at medoid locations indicate no significant associations between K=3 medoids (Figure 4.10-A) but that for K=12 medoids yields associations between neighboring clusters (e.g., clusters 1-4, clusters 5-10, and clusters 11-12) (Figure 4.10-B).
Figure 4.11 shows time series plots for CDO for each cluster where K=3 gives a higher spatial variability in the data than K=12 (Figure 4.12), especially cluster 3. Therefore, for K=3 related climate analyses the standardized data (z-scores) within each cluster is combined by selecting the significant stations and using their SC values as weights. However, for K=12 related climate analyses medoids are still used as representative stations (variables).


Silhouette Coefficients (SC)
88
(A) 557 stations with 30 years data (1983-2012)
(B) 466 stations with 40 years data (1973-2012)
Figure 4.6: The distributions of SC for annual CDO across the U.S. with different data lengths where the average SC is represented by the red line, the solid black lines correspond to the SC median values. The dotted blue line corresponds to the 95 percentile levels.


(A) 557 stations with 30 years data (1983-2012)
89
(B) 466 stations with 40 years data (1973-2012)
• i
(C) 280 stations with 60 years data (1953-2012)
Figure 4.7: Spatial clustering maps of CDO for variable data length where medoids are represented by solid triangles and streamgauge stations are attached to their medoids by gray lines, gray circles correspond to nonsignificant locations.


90
(A)
-120 -100 -80
Figure 4.8: (A) The boxplot summarizes the distribution of silhouette coefficients (SC) for annual CDO across stations within U.S. for 50 years data where the average SC is represented by the red line, the solid black lines correspond to the SC median values, and the dotted blue line corresponds to the 95 percentile levels; (B) Spatial map for clusters where medoids are represented by triangles and stations are attached to their medoids by gray lines, gray circles correspond to nonsignificant locations.


91
(A) 557 stations with 30 years data (1983-2012)
« S -- • » •. v Vr.
• ;■ • jay


(B) 466 stations with 40 years data (1973-2012)
• . •. * . Cluster
f 3* p ; .?v. • *?*it
?*i . . * *.r . «5y • <
..f . ••
• s
• 9
• 10 • 11 • 12
.•'t.V'v
*
V? .C V, \ II
(D) 280 stations with 60 years data (1953-2012)
&M,
v.*- 7 «. ..
- c~-
•: “)y
*V-
PI '■ f i • ^ *
tSv *.y
3 A • # • V
• ‘a *«*

A
Figure 4.9: Spatial clustering maps of annual CDO for K = 12 and different data length where medoids are represented by triangles and stations are attached to their medoids by gray lines, gray circles correspond to nonsignificant locations.
River Intermittency Indicators across the CONUS
Annual total number of days with zero flows (MZD) or annual total number of days with flows going under a specific threshold can be used to characterize severe intermittency of streams [Jaeger etcil., 2014; Engetal., 2016], This study presents spatial variability of median of the MZD (Figure 4.13-A) as well as median of CDO (Figure 4.13-B) at each location of interest. Both indicates severe intermittency within streams across southwestern (California, lower Colorado), western (Rio Grande, Texas-Gulf), some of the central locations of the U.S. (such as Arkansas White red and Ohio), and in some parts of northern U.S. (upper Missouri).


92
Figure 4.10: Correlation coefficients between medoids for (A) K=3, and (B) for K=12. Ml to M12 display medoid 1 to medoid 12, respectively.


Annual CDO (day)
1970 1980 1990 2000 2010
Year
Figure 4.11: Spaghetti plots for time series of CDO for K=3. Medoids are shown by solid thick lines. To be noted that only stations with significant SC are included in this graph.
VO
u>


Annual CDO (day)
1970 1980 1990 2000 2010
1970 1980 1990 2000 2010
1970 1980 1990 2000 2010 1970 1980 1990 2000 2010
Year
Figure 4.12: Spaghetti plots for time series of CDO for K=12. Medoids are shown by solid thick lines. To be noted that only stations with significant SC are included in this graph.
VO


95
Figure 4.13: (A) Median of zero days of streamflow (MZD) within CONUS, (B) Median of stream dry days indicator (CDO) within CONUS.
The positive association between CDO and MZD (Figure 4.14) suggests that CDO can be used to study river intermittency in the headwater streams and its connections with climate, where the impact of zero values at some locations in the wet regions is minimized. Therefore,


96
in the next section, this study explores the connection between the CDO, an important indicator of streamflow drought, and climate anomalies.
Figure 4.14: Correlation map between MZD time series and CDO time series with 50 years data (1963-2012) at each location (upward tringles display positive correlations).
Climate Connections to Streamflow Drought Indicators
Local-Scale Climate
Here, this study examines the correlations and composite analyses between CDO (both combined CDO data (CCDO) for K=3 and medoid CDO data (MCDO) for K=12) and climate indicators derived from precipitation and temperature data (Table 4.1), and Palmer Modified Drought Index (PMDI). Both annual CCDO and MCDO has strong positive correlations with annual cumulative precipitation dry days (CPD) (Figure 4.15-A and Figure 4.16-A) and negative correlations with cumulative annual precipitation (CAP) (Figure 4.15-B, and Figure


97
4.16-B). This indicates that in a year when cumulative annual precipitation quantity and the number of wet days is low, there tends to be significantly higher number of river dry days. On the other hand, both CCDO and MCDO positively correlate with local annual daily maximum temperature magnitude (Tmax), the heat wave index (HWF), and diurnal temperature ranges (DTR) (Figures 4.15-CtoE as well as Figures 4.16-CtoE), meaning that extreme hot years tend to associate with drier rivers. Annual mean daily temperatures (Tave) have significant positive correlations with CCDO but predominantly for cluster 3 (Figure 4.15-F), and that for DUT (total annual number of days going above annual mean daily temperature) and annual minimum temperature (Tmin) are also positive but mainly nonsignificant in all clusters (Figure 4.15-GtoH, and Figure 4.16-GtoH). Overall, these findings indicate that cumulative number of dry days in a river in any region in the CONUS - a primary streamflow drought or river intermittency indicator, associate significantly with extremely hot and precipitation dry years in the same homogenized cluster.
Table 4.1 : Local precipitation and temperature indices used
Climate variable Indicator Description
Precipitation CPD Cumulative precipitation dry days ( Total annual number of days with precipitation less than 1 mm)
CAP Cumulative annul precipitation ( Total annual precipitation)
Tmax Annual maximum value of daily maximum temperature
Temperature HWF 95th percentile of daily maximum temperature over all the years (heat wave index)
DTR Annual mean value of diurnal temperature ranges
Tave Annual mean daily temperature
DUT Total annual number of days going above annual mean daily temperature
Tmin Annual minimum value of daily minimum temperature


98
Furthermore, PMDI is a measure of meteorological drought/floods accounting for evapotranspiration and soil moisture conditions, as well as precipitation in a region. A linear associations is found between CCDO/MCDO and PMDI, which are predominantly negative and significant within each cluster (Figure 4.15-1 and 4.16-1). This further reinforces the associations between meteorological and hydrological droughts in a given year.
Figure 4.17 (and Figure 4.18) shows precipitation and temperature composites over driest years of rivers in the history where within-cluster precipitation and temperature data is considered. Driest years are determined as the years with CCDO/MCDO magnitudes more than their long-term 75th percentile values. Figures 4.19 and 4.20 display the driest years within each cluster, where many being concurrent (e.g., 1988, 2002) or perpetual (e.g., 1991-1995, 1999-2004, 2011-2012) are found. Different shades of blue and red colors (from light to dark) in Figure 4.17 and 4.18 display the average positive and negative deviations of CCDO and MCDO values within the driest years from corresponding long term means, i.e. 0-25%, 25-50%, 50-75%, or greater than 75% deviation values. In general, driest years mostly coincide with positive deviations of annual cumulative precipitation dry days from the mean (Figure
4.17- A), negative deviations of cumulative annual precipitation values from the mean (Figure
4.17- B), and positive temperature deviations from the mean (Figure 4.17-C to 4.17- F). The same analysis but for the data across the entire CONUS (Figure 4.21) indicate that regional precipitation and temperature indices associate better with driest years.


99
(A) CPD: Positive correlation
(D) HWF: Positive correlation
(G) DUT: Positive correlation
(B) CAP: Negative correlation
(E) DTR: Positive correlation
(H) Tmin: Positive correlation
$6
v* -9* % 4 - * A .A* wtAjI#
r-
(C) Tmax: Positive correlation
(F) Tavg: Positive correlation
(I) PMDI: Negative correlation
Cluster
• 1
• 2
• 3
Figure 4.15: Correlations between CCDO and annual regional precipitation and temperature indices, and PMDI falling within the range of each cluster. Solid colored tringles correspond to the climate stations with significant correlations (upward tringles displaying positive correlations and downward tringles negative correlations), where different colors illustrate different clusters. Climate stations with nonsignificant correlations are shown as empty tringles. Grey circles display streamflow station locations with significant SC, indicating cluster borderlines and solid colored circles displaycorresponding medoid locations.


100
(A) CPD: Positive correlation (B) CAP: Negative correlation (C) Tmax: Positive correlation
. |t i % ° *•> t ^ *W*r i T) Jill fid* ,SA. ^ „ »> yit • - v* * ftf « • A .. - ° 1
(D) HWF: Positive correlation (E) DTR: Positive correlation (F) Tavg: Positive correlation
f i»|* * • • k % ■ 1 ■P * * * * * 4 * ■ $*. • . W . ’ • * 4# : • .: . A ;«A S • _JL P .1 / r SRT . * . * tP ;
* S- -a • jy *** l * ^ : i
(G) DUT: Positive correlation (H) Tmin: Positive correlation (I) PMDI: Negative correlation
• • • s? ; a* * ♦ ^ m £ * .-ft.* ,4 vW? • % It / • At ■>» 4*4' 4m*_ <4>4«4iA *4*4i4 44** ****** * ****** ***** >1** ***** ****** ***** '» *************** r;xi ****** o a$*3E^*a ****** ****!* * ************ ****** ***** * ****** M...4—JM*
• It | # • *• r*/ C . ^ • ** * *;* ^ ***y 0
^ ; • ***** ***JB******* >> **** *** ** • O 4*
Cluster • 1 • 2
• 3
• 4
• S
• 6
• 7
• e
• 9
• 10 • 11 • 12
Figure 4.16: Correlations between annual MCDO and annual regional precipitation and temperature indices, and PMDI falling within the range of each cluster. Solid colored tringles correspond to the climate stations with significant correlations (upward tringles displaying positive correlations and downward tringles negative correlations), where different colors illustrate different clusters. Climate stations with nonsignificant correlations are shown as empty tringles. Grey circles display streamflow station locations with significant SCs, and solid colored circles display medoid locations.


Full Text

PAGE 1

STREAMFLOW DROUGHTS IN MAJOR WATERSHED REGIONS OF THE CONTERMINOUS UNITED STATES: INVESTIGATING SPATIO TEMPORAL CHARACTERISTICS AND CLIMATIC CONNECTIONS by MARYAM POURNASIRI POSHTIRI B.S., Isfahan University of Technology , 200 4 M.S., Power and Water Univ ersity of Technology, 2006 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Doctor of Philosophy Civil Engineering Program 201 7

PAGE 2

ii This thesis for Do ctor of Philosophy degree by M aryam Pournasiri Poshtiri has been approved for the Civil Engineering Program by David Mays, Chair Indrani Pal , Advisor Balaji Rajagopalan Ben Livneh Erin Towler Date : December 16, 2017

PAGE 3

iii Pournasiri Posht iri, Maryam (Ph.D., Civil Engineering Program ) Streamflow Droughts in Major Watershed Regions of the Conterminous United States: Investigating Spatio temporal Characteristics and Climatic Connections Thesis directed by Research Assistant Professor Indrani Pal ABSTRACT This PhD thesis evaluates historic patterns and spatial temporal variability of streamflow drought indicators and their hydro meteorological characteristics across the conterminous U.S. (CONUS). Streamflow drought indicators are developed usi ng known approaches, such as low flow magnitude characteristics and deficit characteristics, based on observed daily streamflow records for 603 locations. A comparison of space time variability patterns of different streamflow drought indicators show the n ecessity of evaluating the multi feature characteristics of this natural hazard that has a different onset time and spatial extent from meteorological droughts in a given region of interest. A general increase in drying tendency of streamflow droughts is p resented across the CONUS, corresponding with regionally consistent trends in climatic conditions on an annual and seasonal scales. Specifically, hotspot locations of negative trends are found in the northern Great Lakes region, Pacific Northwest, Southeas t (Atlantic plains) and Southwestern U.S. Regional differences in the low flow trends are also notable. A reversal of trend s in streamflow drought indicators is also seen in the Northeast and Pacific Northwest since 1980s. Patterns in return periods and co rresponding return values of low flows are also reported, suggesting changing risk conditions that are important for water resources decision making. Annual total number of days with zero flows and annual total number of days with flows going under a speci fic threshold were helpful to identify severely intermittent streams, which are found in southwestern (California, lower Colorado), western

PAGE 4

iv (Rio Grande, Texas Gulf), some of the central locations of the U.S. (such as Arkansas White red and Ohio), and in so me parts of northern U.S. (upper Missouri). The spatial clusters of locations were identified and presented, indicating homogeneous regions for streamflow drought indicators with similar variability . Further analyses indicated how these clusters of locatio ns connect to the different regional and large scale climate. There are strong associations between the streamflow drought indicators in the homogeneous regions and historic variations in climate. In general, driest river conditions co occur with dry and w arm regional climatic conditions in an annual scale displaying a significant link with both Pacific and northern Atlantic Ocean warm pools. The form and content of this abstract are approved. I recommend its publication. Approved: Indrani Pal

PAGE 5

v DEDICATION To Mom who has always been a source of inspiration and encouragement to me. To Barbara and Rober t for being my constant support . To Indrani for being ambition and an innovative change maker. To Sara and David for being kind and generous .

PAGE 6

vi ACKNOW LEDGEMENTS First and foremost , I want to thank my advisor Dr. Indrani Pal. Her advice on both research as well as on my skills have been priceless. She has taught me, both consciously and unconsciously, the qualities of a good scientist and researcher. She also has greatly helped me with statistical methods and programming skills, specifically in R and Matlab. I appreciate all her contributions of time, ideas, and guidance to make my Ph.D. experience productive. I am thankful to my committee members, Drs. Ben Livneh, David Mays, Rajagopalan Balaji and Erin Towler for their thoughtful comments and suggestions, which greatly enhanced the quality of my research. I would especially like to thank Dr. T owler, for generously sharing her time, knowledge , and suppo rt in my research. I would also like to thank Dr. Upmanu Lall for his brilliant comments and suggestions on my research . I am extremely grateful to my friends for standing by my side at truly challenging times during the last four years . I am especially thankful to Dr. Maryam Darbeheshti for her great support through my brain injury. I also express my sincere gratitude to the Civil Engineering Department at the University of Colorado Denver for giving me an opportunity to teach two graduate level courses , which not only ease d my financial status for the past two years, but also allowed me to acquire valuable teaching skills and lasting memories. Throughout this process the support of my American host family has been vital. I sincerely thank Barbara and Ro bert for their nonstop support in the last four years.

PAGE 7

vii Last but not least, I would like to say thank you to my family to give me lot s of freedom and respect to my choices, and providing me unconditional love. I am truly grateful to my unique mother who is my role model throughout my life . Mom , you taught me to work hard, be independent, embra c e change s , and show compassion. We did this thesis together. Thank you.

PAGE 8

viii TABLE OF CONTENTS CHAPTER I. INT RODUCTION ................................ ................................ ................................ ........... 15 Drought and its Classification ................................ ................................ ......................... 15 Streamflow Droughts Generation ................................ ................................ ................... 17 The Significance of the Streamflow Drought Studies ................................ .................... 19 Objective and Research Aims ................................ ................................ ......................... 22 Outline of the Thesis ................................ ................................ ................................ ....... 23 I I. STREAMFLOW DROUGHT INDICATORS FOR THE CONTERMINOUS UNITED STATES ................................ ................................ ................................ ................................ 24 Abstract ................................ ................................ ................................ ........................... 24 Introduction ................................ ................................ ................................ ..................... 24 Data ................................ ................................ ................................ ................................ . 27 Method (Indicators Calculation) ................................ ................................ ..................... 29 Low Flow Characteristics ................................ ................................ ............................ 29 Deficit Characteristics ................................ ................................ ................................ . 33 Independent Drought Event Characteristics ................................ ................................ 37 Data Validation ................................ ................................ ................................ ............... 39 Summary and Discussion ................................ ................................ ................................ 44 III. DETECTING NON STATIONARITY IN STREAMFLOW DROUGHT INDICATORS ACROSS THE CONTERMINOUS U.S. ................................ .................... 46 Abstract ................................ ................................ ................................ ........................... 46 Introduction ................................ ................................ ................................ ..................... 47

PAGE 9

ix Data ................................ ................................ ................................ ................................ . 49 Method ................................ ................................ ................................ ............................ 50 Results ................................ ................................ ................................ ................................ ... 52 Trends in Low Flow ................................ ................................ ................................ .... 52 Trends in Return Levels ................................ ................................ .............................. 57 Trends in Deficit Indicators ................................ ................................ ......................... 65 Summary and Discussion ................................ ................................ ................................ 71 IV. SPATIO TEMPORAL VARIABILITY OF STREAMFLOW DROUGHT INDICATORS ACROSS CONTERMINOUS U.S. ................................ ............................. 75 Introduction ................................ ................................ ................................ ..................... 75 Data ................................ ................................ ................................ ................................ . 77 Streamflow Drought Indicators ................................ ................................ ................... 77 Climate Data ................................ ................................ ................................ ................ 78 Method (Spatial Clustering) ................................ ................................ ............................ 78 Results ................................ ................................ ................................ ............................. 80 Spatial Variability of Streamflow Drou ght Indicators ................................ ................ 81 Spatial Variability of Low Flow Indicators (i.e., q7) ................................ ........ 81 Spatial Variability of Deficit Indicators (CDO) ................................ ................ 86 River Intermittency Indicators across the CONUS ................................ ..................... 91 Climate Connections to Streamflow Drought Indicators ................................ ............ 96 Local Scale Climate ................................ ................................ ........................... 96 Large Scale Climate ................................ ................................ ........................ 105 Summary and Discussion ................................ ................................ .............................. 109

PAGE 10

x V. DISSERTATION CONCLUSIONS ................................ ................................ ............ 114 ................................ ................................ ................................ ...... 114 Discussion ................................ ................................ ................................ ..................... 116 Future Work ................................ ................................ ................................ .................. 118 BIBLIOGRAPHY ................................ ................................ ................................ ............... 119 APPENDIX ................................ ................................ ................................ ......................... 130 TRENDS OF ANNUAL Q30 AND ANNUAL CDO WITH 10Q30 THRESHOLD LEVEL ................................ ................................ ................................ .......................... 130

PAGE 11

xi LIST OF TABLE TABLE Table 2.1: Description of streamflow drou ght indicators. ................................ ...................... 30 Table 2.2: Standard deviation of different threshold level a cross all stations within CONUS ................................ ................................ ................................ ................................ ......... 40 Table 2.3: Description of selected streamflow drought indicators ................................ ......... 42 Table 2.4: Statistical differences ( t test results) between variability of different streamflow drought indicators ................................ ................................ ................................ ........... 4 3 Table 3.1: Fraction of total number of stations with different sign and sign ificance of trends. ................................ ................................ ................................ ................................ ......... 5 5 Table 3.2: Major water resources regions and their part numbers (two digit numbers designate major river basins). ................................ ................................ ................................ ......... 60 Table 4.1:Local precipitation and temperature in dices used ................................ .................. 9 7

PAGE 12

xii LIST OF FIGURES FIGURE 2.1 Water resources regions comprising of the major river basins in the CONUS. ........ 2 8 2.2 The flowchart of Streamflow drought indicators calculation. ................................ ... 3 1 2.3 Illustration of the deficit characteristics and pooled streamfow features as defined with the threshold level method . ................................ ................................ ............... 3 5 3.1 Trends of annual 7 days minimum flow (q7) ................................ ............................ 5 4 3.2 Trends of seasonal q7 ................................ ................................ ................................ . 5 6 3.3 Trends of annual q7 where the end years are changing decade by decade. ............... 5 8 3.4 Trends of annual q7: (A) start year to 1979, (B) 1980 to 2012 ................................ . 5 9 3.5 Box plots of annual q7 trend magnitudes for major river basins, before and after 1980. ................................ ................................ ................................ .......................... 6 2 3.6 Return level plots of q7 for different river ................................ ................................ . 6 4 3.7 Trends of annual CDO (total number of days below 1 0q7) . ................................ ...... 6 7 3.8 Trends of seasonal CDO (t otal number of day s below 10q7) . ................................ ... 6 8 3.9 Trends of annual SDS (total number of consecutive dry days below 10q7 and 10q30).. ................................ ................................ ................................ ...................... 6 9 3.10 Trends of annual CDV ( c umulative deficit volume below 10q7 ) ............................ 70 4.1 The boxplot summarizes the distribution of silhouette coefficients (SC) for annual q7 across stations within U.S. for 50 years data. ................................ ............................ 8 2 4.2 Spaghetti plots (standardized ) for all the time series of annual q7 for different clusters (1963 2012) . ................................ ................................ ................................ . 8 3 4.3 Distribution of the occurrence time of q7 over all th e locations within each cluster 8 4

PAGE 13

xiii 4.4 D istribution of SC for annual q7 across the U.S. with different data lengths. ........... 8 5 4.5 Spatial clustering maps for annual q7 for K = 12 and different data length. ............. 8 6 4.6 D istrib ution of SC for annual CDO across the U.S. with different data lengths ....... 8 8 4.7 Spatial clustering maps of CDO for vari able data length . ................................ ......... 8 9 4.8 (A) D istribution of SC for annual CDO across stations within U.S. for 50 years data , (B) Spatial clustering maps for annual CDO for K = 12 ................................ .......... 90 4.9 Spatial clustering maps of annual CDO for K = 12 and different data length. .......... 91 4.10 Correlation coefficients between medoids for (A) K=3, and (B) for K=12. ............ 9 2 4.11 Spag hetti plots for time series of CDO for K =3. ................................ ...................... 9 3 4.12 Spaghetti plots for time series of CDO for K =12. ................................ .................... 9 4 4.13 (A) Median of zero days of streamflow (MZD), (B) Median of stream dr y days indicator (CDO) ................................ ................................ ................................ ......... 9 5 4.14 Correlation map between MZD time series and CDO time series with 50 years data (1963 2012) at each location . ................................ ................................ .................. 9 6 4.15 Correlations between CCDO and annual regional precipitation and temperature indices, and PMDI falling within the range o f each cluster ................................ ..... 9 9 4.16 Correlations between annual MCDO and annual regional precipitation and temperature indices, and PMDI falling within the range of each cluster. .............. 100 4.17 Composite maps (for K=3) , where average deviations of different inter cl uster climate indices are shown . ................................ ................................ ..................... 10 1 4.18 Composite maps (for K =12), where average deviations of different inter clus ter climate indices are shown. . ................................ ................................ .................... 10 2 4.19 Historical occurrence of dry years of CCDO within each clusters for K =3 ........... 10 3

PAGE 14

xiv 4.20 Historical occurrence of dry years of CCDO within each clusters for K=12 ......... 10 3 4.21 Composite m aps (for K=3) , where average deviations of (A) cumulative annual precipitation dry days (CDP) and (B) cumulative annual precipitation magnitudes (CAP) over the driest years of C CDO at each cluster are shown. . .......................... 10 4 4.22 Significant regions of SSTA composites over driest years for the combined CDO data within each cluster for K=3. ................................ ................................ ............ 10 6 4.23 Significant regions of SSTA composites over driest years of CDO in medoid stations for K =12. ................................ ................................ ................................ .... 10 7 4.24 C omposite map of integrated moisture from 500 1000 pressure levels in the history of CDO in (A) W et years and (B) dry years 10 8 A.1 Trend of q30 minimum flow ................................ ................................ .................. 13 1 A.2 Trends of annual CDO from start year to 2012 ................................ ...................... 1 3 2

PAGE 15

15 CHAPTER I INTRODUCTION Abstract Streamflow drought is defined as a sustained period o f below normal water availability . It is part of local hydrology, occurring as a result of the complex interactions between climate, catchment , and anthropogenic factors. It can negatively affect water quality and quantity, such as elevated water temperatu re, decreasing ecosystem functions and services, reservoir storage D espite the recognition that the streamflow droughts have widespread hydrological, eco nomic, and social implications, streamflow drought research has not received much attention in U.S. Drought studies are mainly focused on meteorological or agricultural droughts. Thought a few number of studies have investigated the presence of trends in low str eam flow , comprehensive research synthesis on various aspect of streamflow droughts, and their hydro meteorological characteristics are conspicuously absent from the U.S. literature . Drought and its C lassification Drought is a common natural hazard occurr ing throughout the globe and affecting different aspects of human life including the environment and ecological systems, agriculture, forestry, river transport, energy production, regional infrastructure, economy , and society [ Tallaksen and Van Lanen , 2004; Nalbantis , 2008; Tallaksen and Stahl , 2014; Laaha et al. , 2016] . , so it may occur in both arid areas and humid regions [ Dai , 2011] . The most devastating events in the U.S. have occurred many times in the last century [ Dai , 2011; Seneviratne et al. , 2012; Kam and

PAGE 16

16 Sheffield , 2016] , affect ing every component of the hydrological cycle originating from precipitation and leading to stream flow reduction in surface water systems, reservoir storage, and inflow to reservoirs, or the recharge and storage in the groundwater aquifers [ Nalbantis , 2008] . Although these terms are not easily quantified , depending on the component of i nterest in the hydrological cycle, droughts can be classified into four categories [ WMO , 2009; Dai , 2011] : (1) M eteorological or climatological drought is a period of months to years with below normal precipitation. (2) Agricultural drought is a period with dry soils that results fro m below average precipitation or above normal evaporation, all of which lead to reduce the growth and production of crop, plant, or livestock . (3) Hydrological drought and water storages in aquifers, lakes, or reservoirs fall below long term mean levels. (4) Socioeconomic drought is associated with the impacts of the three above mentioned types. It can refer to a failure of water resources sys tems to meet water demands , and to ecological , or health rela ted impacts of drought. Various drought indicators have been developed t o monitor and quantify drought conditions. For meteorological drought s , precipitation is the key variable in defining the indicators , and typically, surface air temperature is considered as the secondary variable to account for the effect of evaporation , such as the Palmer Drought Severity Index (PDSI) [Dai, 2011] . For agricultural drought, soil moisture content is often used [ Nalbantis and Tsakiris , 2009; Dai , 2011] . Hydrological and socioeconomic droughts are notably difficult to appro ach [ Sung and Chung , 2014] . significan t decrease in the availability of water in all its forms, appearing in the land phase of [ Nalbantis and Tsakiris , 2009] . Therefore, commonly used as the key variable in describing hydrological drought [ Nalbantis and Tsakiris , 2009; Dai ,

PAGE 17

17 2011; Tsakiris et al. , 2013; Van Loon , 2013; Pathak and Dodamani , 2016] , because it considers the surface runoff form surface water system , subsur face runoff from unsaturated zones, and base flow from the groundwater [ Tsakiris et al. , 2013] . S treamflow also has a more rapid response to meteorological drought s in compare to groundwater, and so streamflow proved to be suitable for measuring the hydrological drought [ Tallaksen and Van Lanen , 2004] . A s many human activities are highly dependent on surface water supply [ Vasiliades et al. , 201 1; Li et al. , 2013] , which accounts for nearly a 77% of the total freshwater withdrawal in the U.S., streamflow drought is considered to be one of the most important types of drought in this region, affecting the ecological system and many economic sector s (e.g., drinking water supply, hydropower generation, thermoelectric power plants, recreation, and irrigated agriculture ) . Streamflow D roughts Generation Streamflow drought is part of regional hydrology, occurring as a result of the complex natural and man made processes through atmosphere, hydrological cycle, and land atmosphere feedbacks [ Smakhtin , 2001; WMO , 2009; Van Loon , 2013; Van Loon et al. , 2014; Laaha et al. , 2016] . Water balance equation (Equations 1.1 or 1. 2) shows the different process leading to stream low flow co ndition s , an important indicator of streamflow drought, which are dependent to the wide variety of climatic, topographic , and geological conditions [ Smakhtin , 2001] . ( 1. 1) ( 1. 2)

PAGE 18

18 When is Evapotranspiration, is precipitation, is a watershed storage, is inflow of surface water, is outflow of surface water, is groundwater input, is groundwater output, and is the net streamflow. Various components of the water budget ( Equation 1.1 or 1.2) are connected through the balance of incoming and outgoing energy. The most important climate anomalies contribute to the water budget are precipitation and temperature. Recharge to river system is largely dependent on precipitation. D uring a prolonged period of dry condition ( e.g., precipitation deficit), drainage and runoff are typically low , but potential evapotranspiration may increase due to wind speed, vapor pressure deficit, or increased radiation [ Van Loon and Laaha , 2015] . This can lead to increased actual evapotranspiration and result in an extra loss of water from the soil and open water bodies , which consequently may promote sustained streamf low deficit condition s [ Menzel and Bürger , 2002; Whitfield et al. , 2003; Van Loon and Laaha , 2015; Wang et al. , 2015; Huang et al. , 2016] . Low flows are generally controlled by subsurface flows sourced from groundwater that maintain flows during the dry periods of the year [ Smakhtin , 2001; Van Loon , 2013] . The depletion of soil moisture storage causes a decreased rech arge to the groundwater system and declining groundwater levels. When groundwater levels are low, streamflow drought might lead to a hydrological drought. This spatiotemporal migration of a drought signals due to atmospheric transport of anomalous climate pattern [ Joseph et al. , 2009; Van Loon and Laaha , 2015] . Furthermore, s treamflow drought propagation may be dampened or amplified by catchment characteristics (e.g., catchment area, channel and/or catch ment slope, various soil and geology indicators , length of the main stream, catchment shape and watershed perimeter, mean catchment elevation) [ Vogel and Kroll , 1992; Smakhtin ,

PAGE 19

19 2001; Van Loon , 2013; Laaha et al. , 2016] . S treamflow drought characteristics are also influenced by the local climatic events as well as teleconnected by recurring climate cycles caused by fluctuations in the ocean atmospheric system, such as the El Niño Southern Oscillation or the Pacific Decadal Oscillation [ Lins and Slack , 1999; Van Loon and Van Lanen , 2012; Kingston et al. , 2013; Trenberth et al. , 2014; Van Loon and Laaha , 2015] . T herefore, an integral part of investigating low streamflow characteristics also lies in identifying the connections with climate variability [ Pal et a l. , 2015] . Likewise , streamflow conditions might be accelerated by direct or indirect anthropogenic impacts such as dams construction, water regulation, land use change, groundwater abstraction, artificial drainage, planting, afforestation, irrigation, u rbanization [ Smakhtin , 2001; WMO , 2009; Sadri et al. , 2016; U.S. EPA , 2017] . As an example, modification of the land use may contribute to changes in the infiltration and/or evaporation characteristics, as well as amount of groundwater recharge, which consequently influences th e water balance budget. This thesis aims at investigating climate variability related to streamflow drought. However, it is hard to neglect anthropogenic impacts , as they affect all observed hydro meteorological variables. The Significance of the Streamfl ow Drought Studies The greatest impacts of water related extremes are expected from the highest and lowest amount of streamflow [ Kroll et al. , 2004; Melillo et al. , 2014] . H ydrological droughts predominantly relate to low streamflow conditions [ Keyantash and Dracup , 2002; González and Valdés , 2006] or streamflow deficiency co varying with depletion of groundwater reservoirs and connecting to agricultural droughts (soil water deficiency) and meteorological droughts (precipitation deficiency and increased evaporation and transpiration) [ Tallaksen and Van Lanen , 2004] .

PAGE 20

20 Most drought impacts are not simply caused by lack of precipitation, but it is mainly caused by a lack of available water resources at the specific time such as seasonal anomalies of soil moisture, groundwater, or streamflow [ Laaha et al. , 2016] . Streamflow droughts directly impact the quantity , quality, and reliabilit y of streamflows at various temporal and spatial scales, and consequently affect a range of different water use sectors [ Laaha et al. , 2016] . It determines the adequacy of flow of a river to supply and maintain enough water for disposal of liquid wastes, municipal or industrial supplies, supplemental irrigation, and ecological demand such as suitable condition s for fish [ Riggs , 1972] . Stream low flow influence physical habitat, ecological processes, and biodiversity within and among stream networks [ Falke et al. , 20 11; Rolls et al. , 2012; Jaeger et al. , 2014; Sadri et al. , 2016] . Therma l plant cooling in the U.S., supported by surface water withdrawals , has already been affected in areas where surface water supplies are diminishing, such as the southern United State s [ Van Vliet et al. , 2012; Wilbanks et al. , 2012] . P ersistent low flows during warm seasons can limit the ability of utilities to d ischarge heated water to streams from once through cooled power systems due to regulatory requirements and concerns about how the release of warmer water into rivers and streams affects ecosystems and biodiversity [ Laaha et al. , 2016] . That may also require wastewater facilities to increase treatment to meet stream water quality standards [ EPA , 2011] . Streamflow droughts decrease reliability of streamflows for navigation , which consequently increas e shipping costs and result in commodity price volatility [ Melillo et al. , 2014; Laaha et al. , 2016] . Persistent dry condition in r ivers can alter flow continuity in time and space [ Bogan et al. , 2015; Datry et al. , 2016] and consequently streams may become more intermittent [ Döll and Schmied , 2012; Jaeger et al. , 2014; Reynolds et al. , 2015; Datry et al. , 2016] . Social societal/cross boundary conflicts during low flow periods have also been

PAGE 21

21 reported in the history , especially in the smaller basins where reservoir storage is limited to buffer against hydrologic extremes [ Greenberg , 2009; Kanno and Vokoun , 2010; Grandry et al. , 2013; Ahn et al. , 2017] . Impacts of drought s demands on water and environmental services increase [Wada et al., 2013] . In every state of U.S., k nowledge about low streamflow characteristics is crucial for water quality management; issuing and/or renewing National Pollution Discharge Eliminatio n System (NPDES) permits; planning water supplies, hydropower, and irrigation systems; designing cooling plant facilities, site treatment plants and sanitary landfills; determining waste load allocations; and for assessing the impact of prolonged droughts on aquatic ecosystems [ Kroll and Vogel , 2000; Kroll et al. , 2004] . Despite all these important environmental, economic, and societal implications, stream flow drought resea rch has not received much attention, unlike high flows or flood research [ Kroll et al. , 2004; Tallaksen and Van Lanen , 2004; Van Loon , 2013] . Drought r esearch are mainly focused on meteorological or agricultural droughts , which considers the atmospheric and terrestri al components of the water cycle and the linkages between them ( i.e., precipitation, evapotranspiration, snow accumulation, soil moisture, groundwater, lakes and wetlands, and streamflow ) [ Van Loon , 2015] . It is also important to note that drought is a relative, rather than absolute, condition of the hydrological system [ Van Loon , 2015] . Except a very few existing studies [ Lins and S lack , 1999; Douglas et al. , 2000; Kam and Sheffield , 2016; Sadri et al. , 2016] , comprehensive research synthesis on various aspect of streamflow droughts, and their hydro meteorological characteristics are conspicuously absent from the U.S. literature . Th is knowledge void is noted in a recent opinion article by Pal et al. [ Pal et al. , 2015] . Thus, it is imperative that research regarding different asp ects of streamflow drought

PAGE 22

22 characteristics receives more attention in order to manage water resources efficiently during the times of need, as well as balancing socio economic and environmental objectives. A better underst anding of the behavior of streamfl ow drought characteristics is urgently needed to inform sustainable water management practices and to protect against related risks across scales . Objective and Research Aims This thesis analyze s streamflow drought features within major river basins of the conterminous U.S. based on streamflow observations. The overarching aim is to evaluate historic patterns and spatial temporal variability of various streamflow drought indicators identified to be usable for different water using sectors , as well as to und erstand their hydro meteorological characteristics. Founded on the prior scientific knowledge and research needs, the underlying specific research aims are: (1) Developing streamflow drought indicators across the conterminous U.S. (CONUS) (Chapter II ) . (2) Detecti ng non stationarity in streamflow drought indicators across the CONUS (Chapter III ). (3) Characterizing streamflow droughts across the CONUS : Investigating spatial homogeneity and climate connections (Chapter IV ) . (4) Summary and future research directions (Chapte r V ). Meeting these objectives is crucial to systematically characterize streamflow droughts, which is largely missing in the U.S. literature. The outcomes from thesis can contribute to the scientific knowledge of a critical component of the earth system the hydrological system and its characterization and alterations by climate and other factors.

PAGE 23

23 Outline of t he Thesis After this introductory chapter, Chapter II present s the detailed analysis of a dataset of multiple indicators that can be used to char acterize streamflow droughts in the major watershed regions of CONUS . This study uses the observed daily streamflow records of 603 gauges across U.S. from U.S. Geological Survey (USGS) Hydro Climatic Data Network 2009 (HCDN 2009) to create the indicators . Indicators are developed for three main categories including low flow characteristics, deficit characteristics, and independent drought event characteristics . Different time scales are considered in order to represent drought features under different clima te conditions and streamflow regimes. Chapter III investigates the patterns of historical streamflow droughts across the major river basins of U.S . This research considers low streamflow and flow deficit characteristics, which are important indicators of water scarcity within the broader context of droughts. The significance of this study is to examine non stationarity in streamflow drought indicators , d etect the trend pattern of indicators, and identify manageme nt measures are most needed. Chapter IV explores spatial temporal variability of streamflow drought indicators across the CONUS , and their connection to regional and large scale climate anomalies. This study firstly characterizes intermittency of streams in the headwater locations across the CONUS. It then uses a spatial clustering approach to identify homogeneous groups of locations having similar variability. Finally, it presents streamflow drought connections to regional and large scale climate anomali es. Lastly, Chapter V provides a summary of the key findings of this dissertation as well as marks the thesis limitations, and also propose s its future work .

PAGE 24

24 CHAPTER II STREAMFLOW DROUGHT INDICATORS FOR THE CONTERMINOUS UNITED STATES Abstract This chapte r present s a dataset of multiple indicators that can be used to characterize streamflow droughts in the major watershed regions of the conterminous U.S. (CONUS). The dataset is derived from the daily streamflow data from U.S. Geological Survey (USGS) Hydro Climatic Data Network 2009 (HCDN 2009). This study considers three main categories including low flow characteristics, deficit characteristics, and independent drought event characteristics. The indicators are selected based on their advantages and popula rity in scientific studies and operational usage. Indicators are calculated for annual, seasonal, and monthly time scales to represent drought features under different climate conditions and streamflow regimes. The resulting dataset offers a comprehensive source of streamflow drought features on regional and national scale within CONUS. This data can be used by diverse users to help evaluate the spatial and temporal characteristics of streamflow droughts from region to region. Introduction Streamflow droug ht refers to a significant low streamflow conditions with respect to normal flow with spatial and temporal characteristics that varies significantly from one region to another [ Van Loon , 2013] . Based on this definition, evaluating the low streamflow statistics are urgent from the viewpoint of water quantity and quality [ Reilly and Kroll , 2003; Tallaksen and Van Lanen , 2004; Nalbantis and Tsakiris , 2009] . However, the main focus of most drought researches is to find the best drought indicator [e.g., Heim Jr , 2002; Keyantash and Dracup ,

PAGE 25

25 2002; Ntale and Gan , 2003; Mpelasoka et al. , 2008] , but streamflow droughts have very different causes that cannot be captured by a single indicator [ Wanders et al. , 2010; Van Loon , 2013] . Due to multi attributable feature of this natural hazard, dive rsity of interrelated features should be considered for characterizing the streamflow droughts [ Smakhtinb and BagherzadehC , 2007; Niemeyer , 2008; Shiau and Modarres , 2009; Van Loon , 20 13] . Generally, two main approaches of deriving streamflow drought characteristics can be distinguished. One is to characterize droughts according to their magnitude such as minimum n day streamflow, or a percentile from the flow duration curve (FDC). For example, q7 (dry weather flow) is the most widely used low flow statistic in the U.S. [ Smakhtin , 2001; Kroll et al. , 2004] , or q1 is important for ecological assessments [ Smakhtin , 2001; Pyrce , 2004] . The threshold level method proved to be a common method for both perennial and intermitt ent streams and can be chosen either to represent a specific water demand or the boundary to normal streamflow conditions [ Pyrce , 2004; Tallaksen and Van Lanen , 2004; Fleig et al. , 2006; Yahiaoui et al. , 2009] . However, two streams with, for example, similar FDCs may have very different low flow sequences: one may have a few long intervals below a given discharge, the other many short intervals below the same flow rate [ Smakhtin , 2001] . Thus, streamflow drought may be characterized by more factors than by just low stream flow [ Smakhtin , 2001; Van Loon , 2013] . Therefore, second approach attempts to identify the complete drought event properties (i.e., drought timing, duration, or defi cit volume) [ Smakhtin , 2001; Tallaksen and Van Lanen , 2004; Fleig et al. , 2006; Van Loon , 2013] . While this second approach has been widely used in Europe, it has been adopted less wi dely in the U.S. Further, drought indicators can be calculated on various time scales to represent drought features across different climate conditions and streamflow regimes [ Fleig et al. , 2006; Khaliq

PAGE 26

26 et al. , 2008; Van Loon , 2013; Wang et al. , 2015] . As such, analyzing multiple drought indicators over various time scales can provide valuable references for drought monitoring and asses sment [ Wang et al. , 2015] . This research aimed at synthesizing a range of reliable and robust streamflow dro ught indicators in order to develop time series datasets for 18 watershed regions in the CONUS comprising of the major river basins (Figure 2.1). In general, the choice of a streamflow drought indicator is a subjective choice and depends on the purpose of the study and streamflow characteristics [ Smakhtin , 2001; Tallaksen and Van Lanen , 2004; Fleig et al. , 2006; Sheffield and Wood , 2008; Van Loon and Laaha , 2015] . For example, certain minimal water levels in rivers are needed for navigation and ecosystems, whereas seasonal deficit volume have serious impacts in reservoir management [ Van Loon and Laaha , 2015] . Due to the many attributes and features of this natural hazard, there is no unique and universal indicator [ Morid et al. , 2006; Niemeyer , 2008; Shiau and Modarres , 2009] and consequently, analyzing the various indicators is necessary in evaluating the risk of droughts [ Shiau and Modarres , 2009] . Considering this need in the U.S., this chapter presents a description of a dataset of his torical streamflow drought indicators across CONUS that we have developed. The developed dataset has been published [ Pournasiri Poshtiri et al. , 2016] https://rda.ucar.edu/datasets/ds550.0/ [ Pournasiri Poshtiri et al. , 2016] . The newly developed indicator dataset can be applied to compare the streamflow drought characteristics across time and space and can be used according to the needs of different water u sers (e.g., water supply planning and design, waste load allocation, reservoir storage design, and maintenance of quantity and quality of water for irrigation, recreation, and wildlife conservation).

PAGE 27

27 Data Observed streamflow data is used exclusively as t he variable for assessing streamflow droughts. However, it is difficult to identify stream gauges that are not (or minimally) impacted by humans, as well as having sufficient data length. These two issues can be minimized by examining the Hydro Climatic Da ta Network 2009 (HCDN 2009), developed by the U.S. Geological Survey (USGS) [ Lins , 2012] . HCDN 2009 is a subset of the USGS GAGES II Reference stations, presenting pristine gauges from the headwat er type basins. This data network represents smaller and higher elevation basins, whose flows are minimally impacted by human development [ Lins , 2012] e lower part of th e basins (unregulated basins). Thus, this study uses the daily streamflow data derived by the HCDN 2009 for the undisturbed stations in order to minimize the human induced artifacts on the flow and capture natural variability and changes . To download the daily data from USGS stream gauges and to fill in any missing data, the is used [ Ryberg and Vecchia , 2012] . Only the stations with less than 10% missing data are imported and imputed by the The missing data is filled in by using a linear Gaussian structural time series model. Then, a fixed interval smoothing gives the best estimate of the state at each time point based on the whole observed series [ Ryberg and Vecchia , 2012] . In the end, there are 603 stations with between 25 to 111 years of daily data. Geographic locations of the stations and length of the available da ta in years are presented in Fi gure 2.1.

PAGE 28

28 Souris Red Rainy Mis sour i Arkansas White Red Great Basin Texas Gulf Data Length (Year) Ohi o Figure 2 .1: Water resources regions (pink boundaries) comprising of the major river basins in the CONUS, green bubbles indicate locations of HCDN 2009 stream gauges. The size of the bubbles is proportional to the data length in years.

PAGE 29

29 Though finding the naturalized data is hard and streamflow data are prone to artificial influences and the results may have been affected by this, the spatial consistency in the results fr om this thesis ( e.g., Chapter III and Chapter IV ) indicate the climate systematic factors, successfully, corresponding to several other hydro findings (e.g., Chapter III and Chapter IV ) . To a reasonable degree , the results of this study sh ow that HCDN 2009 is a very good choice for climate relationship study. Furthermore, HCDN 2009 remains a valuable indication of historic natural streamflow data and has been employed in many streamflow and climate change studies even in recent years [e.g., Shiau and Modarres , 2009; Timilsena et al. , 2009; Mar tin and Arihood , 2010; Newman et al. , 2015; Kam and Sheffield , 2016; Pournasiri Poshtiri and Pal , 2016; Rossi et al. , 2016] . Method (Indicators Calculation) As mentioned, various indicators are used in drought studies. In this paper, the focus is to prese nt a description of a suite of streamflow drought indicators that are included in the published dataset [ Pournasiri Poshtiri et al. , 2016] . A s summarized in Table 2.1, three main categories are included in this datasets to quantify low flow characteristic, deficit characteristics, and independent drought event characteristics [ Pournasiri Poshtiri et al. , 2016] . This section describes the detailed calculation of indicators under each of above categories following the sub steps conducted in Figure 2.2. Low Flow Characteristics Low flow characteristics describe the average low flow conditions of a stream and are defined in terms of time series that are derived from the series of daily streamflow data. Low flow characteristic can be derived from (i) percentiles (Qx) from the flow duratio n curve (FDC) and (ii) minimum n day mean flow [ Smakhtin , 2001; Pyrce , 2004; Fleig et al. , 2006] .

PAGE 30

30 Table 2.1: Description of streamflow drought indicators. Category Description Threshold level Time scale Low flow charact eristics (i) Percentiles indicators from FDC: 50th, 75th, 90th and 95th percentile flows (Q50th, Q75th, Q90th, Q95th) (ii) Minimum n day mean flow indicators: minimum 1 , 7 , and 30 day mean flow (q1, q7, q30) Annual Northern Summer Northern Winter Mont hly Deficit characteristics Cumulative streamflow deficit occurrence (CDO) Cumulative streamflow deficit volume (CDV) Minimum streamflow deficit volume (MDV) Streamflow deficit intensity (SDI) 10q7, 10q30, 10q1, Q50th, Q75th, Q90th, Q95th Annual Northern Summer Northern Winter Monthly Independent drought event characteristics Duration of each drought event (dde) Deficit volume of each drought event (vde) First day of each drought event (fde) Number of drought events (nde) 10q7, 10q30, 10q1, Q50th, Q75th, Q90th, Q95th Annual

PAGE 31

31 \ Figure 2 .2: The flowchart of Streamflow drought indicators calculation.

PAGE 32

32 (i) Percentiles (Qx) from the flow duration curve (FDC): The flow duration curve (FDC) plots the empirical cumulative frequency of streamflow as a function of the percentage of time that the streamflow is exc eeded. As such, the curve is constructed by ranking the data, and for each value the frequency of exceedance is computed. Low flow indicators are derived directly for every year from the curve as low flow exceedances. The flow exceedance are in terms of pe rcentile [ Risley et al. , 2008] , for example, 90 percentile flow, or Q90th, is the flow that is exceeded for 90% of the period of record. Becau se of the large percentage of zero flow values found for streams in semi arid regions, 50th, 75th, 90th and 95th percentile flows are included in the dataset for every year (Table 2. 1) to define low flow characteristics, as has also been considered in man y other studies [ Smakhtin , 2001; Tharme , 2003; Kroll et al. , 2004; Pyrce , 2004; Risley et al. , 2008; Van Loon , 2013; Wang et al. , 2015] . (ii) Minimum n day mean flo w: One of the most frequently applied low flow statistics is derived from a series of the annual (or seasonal) minima of the n day average flow. In the United States and in the United Kingdom, a 7 days (or sometimes 1 day and 30 days) averaging period is o ften considered [e.g., Yulianti and Burn , 1998; Douglas et al. , 2000; Reilly and Kroll , 2003; Kroll et al. , 2004; Pyrce , 2004; Small et al. , 2006; Martin and Arihood , 2010; Curran et al. , 2012; Sadri et al. , 2016] . Similarly, this study calculates the minimum 1 , 7 , and 30 day mean flow for the dataset (Table 2. 1) Since low flow statistics in different time scales represent different flow regime and climate conditions, a variety of timescales are included in the dataset, including annual, Northern summer (April May June July August September), N orthern winter (October November December January February March), and monthly (column 4 in the Table 1). Low

PAGE 33

33 flow is calculated on the seasonal or monthly scale by limiting the streamflow data used for the annual series to just the season or month of inte rest [ Risley et al. , 2008] . Deficit Characteristics The low flow characteristics defined in the previous section quantify droughts according to their magnitude and are useful for understanding the hydrological regime of a river [ Tallaksen et al. , 1997; Fleig et al. , 2006] . However, to understand drought pr ocesses and impacts, it is necessary to identify the drought characteristics based on streamflow below a certain threshold level (Q 0 ) . The threshold level method [ Yevjevich , 1967] is the most frequently applied quantitative method to identify drought characteristics from time series variables [ Tallaksen et al. , 1997; Fleig et al. , 2006; Lehner et al. , 2006; Van Loon , 2013; Beyene et al. , 2014] . This method allows for deriving a time series for each drought event and then characterizing each drought by its time of occurrence, duration, and deficit volume (or severity) as illustrated in Figure 2.3. Thus, this dataset includes several deficit indicators (i.e., Cumulative deficit occurrence, Cumulative deficit volume, Minimum deficit volume, and Streamflow deficit intensity) wit h respect to various threshold levels (Table 2.1). The calculation of deficit indicators involved two following steps: (i) Threshold level calculation and (ii) Deficit indicators calculation. (i) Threshold level calculation: as shown in Figure 2.2, the fi rst step of quantifying the deficit indicators is to calculate the threshold level. The type and magnitude of a threshold level depends on the drought impacted sector [ Lehner et al. , 2006; Beyene et al. , 2014] . The threshold can be selected based on minimum n day mean flow or percentile of streamflow from the FDC. Minimum n day mean flow with 10 year s return period (10q1, 10q7, and 10q30) are often used as hydrologically based design flows [ Pyrce , 2004; Risley et al. , 2008; U.S. EPA ,

PAGE 34

34 2017] . On the other hand, the FDC allows for the selection of suitable threshold levels both for perennia l streams without and with a frost season, as well as for intermittent streams. For perennial streams, Q95th, Q90th, or Q75th can be applied, whereas for intermittent streams lower percentiles are chosen [ Tal laksen and Van Lanen , 2004] . In the published dataset, we include Q95th, Q90th, Q75th, Q50th, 10q1, 10q7, and 10q30 as threshold levels (column 3 in Table 2.1). Furthermore, the threshold might be fixed with a constant value for the entire streamflow re cord, or, a variable threshold can be applied to investigate the streamflow deviations during both high and low flow seasons [ Tallaksen and Van Lanen , 2004; Lehner et al. , 2006; Beyene et al. , 2014; Van Loon and Laaha , 2015] . Variable thresholds have been found to be the most approp riate method in catchments with different seasonal patterns [ Nyab eze , 2004; Lehner et al. , 2006; Vidal et al. , 2010; Van Loon , 2013; Beyene et al. , 2014] . The main advantage of the variable threshold level method is that it stays as close as possible to the actual time series of hydro meteorological variables [ Van Loon , 2013] . However, there is no single threshold level that is preferable to another and the selection of a specific threshold level remains a sub jective decision [ Fleig et al. , 2006] . In the developed dataset, fixed thresholds at annual timescales, and variable thresholds taken at sea sonal and monthly timescales (column 4 in Table 1) are used to derive the deficit duration and severity indicators column 2 in Table 2. 1). (ii) Deficit indicators calculation: To calculate deficit indicators (column 2 in Table 2.1), it is firstly needed t o define each deficit event characteristics such as deficit duration (d i ), deficit volume or severity (V i ), and minimum deficit volume (V min,i ). Deficit duration is defined as the number of dry days when the streamflow (Q(t)) is below a certain threshold l evel (Q 0 ) of interest (Figure 2.3) and is calculated using the equations below:

PAGE 35

35 Figure 2. 3 : Illustration of the deficit characteristics and streamflow drought event features as defined with the threshold level method. Duration (di), Deficit volume or severity (Vi), Min imum deficit volume (V min,i ) , Time of occurrence, Mutually dependent droughts (d i ,V i ; d i+1 ,V i+1 ; ) and Minor droughts.

PAGE 36

36 ( 2. 1) ( 2. 2) Here, the variable d(t) indicates a day when the streamflow measurement fell below or above the threshold level and so defines a drought event. t=1 is the beginning of drought event i, T is the end of drought eve nt i, and t is the time step, which is 1 day here. Next, deficit volume or severity (V i ) is calculated by summing up the differences between daily streamflow and the threshold level over the deficit duration d i (Figure 2.3) using equations below: (2.3) (2.4) Here, V(t) is the deviation of daily streamflow from the threshold at time t, and V i is the deficit volume within drought event i. Minimum deficit volume (V m in ,i ) is selected as the minimum value of V(t) over the deficit duration (Figure 2.3). Based on the above definitions, several streamflow deficit indicators are developed, including cumulative deficit occurrence , cumulative deficit volume, minimum def icit volume,

PAGE 37

37 streamflow dry spells, and streamflow deficit intensity. Cumulative deficit occurrence (CDO) is calculated by summing up all the deficit durations over a year, as below: (2.5) Here, n is the number of deficit events in a y ear or any other time scale of interest. Cumulative streamflow deficit volume (CDV) is calculated by summing up deficit volumes in a year or any other time scale of interest, as below: (2.6) Then, minimum streamflow deficit volume (MD V) is calculated by selecting the minimum deviation of the streamflow from the threshold level in a year or any other time scale of interest, as below: (2.7) Similarly, streamflow dry spell (SDS) is cal culated by the selecting the minimum deficit duration in a year or any other time scale of interest, as below: (2.8) Another deficit characteristic is drought intensity (sdi). This is defined as the ratio bet ween drought deficit volume and duration [ Fleig et al. , 2006] , as calculated below: (2.9) Independent Drought Event Char acteristics These deficit indicators , identified in the previous section, allow comparisons between drought characteristics across time and space, but can also be used for meeting various needs of water resource management. Since a single drought event ca n cover a large region,

PAGE 38

38 extending across different climate zones and affecting various human activities [ Fleig et al. , 2006] , it is need ed to identify independent drought event characteristics. Thus, the dataset further identifies the within year drought event characteristics with respect to the streamflow series ( i.e., duration of each drought event, deficit volume of each drought event, fi rst day of each drought event, and number of drought events by considering Q95th, Q90th, Q75th, Q50th, 10q1, and 10q7 as threshold levels) (Table 2.1). This section aims to provide some detail on how to combine mutually dependent droughts, to exclude mino r droughts, and consequently, to recognize all independent drought characteristics included in the dataset. The calculation process is similar to deficit characteristics with minor modifications as follows: Using the threshold level method to define stream flow drought events from a daily streamflow series causes two problems: mutually dependent drought events and occurrence of minor droughts, as shown in Figure 2.3 [ Engeland et al. , 2004; Tallaksen and Van Lanen , 2004; Fleig et al. , 2006; Van Loon and Laaha , 2015] . During a long period of streamflow deficit, excess periods with Q(t) > Q 0 of short duration and small excess volume can occur, which divides the phase of low discharge into several drought events. These mutually dependent droughts (Figure 2.3) can be combined into a long event by pooling procedures [ Tallaksen et al. , 1997; Fleig et al. , 2006; Van Loon , 2013; Beyene et al. , 2014] . The inter event time method (IT method) [ , 1987] is a common procedure to knit such multiple drought events based on an inter event time criterion (IT criterion). When the inter event time (t i ) between two consecutive drought events with duration (d i , d i+1 ) and deficit volum e (V i , V i+1 ) is less than a critical duration (t c ), they are pooled into a single drought with duration (pd i ) and deficit volume (pV i ). (2.9)

PAGE 39

39 (2.10) The value of t c results of the drought propagation [ Van Loon , 2013] . This study consider s an inter event time period of the 5 days, as also recommended by [ Tallaksen et al. , 1997] and [ Fleig et al. , 2006] . Fleig et al. (2006) analyzed a global dataset of streamflow series demonstrating streams from most of the major climate zones with different catchment characteristics and found that the sensitivity curves generally st arted to level out around 5 days. Consequently, the duration of each drought event (dde) and deficit volume of each drought event (vde) (Table 2. 1) are calculated using equations (2.2) and (2.4) for condition when (t i > t c ) or equation (2.9) and (2.10) whe n (t i t c ). Minor droughts are also excluded when their deficit volume is smaller than a certain i max ), or when they are of a short duration (d i min ) [ Tallaksen et al. , 1997; Fleig et al. , 2006] . Fleig et al. (2006) min = 3 days, which we follow in this study. After the pooling procedure is applied and so the minor droughts are eliminated, the first day of each drought event (fde) (Table 2.1) is selecte d as the start day of each event and total number of drought events (nde) (Table 2.1) is calculated by counting the number of d i within a year. Data Validation To examine the derived data developed in this paper, it is helpful to explore the spatial varia bility of the data and to view the results in terms of previous studies to assure that the derived data is suitable for use. Since different threshold levels affect the magnitude, occurrence, and persistence of drought events [Lehner et al., 2006; Van Loon and Van Lanen,

PAGE 40

40 2012] , this study firstly explores the spatial variability of the threshold levels. The probability distribution of threshold levels (10q1, 10q7, 10q30, Q50th, Q75th, Q90th, Q95th) across all stations in the CONUS is highly skewed, so a log scale method is applied to transform that data from the nonlinear to a liner scale of distribution. Then, the logarithm of each station data is stan dardized by subtracting the mean and dividing by the standard deviation at each station. Table 2. 2 presents the variability as measured by the standard deviation of the transformed threshold levels. 10q30 has the lowest temporal variability and Q95th has t he highest variability across the CONUS. The variability in 10q30 is lower as compared to 10q1 and 10q7. T he threshold variability gets higher when moving towards a higher percentile value (e.g., Q50 th to Q95 th ). This has also been identified in previous s tudies over Europe, that a high percentile (e.g., Q95th) threshold level shows the maximum deviation, while the opposite is true for lower percentile such as Q70th [ Van Loon and Van Lanen , 2012] . T he variability in Q75th, Q90th, and Q95th is greater than the variability in 10q1, 10q7, 10q30, which was also validated in the previous research findings [ Feyen and Dankers , 2009] . Table 2 . 2 : Standard deviation of different threshold level across all stations within CONUS. Variable Standard deviation 10q1 2.276 10q7 2.213 10q30 2.080 q50 th 2.126 Q75 th 2 .434 Q90 th 2.799 Q95 th 2.993

PAGE 41

41 Several studies have shown that streamflow drought is a multi feature phenomena [ Tallaksen and Van Lanen , 2004; Lehner et al. , 2006; Sheffield and Wood , 2008; Van Loon , 2013; Wang et al. , 2015] . To test that for the U.S. dataset, the spatial variability of the low flow and deficit characteristics are compared using independent t tests. Because a t test subtracts first sample mi nus second sample, the difference between means also suggests the difference in spatial variability in two variables. To do this, study looks at coefficient of variations (CV) of annual low flow characteristics (q1, q7, q30, Q50th, Q75th, Q90th, Q95th) and annual streamflow deficit occurrences with respect to fixed thresholds 10q1, 10q7, 10q30, Q50th, Q75th, Q90th, Q95th. Table 2. 3 lists indicators used here. The standard deviation gives information about the scatter around the mean whereas the coefficient of variation normalizes that variation, so that variations of properties with different mean magnitudes can be easily compared. To calculate CV, first, calculated the standard deviation (SD) and mean of each indicator at each station are calculated, and th en divided by the respective mean. As the CV data for all stations pooled together is skewed, log of CV is considered to study the relative variability across the CONUS. The t test results between different low flow indicators is shown in Table 2.4. The d ifferences in mean of spatial variability measured by minimum n day mean flow (q1, q7, and q30) is not significant (row 1 to 3 in Table 2.4), while differences in mean of spatial variability measured by percentiles from the flow duration curve (Q50th, Q75t h, Q95th) are significant (row 4 to 6 and 8 in Table 2.4). In addition, spatial variability of several minimum n day mean flow are significantly different from the percentile low flow indicators (row 9 to 10, row 13 to 14, and row 17 to 19 in Table 2.4). A bove all, spatial variability of deficit indicators are significantly different from the low flow magnitude indicators (row 21 to 27 in Table 2.4). Thus, it is necessary to evaluate the multi feature

PAGE 42

42 characteristics of streamflow droughts, in an agreement with European studies [e.g., Tallaksen and Van Lanen , 2004; Lehner et al. , 2006; Sheffield and Wood , 2008; Van Loon , 2013; Wang et al. , 2015; Laaha et al. , 2016] . Table 2 .3: Description of selected streamflow drought indicators Approach Indicator Description Low flow magnitude q1 q7 q30 Q50th Q75th Q90th Q95th Minimum 1 day mean streamflow Minimum 7 days mean streamflow Minimum 30 days mean streamflow Q50th percentile s treamflow Q75th percentile streamflow Q90th percentile streamflow Q95th percentile streamflow Deficit approach CDO10q1 CDO10q7 CDO1030 CDOQ50th CDOQ75th CDOQ90h CDOQ95th Cumulative streamflow deficit occurrence for10q1 threshold level Cumulative st reamflow deficit occurrence for10q7 threshold level Cumulative streamflow deficit occurrence for 10q30 threshold level Cumulative streamflow deficit occurrence for Q50th threshold level Cumulative streamflow deficit occurrence for Q75th threshold level Cumulative streamflow deficit occurrence for Q90th threshold level Cumulative streamflow deficit occurrence for Q95th threshold level

PAGE 43

43 Table .4: Statistical differences (t test results) between variability of different streamflow drought indicato rs (indicators in Table 2.3). Bolded p value shows that the difference between means of two indicators is significance (p<0.05). Row Variables t test results p value Mean of the first variable Mean of the second variable 1 q1 vs q7 0.722 0.470 0.486 2 q1 vs q30 0.945 0.470 0.467 3 q7 vs q30 0.670 0.486 0.467 4 Q50th vs Q75th 7.28e 4 0.740 0.619 5 Q50th vs Q90th 1.31e 06 0.740 0.554 6 Q50th vs Q95th 8.99e 08 0.740 0.530 7 Q75th vs Q90th 0.113 0.619 0.554 8 Q75th vs Q95th 0.033 0.619 0.530 9 q1 vs Q50th 7.06e 12 0.470 0.740 10 q1 vs Q75th 4.57e 4 0.47 0.619 11 q1 vs Q90th 0.055 0.470 0.554 12 q1 vs Q95th 0.178 0.470 0.530 13 q7 vs Q50th 1.27e 10 0.486 0.740 14 q7 vs Q75th 1.48e 3 0.486 0.619 15 q7 vs Q90th 0.121 0.48 6 0.554 16 q7 vs Q95th 0.324 0.486 0.530 17 q30 vs Q50th 3.20e 12 0.467 0.740 18 q30 vs Q75th 2.53e 4 0.467 0.619 19 q30 vs Q90th 0.046 0.467 0.554 20 q30 vs Q95th 0.156 0.467 0.530 21 q1 vs CDO10q1 5.50e 12 0.470 0.246 22 q7 vs CDO10q7 1 .31e 4 0.486 0.363 23 q30 vs CDO10q30 2.42e 08 0.467 0.645 24 Q50th vs CDOQ50th < 2.20e 16 0.740 1.08 25 Q75th vs CDOQ75th < 2.20e 16 0.740 0.412 26 Q90th vs CDOQ90th < 2.20e 16 0.554 0.183 27 Q95th vs CDOQ95th < 2.20e 16 0.530 0.519 28 CD O10q1 vs CDO10q7 < 2.20e 16 0.246 0.363 29 CDO10q1 vs CDO10q30 < 2.20e 16 0.246 0.645 30 CDO10q7 vs CDO10q30 < 2.20e 16 0.363 0.645 31 CDOQ50th vs CDOQ75th < 2.20e 16 1.08 0.412 32 CDOQ75th vs CDOQ90th < 2.20e 16 0.412 0.183

PAGE 44

44 Summary and Di scussion This study uses statistical techniques to develop a dataset streamflow drought indicators across the CONUS. The dataset is derived from the daily streamflow data from HCDN 2009 at varying time scales from 1901 to 2012 (as summarized in Table 2.1). This dataset simplifies comprehensive information regarding magnitude, timing, severity, duration, and frequency of streamflow droughts and has a wide range of potential users in academic fields or at strategic and operational levels. A comparison of vari ability patterns of different streamflow drought indicators show the necessity of evaluating the multi feature characteristics of this natural hazard since one indicator might include different information from another one. T his set of indicators can be us ed for a wide variety of applications and studies in regional or national scales for different time scales, and the methodology can be transferred to any other region of interest. It can allow researchers from diverse disciplines to identify potential driv ers (e.g., land use or climate) associated with streamflow droughts. However, the choice of indicator and its implementation are important as they can result in different conclusions, especially in the light of magnitude, spatial variability, and trends. D ata are available in two and three dimensional files as CSV and MAT format (zip files) to the public through an unrestricted repository at http://dx.doi.org/10.5065/D6QR4V9X [ Pournasiri Poshtiri et al. , 2016] . The low flow characteristics and deficit characteristics organized in 2 dimensional CSV files as described in Table 2 and Table 3 of the Documentation section in the repository [ Pournasiri Poshtiri et al. , 2016] . Each row in the file represents one year data and each column rep resents one station. The independent drought event characteristics are arranged in either two dimensional CSV files representing each drought event in each year for a station or three dimensional MAT files including each drought

PAGE 45

45 event in each year for all stations, as described in Table 4 of the Documentation section in the repository [ Pournasiri Poshtiri et al. , 2016] . These datasets can be im ported to any analysis software packages ( e.g. R, Matlab, ArcGIS, etc.). This dataset helps us to proceed our further streamflow drought research in the U.S., aiming to advance our knowledge and skills in low flows. For example, this dataset is used to ex amine non stationarity in extreme low flow pattern in Major U.S. river basins and to detect the trend pattern of streamflow drought (Chapter III ), to identify homogeneous groups of hydro climatic regions and their connections with large scale climate (Chap ter IV ) .

PAGE 46

46 CHAPTER III DETECTING NON STATIONARITY IN STREAMFLOW DROUGHT INDICATORS ACROSS THE CONTERMINOUS U.S. Abstract C limate non stationarity, changes in land use and water management practices affect regional hydrological extremes. This study consi and flow deficit characteristics, which are important indicators of water scarcity within the broader context of droughts. The results indicate the fraction of total number of stations with negative trend s has been increasing in the last four decades while the fraction of total number of stations with positive trends have been decreasing, making a gradual increase in the drying tendency in low flows in the conterminous U.S. Regional differences in low flow trends are notable, which echo the local climatic changes and soil moisture trends documented in the recent National Climate Assessment (NCA), as well as the changes in cropping and irrigation practices, and creation of forest plantations. A reversal of t rend is seen Northeast and Pacific Northwest since 1980s. Patterns in return periods and corresponding return values of low flows are also examined, which suggests changing risk conditions that are important for water resources decision making . Persistent low flow conditions in a river can lead to chronic water scarcity a main driver of societal and cross boundary conflicts. Thus, this study identifies

PAGE 47

47 Introduction Historic ob servational hydro climatologic data are the most popular information sources for water resources project planning and designing [ Y enigun and Ecer , 2013] , and prediction modeling [ Sivapalan et al. , 2011] . Approaches in water resources are generally founded on the assumption of spatiotemporal stationarity in the hydro climatic systems [ Sivapalan et al. , 2011; Yenigun and Ecer , 2013; Milly et al. , 2015] . However, global climate and land use changes can profoundly impact regional hydrolo gical cycles, including the variability of precipitation and stream flows [ Wagener et al. , 2010; Yenigun and Ecer , 2013] . The recent (third) National Climate Assessment (NCA) report concluded that human induced climate change continu es to intensify and that impacts are increasing across the U.S., particularly exaggerating extremes such as magnitudes and frequency of droughts in certain regions [ Melillo et al. , 2014] . Similarly, human changes to the land surface can have multiple consequences for biophysical systems at all scales, including alteration of streamflow patterns [ DeFries and Eshleman , 2004] . Therefore, stationary assumption is no longer valid with the expected changes in hydro climatologic variables . Such changes in hydrology present unfamilia r risks, and of course, result in major problems in regional and national water resources management. Thus, to formulate and implement effective risk management strategies, it is essential to detect where and when observed hydrological extremes have alread y been changing and to what extent. These findings can then be shared with engineers, communities, and stakeholders to better inform them as they make water management decisions. The recent NCA confirmed that annually averaged temperatures have increased i n the U.S. since 1895, but this has not been constant over time [ Melillo et al. , 2014] . In particular,

PAGE 48

48 temperatures generally rose until about 1940, declined slightly around 1970, and then increased rapidly thereafter, leading to a shift in climatic conditions. Furthermore, observational studies have shown that decadal fluctuations in average precipitation changes have occurred in most areas of the U.S . during the last century [ Melillo et al. , 2014] . Fluctuations in streamflow are dominated by precipitation in the U.S. [ Douglas et al. , 2000; Déry and Wood , 2005; Melillo et al. , 2014] , thus, it is likely that there would be some decadal fluctuations in streamflow drought indicators as well. Therefore, it is als o of interest to identify any correspondence between changes in observed streamflow drought indicators and regional climatic changes. A few number of studies in the U.S. have investigated the presence of trends in low streamflow [e.g., Lins and Slac k , 1999; Douglas et al. , 2000; Small et al. , 2006; Kam and Sheffield , 2016; Sadri et al. , 2016] . However, the results of those studies vary depending on the spatio temporal scales and locations of the study area. Lins and Slack (1999) reported an increasi ng trends in lower magnitude streamflow quantiles (annual minimum through the 70th quantile) across the U.S. [ Lins and Slack , 1999] . Douglas et al. (2000) found upward trends in 7 day low flows across the Midwest, but not in the eastern U.S. [ Douglas et al. , 2000] . Small et al. (2006) analyzed trends in annual 7 day low flow, average, and high flows along with seasonal precipitation over individual basins in the U.S. for 1948 1997. They found statistically significant trends in both fall precipitation and 7 day low flow in the upper Mississippi and Great Lakes regions of the country [ Small et al. , 2006] . Recently, Kam and Sheffield (2016) investigated trends and variability in annual 7 day low flows over the eastern U.S. and their attribution in a changing climate. In general, they found a north south (increasing decreasing) dipole pattern in low flow trends for 1962 2011. Sadri et al. (2016) investigated four variants of low flows (1 day, 7 day, 30 day, and 90 day annual low flow) in

PAGE 49

49 the eastern U.S. over the time period 1951 2005. They d etected a general pattern of increasing low flows in the northeast and decreasing low flows in the southeast [ Sadri et al. , 2016] . As such, a comprehensive research synthesis on changes in streamflow droughts, especially using updated data, where the entire CONUS is considered, is clearly needed. The goal of this study is to examine non stationarity in streamflow drought indicators across the major river basins of U.S. (Figure 2.1) . Detecting the trend pattern of streamflow drought indicators can provide valuable in sights and understanding for climate change impacts, risk assessment of extreme events, and water resources decision making that serve society and the economy (e.g., public water supply, irrigation, energy, navigation, and industry) . Data For this study , low flow characteristics and deficit characteristics are selected from the dataset developed in Chapter II . Indicators are selected based on their popularity in the scientific studies and operational usage in the U.S. as follows: Low flow characteristics: In the U.S., the most widely used low flow indicators are q7 (7 day low flow) and q30 (30 day low flow) [ Ries and Friesz , 2000; Flynn , 2003; Reilly and Kroll , 2003; Pyrce , 2004; Martin and Arihood , 2010; Curran et al. , 20 12; U.S. EPA , 2017] , which are considered for this study. Streamflow deficit characteristics: Three different deficit characteristics are selected including (1) total number of streamflow dry days (CDO): total number of occurrences of dry days, i.e. days when the flow is less than or equal to a threshold for a given period of interest; (2) streamflow dry spells (SDS): the maximum duration of stream flow (in days) below a certain threshold, and (3) cumulative streamflow deficits (CDV): the difference in str eamflow

PAGE 50

50 magnitude that fell below a threshold. Here, we consider 10q7 and 10q30 thresholds, in light of their popularity in the U.S. Method Non parametric Mann Kendall (MK) trend test is used to detect monotonic trends and the corresponding slopes is esti mated using Sen's method [ Sen , 1968] . The MK test is applicable to the detection of a monotonic trend in a time series with no seasonal or other cycle. Mann (1945) formulated this non parametric test for monotonic trend detection [ Mann , 1945] , and Kendall (1975) derived the test statistic di stribution for testing non linear trend and turning point [ Kendall , 1975] . The output of the Mann S statistic and the p value. The S statistic checks whether to reject the null hypothesis ( H 0 ) that there is no trend, or accept the alternative hypothesis ( H a ), that there is a trend (upward or downward). If x 1 , x 2 n represent n n , Mann Kenda ll statistic ( S ) is defined as: (3.1) The sign of S indicates the slope of the trend. The positive value of S indicates an increasing trend and negative value indicates a decreasing trend. To statistically quantify the significance of the trend, the probability (Equation 3.2) associated with S and the sample size ( n ) is calculated and compared to the probability at a significant level. (3.2) Z is a normal approximation distribution, defined by Kenda ll (1975) as:

PAGE 51

51 (3.3) In this study, trends with p values < 0.1 are considered statistically significant. It is to be note d that when the slope is zero, the null hypothesis of zero slope over time is accepted that there is no trend ( p value > significant level conclude that there is a trend. This method allows for high frequ ency (i.e. multiple change point) variations to be ignored. Since there are often outliers in the data of interest, non parametric MK test is useful because its statistic is based on the sign of differences, not directly on the values of the random variabl e. Therefore, outliers have less effect on the trends determined. Trend assessment outcomes are sensitive to climate anomalies occurring in the early as well as end years of temporal window being studied. To check for the shift in monotonic trend patterns or detect inter decadal variations, trend tests are repeated for each station keeping the start year of record constant and varying the end years to 1960, 1970, 1980, 1990, 2000, and 2012. Before studying the trends of CDO, SDS and CDV, It is examined whe ther the threshold values corresponding to different return periods have also changed in different river basins. In addition to annual analysis, four seasons are analyzed , namely Dec Jan Feb (DJF), Mar Apr May (MAM), Jun Jul Aug (JJA), and Sep Oct Nov (SON ) using the same daily streamflow d ata from the HCDN 2009 . Furthermore, regional climates differ along the latitudes and longitudes; therefore study also examine s the variations in trend results along the latitudes and longitudes, dividing them into three parts (western/central/eastern) along the

PAGE 52

52 longitudes and two parts along the latitudes (northern/southern): In particular: Long 1 (western): 124.251033 till 105.3189; Long 2 (central): 104.970003 till 86.7057; Long 3 (eastern): 86.405518 till 67.935 242; Lat 1 (South): 27.199495 till 38.018968; and Lat 2 (North): 38.042347 till 48.822925. Results Trends in Low Flow Figure 3.1 A depicts a map of monotonic trends in annual q7 magnitudes the first streamflow drought indicator considered at each gauge lo cation of interest (Figure 2.1). Figure 3.1 A shows that the locations in the Northeast, Midwest, and eastern Great Plains region largely show positive significant trends, while the opposite is true for northern Great Lakes region, Pacific Northwest, the S outheast (Atlantic plains) and parts of Southwestern U.S. (western side of upper and lower Colorado River basins). Similar results are also observed for annual q30, which is shown in Appendix (Figure A.1). The majority of the probability density functions in Figure 3.1 A are bimodal as there are two peaks of trends, with a number of stations getting positive trends and many getting negative trends and some stations getting zero trends. Thus, one group of stations in a region is under represented for the na tionwide trends and one is over represented. For example, western and eastern U.S. basins have higher tendency of decreasing q7 (negative trends), while central U.S. basins have higher tendency of increasing q7 (positive trends). On the other hand, norther n U.S. basins seem to have slightly more wetting tendencies as compared to the same in southern basins. Therefore, overall, longitudinal variation in trending patterns of q7 and q30 is more pronounced than the latitudinal variations.

PAGE 53

53 (A)

PAGE 54

54 Figure 3 .1: (A) Trends of annual 7 days minimum flow (q7). Color bubbles indicate location of the stations, and sign a nd significance of trend estimates (in cubic feet per second (cfs) per day per year). The size of the bubble is proportional to the magnitude of trend value. The probability density functions of q7 trends are determined pooling together all the stream gaug es located in the western, central, and eastern U.S. (top three panels), as well as northern and southern U.S. (right two panels ); (B) Box plot showing distributions of trend values (in cubic feet per second (cfs) per day per year) over all locations havin g at least 30 years of data length, pooling together all the stations across CONUS, for different end year (decadal shift in trends). Numbe r of stations considered for each decadal analysis is stated in the brackets. For example, only 81 stations are inclu ded for the analysis when 1960 is the end year, while 570 stations are included for 2012. Also shown in color bar plot is fraction of tot al number of stations showing positive or negative trends in the box plot. ( B )

PAGE 55

55 Seasonal patterns are also important because trends might differ during different times of a year. Figure 3.2 displays seasonal trends in q7. Summer and autumn (JJA and SON) show similar trends as annual q7 (and q30 ), but Pacific Nor thwest shows positive q7 trends in winter (DJF) and spring (MAM) the wet seasons of this region. According to Figure 3.1 B, there are more than twice as many stations having positive significant trends than negative significant trends in every decade exce pt when the full data length was taken into consideration (i.e. start year to 2012). However, the fraction of total number of stations with negative trends has been increasing in the last four decades while the fraction of total number of stations with pos itive trends have been decreasing, making a gradually decreasing wetting tendency in low flows in the conterminous U.S. (Table 3.1). This is also confirmed by the color bar plot in Figure 3.1 B, which indicates an increasing fraction of total number of sta tions having negative trends (significant and non significant) and a decreasing fraction of total number of stations having positive trends (significant and non significant). In addition, the fraction of total number of stations with zero trends has also i ncreased. Table 3 .1: Fraction of total number of stations with different sign and significance of trends. Decade zero insig + insig sig + sig 1960 6.17 24.69 43.21 6.17 19.75 1970 3.70 43.83 20.99 13.58 17.90 1980 5.65 21.77 37.50 4.84 30.24 199 0 8.65 16.71 41.21 6.34 27.09 2000 10.47 23.00 32.65 8.21 25.67 2012 12.63 28.60 21.93 17.02 19.82

PAGE 56

56 Figure 3 .2: Trends of seasonal q7. Color bubbles indicate location of the stations, sign and significance of trend estimates ( in cubic feet per second (cfs) per day per year). The size of the bubble is proportional to the magnitude of trend value. Figure 3.3 shows temporal shifts in trends, like Figure 3.1 B, but for each station. This figure suggests a reversal in the sign of tr ends over some regions in the CONUS , specifically, the Pacific Northwest (shift after 1980) and the Northeast (shift after 1970), which is not visible in Figure 3.1 B when all the stations are pooled together. To confirm this shift before and after 1980s, Figure 3.4 shows trends in annual q7 magnitudes, considering start year 1979 and 1980 2012. This figure displays a reversal in the sign of trends over most regions in the CONUS after 1979, mostly shifts from positive significant trend to negative signi ficant trend. Overall, this figure confirms the reduced wetting tendency in low flows in the recent decades. Figure 3.5 shows boxplots specific to each river basin (river basins information is in Table re and after 1980. Majority of the basins under study show drying trends (shift from red color to blue color, before and after 1980), DJF MAM JJA SON

PAGE 57

57 where the median trend value decreased. Hence, Figure 3.5 again confirms the shifts in sign of trends specific for each ri ver basin considered for this study . Trends in Return Levels Figure 3.6 displays return level plots of q7 for different river basins , where one representative station from each river basin is selected. Information of those selected stations is in Table 3.2 . Different colors indicate different end year of analysis. Bigger graph shows return periods of q7 where 2 yr, 5 yr and 10 yr return levels are marked by dotted lines, and the graph in the inset shows the changes of 2q7, 5q7, and 10q7 values over the deca des. Figure 3.6 indicates that the stations with significant positive trends in q7 for the start year to 2012 (Figure 3.1 A) also show increasing trends in 2q7, 5q7, and 10q7 from decade to decade, while the opposite is true for the stations having negativ e trends in q7 magnitude (also in Table 3. 2). An exception is basin 7 where we found a decreasing trend first and then an increasing trend, which is also true for all other stations analyzed for this basin.

PAGE 58

58 Start year 1950 Start year 1960 Start ye ar 1970 Start year 1980 Start year 1990 Start year 2000 Figure 3 .3: Trends of annual q7 where the end years are changing decade by decade. Color bubbles indicate location of the stations, sign and significance of trend estimates (in cubic feet per second (cfs) per day per year). The size of the bubble is proportional to the magnitude of trend value.

PAGE 59

59 (A) Start year 1979 (B) 1980 2012 Figure 3 .4: Trends of annual q7: (A) start year to 1979, (B) 1980 to 2012. Color bubbles indicate location of the stations, sign and significance of trend estimates (in cubic fe et per second (cfs) per day per year). The size of the bubble is proportional to the magnitude of trend value.

PAGE 60

60 Table 3 .2: Major water resources regions and their part numbers (two digit numbers designate major river basins). The information for selected stations in Figure 3.7 within corresponding water resources region is presented from columns 3 7. Water resources regions USGS Station Number Longitude Latitude Trend Value for start year to 2012 (cubic feet per second (cfs) per day per year) Level of 2q7 , 5q7, 10q7 New England an Mid Atlantic 01144000 72.42 43.71 + 0.95 Increasing South Atlantic Gulf 02245500 81.85 29.98 0.26 Decreasing Ohio and Tennessee 03070500 79.70 39.62 + 0.07 Increasing Great lakes 04233000 76.54 42.39 + 0.03 Increasing Upper Mississippi and Souris Red Rainy 05362000 90.96 45.31 + 0.3 Increasing Missouri 06354000 100.93 46.38 + 0.01 Increasing Lower Mississippi and Arkansas White Red 07066000 91.36 37.15 + 0.35 Increasing Texas Gulf and Rio Grande 08190000 100.0 0 29.43 + 0.21 Increasing Upper Colorado and Lower Colorado 09430500 108.54 33.06 0.07 Decreasing Great Basin 10234500 112.57 38.28 0.02 Decreasing California 11098000 124.08 41.79 0.08 Decreasing Washington and upper Columbia River basin 12 010000 123.74 46.37 0.09 Decreasing Snake River basin 13083000 113.98 42.17 0.1 Decreasing Oregon and lower Columbia River basin 14305500 123.89 44.72 0.15 Decreasing

PAGE 61

61 0 1 0 2 0 5 0 6 0 4 0 3 0 7 0 8

PAGE 62

62 Figure 3.5: Box plots of annual q7 trend magnitudes for major river basins, before and after 1980. Most of the basins show drying trends (red color before 1980 to blue color after 1980), where median trend value decreased, only except basin 9. Number of stations included in the analyses ar e indicated in the brackets. Stations having at least 30 y ears of data availability were included. 0 9 10 11 12 13 14

PAGE 63

63 03 04 05 06 02 01 07 08

PAGE 64

64 Figure .6: Return level plots of q7 for different river. Different colors indicate different end year of analysis. Bigger graph shows return periods of q7 where 2 yr, 5 yr and 10 yr return levels are marked by dotted lines, and the graph in the inset shows the changes of 2q7, 5q7, and 10q7 values over the decades. 10 11 09 12 14 13

PAGE 65

65 Trends in Deficit Indicato rs The map of monotonic trends in CDO is shown in Figure 3.7. It is expected that the trends in CDO would be similar to those in q7 and q30, but opposite in sign, which is found in Figure 3.7. In a location where low flows are declining, the occurrence of dry days are increasing . For example: Northern Great Lakes region, Pacific Northwest, the Southeast (Atlantic plains) and parts of Southwestern U.S. (western side of upper and lower Colorado River basins) have positive significant trends in CDO, which make adaptive management measures are most needed. Thus, the results in Figure 3.7 are consistent with those from Figure 3.1 (and so is also found for q30, Figure A2 in Appendix). Seasonal trends in CDO corresponding to the 10q7 threshold are shown in Figure 3.8 . C onsistent with q7 in Figure 3.2, summer and autumn (JJA and SON) show similar outcomes as annual but Pacific Northwest shows positive trends in the spring (MAM). In order to confirm the decadal shifts in the sign of trends, Figure 3.7 B displays how the boxplot of the CDO trend values shifted in the previous decades. The median trend values, although remained negative, increased in the last four decades in the overall CONUS. This confirms that the natural flo ws of the major rivers overall are becoming drier, most notably since 1980. This reduced wetting pattern is also confirmed in the increasing number of stations having positive trends in CDO in Figure 3.7 B color bar plot.

PAGE 66

66 (B) ( A )

PAGE 67

67 (B) Figure 3 . 7 : (A) Trends of annual CDO (total number of days below 10q7). Color bubbles indicate location of the stations, and sign and significance of trend estimates (in days p er year). The size of the bubble is proportional to the magnitude of trend value. The probability density functions are determined pooling together all the stream gauges located in the western, central, and eastern U.S. (top three panels), as well as north ern and southern U.S. (right two panels); (B) Box plot showing distributions of trend values (days per year) over all locations having at least 30 years of data length, pooling together all the stations across the CONUS, for different end year (decadal shi ft in trends). Number of stations considered for each decadal analysis is stated in the brackets. For example, only 81 stations are included for the analysis when 1960 is the end year, while 570 stations are included for 2012. Also shown in color bar plot is fraction of total number of stations showing positive or negative trends in the box plot.

PAGE 68

68 DJF MAM JJA SON Figure 3 .8: Trends of seasonal CDO (total number of days below 10q7). Color bubbles indicate location of the stations, and sign and significance of trend estimates (in days per year). The size of the bubble is proportional to the magnitude of trend value. Patter ns in annual SDS and CDV are also found to be similar to annual CDO in Figure 3.9 and 3.10, respectively. Patterns in SDS, in Figure 3.9, indicate longer deficit periods in the Southeast, northern Great Lakes, and the western U.S. Therefore, annual persis tence of dry days, indicated by SDS, and cumulative flow deficit also confirm changes found for the low flow magnitudes and frequency of dry days. Thus, multiple indicators of hydrological droughts considered in this study consistently depict a reduced wet ting tendency and often drying tendency of the major watershed regions comprising of the key rivers in the mainland U.S.

PAGE 69

69 Figure 3.9: (A) Trends of annual SDS (total number of consecutive dry days below 10q7 and 10q30) . Color bubbles indicate location of the stations, and sign and significance of trend estimates (in days per year). The size of the bubble is proportional to the magnitude of trend value; (B) Proportions of total number of stations showing different types of trends in annual persistence of dry days corresponding to 10q7 and 10q30. (A) 10q7 10q30 (B)

PAGE 70

70 (A) (B) Figure 3 . 10 : (A) Trends of annual CDV. Color bubbles indicate location of the stations, and sign and significance of trend estimates (in cfs per year). T he size of the bubble is proportional to the magnitude of trend value. (B)

PAGE 71

71 Summary and Discussion This national scale study on streamflow drought indicators is based on natural daily flow records from 603 USGS stream gauge stations that are minimally influenced by upstream water uses, diversions, impoundments, or land use changes. A non parametric trend analysis of different indicators yields a general reduced wetting trend in the U.S., as opposed to increase in average U.S. precipitation since 1900 [ Melillo et al. , 2014] , increase in the wet days [ Pal et al. , 2013] and increase in water availability within the U.S. [ McCabe and Wolock , 2002] . Furthermore, patterns in streamflow drought indicators along the latitudes and longitudes differ, specifically, more drying patterns are observed over the southern U.S., which corresponds to previous precipitation studies [ Milly et al. , 2005] . The signif icant spatial clustering of similar sign trends in streamflow drought indicators shown in this study suggests systematic causes. Increases in the dry conditions could be driven by different environmental changes [ DeFries and Eshleman , 2004] . Generally, the reductions in precipitation with increased evaporation along with higher temperatures can trigger the changes in streamflow d rought indicators [ Sheffield and Wood , 2008] . The streamflow drought indicators in this study are associated with the regionally consistent trends in climatic conditions, as synthesized in the NCA. For example, increases in the 7 day minimum flow magnitudes and decreases in the frequency of dry days in the Midwest and Northeast are associated with average increa ses in the annual precipitation and runoff patterns in that region [ Melillo et al. , 2014] . Further, runoff was shown to increase in the Mississippi B asin and the Northeast, but a declining trend was documented in annual runoff in the Colorado River Basin [ Melillo et al. , 2014] . Heavy downpours wer e also reported to be increasing over the last three to five decades in the CONUS, with the largest increases in the Midwest and Northeast, and

PAGE 72

7 2 upper Great Plains [ Melillo et al. , 2014] . Although attribution of trends of streamflow drought indicators cannot simply be explained without considering decadal to multi decadal variability [ Novotny and Stefan , 2007] , there is significant correspondence between the changes in annual precipitation, runoff, and surface soil moisture [ Melillo et al. , 2014] , and streamflow drought indicators reported in this study. Furthermore, this study finds an increase in streamflow dry days (CDO), streamflow dry spells (SDS), and cumulative streamflow deficits (CDV), and a decrease in low flow magnitudes (q7 and q30) in the Southwest, where according to NCA, surface soil moisture has exhibited drying trends [ Melillo et al. , 2014] . Rece nt soil moisture trends also show moistening in the Northeast and northern Great Plains and Midwest [ Melillo et al. , 2014] . S ome regions went throu gh a complete shift in trends before and after 1980, for example, the Northeast and Pacific Northwest. A comparison between the probability density functions of magnitude and sign of trends for every basin in this study, before 1980 and after 1980, yield r educed wetting and often drying patterns for most basins. Yulianti and Burn ( 1998) noticed opposite streamflow trends in Canada before and after 1970 [ Yulianti and Burn , 1998] . Milly et al. (2005) selected 1970 as a turning point year because of noticeable changes in global mean measures of hydro climate around this time [ Milly et al. , 2005] . Although climate is an important dri ver of patterns in streamflow drought indicators, it would be inappropriate to attribute these hydrologic trends to climate change alone, without consideration of how land use and water management may have influenced the trends [ DeFries and Eshlema n , 2004; Schilling et al. , 2008; Tomer and Schilling , 2009] . Given that the HCDN is not fully free of changes in water and land management, these factors could also attribute to the changes in streamflow drought indicators. For example: in the northern mi dwest, increasing

PAGE 73

73 low flows may be attributable to increased cropping [ Schilling et al. , 2008; Tomer and Schilling , 2009] and/or increased irrigation because agricultural changes could be associated with ecohydrologic shifts [ Schilling et al. , 2008; Tomer and Schilling , 2009] and irrigation may alter the atmospheric transport of water vapor via partitioning of surface energy fluxes [ Huber et al. , 2014] . In the Sout heast, Pacific Northwest, and Rocky Mountains, the creation of forest plantations may have contributed to decreasing late summer low flows because higher leaf area index may result in lower streamflow and higher evapotranspiration [ Stohlgren et al. , 1998; VanShaar et al. , 2002] . Similarly, in the northeast, increasing low flows may also be the result of forest succession. Finally, there are important implication s of this work for the different regions. The combination of longer low flow periods with rising stream temperatures, especially during summer, may affect the efficiency of thermoelectric power production, make electric power plant cooling water withdrawal s unreliable [ Van Vliet et al. , 2012] , may degrade habitats and favor invasive, non n ative species, and thus affecting aquatic and riparian ecosystems [ Falke et al. , 2011] . It also limits the ability of utilities to discharge heated water to str eams from once through cooled power systems due to regulatory requirements and concerns about how the release of warmer water into rivers and streams affects ecosystems and biodiversity. The longer low flow periods are observed in the Southeast, northern G reat Lakes, and the western U.S.; thus, if stream temperatures increase in these regions, it could lead to potential threats for electric power plants. In addition, a combination of a decrease in magnitude and an increase in the frequency of low flow resul ts in poor water quality conditions in a river, with negative implications for aquatic life. This brings another concern for the Southeast, northern Great Lakes, and the western U.S. locations.

PAGE 74

74 Furthermore, water sector withdrawals and uses vary significa ntly by region and correlate with climate [ Melillo et al. , 2014] . The Southwest, Great Plains, and Southeast are particularly vulnerable to changes i n water supply and demand. Irrigation is dominant along the Mississippi Valley, in Florida, and in southeastern Texas, where dry hydrological extremes are trending to be drier (Figure 3.1). Groundwater withdrawals are especially intense in parts of the Sou thwest, Southeast, Northwest, and Great Plains, the Mississippi Valley, Florida and South Georgia, and near the Great Lakes. Drier conditions in these regions could drive greater groundwater usage. Overall, this research presents new findings on the regio nal changes and decadal shifts in patterns in streamflow drought indicators in major watershed regions in the CONUS. This work provides additional evidence that water resources could possibly being systematically impacted by climate and environmental chang es. Therefore, water resources managers, engineers and planners may face changes in the risks and impacts, as well as opportunities, which may not be properly identified and incorporated in the existing practices. Water management should thus rely less on historical practices and responses and more on robust, risk based, and adaptive decision approaches [ Milly et al. , 2015] .

PAGE 75

75 CHAPTER IV SPATIO TEMPORAL VARIABILI TY OF STREAMFLOW DROUGHT INDICATORS ACROSS CONTERMINOUS U.S. Abstract This study presents spatial temporal variability of streamflow drought indicators across the conterminous United States (CONUS) spanning the period from 1963 2012, and its connection to regional and large scale climate anomalies. This study uses Partitioning Around Medoids (PAM) algorithm along with F Madogram, a non parametric clustering method, to systematically identify the homogeneous regions for streamflow drought indicators . There are strong associations between the streamflow drought indicator s and historic variations in regional climate indices, including PMDI, a known meteorological drought indicator. In general, driest river conditions co occur with dry and warm regional climati c conditions. Hence, in an annual scale, streamflow drought events show coincidence with known historic significant link with the combination of Pacific and Atlantic Ocea n warm pools. Introduction The significant spatial patterns of similar sign trends in streamflow drought indicators (Chapter III ) echo the local climatic changes and soil moisture trends documented in the recent National Climate Assessment (NCA) [ Pournasiri Poshtiri and Pal , 2016] . In the last four decades, low flow magnitudes have shown negative patterns in most river basins wi thin the conterminous U.S. (CONUS), particularly, in the northern Great Lakes region, Pacific Northwest, southeast (Atlantic plains) and southwestern U.S. [ Pournasiri Poshtiri and Pal , 2016] , indicating a gradual increase in the drying tendency of rivers across the CONUS.

PAGE 76

76 Spatial and temporal configurations of stream low flow influence physical habitat, ecological processes, and biodiversity within and among stream networks [ Rolls et al ., 2012; Jaeger et al., 2014]. Tests of regional homogeneity form an integral part of river flow regionalization procedures, such as homogeneous low flow regions, will extract relevant informati on hidden in the complex spatial temporal low flow data, and be used to transfer those statistics from gaged to ungaged sites [ Ries , 2007] . Numerous researchers have used and documented statistical approaches to regionalize drought indices, predominantly in the context of climate [e.g., Fovell , 1997; Lund and Li , 2009] , or annual streamflows [McCabe and Wolock, 2014]. Despite the recognition that the low flows have widespread hydrological, economic, and social implications, to our knowledge, no study so far has presented spatio temporal characteristics of streamflow drought features for U.S. rivers. Persistent dry condition in rivers can alter flow continuity in time and space [ Bogan et al. , 2015; Datry et al. , 2016] . As a resul t, under climate change and landuse modifications there can be a growing tendency of potential shift in perennial to intermittent streams [ Döll and Schmied , 2012; Jaeger et al. , 2014; Reynolds et al. , 2015; Datry et al. , 2016] with long term effects on biodiversity and ecosystem functions [ Bogan et al. , 2015; Datry et al. , 2016; Vander Vorste et al. , 2016; Leigh and Datry , 2017] . In fact, many of the CONUS rivers are already experiencing increasing intermittency [ Jaeger et al. , 2014; Reynolds et al. , 2015] . However, headwater streams are still poorly characterized in the U.S. intermittency [ Jae ger et al. , 2014; Reynolds et al. , 2015] . Streamflow drought is part of local hydrology . During dry weather, runoff is usually low, but potential evapotranspiration may be high due to increased radiation, wind speed, or vapor pressure deficits. This can l ead to increased actual evapotranspiration, resulting in an extra loss

PAGE 77

77 of water from the open water bodies, and consequently, sustained streamflow deficit conditions [ Menzel and Bürger , 2002; Whitfield et al. , 2003; Van Loon and Laaha , 2015; Wang et al. , 2015; Huang et al. , 2016] . On the other hand, spatiotemporal variability of streamflow dro ught characteristics can be associated with large scale atmospheric or oceanic patterns (e.g., El Niño Southern Oscillation (ENSO), North Atlantic Oscillation (NAO)) [ Lins and Slack , 1999; Van Loon and Van Lanen , 2012; Van Loon , 2015] . Occurrence of many extreme droughts an d megadroughts over most of the CONUS have been linked with cool SSTA (e.g., La Nina phase) in the middle and eastern tropical Pacific (ENSO region) or the warm SSTA in northern Atlantic Ocean [ , 1989; McCabe et al. , 2004; Cook et al. , 2010; Dai , 2013; Peterson et al. , 2013] . However, no study so far has looked into how regional and large scale climate directly links with regional river drying or streamflow drought patterns. The main objective of this study are to ( 1 ) use a spatial clustering approach for important streamflow drought indicators and identify homogeneous groups of locations having similar variability, (2) characterize intermittency of streams in the headwater locations across the CONUS, and (3) present streamflow drought connections with regional and large scale climate. This study identifies region specific climate indices those associate with streamflow drought occurrence and magnitudes within the CONUS such that water managers are able to forecast streamflow drought magnitudes once climate model forecasts become available some time in advance. Data Streamflow Drought I ndicators For this study , annual minimum 7 day mean flow ( q7) and annual cumulative stream dry days occurrence (CDO) are selected from the dataset developed in Chapter II . q7 is most

PAGE 78

78 popularly used regulatory tool by the USGS and the US EPA for low flow frequency analysis and water quality studies library [ Pournasiri Poshtiri et al. , 2016] . CDO is defined as the annual total number of days falling below 10 year q7 low flow. Climate Data Regional climate data (Precipitation and Temper ature data): daily temperature (maximum and minimum values) and precipitation data are selected from United States Historical Climatology Network (USHCN) [ Menne et al. , 2010] . The USHCN stations represent the long term high quality daily data from 1218 observing stations across the contiguous U.S. [ Menne et al. , 2010] . Palmer Modified Drought Index (PMDI ): Gridded instrumental Palmer Modified Drought Index (PMDI) data is a small variant of the original PDSI that enables real time drought monitoring. PMDI accounts for evapotranspiration, soil moisture conditions, and precipitation. This study uses 1 degre e gridded monthly data across the CONUS covering the period of 1895 to 2004 [ Heim et al. , 2007] . Large scale climate data: To study large scale climatological patterns connected to st reamflow droughts, Extended Reconstructed Sea Surface Temperature anomalies is slecelted from NOAA NCDC ERSST version3b, obtained in ready to analyze format from the IRI Data Library ( http://iridl.ldeo.columbia .edu ). Method (Spatial C lustering) Streamflow drought indicators are clustered using Partitioning Around Medoids (PAM) algorithm based on F madogram [ Bernard et al. , 2 013] . PAM algorithm generates clusters around the representative stations called medoids. In this study, a medoid is a station where CDO time series mimics the statistical distribution of the data within the other stations in the

PAGE 79

79 cluster, and consequently , each data point keeps its properties [ Kaufman and Rousseeuw , 2009; Bernard et al. , 2013] . A nonparametric approach called F Madogram is used for measuring pairwise dependenc e among distributions (or time series or variables) and can be used as distance matrix in the PAM clustering algorithm [ Bernard et al. , 2013] . For a sample of variable s (M(t) i , M(t) j ) T from two locations i and j at T different time units, madogram estimator d ij is defined as: ( 4. 1) Where is the empirical distribution function, de fined as [ Van der Vaart , 2000] : ( 4. 2) In this way, variable M i is applied to its own distribution , and hence the entire function returns the empirical cumulative distribution, a p roportion of the number of data points less than or equal to value u [ Bernard et al. , 2013] . PAM moves around K number of medoids and try to minimize total intra clus ter distance. Medoids form the center of the cluster representing the valid annual q7 (or CDO) at each step of the algorithm. To select the optimal number of clusters (K) and to distinguish the well classified station, silhouette coefficient can be used, which compares the tightness and separation of clusters. The silhouette coefficient (SC) for a station i is defined as: ( 4. 3) Where d iK is intracluster distance between medoid K an d station i and d i, K is the smallest distance between station i and all the other medoids but K. SC i (K) has a value between 1 and

PAGE 80

80 +1, where SC i (K) ~ +1 indicates that the intra cluster distance is much smaller than the inter cluster distance and the sta tion i is well classified; if SC i (K) ~ 0 then it is viewed as non informative. After running the PAM algorithm for a given number of clusters K , SC is calculated for each station. The average of all SCs helps evaluating the quality of a partitioning into K clusters. For a more detailed theoretical description of both PAM and FM algorithm applied here, the readers are kindly referred to [ Cooley et al. , 2006] , [ Naveau et al. , 2009] , and [ Lehner et al. , 2006] . Finally, 95 th perce ntile significance levels of SCs for a given K is determined by shuffling the q7 (and CDO) data at each station, sampling that randomly, and running the PAM algorithm on it. This process breaks down any spatial and temporal dependence in the data. This alg orithm is repeated 50 times. The 95 th quantile values from 50 average SCs from randomized experiments are considered as significance levels, above which a station is marked as significant. [ Bernard et al. , 2013] . Results This section presents the main findings and results of the study as follows: Section 4.4. 1 presents spatio temporal variability lowflow indicators (i.e., q7) and deficit indicators (i.e., CDO) . Section 4 .4.2 presents the river intermittency behavior within CONUS. S ection 4.4.3 presents the association between river intermittency indicators (i.e., CDO) and climate anomalies.

PAGE 81

81 Spatial Variability of Streamflow Drought Indicators Spatial Variability of Low Flow Indicators ( i.e., q7) Figure 4. 1 A summarizes SC distributions for q7 with 50 years data, where is varying from 3 to 25. The red line represents average SCs among all stations for a given K , the solid black lines correspond to the median values, and dotted blue lines displays 95 th percentile values of SCs. Maximizing the mean SC and minimizing the number of negative SC values helps to select a K = 12. Figure 4. 1 B shows the spatial clustering pattern of q7 for K = 12, which indicates that the low flo w distribution in one cluster is different from the low flows in another cluster, as the spatial relationships between neighboring stations i.e. the bivariate distributions are stronger within individual cluster. Figure 4. 2 shows the time series of q7 at s tations corresponding to each cluster. The spatial relationships between clusters are variable due to the timing of occurrence of annual low flows. Figure 4. 3 indicates a range of occurrence timings, often within a given cluster itself, where the spread is lower for the northeastern clusters (1, 2, 3), cluster 8 and 9, and Pacific northwest cluster (12), where low flows occur around September on an average, but the same for cluster 10 has the largest spread with a median timing of q7 occurrence in the fall. It is also checked if the spatial patterns in Figure 4.1 B remain consistent for the variable data length, as we considered a tradeoff between the number of stations included in the analysis and their time lengths. Figures 4.4 and 4.5 show SC and spatial clustering patterns for the different lengths of low flow time series. An overall general consistency is noticed between Figure 4.4 and Figure 4.1 B, despite the differences in the data length and number of stations and often differing K (varied between 10 and 12).

PAGE 82

82 (A) (B) Figure 4 .1: (A) The distribution of silhouette coefficients (SC) for annual q7 across stations within U.S. for 50 years data where the average SC is represented by the red line, the solid black lines correspond to the SC median values, and the dotted blue line c orrespon ds to the 95 percentile levels; (B) Spatial map for clusters where medoids are represented by triangles and stations are attached to their medoids by gray lines, gray circles correspond to nonsignificant locations.

PAGE 83

83 Figure 4 .2: Spaghetti plots (standardized) for all the time series of annual q7 for different clusters (1963 2012). Medoids are indicated as solid black line and gray colors indicate all the other stations.

PAGE 84

84 Figure 4 .3: Distribution of the occurrence time of q7 over all the locations within each cluster. Black line in the middle of the box is the median.

PAGE 85

85 Figure 4.4: The distributions of SC for annual q7 across the U.S. with different data lengths where the average SC is represented by the red line, the solid black lines correspond to the SC median values. The dotted blue line corresponds to the 95 percentile level SC. (A) 557 stations with 30 years data (1983 2012) (B) 466 stations with 40 years data (1973 2012) (C) 280 stations with 60 years data (1953 2012) (D) 196 stations with 70 years data (1943 2012)

PAGE 86

86 Figure 4 . 5 : Spatial clustering maps for annual q7 for K = 12 and different data length where m edoids are represented by triangles and stations are attached to their medoids by gray lines, gray circles correspond to nonsignificant locations. Spatial Variability of Deficit Indicators (CDO) Figure 4.6 summarizes SC distributions for different data len gth of CDO where K = 2 to 25 are checked. Similar to q7, 50 years data is selected in the end for further analyses because this data length provides a good tradeoff between the number of years used in the study and corresponding number of stations across t he CONUS providing a fair spatial distribution for clustering analysis. Nonetheless, we find a nice consistency between spatial clusters corresponding to the data having different lengths (Figure 4.7). Figure 8 A shows an enlarged version of the distributi 2012 and 393 stations) (also in Figure 4.6) where the red line represents average SC values among all stations (A) 557 stations with 30 years data (1983 2012) (B) 466 stations with 40 years data (1973 2012) (C) 280 stations with 60 years data (1953 2012) (D) 196 stations with 70 years data (1943 2012)

PAGE 87

87 for each given K and the solid black lines correspond to the median values. K is selected based on where the SC has the highest average value and the lowest number of stations having negative SC values. Such condition yields for a K=3, where intra cluster heterogeneity (i.e. higher intra cluster distance) is also maintained, which is a conditi on to be checked in addition to average SC , as suggested by previous researchers [e.g., Bernard et al. , 2013] . As the second best K , this study also presents the spatial clustering pattern corresponding to K=12 to compare results with K=3. Figure 4.8 A (and Figure 4.6) displays 95 th percentile values of SCs from 50 average samples as dotted blue lines. The stations having SCs below that threshold are considered as non significant. Figure 8 B corresponds to the spatial clustering pattern for K=3 and 50 years data, while Figure 4.9 displays that for the other data length. Non significant stations are displayed by grey colors and the me doids are displayed by triangles. Figure 8 B indicates that K=3 divides the CONUS into three large regions (east, midwest, and west), while K=12 in Figure 9 C creates smaller regions (6 regions in total in the east, 4 regions in the midwest, and 2 in the w est). A cross correlation analysis between CDO time series at medoid locations indicate no significant associations between K=3 medoids (Figure 4 .1 0 A ) but that for K=12 medoids yields associations between neighboring clusters (e.g., clusters 1 4, clusters 5 10, and clusters 11 12) (Figure 4 .1 0 B ). Figure 4.11 shows time series plots for CDO for each cluster where K=3 gives a higher spatial variability in the data than K=12 (Figure 4.12), especially cluster 3. Therefore, for K=3 related climate analyses the st andardized data (z scores) within each cluster is combined by selecting the significant stations and using their SC values as weights. However, for K=12 related climate analyses medoids are still used as representative stations (variables).

PAGE 88

88 (A ) 557 sta tions with 30 years data (1983 2012) (B ) 466 stations with 40 years data (1973 2012) (D) 280 stations with 60 years data (1953 2012) 393 stations with 50 years data (1963 2012)

PAGE 89

89 (A) 557 stations with 30 years data (1983 2012) (B) 466 stations with 40 years data (1973 2012) (C) 280 stations with 60 years data (1953 2012) Cluste r

PAGE 90

90 Figure .8: (A) The boxplot summarizes the distribution of silhouette coeffi cients (SC) for annual CDO across stations within U.S. for 50 years data where the average SC is represented by the red line, the solid black lines correspond to the SC median values, and the dotted blue line corresponds to the 95 percentile levels; (B) Sp atial map for clusters where medoids are represented by triangles and stations are attached to their medoids by gray lines, gray circles correspond to nonsignificant locations.

PAGE 91

91 Figure 4 . 9 : Spatial clustering maps of annual CDO for K = 12 and different data length where medoids are represented by triangles and stations are attached to their medoids by gray lines, gray circles correspond to nonsignificant locations. River Intermittency Indicators across the CONUS Annual total number of days with zero flows ( MZD) or annual total number of days with flows going under a specific threshold can be used to characterize severe intermittency of streams [ Jaeger et al. , 2014; Eng et al. , 2016] . This study presents spatial variability of median of the MZD (Figure 4.13 A) as well a s median of CDO (Figure 4.13 B) at each location of interest. Both indicates severe intermittency within streams across southwestern (California, lower Colorado), western (Rio Grande, Texas Gulf), some of the central locations of the U.S. (such as Arkansas White red and Ohio), and in some parts of northern U.S. (upper Missouri). (A) 557 stations with 30 years data (1983 2012) (B) 466 stations with 40 years data (1973 2012) (D) 280 stations with 60 years data (1953 2012) Cluster ( C ) 393 stations with 50 years data (1963 2012)

PAGE 92

92 Figu re 4 . 10 : Correlation coefficients between medoids for (A) K=3, and (B) for K=12. M1 to M12 display medoid 1 to medoid 12, respectively. (A) (B)

PAGE 93

93 Figure : Spaghett i plots for time series of CDO for K=3. Medoids are shown by solid thick lines. To be noted that only stations with significant SC are included in this graph.

PAGE 94

94 Figure 4.12: Spaghetti plots for time series of CDO for K=12. Medoids are shown by solid thick l ines. To be noted that only stations with significant SC are included in this graph.

PAGE 95

95 (A) (B) Figure 4.13 : (A) Median of zero days of streamflow (MZD) within CONUS, (B) Median of stream dry days indicator (CDO) within CONUS. The positive association between CDO and MZD (Figure 4.14) suggests that CDO can be used to study river intermittency in the headwater streams and its connections with climate, whe re the impact of zero values at some locations in the wet regions is minimized. Therefore,

PAGE 96

96 in the next section, this study explores the connection between the CDO, an important indicator of streamflow drought, and climate anomalies. Figure 4.14: Correla tion map between MZD time series and CDO time series with 50 years data (1963 2012) at each location (upward tringles display positive correlations). Climate Connections to Streamflow Drought Indicators Local S cale C limate Here, this study examines the c orrelations and composite analyses between CDO (both combined CDO data (CCDO) for K=3 and medoid CDO data (MCDO) for K=12) and climate indicators derived from precipitation and temperature data (Table 4.1 ), and Palmer Modified Drought Index (PMDI). Both an nual CCDO and MCDO has strong positive correlations with annual cumulative precipitation dry days ( CPD) (Figure 4.1 5 A and Figure 4.1 6 A) and negative correlations with cumulative annual precipitation (CAP) (Figure 4.1 5 B, and Figure

PAGE 97

97 4.1 6 B). This indica tes that in a year when cumulative annual precipitation quantity and the number of wet days is low, there tends to be significantly higher number of river dry days. On the other hand, both CCDO and MCDO positively correlate with local annual daily maximum temperature magnitude (Tmax), the heat wave index (HWF), and diurnal temperature ranges (DTR) (Figures 4.1 5 CtoE as well as Figures 4.1 6 CtoE), meaning that extreme hot years tend to associate with drier rivers. A nnual mean daily temperatures ( Tave) have s ignificant positive correlations with CCDO but predominantly for cluster 3 (Figure 4.1 5 F), and that for DUT ( total annual number of days going above annual mean daily temperature ) and annual minimum temperature (Tmin) are also positive but mainly nonsigni ficant in all clusters (Figure 4.1 5 GtoH, and Figure 4.1 6 GtoH). Overall, these findings indicate that cumulative number of dry days in a river in any region in the CONUS a primary streamflow drought or river intermittency indicator, associate significan tly with extremely hot and precipitation dry years in the same homogenized cluster. Table 4.1: Local precipitation and temperature indices used Climate variable Indicator Description Precipitation CPD Cumulative precipitation dry days ( Total annual num ber of days with precipitation less than 1 mm) CAP Cumulative annul precipitation ( Total annual precipitation) Temperature Tmax Annual maximum value of daily maximum temperature HWF 95th percentile of daily maximum temperature over all the yea rs (heat wave index) DTR Annual mean value of diurnal temperature ranges Tave Annual mean daily temperature DUT Total annual number of days going above annual mean daily temperature Tmin Annual minimum value of daily minimum temperature

PAGE 98

98 Furth ermore, PMDI is a measure of meteorological drought/floods accounting for evapotranspiration and soil moisture conditions, as well as precipitation in a region. A linear associations is found between CCDO/MCDO and PMDI, which are predominantly negative and significant within each cluster (Figure 4.15 I and 4.16 I). This further reinforces the associations between meteorological and hydrolog ical droughts in a given year. Figure 4.17 (and Figure 4.18) shows precipitation and temperature composites over dries t years of rivers in the history where within cluster precipitation and temperature data is considered. Driest years are determined as the years with CCDO/MCDO magnitudes more than their long term 75 th percentile values. Figures 4.19 and 4.20 display the d riest years within each cluster, where many being concurrent (e.g., 1988, 2002 ) or perpetual (e.g., 1991 1995, 1999 2004, 2011 2012) are found. Different shades of blue and red colors (from light to dark) in Figure 4.17 and 4.18 display the average positiv e and negative deviations of CCDO and MCDO values within the driest years from corresponding long term means, i.e. 0 25%, 25 50%, 50 75%, or greater than 75% deviation values. In general, driest years mostly coincide with positive deviations of annual cumu lative precipitation dry days from the mean (Figure 4.17 A), negative deviations of cumulative annual precipitation values from the mean (Figure 4.17 B), and positive temperature deviations from the mean (Figure 4.17 C to 4.17 F). The same analysis but fo r the data across the entire CONUS (Figure 4.21) indicate that regional precipitation and temperature indices associate better with driest years.

PAGE 99

99 (A ) CPD: Positive correlation (B ) CAP: Negative correlation (C ) Tmax : Positive correlation (D ) HWF: P ositive correlation (E ) DTR: Positive correla tion (F ) Tavg: Positive correlation (G) DUT : Positive correlation (H ) Tmin : Positive correlation (I ) PMDI : Negative correlation Figure 4.15 : Correlations between CCDO and annual regional precipitation and temperature indices, and PMDI falling within the range of each cluster. Solid colored tringles correspond to the climate stations with significant correlations (upward tringles displayi ng positive correlations and downward tringles negative correlations), where different colors illustrate different clusters. Climate stations with nonsignificant correlations are shown as empty tringles. Grey circles display streamflow station locations wi th significant SC, indicating cluster borderlines and solid colored circles displaycorresponding medoid locations.

PAGE 100

100 (A ) CPD : Positive correlation (B ) CAP : Negative correlation (C ) Tmax : Positive correlation (D ) HWF: Positive correlation (E ) DTR : Positive correlation (F ) Tavg : Positive correlation (G) DUT : Positive correlation (H ) Tmin : Positive correlation (I ) PMDI : Negative correlation Figure 4.16: Correlations between annual MCDO and annual regional precipitation and temperature indi ces, and PMDI falling within the range of each cluster. Solid colored tringles correspond to the climate stations with significant correlations (upward tringles displaying positive correlations and downward tringles negative correlations), where different colors illustrate different clusters. Climate stations with nonsignificant correlations are shown as empty tringles. Grey circles display streamflow station locations with significant SCs, and solid colored circles display medoid locations.

PAGE 101

101 (A) CPD (B ) CAP (C) Tmax (D) HWF (E) DTR (F) Tavg (G) DUT (H) Tmin Figure 4 .1 7 : Composite maps (for K=3), where average deviations of different inter cluster climate indices are shown. Black tringles display medoid stations, blue circles corresponds to the positive average deviation values, and red circles corresponds to the ne gative average deviation values. Different shades of blue and red colors from light to dark display the 0 25, 25 50, 50 75 , and 75 100 percent deviation values.

PAGE 102

102 (A) CPD (B ) CAP (C) Tmax (D) HWF (E) DTR (F) Tavg (G) DUT (H) Tmin Figure : Composite maps (for K=12), where average deviations of different inter cluster climate indices are shown. Black tringles display medoid stations, blue circles corresponds to the positive average deviation values, and red circles correspond s to the negative average deviation values. Different shades of blue and red colors from light to dark display the 0 25, 25 50, 50 75 , and 75 100 percent deviation values.

PAGE 103

103 Figure 4.19 : Historical occurrence of dry years of CCDO within each clusters for K=3. Driest years were selected as years with CCDO values more than 75th percentile values within each cluster. Figure 4.20 : Historical occurrence of dry years of CCDO within each clusters for K=12. Driest years were selected as years with CCDO value s more than 75th percentile values within each medoid location.

PAGE 104

104 (A) Cluster 1 (B) Cluster 1 Cluster 2 Cluster 2 Cluster 3 Cluster 3 Figure 4.21: Composite maps, where average deviations of (A) cumulative annual precipitation dr y days (CDP) and (B) cumulative annual precipitation magnitudes (CAP) over the driest years of CCDO at each cluster are shown. Black tringles display medoid stations, blue circles corresponds to the positive average deviation values, and red circles corres ponds to the negative average deviation values. Different shades of blue and red colors from light to dark display the 0 25, 25 50, 50 75, and 75 100 percent deviation values.

PAGE 105

105 Large S cale C limate Figure 4. 22 (and Figure 4. 2 3 ) shows sea surface temperatur e anomaly (SSTA) composites over driest years in the history, for K=3 (and K=12) analysis. Significance testing is conducted by selecting any 15 years randomly at a time and determining average SSTA values. SSTA composite magnitudes falling outside the 5 th and 95 th percentile ranges are considered significant. Both Figure 4. 22 and 4. 23 present distinct composite patterns of SSTA for each cluster where extreme river dry days in different regions of the U.S. coincidence with warm SSTA pools in the Pacific and Atlantic Oceans. In particular, streamflow drought events or intermittency of rivers in cluster 1 coincide with warm SSTA in the Northwestern Pacific Ocean, extending from the eastern coast of China to the central Northern Pacific (lat.: 0 45 N & long.: 150E 120W) as well as the same in Northern Atlantic Ocean from the southern coast of Greenland to Equator (Figure 4.22 A) . SSTA composites for cluster 2 is predominant in the central North Pacific (lat.: 20 45 N & long.: 165E 120W) as well as along th e coast of Greenland in Northern Atlantic. For the western U.S. (cluste r 3), warm SSTA is noted in the ENSO region (eastern tropical Pacific to the coast of southern U.S. (lat.: 35S 0 & long.: 115E 90W), and in the Northern Atlantic Ocean (lesser magnit ude as compared to cluster 1). Figure 4.24 illustrates the composite map of the vertically integrated moisture from 500 1000 pressure levels over the wet (i.e., years with CDO less than 25%) and dry years (i.e., years with CDO higher than 75%) over combine d stations of CDO for K=3. One important feature of these maps is the prevailing moisture pattern in wet and dry years. In the cluster 1, strong advection of moisture dominates the supply of atmospheric moisture over much of clusters , mainly coming from the south, specifically the Gulf of Mexico over the wet years (Figure 4.2 4 A, panel 1), while the opposite pattern is seen over dry years in these regions

PAGE 106

106 (Figure 4.2 4 B, panel 1). In the cluster 2, strong moisture advection is coming from the Pacific wes t over the wet years (Figure 4.2 4 A, panel 2). The similar pattern is seen over the dry years but with reduced intensity and perhaps more dry wind from the north (Figure 4.2 4 B, panel 2). (A) Cluster 1 (B) Cluster 2 (C) Cluster 3 Figure 4.22 : Significant regions of SSTA composites over driest years for the combined CDO data within each cluster for K=3.

PAGE 107

107 Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 Cluster 7 Cluster 8 Cluster 9 Cluster 10 Cluster 11 Cluster 12 Figure 4.23 : Significant regions of SSTA composites over driest years of CDO in medoid stations for K=12.

PAGE 108

108 (A) W et years (i.e., years with CDO less than 25%) (B) Dry years (i.e., years with CDO higher than 75%) F igure 4.24 : Composite map of integrated moisture from 500 1000 pressure levels in the history of CDO in (A) Wet years (i.e., years with CDO less than 25%), (B) dry years (i.e., years with CDO higher than 75%). Red Circles indicate the cluster locations.

PAGE 109

109 S ummary and Discussion A nonparametric data driven clustering method is used to systematically identify the homogeneous regions of annual lowflow (q7) variability and annual streamflow dry days (CDO) variability . This approach is capable to recognize identi cal patterns of q7 and CDO based on silhouette coefficient distributions. Study highlights that the timing of the low flows may be differ, with important water management implications for regions such the California and the Great Basins, as well as upper M ississippi where low flows may last from late summer to next winter. This study presents severely intermittent locations at headwater streams in southwestern (California, lower Colorado), western (Rio Grande, Texas Gulf), some of the central locations of the U.S. (such as Arkansas White red and Ohio), and in some parts of northern U.S. (upper Missouri), which are in agreement with intermittent streams found by previous studies [e.g., Jaeger et al. , 2014; Reynolds et al. , 2015; Eng et al. , 2016] . The spatial variabi lity of annual number of zero days and annual number of stream dry days (CDO) is si milar . Stream drying might results in more frequent and longer period of zero flow days and consequently hydrological disconnectivity in streams during lowflow periods, as i ndicated by Jaeger et al. (2014) and Reynolds et al. (2015) for Colorado River Basin . (CDO) and historic variations in regional climate (within same cluster boundary). In particular, hot and dry regional conditions (inter cluster) associate better with driest river conditions within each cluster (and therefore CONUS) leading to severe ecological stress. The associations between dry river conditions and regional cumulative precipitatio n dry days (CPD) and cumulative annual precipitation magnitudes (CAP) within each cluster are consistent with previous

PAGE 110

110 precipitation deficit studies [ Muelle r and Seneviratne , 2012; Griffin and Anchukaitis , 2014; Mao et al. , 2015; Shukla et al. , 2015; Donat et al. , 2016; Livneh and Hoerling , 2016; Luo et al. , 2017] , where they indicate that local precipitation deficit is a main factor in meteorological drough t development and temperature is the response variable to precipitation deficit (for example: 2012 drought in Great Plains, 2011 extreme events in Texas, and 1982 2015 hot days and heat events in eastern U.S.) while the combined effect of the both as well as regional land surface conditions (e.g. soil moisture and ET) affect runoff and streamflow magnitudes. Study here shows that hydrological response to meteorological drought development on the river flows. Hence, there are strong connections between CDO a nd concurrent year regional temperature indices, in addition to precipitation deficits [ Cook et al. , 2010; AghaKouc hak et al. , 2014; Griffin and Anchukaitis , 2014; Mao et al. , 2015; Shukla et al. , 2015; Donat et al. , 2016; Luo et al. , 2017] . Finally, a strong inverse relationship exists between CDO and PMDI, confirming that river dry years co occur with dry and warm r egional climatic conditions, which is also in an agreement with recent studies [ Reynolds et al. , 2015] . Reynolds et al. , 2015 declared that local coexistence of dry and warm periods gradually create longitudinally discontinuous flow and intermittent streams i n Colorado River Basin [ Reynolds et al. , 2015] . Due to the combination of the dry period and high soil moisture demand, run off is generally produced only after sustained periods of precipitation and minimal evapotranspiration [ Eng et al. , 2016] . This can be explained through local energy balance considerations where precipitation deficits reduce surface moisture and the associated upward turbulent flux of latent energy. As a consequence, in the absence of rainfall and thus a reduction in surface moisture, a greater fraction of inco ming solar radiation is balanced by increased upward turbulence,

PAGE 111

111 thereby leading to high surface temperatures [e.g., Mueller and Seneviratne , 2012; Donat et al. , 2016; Livneh and Hoerling , 2016] . Recharge to river systems is largely dependent on precipitation and how that runs off through the gr ound. During a prolonged period of dry condition (i.e., precipitation deficit), drainage and runoff are typically low, but increased evapotranspiration result in an extra loss of water from the soil and open water bodies, which consequently promote to warm conditions and lead to river drying [Menzel and Bürger, 2002; Whitfield et al., 2003; Van Loon and Laaha, 2015; Wang et al., 2015; Huang et al., 2016]. L ow flows are generally controlled by subsurface flows sourced from groundwater that maintains flows du ring the dry periods of the year [Smakhtin, 2001; Van Loon, 2013] . Increases in evaporation may result in depletion of soil moisture storage, and consequently decline the groundwater recharge and streamflow [ Dudley and Hodgkins , 2013] . For example, recent studies reported that high evaporation rate across the eastern U.S. [e.g., Brutsaert , 2010; Huntington and Billmire , 2014] played an important role in prolonged declines in g roundwater levels, which have altered low flow conditions (e.g., reducing base flow levels and subsurface flow) [ Brutsaert , 201 0; Kanno and Vokoun , 2010] . Because dryness of rivers correlate with PMDI in every cluster, it is also noted that all the driest years of rivers in the history nicely coincide with known meteorological drought events in the U.S. For example, 1963 1966 dr ought in the northeast [ Cook and Jacoby , 1977; Andreadis et al. , 2005] and 1980 1981 drought in the south and southeast [ Karl and Quayle , 1981] correspond to clusters 1 and 2 in Figure 4. 19 (and clusters 1 4, 6 11 for K=12 and to clusters 5 and 7 for K=12 respectively in Figure 4. 20 ); widely reported 1988 drought in different parts of the U.S. (e.g., Westcoast, Grea t Plains, Midwest, and northwest) [ Trenberth et al. , 1988; , 1989; Atlas et al. , 1993; Kam et al. , 2014] sh ow occurrence

PAGE 112

112 in all three clusters in Figure 4. 20 (and clusters 3,5,7,9,10,11 for K=12 in Figure 4. 21 ); 2002 drought covering the 50% of the CONUS under moderate to severe conditions [ Cook et al. , 2010] is found in all three clusters in Figure 4. 20 with more temporal extent (and clusters 1, 3, 4, 11, 12 for K=12 in Figure 4. 21 ); 1999 2004 drought in the southwest [ Piechota et al. , 2004; Andreadis et al. , 2005; Easterling et al. , 2007] correspond with cluster 1 and 3 with more spatial extent in Figure 4. 19 (and to cluster 11 for K=12 in Figure 4.2 0 ); 2007 drought in the east [ Li et al. , 2008] and in the west and southeast [ Luo and Wood , 2007] is seen within all the three clusters in Figure 4. 19 (and cluster 3 to 6, and 11 for K=12 in Figure 4.2 0 ); the drought beginning in 2011 2012 in California [ Mao et al. , 2015] and in the Great Plains [ Livneh and Hoerling , 2016] is recognized within cluster 2 in Figure 4. 19 (and 9 and 11 for K=12 in Figure 4.2 0 ). These results validate that streamflow stations are well cla ssified to display known meteorological drought events across the CONUS. This study also presents strong associations between river dry days indicator and historic variations in large scale climate where we note distinct sea surface temperature anomaly co mposites. Specifically, ENSO years coincide with dry river conditions in the west, along with warm SSTA in the northern Atlantic and northern Pacific. Warm SSTA in the northern Pacific and Atlantic occur with dry conditions in the midwest and eastern U.S. Occurrence of warm SSTA in the northwestern Pacific and northern Atlantic Oceans likely shift the atmospheric pressure patterns, and consequently, northerly wind anomalies reducing the flow of moist air into the continent and leading to reduced precipitati on, and consequently result in high exceedances in the frequency of hot days over much of U.S. [Donat et al., 2016] . Although this study presents climate associations with streamflow droughts, it is important to mention that s tream drying development interact with many other physical

PAGE 113

113 characteristics of the catchments (i.e. area, soils, geology, land use, and grou ndwater hydrology) [ Reynolds et al. , 2015; Van Loon , 2015; Van L oon and Laaha , 2015] . Thus, each stream may vary in their susceptibility to drying under climate change or whether a hydrological drought would make an onset with meteorological drought or would have the same spatial extent [ Tallaksen and Van Lanen , 2004; Mishra and Singh , 2010; Van Loon , 2013; Reynolds et al. , 2015] , both having implications on regional water management. Therefore, this study op ens an argument to take into account the catchment properties to understand the onset as well as spatial extent of streamflow/hydrological droughts. There are important implications of this study for different regions. The combination of sustained stream d ry periods with rising stream temperatures in warm seasons may lead to harsh environmental conditions, and could further threaten regional biodiversity in riverine systems and induce community and ecosystem shifts [Jaeger et al., 2014; Reynolds et al., 2015] , which consequently degrade aquatic and riparian ecosystems functions and services. River drying may reduce stream connectivity within seasons, and consequently, affecting different water using sectors throughout the riverine network [Jaeger et al., 2014]. Balancing aquatic conservation and water supply could be a major issue, specifically in the locations where dry hydrological extremes are trending to be drier [ Pournasiri Poshtiri and Pal , 2016] . Thus, river drying presents significant challenges for both river dependent ecosystems and water resources and hence this research takes a step forward to understand their variability across CONUS, providing the groundwork to develop prediction models for streamflow drought occurrence and magnitudes at a regional level.

PAGE 114

114 CHAPTER V DISSERTATION CONCLUSIONS S ummary This thesis aims at enhancing the understanding of streamflow drought characteristics in the major watershed regions of the conterminous U.S. Research findings offer a broad information on the historic patterns and spatial temporal variability of streamflow droughts , as well as their connections to climat e. Results consistent with the meteorological drought research findings are also presented and discussed. Chapter II presents a list of existing streamflow drought indicators derived from daily streamflow time series, and then develop s a dataset of histor ical indicators across the CONUS. Multiple indicators across various time scales can provide us valuable references for drought monitoring and assessment, usable for different water using sectors . The indicators are derived using observed daily flow record s of 603 USGS Hydro Climatic Data Network 2009 (HCDN 2009) stream gauges that are minimally influenced by upstream water uses, diversions, impoundments, or land use changes. Chapter III presents trend analysis of different indicators in Chapter II , yieldi ng a general increase in drying tendency across the CONUS. This finding corresponds with regionally consistent trends in climatic conditions. Low flow characteristics have demonstrated negative trends in most regions within the CONUS, specifically in north ern Great Lakes region, Pacific Northwest, Southeast (Atlantic plains) and Southwestern U.S. Regional differences in low flow trends are notable, which echo the local climatic changes and soil moisture trends documented in the National Climate Assessment ( NCA), as well as the changes in cropping and irrigation

PAGE 115

115 practices, and creation of forest plantations. A reversal of sign of trend (mainly from positive trend to negative trend) is seen in the Northeast and Pacific Northwest since 1980s . Patterns in return periods and corresponding return values of low flows suggest changing risk conditions that are important for water resources decision making locations where suitable adaptive management measures are most needed. Cha pter VI presents spatio temporal variability of streamflow drought indicators across the CONUS. Lowflow indicator (i.e., q7), deficit indicator (annual cumulative stream dry days occurrence) and annual total number of days with zero flows are used for thi s study . The spatial variability of annual number of zero days and annual number of stream dry days is similar. Severely intermittent locations of rivers are found at the headwater of the rivers within southwestern (California, lower Colorado), western (Ri o Grande, Texas Gulf), some of the central locations of the U.S. (such as Arkansas White red and Ohio), and in some parts of northern U.S. (upper Missouri). Homogeneous regions for streamflow drought indicators are also presented, which show identical patt erns of streamflow drought indicators with similar variability . The results also highlight that the timing of the low flows differ, with important water management implications for regions such the California and the Great Basins, as well as upper Mississi ppi where low flows may last from late summer till next winter. There are strong associations between the streamflow drought indicators and historic variations in regional climate indicators, including PMDI, a known meteorological drought indicator. In gen eral, driest river conditions co occur with dry and warm regional climatic conditions. Specifically, historically, in an annual scale, streamflow drought events show coincidence with known meteorological drought events. Furthermore, streamflow drought indi cators display a significant link with the combination of Pacific and Atlantic Ocean warm pools.

PAGE 116

116 Discussion There are a number of important fields where the outcomes of this research can be applied . A firm understanding of the different feature of streamfl ow droughts (e.g., magnitude, frequency of low flows) is important for society such as water supply planning and design, waste load allocation, reservoir storage design, and maintenance of quantity and quality of water for irrigation, recreation, wildlife conservation, and affecting a diversity of economic and social activities, and consequently, will enable improved characterization of risks and subsequent water resources decision considerations. On the other hand, quantifying historical spatial temporal c hanges in hydrological data can provide useful information as to how water resources are affected by climate, as well as create an understanding of potential future variability in the hydrologic regime in regional and national scale. This study presents cl imate associations with streamflow droughts but it is important to mention that the development of streamflow drought events interact with many other physical characteristics of the catchments (i.e. area, soils, geology, land use, and groundwater hydrology ) [ Reynolds et al. , 2015; Van Loon , 2015; Van Loon and Laaha , 20 15] . Thus, each stream may vary in their susceptibility to drying under climate change or whether a hydrological drought would make an onset with meteorological drought or would have the same spatial extent [ Tallaksen and Van Lanen , 2004; Mishra and Singh , 2010; Van Loon , 2013; Reynolds et al. , 2015] , both having implications on regional water management. The limitation of this study can also be related to the limitation of the data resources, data collection error, and methods, as follows: Data Resource: This thesis used the pristine data, headwater type basins, from HCDN 2009, representing smaller and higher elevation basins, whose flows are mi nimally impacted

PAGE 117

117 by human development [ Lins , 2012] of the basins (unregulated basins). Even so finding the naturalized data is hard and streamflow data are prone to artificial influences and the results may have been affected by this, the spatial consistency in the results from this research does indicate the climate systematic factors, successfully, corresponding to several other hydro c To a reasonable degree , our results have indicated that HCDN 2009 is a very good choice for climate relationship study. The data collection errors: One of the potential issue with measuring lowflow in the stream are the stream wi th very lowflow or zero flow. Measured or observed zero flow indicates the perennial or intermittent status of a stream site [ Bent and Archfield , 2002] . However, the observed zero stream sites might be biased for several reasons such as too shallow streams for flow measurement; flows might not fill the normal stream channel; water would not reach the base of both banks; portions of the stream channel might be dry; flow might be confined to one side of t he channel; uneven streamflow with irregular bottom with large rocks, protruding obstructions, or vegetation. Methods: Limitations also include methods for example, Mann Kendall trend method linear changes, those cou ld be done by ensemble empirical mode decomposition (EEMD) method for example), which we incor porated later in another paper [ Pal et al. , 2017] . Second, PAM algorithm along with F madogram is a data driven method, so we could get a better clustering pattern in southwest U.S. (specifically California a nd Great basins), if we had more stations added for this region.

PAGE 118

118 Future Work While this research focused mainly on climate variability, there are a multitude of relevant processes underlying the development of streamflow droughts. Thus, it would be inter esting to investigate the possibility of extending the streamflow drought generation to include a number of hydrological processes (e.g., catchment controlled types, groundwater hydrology) and also anthropogenic effects . According to several studies in Eur ope [e.g. Van Loon , 2015] , the influence of c atchment characteristics on streamflow drought characteristics is as large as the influence of climate. It would also be of great interest to explore how the results of this thesis can be applied to, for example, streamflow drought forecasting or predictio n in ungauged basins. Improvement of the seasonal forecasting of drought is a prerequisite for adequate operational water management (e.g. reservoir operation, irrigation abstractions, or management of wetlands). For long term water management (e.g. reserv oir design or policy development), information on larger time and spatial scales is needed.

PAGE 119

119 BIBLIOGRAPHY AghaKouchak, A., L. Cheng, O. Mazdiyasni, and A. Farahmand (2014), Global warming and changes in ri sk of concurrent climate extremes: Insights from the 2014 California drought, Geophys. Res. Lett. , 41 (24), 8847 8852. Ahn, K., R. Palmer, and S. Steinschneider (2017), A hierarchical Bayesian model for regionalized seasonal forecasts: Application to low fl ows in the northeastern United States, Water Resour. Res. , 53 (1), 503 521. Andreadis, K. M., E. A. Clark, A. W. Wood, A. F. Hamlet, and D. P. Lettenmaier (2005), Twentieth century drought in the conterminous United States, J. Hydrometeorol. , 6 (6), 985 1001 . Atlas, R., N. Wolfson, and J. Terry (1993), The effect of SST and soil moisture anomalies on GLA model simulations of the 1988 US summer drought, J. Clim. , 6 (11), 2034 2048. Bent, G. C., and S. A. Archfield (2002), A Logistic regression equation for esti mating the probability of a stream flowing perennially in Massachusetts , US Dept. of the Interior, US Geological Survey. Bernard, E., P. Naveau, M. Vrac, and O. Mestre (2013), Clustering of maxima: Spatial dependencies among heavy rainfall in France, J. Cl im. , 26 (20), 7929 7937. Beyene, B. S., A. F. Van Loon, H. A. J. Van Lanen, and P. Torfs (2014), Investigation of variable threshold level approaches for hydrological drought identification, Hydrol. Earth Syst. Sci. Discuss. , 11 (11), 12765 12797. Bogan, M. T., K. S. Boersma, and D. A. Lytle (2015), Resistance and resilience of invertebrate communities to seasonal and supraseasonal drought in arid land headwater streams, Freshw. Biol. , 60 (12), 2547 2558. Brutsaert, W. (2010), Annual drought flow and groundwat er storage trends in the eastern half of the United States during the past two third century, Theor. Appl. Climatol. , 100 (1 2), 93 103. Cook, E. R., and G. C. Jacoby (1977), Tree ring drought relationships in the Hudson Valley, New York, Science (80 . ). , 198 (4315), 399 401. Cook, E. R., R. Seager, R. R. Heim, R. S. Vose, C. Herweijer, and C. Woodhouse (2010), Megadroughts in North America: Placing IPCC projections of hydroclimatic change in a long term palaeoclimate context, J. Quat. Sci. , 25 (1), 48 61. Cooley, D., P. Naveau, and P. Poncet (2006), Variograms for spatial max stable random fields, Depend. Probab. Stat. , 373 390. Curran, C. A., K. Eng, and C. P. Konrad (2012), Analysis of low flows and selected methods for estimating low flow characteristics at partial record and ungaged stream sites in western Washington , US Geological Survey.

PAGE 120

120 Dai, A. (2011), Drought under global warming: a review, Wiley Interdiscip. Rev. Clim. Chang. , 2 (1), 45 65. Dai, A. (2013), The influence of the inter decadal Pacific oscillation on US preci pitation during 1923 2010, Clim. Dyn. , 41 (3 4), 633 646. Datry, T., K. Fritz, and C. Leigh (2016), Challenges, developments and perspectives in intermittent river ecology, Freshw. Biol. , 61 (8), 1171 1180. DeFries, R., and K. N. Eshleman (2004), Land use ch ange and hydrologic processes: A major focus for the future, Hydrol. Process. , 18 (11), 2183 2186. Déry, S. J., and E. F. Wood (2005), Decreasing river discharge in northern Canada, Geophys. Res. Lett. , 32 (10). Döll, P., and H. M. Schmied (2012), How is the impact of climate change on river flow regimes related to the impact on mean annual runoff? A global scale analysis, Environ. Res. Lett. , 7 (1), 14037. Donat, M. G., A. D. King, J. T. Overpeck, L. V Alexander, I. Durre, and D. J. Karoly (2016), Extraordina ry heat during the 1930s US Dust Bowl and associated large scale conditions, Clim. Dyn. , 46 (1 2), 413 426. Douglas, E. M., R. M. Vogel, and C. N. Kroll (2000), Trends in floods and low flows in the United States: impact of spatial correlation, J. Hydrol. , 240 (1), 90 105. Dudley, R. W., and G. A. Hodgkins (2013), Historical groundwater trends in northern New England and relations with streamflow and climatic variables, JAWRA J. Am. Water Resour. Assoc. , 49 (5), 1198 1212. Easterling, D. R., T. W. R. Wallis, J . H. Lawrimore, and R. R. Heim (2007), Effects of temperature and precipitation trends on US drought, Geophys. Res. Lett. , 34 (20). Eng, K., D. M. Wolock, and M. D. Dettinger (2016), Sensitivity of intermittent streams to climate variations in the USA, Rive r Res. Appl. , 32 (5), 885 895. Engeland, K., H. Hisdal, and A. Frigessi (2004), Practical extreme value modelling of hydrological floods and droughts: a case study, Extremes , 7 (1), 5 30. EPA, U. S. (2011), Climate Change Vulnerability Assessments: Four Case Studies of Water Utility Practices , Washington, DC. Falke, J. A., K. D. Fausch, R. Magelky, A. Aldred, D. S. Durnford, L. K. Riley, and R. Oad (2011), The role of groundwater pumping and drought in shaping ecological futures for stream fishes in a dryland river basin of the western Great Plains, USA, Ecohydrology , 4 (5), 682 697. Feyen, L., and R. Dankers (2009), Impact of global warming on streamflow drought in Europe, J. Geophys. Res. Atmos. , 114 (D17). Fleig, A. K., L. M. Tallaksen, H. Hisdal, and S. Demu th (2006), A global evaluation of

PAGE 121

121 streamflow drought characteristics, Hydrol. Earth Syst. Sci. , 10 (4), 535 552. Flynn, R. H. (2003), Development of regression equations to estimate flow durations and low flow frequency statistics in New Hampshire streams , US Department of the Interior, US Geological Survey. Fovell, R. G. (1997), Consensus clustering of US temperature and precipitation data, J. Clim. , 10 (6), 1405 1427. González, J., and J. B. Valdés (2006), New drought frequency index: Definition and compara tive performance analysis, Water Resour. Res. , 42 (11). Grandry, M., S. Gailliez, C. Sohier, A. Verstraete, and A. Degré (2013), A method for low flow estimation at ungauged sites: a case study in Wallonia (Belgium), Hydrol. Earth Syst. Sci. , 17 (4), 1319. G reenberg, M. R. (2009), Water, conflict, and hope, Griffin, D., and K. J. Anchukaitis (2014), How unusual is the 2012 2014 California drought?, Geophys. Res. Lett. , 41 (24), 9017 9023. Heim, R. R., R. S. Vose, J. H. Lawrimore, and E. R. Cook (2007), Putting current North America drought conditions into a multi century perspective. Part 2: Using the blended product in operational drought monitoring, in AGU Fall Meeting Abstracts . Heim Jr, R. R. (2002), A review of twentieth century drought indices used in the United States, Bull. Am. Meteorol. Soc. , 83 (8), 1149 1165. Huang, S., Q. Huang, J. Chang, and G. Leng (2016), Linkages between hydrological drought, climate indices and human activities: a case study in the Columbia River basin, Int. J. Climatol. , 36 (1), 280 290. Huber, D. B., D. B. Mechem, and N. A. Brunsell (2014), The effects of Great Plains irrigation on the surface energy balance, regional circulation, and precipitation, Climate , 2 (2), 103 128. Huntington, T. G., and M. Billmire (2014), Trends in prec ipitation, runoff, and evapotranspiration for rivers draining to the Gulf of Maine in the United States, J. Hydrometeorol. , 15 (2), 726 743. Jaeger, K. L., J. D. Olden, and N. A. Pelland (2014), Climate change poised to threaten hydrologic connectivity and endemic fishes in dryland streams, Proc. Natl. Acad. Sci. , 111 (38), 13894 13899. Joseph, S., A. K. Sahai, and B. N. Goswami (2009), Eastward propagating MJO during boreal summer and Indian monsoon droughts, Clim. Dyn. , 32 (7 8), 1139 1153. Kam, J., and J. S heffield (2016), Changes in the low flow regime over the eastern United States (1962 2011): variability, trends, and attributions, Clim. Change , 135 (3 4), 639 653. Kam, J., J. Sheffield, X. Yuan, and E. F. Wood (2014), Did a skillful prediction of sea surf ace

PAGE 122

122 temperatures help or hinder forecasting of the 2012 Midwestern US drought?, Environ. Res. Lett. , 9 (3), 34005. Kanno, Y., and J. C. Vokoun (2010), Evaluating effects of water withdrawals and impoundments on fish assemblages in southern New England strea ms, USA, Fish. Manag. Ecol. , 17 (3), 272 283. Karl, T. R., and R. G. Quayle (1981), The 1980 summer heat wave and drought in historical perspective, Mon. Weather Rev. , 109 (10), 2055 2073. Kaufman, L., and P. J. Rousseeuw (2009), Finding groups in data: an i ntroduction to cluster analysis , John Wiley & Sons. Kendall, M. G. (1975), Rank Correlation Methods; Charles Griffin: London, 1975, There is no Corresp. Rec. this Ref. Keyantash, J., and J. A. Dracup (2002), The quantification of drought: an evaluation of drought indices, Bull. Am. Meteorol. Soc. , 83 (8), 1167 1180. Khaliq, M. N., T. B. M. J. Ouarda, P. Gachon, and L. Sushama (2008), Temporal evolution of low flow regimes in Canadian rivers, Water Resour. Res. , 44 (8). Kingston, D. G., A. K. Fleig, L. M. Tall aksen, and D. M. Hannah (2013), Ocean atmosphere forcing of summer streamflow drought in Great Britain, J. Hydrometeorol. , 14 (1), 331 344. Kroll, C., and R. Vogel (2000), A National Assessment of Low Streamflow Estimation Using a Physically Based Statistic al Methodology . Kroll, C., J. Luz, B. Allen, and R. M. Vogel (2004), Developing a watershed characteristics database to improve low streamflow prediction, J. Hydrol. Eng. , 9 (2), 116 125. Laaha, G., T. Gauster, L. M. Tallaksen, J. P. Vidal, K. Stahl, C. Pru dhomme, B. Heudorfer, R. Vlnas, M. Ionita, and H. A. J. Van Lanen (2016), The European 2015 drought from a hydrological perspective, Hydrol. Earth Syst. Sci. Discuss. Lehner, B., P. Döll, J. Alcamo, T. Henrichs, and F. Kaspar (2006), Estimating the impact of global change on flood and drought risks in Europe: a continental, integrated analysis, Clim. Change , 75 (3), 273 299. Leigh, C., and T. Datry (2017), Drying as a primary hydrological determinant of biodiversity in river systems: A broad scale analysis, Ecography (Cop.). , 40 (4), 487 499. Li, H., L. Luo, and E. F. Wood (2008), Seasonal hydrologic predictions of low flow conditions over eastern USA during the 2007 drought, Atmos. Sci. Lett. , 9 (2), 61 66. Li, S., L. Xiong, L. Dong, and J. Zh ang (2013), Effects of the Three Gorges Reservoir on the hydrological droughts at the downstream Yichang station during 2003 2011, Hydrol. Process. , 27 (26), 3981 3993. Lins, H. F. (2012), USGS hydro climatic data network 2009 (HCDN 2009) , US Geological

PAGE 123

123 Sur vey. Geophys. Res. Lett. , 26 (2), 227 230. Livneh, B., and M. P. Hoerling (2016), Livneh, B., and M. P. Hoerling (2016), The physics of drought in the US central great plains, J. Clim., 29(18), 6783 6804.The physics of drought in the US central great plains, J. Clim. , 29 (18), 6783 6804. Van Loon, A. F. (2013), On the propagation of drought: how climate and catchment characteristics influence hydrological drought development and recovery . Van Loon, A. F. (2015), Hydrological drought explained, Wiley Interdiscip. Rev. Water , 2 (4), 359 392. Van Loon, A. F., and G. Laaha (2015), Hydrological drought severity explained by climate and catchment characteristics, J. Hydrol. , 526 , 3 14. Van Loon, A. F., and H. A. J. Van Lanen (2012), A process based typology of hydrological drought, Hydrol. Earth Syst. Sci. , 16 (7), 1915. Van Loon, A. F., E. Tijdeman, N. Wanders, H. A. Van Lanen, A. J. Teuling, and R. Uijlenhoet (2014), How climate seasona lity modifies drought duration and deficit, J. Geophys. Res. Atmos. , 119 (8), 4640 4656. Lund, R., and B. Li (2009), Revisiting climate region definitions via clustering, J. Clim. , 22 (7), 1787 1800. Luo, L., and E. F. Wood (2007), Monitoring and predicting the 2007 US drought, Geophys. Res. Lett. , 34 (22). Luo, L., D. Apps, S. Arcand, H. Xu, M. Pan, and M. Hoerling (2017), Contribution of temperature and precipitation anomalies to the California drought during 2012 2015, Geophys. Res. Lett. , 44 (7), 3184 3192. Mann, H. B. (1945), Non Parametric Tests against Trend. Econmetrica, 13, 245 259, Mao, Y., B. Nijssen, and D. P. Lettenmaier (2015), Is climate change implicated in the 2013 2014 California drought? A hydrologic perspective, Geophys. Res. Lett. , 42 (8), 28 05 2813. Martin, G. R., and L. D. Arihood (2010), Methods for estimating selected low flow frequency statistics for unregulated streams in Kentucky , US Geological Survey. McCabe, G. J., and D. M. Wolock (2002), Trends and temperature sensitivity of moistur e conditions in the conterminous United States, Clim. Res. , 20 (1), 19 29. McCabe, G. J., M. A. Palecki, and J. L. Betancourt (2004), Pacific and Atlantic Ocean influences on multidecadal drought frequency in the United States, Proc. Natl. Acad. Sci. , 101 (1 2), 4136 4141.

PAGE 124

124 Melillo, J. M., T. T. Richmond, and G. Yohe (2014), Climate change impacts in the United States, Third Natl. Clim. Assess. Menne, M. J., C. N. Williams Jr, and R. S. Vose (2010), Long term daily and monthly climate records from stations acro ss the contiguous United States, Website http//cdiac. ornl. gov/epubs/ndp/ushcn/access. html [accessed 23 Sept. 2010] . Menzel, L., and G. Bürger (2002), Climate change scenarios and runoff response in the Mulde catchment (Southern Elbe, Germany), J. Hydrol . , 267 (1), 53 64. Milly, P. C. D., K. A. Dunne, and A. V Vecchia (2005), Global pattern of trends in streamflow and water availability in a changing climate, Nature , 438 (7066), 347 350. Milly, P. C. D., J. Betancourt, M. Falkenmark, R. M. Hirsch, Z. W. Kun dzewicz, D. P. Lettenmaier, R. J. Stouffer, M. D. Dettinger, and V. Krysanova (2015), On critiques of Water Resour. Res. , 51 (9), 7785 7789. Mishra, A. K., and V. P. Singh (2010), A review of drought concep ts, J. Hydrol. , 391 (1), 202 216. Morid, S., V. Smakhtin, and M. Moghaddasi (2006), Comparison of seven meteorological indices for drought monitoring in Iran, Int. J. Climatol. , 26 (7), 971 985. Mpelasoka, F., K. Hennessy, R. Jones, and B. Bates (2008), Comp arison of suitable drought indices for climate change impacts assessment over Australia towards resource management, Int. J. Climatol. , 28 (10), 1283 1292. Mueller, B., and S. I. Seneviratne (2012), Hot days induced by precipitation deficits at the global s cale, Proc. Natl. Acad. Sci. , 109 (31), 12398 12403. Nalbantis, I. (2008), Evaluation of a hydrological drought index, Nalbantis, I., and G. Tsakiris (2009), Assessment of hydrological drought revisited, Water Resour. Manag. , 23 (5), 881 897. Naveau, P., A. Guillou, D. Cooley, and J. Diebolt (2009), Modelling pairwise dependence of maxima in space, Biometrika , 96 (1), 1 17. Newman, A. J., M. P. Clark, K. Sampson, A. Wood, L. E. Hay, A. Bock, R. J. Viger, D. Blodgett, L. Brekke, and J. R. Arnold (2015), Develop ment of a large sample watershed scale hydrometeorological data set for the contiguous USA: data set characteristics and assessment of regional variability in hydrologic model performance, Hydrol. Earth Syst. Sci. , 19 (1), 209. Niemeyer, S. (2008), No Title , Options Méditerranéennes , (80), 267 274. Novotny, E. V, and H. G. Stefan (2007), Stream flow in Minnesota: indicator of climate change, J. Hydrol. , 334 (3), 319 333. Ntale, H. K., and T. Y. Gan (2003), Drought indices and their application to East Africa, Int.

PAGE 125

125 J. Climatol. , 23 (11), 1335 1357. Nyabeze, W. R. (2004), Estimating and interpreting hydrological drought indices using a selected catchment in Zimbabwe, Phys. Chem. Earth, Parts A/B/C , 29 (15), 1173 1180. Pal, I., B. T. Anderson, G. D. Salvucci, and D . J. Gianotti (2013), Shifting seasonality and increasing frequency of precipitation in wet and dry seasons across the US, Geophys. Res. Lett. , 40 (15), 4030 4035. Pal, I., E. Towler, and B. Livneh (2015), How can we better understand low river flows as cli mate changes, Eos (Washington. DC) , 96 . evolution of trends, and the periodicity of low flows across the CONUS watersheds, Geophys. Res. Lett. , In press . Palmer, T. temperature, Nature , 338 (6210), 54 57. Pathak, A. A., and B. M. Dodamani (2016), Comparison of two hydrological drought indices, Perspect. Sci. , 8 , 626 628. Peterson, T. C., R . R. Heim Jr, R. Hirsch, D. P. Kaiser, H. Brooks, N. S. Diffenbaugh, R. M. Dole, J. P. Giovannettone, K. Guirguis, and T. R. Karl (2013), Monitoring and understanding changes in heat waves, cold waves, floods, and droughts in the United States: state of kn owledge, Bull. Am. Meteorol. Soc. , 94 (6), 821 834. Piechota, T., J. Timilsena, G. Tootle, and H. Hidalgo (2004), The western US drought: How bad is it, Eos (Washington. DC). , 85 (32), 301 304. Pournasiri Poshtiri, M., and I. Pal (2016), Patterns of hydrolog ical drought indicators in major US River basins, Clim. Change , 134 (4), 549 563. Pournasiri Poshtiri, M., E. Towler, and I. Pal (2016), Streamflow drought indicators across conterminous United States, , doi:10.5065/D6QR4V9X. Pyrce, R. (2004), Hydrological low flow indices and their uses, Watershed Sci. Centre,(WSC) Rep. , (04 2004). Reilly, C. F., and C. N. Kroll (2003), Estimation of 7 day, 10 year low streamflow statistics using baseflow correlation, Water Resour. Res. , 39 (9). Reynolds, L. V, P. B. Shafroth, and N. L. Poff (2015), Modeled intermittency risk for small streams in the Upper Colorado River Basin under climate change, J. Hy drol. , 523 , 768 780. Ries, K. G. (2007), The national streamflow statistics program: A computer program for estimating streamflow statistics for ungaged sites , DIANE Publishing. Ries, K. G., and P. J. Friesz (2000), Methods for estimating low flow statisti cs for Massachusetts streams , US Department of the Interior, US Geological Survey.

PAGE 126

126 Riggs, H. C. (1972), Low flow investigations , US Government Printing Office. Risley, J., A. Stonewall, and T. Haluska (2008), Estimating Flow Duration and Low Flow Frequency Statistics for Unregulated Streams in Oregon . Rolls, R. J., C. Leigh, and F. Sheldon (2012), Mechanistic effects of low flow hydrology on riverine ecosystems: ecological principles and consequences of alteration, Freshw. Sci. , 31 (4), 1163 1186. Rossi, M. W., K. X. Whipple, and E. R. Vivoni (2016), Precipitation and evapotranspiration controls on daily runoff variability in the contiguous United States and Puerto Rico, J. Geophys. Res. Earth Surf. Ryberg, K. R., and A. V Vecchia (2012), waterData -An R pack age for retrieval, analysis, and anomaly calculation of daily hydrologic time series data, version 1.0 , US Geological Survey. Sadri, S., J. Kam, and J. Sheffield (2016), Nonstationarity of low flows and their timing in the eastern United States, Hydrol. Ea rth Syst. Sci. , 20 (2), 633 649. Schilling, K. E., M. K. Jha, Y. Zhang, P. W. Gassman, and C. F. Wolter (2008), Impact of land use and land cover change on the water balance of a large agricultural watershed: Historical effects and future directions, Water Resour. Res. , 44 (7). J. Am. Stat. Assoc. , 63 (324), 1379 1389. Seneviratne, S. I., N. Nicholls, D. Easterling, C. M. Goodess, S. Kanae, J. Kossin, Y. Luo, J. Marengo, K. McIn nes, and M. Rahimi (2012), Changes in climate extremes and their impacts on the natural physical environment, Manag. risks Extrem. events disasters to Adv. Clim. Chang. Adapt. , 109 230. Sheffield, J., and E. F. Wood (2008), Projected changes in drought occ urrence under future global warming from multi model, multi scenario, IPCC AR4 simulations, Clim. Dyn. , 31 (1), 79 105. Shiau, J. T., and R. Modarres (2009), Copula based drought severity duration frequency analysis in Iran, Meteorol. Appl. , 16 (4), 481 489. Shukla, S., M. Safeeq, A. AghaKouchak, K. Guan, and C. Funk (2015), Temperature impacts on the water year 2014 drought in California, Geophys. Res. Lett. , 42 (11), 4384 4393. Sivapalan, M., S. E. Thompson, C. J. Harman, N. B. Basu, and P. Kumar (2011), Wat er cycle dynamics in a changing environment: Improving predictability through synthesis, Water Resour. Res. , 47 (10). Smakhtin, V. U. (2001), Low flow hydrology: a review, J. Hydrol. , 240 (3), 147 186. Smakhtinb, S. M. V., and K. BagherzadehC (2007), Drought networks and time series of drought indices, Int. J. Clim. , 27 , 2103 2111.

PAGE 127

127 Small, D., S. Islam, and R. M. Vogel (2006), Trends in precipitation and streamflow in the eastern US: Paradox or perception?, Geophys. Res. Lett . , 33 (3). Stohlgren, T. J., T. N. Chase, R. A. Pielke, T. G. F. Kittel, and J. Baron (1998), Evidence that local land use practices influence regional climate, vegetation, and stream flow patterns in adjacent natural areas, Glob. Chang. Biol. , 4 (5), 495 50 4. Sung, J. H., and E. S. Chung (2014), Development of streamflow drought severity duration frequency curves using the threshold level method, Hydrol. Earth Syst. Sci. , 18 (9), 3341 3351. Tallaksen, L. M., and H. A. J. Van Lanen (2004), Hydrological drought : processes and estimation methods for streamflow and groundwater , Elsevier. Tallaksen, L. M., and K. Stahl (2014), Spatial and temporal patterns of large scale droughts in Europe: Model dispersion and performance, Geophys. Res. Lett. , 41 (2), 429 434. Tall aksen, L. M., H. Madsen, and B. Clausen (1997), On the definition and modelling of streamflow drought duration and deficit volume, Hydrol. Sci. J. , 42 (1), 15 33. Tharme, R. E. (2003), A global perspective on environmental flow assessment: emerging trends i n the development and application of environmental flow methodologies for rivers, River Res. Appl. , 19 (5 6), 397 441. Timilsena, J., T. Piechota, G. Tootle, and A. Singh (2009), Associations of interdecadal/interannual climate variability and long term colorado river basin streamflow, J. Hydrol. , 365 (3), 289 301. Tomer, M. D., and K. E. Schilling (2009), A s imple approach to distinguish land use and climate change effects on watershed hydrology, J. Hydrol. , 376 (1), 24 33. Trenberth, K. E., G. W. Branstator, and P. A. Arkin (1988), Origins of the 1988 North American drought, Science (80 . ). , 242 (4886), 1640 1 646. Trenberth, K. E., A. Dai, G. Van Der Schrier, P. D. Jones, J. Barichivich, K. R. Briffa, and J. Sheffield (2014), Global warming and changes in drought, Nat. Clim. Chang. , 4 (1), 17 22. Tsakiris, G., I. Nalbantis, H. Vangelis, B. Verbeiren, M. Huysmans , B. Tychon, I. Jacquemin, F. Canters, S. Vanderhaegen, and G. Engelen (2013), A system based paradigm of drought analysis for operational management, Water Resour. Manag. , 27 (15), 5281 5297. U.S. EPA (2017), Definition and Characteristics of Low Flows fro m DFLOW, U.S. EPA . Available from: https://www.epa.gov/waterdata/definition and characteristics low flows dflow Van der Vaart, A. W. (2000), Asymptotic statistics , Cambridge university press. VanShaar, J. R., I. Haddeland, and D. P. Lettenmaier (2002), Ef fects of land cover changes on

PAGE 128

128 the hydrological response of interior Columbia River basin forested catchments, Hydrol. Process. , 16 (13), 2499 2520. Vasiliades, L., A. Loukas, and N. Liberis (2011), A water balance derived drought index for Pinios River Bas in, Greece, Water Resour. Manag. , 25 (4), 1087 1101. Vidal, J. P., E. Martin, L. Franchistéguy, F. Habets, J. M. Soubeyroux, M. Blanchard, and M. Baillon (2010), Multilevel and multiscale drought reanalysis over France with the Safran Isba Modcou hydrometeo rological suite, Hydrol. Earth Syst. Sci. Discuss. , 14 (3), 459 478. Van Vliet, M. T. H., J. R. Yearsley, F. Ludwig, S. Vögele, D. P. Lettenmaier, and P. Kabat (2012), Vulnerability of US and European electricity supply to climate change, Nat. Clim. Chang. , 2 (9), 676 681. Vogel, R. M., and C. N. Kroll (1992), Regional geohydrologic geomorphic relationships for the estimation of low flow statistics, Water Resour. Res. , 28 (9), 2451 2458. Vander Vorste, R., F. Mermillod Blondin, F. Hervant, R. Mons, M. Forcelli ni, and T. Datry (2016), Increased depth to the water table during river drying decreases the resilience of Gammarus pulex and alters ecosystem function, Ecohydrology , 9 (7), 1177 1186. Wada, Y., L. P. H. Van Beek, N. Wanders, and M. F. P. Bierkens (2013), Human water consumption intensifies hydrological drought worldwide, Environ. Res. Lett. , 8 (3), 34036. Wagener, T., M. Sivapalan, P. A. Troch, B. L. McGlynn, C. J. Harman, H. V Gupta, P. Kumar, P. S. C. Rao, N. B. Basu, and J. S. Wilson (2010), The future o f hydrology: An evolving science for a changing world, Water Resour. Res. , 46 (5). Wanders, N., H. A. J. Van Lanen, and A. F. van Loon (2010), Indicators for drought characterization on a global scale , Wageningen Universiteit. Wang, Y., J. Li, P. Feng, and F. Chen (2015), Effects of large scale climate patterns and human activities on hydrological drought: a case study in the Luanhe River basin, China, Nat. Hazards , 76 (3), 1687 1710. Whitfield, P. H., J. Y. Wang, and A. J. Cannon (2003), Modelling future str eamflow extremes floods and low flows in Georgia basin, British Columbia, Can. Water Resour. J. , 28 (4), 633 656. Wilbanks, T., S. Fernandez, G. Backus, P. Garcia, K. Jonietz, P. Kirshen, M. Savonis, W. Solecki, L. Toole, and M. Allen (2012), Climate change and infrastructure, urban systems , Springer. WMO (2009), WMO statement on the status of the global climate in 2009 . Yahiaoui, A., B. Touaïbia, and C. Bouvier (2009), Frequency analysis of the hydrological drought regime. Case of oued Mina catchment in wes tern of Algeria, Nat. Technol. , (1), 3.

PAGE 129

129 Yenigun, K., and R. Ecer (2013), Overlay mapping trend analysis technique and its application in Euphrates Basin, Turkey, Meteorol. Appl. , 20 (4), 427 438. Yevjevich, V. M. (1967), An objective approach to definitions and investigations of continental hydrologic droughts, Hydrol. Pap. (Colorado State Univ. no. 23 . Yulianti, J. S., and D. H. Burn (1998), Investigating links between climatic warming and low streamflow in the Prairies region of Canada, Can. Water Resour. J. , 23 (1), 45 60. Water Resour. Res. , 23 (1), 156 168.

PAGE 130

130 APPENDIX Trends of Annual Q30 c nd Annual C DO With 10q30 Threshold Level This appendix contains figures that show the tr end of annual q30 (Figure A.1) and annual CDO with respect to 10q30 threshold level (Figure A.2) . Trend pattern of q30 (Figure A.1) is similar to trend pattern of annual q7 (Figure 3.1), presented in Chapter III . Trend pattern of CDO is same for both 10q7 threshold level (Figure 3.7) and 10q30 threshold level (Figure A.2).

PAGE 131

131 (A) (B) Figure A.1: (A) Trends of annual 30 days minimum flow (q30). Color bubbles indicate location of the stations, and sign and significance of trend estimates (in cubic f eet per second (cfs) per day per year). The size of the bubble is proportional to the magnitude of trend value; (B) Proportions of total number of stations showing different trends in annual q7 and q30.

PAGE 132

132 (A) (B) Figure A . 2: (A) Trends of annual C DO from start year 2012. Color bubbles indicate location of the stations, and sign and significance of trend estimates (in days per year). The size of the bubble is proportional to the magnitude of trend value. (B) Proportions of total number of stations s howing different trends in annual CDO corresponding to 10q7 and 10q30.