Water Supply and Water Quality: Putting Together How Natural and Human Factors Affect These, Using Satellites

WATER SUPPLY AND WATER QUALITY: PUTTING TOGETHER HOW NATURAL AND HUMAN FACTORS AFFECT THESE, USING SATELLITES

Mejs Hasan

A dissertation submitted to the faculty at the University of North Carolina at Chapel Hill in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Geological Sciences in the College of Arts and Sciences.

Chapel Hill 2018

Approved by: Larry Benninger Aaron Moody John Bane Colin West Xiaoming Liu

ABSTRACT

Mejs Hasan: Water supply and water quality: Putting together how natural and human factors affect these, using satellites (Under the direction of Larry Benninger and Aaron Moody)

This dissertation explores the interplay of human and natural factors upon water resources in the

Chesapeake Bay, the Tigris and Euphrates Rivers, and the marshes of southern Iraq. I used high-quality but expensive-to-collect fieldwork data of water status. I combined that data with regularly-produced satellite images covering large sections of the earth. Combining high-quality fieldwork data with global, continuous satellite data can produce datasets that are richer than the sum of their two parts.

I first studied the effect of storms on water quality in the Chesapeake Bay by using a relationship between satellite-measured red light reflectance and ground measurements of total suspended solids

(TSS). This resulted in viable reflectance-TSS relationships for five major Western Shore rivers.

Modeling a single reflectance-TSS relationship for the entire estuary produced poorer models with less significance compared to treating each channel separately. After studying the aftermath of 2800 rain events, I found some evidence that higher rainfall corresponds to a lower distribution of TSS concentrations one day following the storm in forested, compared to urban, watersheds.

In chapter 2, I studied how conflicts and drought disrupt water supply on dams and barrages along the Tigris and Euphrates rivers. I used a satellite-based algorithm, the normalized difference water index (NDWI), to monitor changes in the extent of surface reservoirs (1985-present). The most sudden changes in water supply occurred during conflict, but conflict was not often a cause of the greatest absolute changes to reservoir area. In chapter 3, I again used NDWI and similar algorithms to examine how the seasonal cycle of marshes at the southern end of the Tigris and Euphrates has been affected by

iii drought, development, and conflict. I found some evidence that the yearly timing of the marsh peak size has become more variable after exposure to these stressors.

No matter the stressor – from storms to drought to war – water quality and supply are highly affected by human and natural factors. In combination with ground information, satellite data can fill in data gaps and offer further insights.

I dedicate my dissertation to the three UNC-Chapel Hill and N.C. State students killed in 2015:

Yusor Mohammad Abu-Salha, her sister Razan Mohammad Abu-Salha, and her husband Deah Shaddy

Barakat; and to our former student body president, Eve Marie Carson, killed in 2008.

I also dedicate it to the many humanitarian workers who traveled to Iraq or neighboring countries in order to build hospitals, set up voting systems, or support education of girls, and instead were killed.

Also, this dissertation is dedicated to the many Iraqi children who have died or suffered through the last three decades of war.

ACKNOWLEDGEMENTS

This work was supported and funded by the Graduate School at the University of North Carolina at Chapel Hill (UNC-CH), Royster Society of Fellows, the Department of Geology at UNC-CH, and the

Martin Funding Program at the Department of Geology.

I was supported through this process by my terrific advisors, Dr. Larry Benninger and Dr. Aaron

Moody, as well as Dr. John Bane, Dr. Colin West, and Dr. Xiaoming Liu who comprised the rest of my committee. They helped me out of very hard situations, heedless of the extra time and work it meant for them. They were patient, excited about my progress, and devoted many conversations to moving the research forward. Their support meant that I was able to successfully complete my degree, and to experience the amazing opportunities that fell into my lap over the past few years. I have always appreciated that my entire committee never hesitated for a second when I asked them for help, that they always showed a lot of confidence in me, and that they always focused on the good. Thank you.

The support from the Royster Society of Fellows – Sandra Hoeflich, Teresa Phan, Jennifer Gerz-

Escandon, Marsha Collins – was also so wonderful and so appreciated.

This study was possible due to governmental and institutional support for open access to environmental data. The satellite data on which this dissertation depended so heavily were freely acquired from American government servers. Specifically, Chesapeake Bay MODIS data was retrieved from the

Level-1 and Atmosphere Archive & Distribution System (LAADS) Distributed Active Archive Center

(DAAC), located in the NASA Goddard Space Flight Center in Greenbelt, Maryland. Landsat and ESA images of the Middle East were retrieved from Google Earth Engine servers, to which the satellite images have been uploaded, processed, and corrected by NASA and the U.S. Geological Survey. Other sources of

vi freely available data used include the Chesapeake Bay Program, NOAA rain gages, and USGS discharge stations.

UNC-CH libraries also played a central role by offering statistical consulting help (especially

Chris Wiesen in Davis Library), providing access to books and journals, and procuring old and difficult- to-locate reports which were especially key for chapter 2.

Ron Vogel of NOAA provided a very helpful review of the portions of the dissertation dealing with the Chesapeake Bay, and Ali Hasan of IBM provided statistical insight on the same chapter. Helpful email communications and documents from Nadhir Al-Ansari of Luleå University of Technology gave great insight and needed information on Mosul Dam for chapter 2. I thank them all for their help.

Finally, thank you to all the family, friends, and relatives who took an interest in my research and whose good wishes sped me along.

vii

TABLE OF CONTENTS

LIST OF TABLES ...... xi

LIST OF FIGURES ...... xii

LIST OF ABBREVIATIONS ...... xiv

INTRODUCTION ...... 1

CHAPTER 1: RESILIENCY OF THE WESTERN CHESAPEAKE BAY TO TOTAL SUSPENDED SOLID CONCENTRATIONS FOLLOWING STORMS AND ACCOUNTING FOR LAND-COVER ... 5

Section 1: Introduction ...... 5

Section 2: Methods ...... 9

Fieldwork data ...... 9

Satellite data ...... 11

Storm and land cover data ...... 13

Section 3: Results and Discussion ...... 15

Relationships for separate channels ...... 15

Error within channels ...... 21

Storms in the Chesapeake ...... 24

Section 4: Conclusions ...... 34

CHAPTER 2: HOW WAR, DROUGHT, AND DAM MANAGEMENT IMPACT WATER SUPPLY IN THE TIGRIS AND EUPHRATES RIVERS ...... 36

Section 1: Introduction ...... 36

Section 2: Materials and Methods ...... 39

Study area ...... 39

Data ...... 43

viii

Methods...... 45

Section 3: Results ...... 47

Extreme events ...... 47

Classification error ...... 51

Altimeters for gaps ...... 51

River surface area for discharge ...... 52

Droughts ...... 52

Conflicts ...... 53

Other instances of significance ...... 53

Relations to other reservoirs ...... 54

Section 4: Discussion ...... 56

Seasonality ...... 56

Droughts and Conflicts ...... 57

Upstream dams ...... 60

Managing for dam failure ...... 60

Uncertainties ...... 61

Reverse pathway of water to conflict ...... 61

Section 5: Conclusions ...... 62

CHAPTER 3: SEASONAL CHANGES IN THE MARSHES OF SOUTHERN IRAQ DUE TO DROUGHT, DEVELOPMENT, AND CONFLICT ...... 63

Section 1: Introduction ...... 63

Section 2: Data and Methods ...... 68

Study area ...... 68

Data ...... 68

Methods...... 69

Section 3: Results ...... 71

Error analysis ...... 71

Size of the marsh ...... 73

Seasons ...... 78

Human and natural factors ...... 82

Section 4: Discussion ...... 83

Section 5: Conclusions ...... 87

CONCLUSION ...... 89

APPENDIX 1: EXTRA NOTES ON TSS ...... 91

APPENDIX 2: SUPPLEMENTARY FIGURES ...... 93

APPENDIX 3: SUPPLEMENTARY TABLES ...... 116

REFERENCES ...... 117

LIST OF TABLES

Table 1. NOAA rain gages...... 14

Table 2. Model parameters...... 20

Table 3. Accuracy analysis...... 22

Table 4. Dams and barrages along the Tigris and Euphrates...... 38

Table 5. Tigris/Euphrates error matrix and accuracy results...... 47

Table 6. Summary of substantial changes in water supply along the Tigris and Euphrates...... 50

Table 7. Correlations between dam lake sizes...... 55

Table 8. Different periods of marsh surface area size...... 73

Table 9. Correlations between upstream reservoirs and Hawiza marsh area...... 83

Table S1. List of conflicts...... 116

Table S2. Landsat path/row for each reservoir, and number of days with data...... 116

LIST OF FIGURES

Figure 1. Chesapeake Bay study area...... 6

Figure 2. Satellite reflectance-TSS concentration relationships...... 18

Figure 3. Chesapeake TSS time series...... 20

Figure 4. TSS estimates derived from MODIS...... 25

Figure 5. Box-plots of post-storm TSS estimates...... 28

Figure 6. Peaks in the MODIS-derived TSS estimates...... 30

Figure 7. MODIS-derived TSS estimates following high rainfall...... 33

Figure 8. Map of the Tigris/Euphrates study area...... 41

Figure 9. The head of Mosul Dam illustrated in Landsat 8 imagery...... 43

Figure 10. Surface area of Haditha (a) and Mosul (b) reservoirs...... 48

Figure 11. Rates of change in reservoir lake-area of Haditha (a) and Mosul (b) reservoirs...... 51

Figure 12. Scatterplots of reservoir size...... 56

Figure 13. A map and a conceptual diagram of the marshes in southern Iraq...... 67

Figure 14. Marsh error analysis...... 72

Figure 15. Timeline of the total marsh surface area...... 74

Figure 16. Landsat images of the marshes...... 77

Figure 17. The range of marsh extent by month...... 79

Figure 18. Timing of marsh peak and minimum extent...... 80

Figure 19. The monthly extent of vegetated and open water marsh area...... 81

Figure 20. The yearly ratio of the peak vegetated marsh area to minimum marsh area...... 82

Figure S1. Percent of various land covers for the 12 gage-station sites in the Chesapeake Bay; ...... 93

Figure S2. Significance of the reflectance-TSS models...... 94

Figure S3. Reflectance-TSS relationships for Mainstem stations...... 95

Figure S4. CBP measures of TSS concentrations versus MODIS-derived estimates...... 97

Figure S5. Box-plots of the mean rainfall distribution...... 98

xii

Figure S6. Cumulative precipitation based on PERSIANN estimates, 1983-2016...... 99

Figure S7. Distribution of dates with satellite data by month and by year, for Mosul reservoir...... 100

Figure S8. NDWI distributions for land and reservoir polygons...... 102

Figure S9. Geographical position of points used in validation analysis...... 103

Figure S10. Altimeter-Landsat scatterplots...... 104

Figure S11. Close-ups of significant episodes of lake surface area...... 106

Figure S12. Close-ups of significant episodes of lake surface area...... 108

Figure S13. Images representing changes in Mosul and Haditha reservoirs...... 110

Figure S14. Floods in near Abu Ghraib...... 111

Figure S 15. Median monthly temperatures in the Tigris/Euphrates...... 112

Figure S16. The discharge at Kut, Iraq...... 113

Figure S17. The month during which Hawiza marsh has its peak area of open water, by year...... 114

Figure S18. Scatterplot of same-month marsh water extent and discharge at Kut station...... 115

xiii

LIST OF ABBREVIATIONS

CBP Chesapeake Bay Program

CDOM Colored dissolved organic matter

DAAC Distributed Active Archive Center

ETM Estuarine turbidity maximum

ETM+ Enhanced Thematic Mapper Plus

LAADS Level-1 and Atmosphere Archive and Distribution System

GRACE Gravity Recovery and Climate Experiment

HUC Hydrologic unit codes

MODIS Moderate Resolution Imaging Spectroradiometer

NDVI Normalized difference vegetation index

NDWI Normalized difference water index

NIR Near-infrared

OLI Operational Land Imager

PERSIANN Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks RIM River-Input Monitoring

RMSE Root mean square error

SAV Submerged aquatic vegetation

SeaWiFS Sea-viewing Wide Field-of-view Sensor

SNAP Sediment and Nutrient Assessment Program

xiv

TOA Top-of-the-atmosphere

TM Thematic Mapper

TSS Total suspended solids

UNC-CH University of North Carolina at Chapel Hill

INTRODUCTION

Availability of clean water is expected to be one of the most important issues faced by 21st century society. The Millennium Ecosystem Assessment (2005) suggests that use of fresh water for drinking, industry and irrigation has reached unsustainable levels, and particularly in the Middle East, a full third of all water-use is sourced unsustainably. Such precarious water sources become even scarcer during droughts and other environmental variability. For example, the Middle East underwent its most severe drought on record between 2007 and 2009 (Trigo et al. 2010), and it is believed that the current

Syrian Civil War was in part driven by the drought-linked agricultural decline (Kelley et al. 2015). As the conflict spread to neighboring Iraq, water has often been used as a weapon of war (Collard 2014). Closer to home, the Chesapeake Bay and other coastal ecosystems face challenges in preserving water quality to sufficient standards to meet ecosystem and recreational needs, while also allowing room for upstream development activities and land-use (Aighewi et al. 2013).

Satellite remote sensing is a valuable tool for studying and addressing these issues. For example, the extent of marshes at the confluence of the Tigris and Euphrates rivers has been mapped by three generations of Landsat satellite data over a 30-year time period (Al-Handal & Hu 2014, Becker 2014,

Ghobadi et al. 2015). Trends in groundwater data in the Middle East have been described using GRACE satellite data (Voss et al. 2013, Tourian et al. 2015). In the Chesapeake Bay, color reflectance properties of various water constituents have been analyzed and can yield models describing suspended sediment concentrations (Tzortziou et al. 2007).

In this dissertation, I use remote sensing and ground data to investigate changes in water supply and quality caused by human-driven land-use change and by human conflict. First, I investigate how land- use, such as farms and urban areas, plays a role in determining suspended sediment levels in the

Chesapeake Bay following rainfall events of all magnitudes (chapter 1). Secondly, I study how water

1 supply in Tigris/Euphrates basin has changed over the last three decades, both as a reaction to conflicts and to development (chapter 2 and 3).

The Chesapeake Bay is the nation’s largest estuary, home to one of its most productive ecosystems, and is an icon of American heritage and history (Kemp et al. 2005). Nutrient and total suspended solid (TSS) concentrations have exceeded Clean Water Act regulations in recent decades, resulting in sediment-driven water clarity impairment and light attenuation. How storms affect TSS concentrations is still not well-understood, since studies are usually small in scope and give conflicting results (Ward 1985, Gellis et al. 2009, Sutton et al. 2009). This can be analyzed in further depth by using satellite-derived reflectance properties of total suspended solids (TSS) in the Bay in order to estimate TSS concentrations. Such relationships enable us to estimate TSS over longer time intervals and at more locations than is possible from fieldwork. Some such relationships have already been published, but there is room for improvement. Current relationships are limited to a narrow TSS concentration range, so very high or very low values cannot be estimated. Also, relationships devised for the Chesapeake Mainstem are extrapolated to all tributaries of the Bay, regardless of whether separate tributaries may have individual sediment signatures with different reflectance properties (Tzortziou et al. 2007, Ondrusek et al.

2012, Son & Wang 2012, Zheng et al. 2015).

Aspects of Tigris/Euphrates basin water resources have also been studied via remote sensing. For example, the effects on vegetation of both a 1998-2000 and a 2007-2009 drought have been examined using the satellite-derived normalized difference vegetation index (NDVI) (Trigo et al. 2010, Villa et al.

2014). Water level estimates for major lakes and reservoirs in this region also exist through satellite altimeters, and show how water levels recede in response to drought (Trigo et al. 2010, Voss et al. 2013,

Joodaki et al. 2014). How conflict and development intertwine with drought and affect water supply has not been studied.

This dissertation consists of three chapters:

Chapter 1: Study of how storms and land-use affect TSS concentrations in the Chesapeake

Bay using satellite reflectance properties. The Chesapeake Bay has an impressive monitoring program

2 that has sampled the Bay for TSS since 1986, but many gaps in time and space exist. I looked for relationships between red light reflectance (measured by satellite) and total suspended solid concentrations (measured from a boat) in order to fill in these gaps. Red light reflectance is retrieved from the Terra sensor on the NASA satellite MODIS. I assessed whether Chesapeake tributaries are better served by their own reflectance-TSS relationships, rather than reliance on a single Bay-wide model; used these relationships to greatly expand post-storm TSS estimates in the Chesapeake and address the ambiguity in current studies on how storms affect TSS; and evaluated whether different land-uses affect the storm-TSS patterns. I found that post-storm river TSS is higher in urban and agricultural areas compared to forests.

Chapter 2: Investigate how the different and at-times coupled impacts of drought, war, and water management have affected both quantity and quality of water resources in the

Tigris/Euphrates area. I used altimeter-derived measures of water depth in lakes and reservoirs, and

Landsat-derived water surface area measurements (via the normalized difference water index algorithms), in order to create timelines of water supply behind dams and in rivers. I also used globally compiled datasets of precipitation and temperature, and Landsat-derived upstream reservoir surface area in order to control for rainfall and snowmelt, upstream dam management, and evaporation. Combining all this information in a crude model, I studied if anomalous changes in reservoir surface area occur during battles and wars. I found that conflict did produce some very sharp changes in reservoir surface area, though not all conflict did; drought also produced large changes, though at a slower pace.

Chapter 3. Investigate how coupled human-natural factors have affected the southern marshes on the Tigris/Euphrates in terms of the seasonal patterns. I developed satellite-based monthly surface area trends of the Tigris/Euphrates southern marshes. From these, I determined the seasonality in marsh lakes, and quantify when the ‘wet season’ occurs and how the onset and duration of this season is affected by upstream dam development and/or drought. I also determined the difference in the marsh extent during the dry versus wet seasons. I found some evidence that the difference in marsh

3 extent between the peak and minimum seasons is declining, and that the timing of the wet/dry seasons has become more variable.

Water is a vital resource for all forms of life, and is connected to survival, sanitation, food production, commerce, recreation, and ecosystems. Growing populations make it more difficult for competing stakeholders to meet all their needs while also preserving the environmental flows needed for biodiversity, habitats, and ecosystems. Disturbances such as storms, drought, and war can all make this competition even more acute. The research described here will increase knowledge of how water supply and some aspects of quality are affected by such disturbances. It provides information on which types of human activities are most stabilizing for water supply and quality, and which are not.

CHAPTER 1: RESILIENCY OF THE WESTERN CHESAPEAKE BAY TO TOTAL SUSPENDED SOLID CONCENTRATIONS FOLLOWING STORMS AND ACCOUNTING FOR LAND-COVER

In the first chapter of this dissertation, I discuss the use of satellite images to study certain aspects of water quality in the Chesapeake Bay. People are often surprised to learn that a satellite image can shed light on water quality, but it is due to the fact that the more impurities water contains, the more light it reflects. Satellites can detect that change in reflectance. I created relationships between satellite reflectance and total suspended solids (TSS), and then used those relationships to explore how TSS changes in the Bay following major storms. I showed that by coupling two very different datasets – high quality ground data with global and continuous satellite data – the resulting dataset is more comprehensive than the sum of its two parts.

This paper was published as: Hasan, M., and L. Benninger. 2017. Resiliency of the western

Chesapeake Bay to total suspended solid concentrations following storms and accounting for land-cover.

Estuarine, Coastal and Shelf Science 191. doi:10.1016/j.ecss.2017.04.002.

Section 1: Introduction

The Chesapeake Bay is the nation’s largest estuary, home to one of its most productive ecosystems, and is an icon of American heritage and history (Kemp et al. 2005). It extends across a large swath of the mid-Atlantic coast, from the mouth of the Susquehanna River near Havre de Grace,

Maryland, to its juncture with the Atlantic Ocean near Virginia’s Norfolk metropolitan area (Figure 1).

However, nutrient and total suspended solids (TSS) have exceeded Clean Water Act regulations in recent decades, resulting in sediment-driven water clarity impairment and light attenuation. This inhibits solar energy from reaching submerged aquatic vegetation (SAV) and oyster reefs and the associated fish nurseries, some of the most crucial components of the ecosystem (Jordan et al. 1997, Focazio et al. 1998,

Gellis et al. 2009). In this paper, we use satellite data to study the effect of storms and land cover on TSS levels.

Figure 1. Chesapeake Bay study area. Black squares represent the CBP stations and the extent of the study area upstream each channel. CBP stations identified in the paper are labeled in pink. Location and name of NOAA rain gages are represented by red triangles/white text bubbles. Light blue text labels indicate channel names. Orange text labels indicate NOAA “tides and currents” stations. The lower green dash is the Virginia-Maryland border; the upper green dash is the Bay Bridge.

Uncoordinated research on storm-effects in the Bay has been performed in a patchwork way at limited study sites and times. Some studies found TSS and nutrients increased following storms, but this is not universal and depends on sediment size, organic/mineral content, compaction, invertebrate mucus secretions, land slope, and water velocity (Ward 1985, Gellis et al. 2009, Sutton et al. 2009). Furthermore, wet weather is not alone in increasing TSS levels, as dry, windy weather can lead to increased erosion and wave-caused resuspension acting on normally submerged river beds (Stevenson et al. 1985, Ward &

Twilley 1986). Storm-induced TSS concentrations are not necessarily runoff-based but may be wind or tidal-driven bottom resuspension, particularly in areas without buffering features like SAV beds (Ward et al. 1984, Stevenson et al. 1985, Ward 1985). The settling periods for post-storm TSS levels before a return to normal conditions can be less than 24 hours for smaller events, but as high as weeks for large hurricanes (Ward et al. 1984, Stevenson et al. 1985, Liu & Wang 2014). Land cover variables, such as proximity and size of urban areas, forests, agriculture and wastewater treatment plants, play an important role in how the Bay responds to stressors (Aighewi et al. 2013). Heavily forested watersheds can sustain post-storm TSS concentration declines due to dilution if water is added more rapidly than suspended solids, while urban watersheds may become more susceptible to storm-driven TSS increases due to streambank erosion (Stevenson et al. 1985, Ward & Twilley 1986, Devereux et al. 2010).

To investigate effects of storms and land cover on TSS levels, data must be gathered directly following storms in river basins with different land cover type dominance. Such data is difficult to come by. The EPA Chesapeake Bay Program (CBP) has systematically sampled the Bay for nutrients and TSS since 1986, but tidal river measurements are often made once a month at routine station stops where the schedule does not shift to emphasize different weather conditions (Olson et al. 2012). Indeed, sampling is less likely to occur during stormy weather due to harsher conditions. Lacking data, TSS may be estimated after the fact using water quality models that rely on rainfall and hydrodynamic principles. However, a review of ten such models in the Chesapeake Bay found them afflicted with conflicting results, and that while a model might be suited to a specific river, it could not be extrapolated to the entire watershed

(Boomer et al. 2013). Alternatively, the USGS and other researchers have used daily streamflow

7 monitoring to create load-flow relationships based on time, discharge, and season (Moyer et al. 2012,

Zhang et al. 2015, Chanat et al. 2016).

Satellite light reflectance data provide the potential to improve both spatial and temporal coverage of surface-water TSS levels. Since the 1970s, water light reflectance models of sediment concentrations have been developed for several watersheds world-wide. Particle size, shape, and composition all combine to reflect light from the water surface distinctly as compared to pure water

(Novo et al. 1989, Binding et al. 2005). Per unit mass, larger particles reflect less because of the decrease in surface area to mass ratio, while finer particles reflect more light (Stumpf & Pennock 1989). The light reflectance-sediment concentration relationship is often predicated on red light (around 600-700 nm), as in this wavelength range water reflects the least while suspended particles reach peak reflectance (Binding et al. 2005, Tzortziou et al. 2007). Advantageously, red light wavelengths are also most likely to escape interference from atmospheric aerosol absorption. Some models use a ratio of red and near-infrared (NIR) light, but NIR light requires higher TSS concentrations in order to overcome water absorbance of NIR.

Thus, inclusion of NIR is not always effective, and can in fact obscure reflectance-TSS relationships in estuaries where TSS normally remains below 50 mg/L (Binding et al. 2005). Relationships between suspended material concentrations and light reflectance can be linear (Miller & McKee 2004, Bowers &

Binding 2006). More often, models describing relationships are somewhat curved so as to account for saturation in light reflectance that occurs at high levels of particle concentration due to particle shading

(Petus et al. 2010, Ondrusek et al. 2012). In areas like the Irish Sea and the Mississippi Sound/northern

Gulf of Mexico, the red satellite reflectance from SeaWiFS and MODIS explained suspended material concentrations with an r2 of 0.89 or higher, with a maximum concentration of approximately 60 mg/L

(Miller & McKee 2004, Binding et al. 2005).

In the Chesapeake itself, Tzortziou et al. 2007 studied water reflectance properties and confirmed that unlike in open ocean waters, Chesapeake non-algal particles, dissolved organic matter, and phytoplankton vary independently, and each exhibits its own reflectance and absorption properties. That paper and subsequent ones then attempted to describe Chesapeake reflectance-suspended material

8 relationships using various MODIS products, with models ranging from linear to polynomial, and r2 values that ranged from 0.44 to 0.79 (Tzortziou et al. 2007, Ondrusek et al. 2012, Son & Wang 2012,

Zheng et al. 2015). Maximum suspended material concentrations in these studies ranged from 15 mg/L to

25 mg/L. Fieldwork for these studies was conducted solely in the Mainstem and then often applied to individual Bay tributaries such as the Susquehanna River (Ondrusek et al. 2012, Son & Wang 2012, Liu

& Wang 2014). However, as Tzortziou et al. 2007 point out, the optical properties of the Bay, including particulate backscatter, vary widely both in space and from season to season, so extrapolation of a model from one section to others may be inadvisable.

In this paper, we used a relationship between MODIS-Terra red light reflectance and total suspended solid (TSS) concentrations in order to study the effect of storms on TSS levels in the

Chesapeake Bay. The objectives of the study were to: (i) assess whether Chesapeake tributaries are better served by their own reflectance-TSS relationships, rather than reliance on a single Bay-wide model; (ii) investigate sources of error in these relationships; (iii) use these relationships to greatly expand post-storm

TSS estimates in the Chesapeake and address the ambiguity in current studies on how storms affect TSS in five Western Shore rivers; and (iv) evaluate whether different land-uses affect the storm-TSS patterns.

Section 2: Methods

Fieldwork data

TSS concentration data were obtained from the CBP monitoring database for the entire Bay from

2000 onwards, the year that marks the launch of the MODIS-Terra sensor (Olson et al. 2012). CBP data has been used to validate previous sediment concentration-light reflectance models and is also used as inputs in eutrophication models (Cerco et al. 2004, Ondrusek et al. 2012, Zheng et al. 2015). Although suspended material concentrations measured via the “suspended sediment concentration” laboratory method (as opposed to the “total suspended solids (TSS)” method) have been shown to be more accurate, the CBP database is populated almost exclusively with TSS data points in describing both Mainsteam and tributary sediment concentrations (Glysson et al. 2000, Gray et al. 2000; see also Appendix 1). CBP

9 samples in this analysis were restricted to those collected within one meter of the water surface because red light most effectively samples depths of 1 meter or less. We eliminated all CBP stations located in areas where the 250 meter MODIS surface reflectance pixel did not contain purely open water, leaving us with 240 stations out of a total of over 600 (Figure 1). Establishing that a pixel was free of land coverage was first done by manually overlaying satellite images on Chesapeake Bay maps; all reflectance data was then checked again to ensure it was not flagged as a shoreline pixel according to MODIS datastate variables. Eliminating land contamination meant narrower reaches of rivers were excluded from the study area. For example, stations upstream of Washington D.C. on the Potomac River were generally disqualified, though those shortly downstream were usable. For the large Western Shore rivers (Potomac,

James, Rappahannock, York, Patuxent), at least 10 stations were utilized. Smaller rivers had around five stations, though for many, only one or two of those stations had been continuously sampled since 2000.

All stations utilized fell within the tidal portion – defined as the area below the fall line (the point where the rivers plunge from the piedmont to the coastal elevation) of various tributaries - of the Chesapeake and this is the area for which the results are applicable. About 20% of Chesapeake freshwater volume originates in the tidal portions of Bay tributaries, while the contributions of nutrients these flows carry are as high as one-third of the total (Zhang et al. 2015).

Other sources of ground data were also considered, most particularly TSS concentrations measured by USGS River-Input Monitoring (RIM) stations on the fall-line of the nine major Chesapeake tributaries, and data from the Susquehanna Sediment and Nutrient Assessment Program (SNAP). SNAP stations were located in the upstream portions of the Susquehanna, and all were disqualified due to land contamination (McGonigal 2010, Zhang et al. 2016). Of the USGS RIM stations, their position on the fall-line likewise meant they were too far upstream to accommodate the MODIS pixel width-wise, with the exception of the Susquehanna; but no strong RIM TSS-reflectance relationship with minimal error emerged in the Susquehanna so this station was ultimately not used. For the other 8 RIM stations, we attempted to find a relationship between TSS at the RIM site and reflectance at the mouth of the river.

With the exception of the Potomac, all rivers yielded nonexistent or excessively noisy relationships, so no

10 further information could be gleaned about CBP data-sparse rivers like the Appomattox, Pamunkey, and

Mattaponi.

Using data from 48 Chesapeake tributaries, this study carried out an analysis of reflectance-TSS relationships over all those tributaries, and then further examined rainfall-TSS relationships in the five

Western Shore rivers in which data were most plentiful (James, York, Rappahannock, Potomac,

Patuxent). The watersheds of these rivers lie across the Coastal Plains, Piedmont, and Valley and Ridge physiographical regions (Zhang et al. 2015).

Satellite data

We relied on MODIS surface reflectance from the NASA MOD09 series as our source for reflectance data. This corresponds to water-leaving radiance at the surface in ratio to down-welling solar energy at sea level (Doxaran et al. 2009). The product we used is already corrected via the 6S radiative transfer theory for atmospheric gases and aerosols upon delivery, and is also radiometrically calibrated and geolocated (Petus et al. 2010, Vermote et al. 2015). We assume that the MODIS surface reflectance data accurately replicate, within the bounds of spatial aggregation, reflectance measurements that otherwise would be made from a boat. We downloaded MODIS-Terra and MODIS-Aqua red wavelength

(620-720 nm) images at 250 meter pixels published on NASA’s LAADS DAAC website and pertaining to the period between September 2000 and December 2014, for which the image was not completely cloudy. Of these, 1142 images coincided with CBP fieldwork monitoring that occurred on the same day as one of the downloaded images, for at least one station that was not masked by clouds. The particular product we used was the MODIS-Terra surface 250-meter resolution reflectance data, although it is sometimes eschewed in favor of Aqua products, due to greater noisiness in Terra reflectance (Chen et al.

2007). We created reflectance-TSS relationships for both Terra- and Aqua-based MODIS surface reflectance based on the 301 images from each satellite with same-day data, and found that Aqua-based relationships were not consistently superior across all Chesapeake Bay channels; thus, we ultimately carried out our research using those MODIS-Terra surface reflectance images, as MODIS-Terra’s morning overpass is coincident to most CBP fieldwork collection times, yielding more temporal matches

11 between satellite and field data. Having been launched two years in advance of Aqua, MODIS-Terra also covered a crucial 2000-2002 period of intense CBP data collection. Finally, there is precedence in using

MODIS-Terra to study TSS in both the northern Gulf of Mexico and the Gironde Estuary (Miller &

McKee 2004, Doxaran et al. 2009). These coincident ‘station-days’ on which MODIS red light reflectance coincided with CBP measures of TSS formed the backbone of the analysis. Reflectance-TSS relationships were first tested for the entire Chesapeake Bay, and then individually for the Mainstem and thirty tributaries. We used a pixel window, in our case a 3x3 window centered over the CBP station, to capture average reflectance and to smooth out inaccurate reflectance data within that window (Bailey &

Werdell 2006, Ondrusek et al. 2012). The average reflectance of the pixel window was calculated only if at least 4 of the pixels passed the cloud and land-cover tests.

Previous papers studying light reflectance-sediment concentration relationships along mid-

Atlantic Coast estuaries restricted fieldwork TSS measurements to within “a few hours”, usually three or less, of the satellite sensor overpass (Bailey & Werdell 2006, Ondrusek et al. 2012). Although our data showed that restricting CBP data to within three hours of the MODIS Terra or Aqua overpass did not consistently reduce error in relationships for the rivers tested, we adopted a three-hour time gap nevertheless to conform with general practice.

Sun glints were detected, and then removed, by their discolorations on the satellite imagery (Kay et al. 2009, Steinmetz et al. 2011). On certain summer days, we found high sun glints afflicted most stations, especially those lying in the north-south orientation of the Mainstem and the upper branch of the

Potomac River. These images also had a high occurrence of water pixels erroneously flagged by MODIS state data as ‘salt pans’, or stretches of desert sand that also reflect bright light. Out of 126 summer images (May through September), 12 showed evidence of widespread sun glints and were removed.

Reflectances recorded under high aerosol conditions flagged in MODIS datastate variables were often outliers and TSS-reflectance models with and without high aerosol conditions were significantly different, leading us to mask out high aerosol conditions. High aerosols were not a common occurrence in the Bay

(fewer than 10 out of a total of 152 datapoints in the Rappahannock River, for example), but when present, aerosols scatter, absorb, and reflect light over a range of wavelengths (Tzortziou et al. 2007).

CDOM, chlorophyll a and other pigments, and dissolved and particulate nutrients were collected simultaneously with TSS data by the CBP, and none of these water constituents covaried with each other.

We tested whether by removing CBP TSS measurements for which simultaneous measurements of the other constituents were high outliers, or greater than 1.5 standard deviations from the mean, resulted in a significantly improved model fit. Precedence for this has been noted in previous research in that CDOM and other water constituents have independent light absorption and reflectance patterns, and the patterns are not always additive or simply related to TSS (Stumpf & Pennock 1989, Carder et al. 1989, Binding et al. 2005). Pigments like chlorophyll a and its degraded product, pheophytin, absorb light at the red wavelength, in direct opposition to the elevated reflectance of TSS at this wavelength (Stumpf & Pennock

1989). We tested multiple linear regression models between TSS and MODIS red reflectance to examine whether inclusion of CDOM, chlorophyll a, and other constituent concentrations was significant.

Other data collected simultaneously with TSS concentrations, such as the agency that collected fieldwork under the CBP auspices, station depth, weather conditions, tidal stage, and wind speed, were also tested to determine if inclusion of these factors led to significantly different TSS-reflectance models, but we ultimately found no evidence of this.

Storm and land cover data

To study storm effects on TSS, both wind and rain data were downloaded from NOAA’s Climate

Data Online website. Twelve rain gages were chosen that lie on the lower Western Shore (Figure 1).

Gages were chosen based on continuity of data over time, with preference given those gages that collected rain measurements for the entire study period (2000-2014) but supplemented by gages of a shorter duration as well (Table 1). The gages were located in urban areas (Norfolk and Washington D.C.) as well as in forested and farmed areas. CBP TSS-measuring stations were assigned to rain gages based on proximity, creating rain gage-TSS station areas. Out of 109 CBP stations in this area, over 70% were less than 40 km from their assigned rain gages, and only 7 were more than 60 km away. We

13 complemented the rain gages with the two USGS discharge gages in the Lower Western Shore with long data records. Additional wind data was downloaded from five NOAA “tides and currents” stations

(Figure 1). The gage-station land areas were defined as the hydrologic unit codes (HUC) in which the

TSS stations and gage were located, as well as the surrounding HUCs feeding in. Land cover was then determined by summing different land cover types based on the National Land Cover Dataset for 2011 over each gage-station land area (NLCD 2011; see Figure S1).

River Gage Long. Lat. Start of End of data Data James WILLIAMSBURG 3.2 W VA US -76.76 37.27 6/26/2006 3/2/2016 James HOPEWELL VA US -77.28 37.30 1/1/2000 4/8/2016 James NORFOLK NAS VA US -76.29 36.94 10/1/2000 4/8/2016 James SMITHFIELD 2.6 E VA US -76.57 36.98 5/19/2008 8/11/2013 York WILLIAMSBURG 2 N VA US -76.70 37.30 1/1/2000 4/5/2016 York WEST POINT 2 NW VA US -76.80 37.57 1/1/2000 4/2/2016 Rappahannock URBANNA 6.2 NNE VA US -76.52 37.72 6/27/2006 4/5/2016 Rappahannock WARSAW 2 NW VA US -76.78 37.99 1/1/2000 2/29/2016 Potomac QUANTICO MCAS VA US -77.31 38.50 1/1/2001 1/1/2016 Potomac MONTROSS 5.2 ESE VA US -76.73 38.08 5/9/2009 1/1/2016 Potomac ST INIGOES WEBSTER NAVAL -76.43 38.14 11/3/2006 1/1/2016 OUTLYING FIELD MD US Potomac WASHINGTON REAGAN -77.03 38.85 1/1/2000 1/1/2016 NATIONAL AIRPORT VA US Table 1. NOAA rain gages. Their location on lower Western Shore rivers, and the minimum and maximum dates of their data collection period. Previous satellite-based attempts to study Chesapeake storm after-effects on water quality rarely consider accuracy of the predictions, even when a model is used in circumstances outside the conditions for which it was created (Ondrusek et al. 2012, Son et al. 2012). For our reflectance-TSS relationships, we created prediction intervals that define the error bounds associated with TSS concentrations estimated via our models. These intervals take account of both error in the population mean and dispersion in the data, and allowed us to base our conclusions solely on those results where statistically significant changes in

TSS were detected. We emphasize, however, that in order to expand the set of significant datapoints, our prediction intervals were set at the 80% confidence level. This trade-off reflects some of the uncertainty inherent in the use of satellite reflectance data.

All fieldwork data, rain gage data, and wind data were organized and analyzed using scripts written in the R project for Statistical Computing and the PostgreSQL/postgis spatial database. Land cover data was quantified for each gage-station area through a combination of QGIS and R scripts.

Satellite reflectance data was retrieved from NASA geotiff files using Python scripts. All subsequent analyses were performed in R.

Section 3: Results and Discussion

For channels with viable red reflectance-TSS relationships, exponential models were a natural fit.

Two estimated parameters, k and b, are meant to approximate the true model parameters k and β in the form [TSS]= β*exp (k * ρ) + ε, where [TSS] is concentration of total suspended solids, ρ is MODIS red light reflectance, β and k are parameters calculated via regression, and ε is the error term. Previous work in the Chesapeake has characterized the reflectance-TSS concentration relationship either as a linear or a third order polynomial model (Tzortziou et al. 2007, Ondrusek et al. 2012). However, these studies had limited TSS ranges (up to 15 mg/L) compared to the data we retrieved from CBP databases, in which the

TSS range extends comfortably into the 60-70 mg/L range for many rivers. Our exponential models follows in the tradition of those used to describe reflectance-suspended material relationships in the

Gironde Estuary of France (Doxaran et al. 2003, Doxaran et al. 2009).

Relationships for separate channels

Strong red reflectance-TSS concentration relationships were mainly found in the lower Western

Shore rivers and the Mainstem itself (Figure 2a-f). Among these, the Mainstem, Rappahannock, and

Potomac had the highest correlations and smallest standard error for both parameters, perhaps owing to their large range of TSS values (Table 2). We took this one step further by computing the ratio between the standard errors and their parent parameters, and by summing both measures; again, the Mainstem and

Potomac performed the best. The York performed the worst on these measures out of major Western

Shore rivers, perhaps owing to its narrow channel and overall limited range of TSS readings. Moving to other tributaries of the Chesapeake, like the Chester, Choptank, and Little Choptank on the Eastern Shore

15 as well as small Western Shore rivers (Middle, Rhode, Severn), we found weaker relationships. This was not necessarily because the data were noisier, but due to fewer data points and a maximum TSS that did not exceed 20 or 30 mg/L. The Susquehanna River had a poorer TSS-reflectance relationship compared to the lower Western Shore rivers. This might be because the Susquehanna River is both narrower and shallower than its lower Western Shore counterparts, even though it provides more discharge to the

Mainstem than any other river, and even though the Susquehanna maximum TSS is close to double that for the lower Western Shore rivers. In addition, the Susquehanna differs from other channels because several reservoirs capture a good proportion of particulate matter before reaching the downstream CBP stations (Cerco 2016). Lower Eastern Shore bodies like Tangier Sound and Fishing Bay, where stations are > 19 meters in depth, had differing results: Fishing Bay had a generally clean relationship, truncated at about 40 mg/L, while the Tangier Sound exhibited no reflectance-TSS correlation at all. Depth seemed to play a role in shallow rivers: in the Nanticoke, where all stations lie in 4 meters of water or less, no relationship emerged (Figure 2g). Bottom backscatter can cause superfluous reflectance, particularly in shallow areas, and is likely playing a role here (Chen et al. 2007, Tzortziou et al. 2007, Cannizzaro et al.

2013). Previous papers have used satellite reflectances to make predictions on Eastern Shore water quality, extrapolating from models created for the mid-Mainstem, but the evidence here suggests that this procedure might be unreliable, particularly for the mouth of the Nanticoke and Tangier Sound (Ondrusek et al. 2012, Liu & Wang 2014).

16 a) b)

c) d)

17 e) f)

Figure 2. Satellite reflectance-TSS concentration relationships. Relationships between MODIS-Terra red light reflectance and TSS concentrations in the four major Western Shore rivers and the Mainstem, where exponential relationships emerge. The eastern Nanticoke River, where the deepest station lay in 4 meters of water, has no clear relationship. TSS concentrations are in mg/L. Filled large circles denote the fitted regression model, with filled small circles indicating prediction intervals with 80% confidence.

Most CBP-measured TSS levels are below 90 mg/L; we did not ignore TSS concentration data above this level, but think it is prudent not to make strong conclusions for data far in excess of this

(Stumpf & Pennock 1989). The CBP and MODIS data feeding these relationships have been collected over a period of 14 years, but we did not find relationship divergences for later years, or that any

18 particular year stood out as an outlier. This is evidence that both ground and satellite data sources have been calibrated and processed consistently over the time period.

The F statistic is a measure of the strength of the TSS-reflectance relationship given by mean squares of regression over mean squares of error. We looked for a correlation between the F statistic of each tributary model and the number of data points, average station depth, and maximum TSS (Figure

S2). Number of data points used in the model was strongly linked to a highly significant F statistic.

Station depth and maximum TSS measured both played a role in some individual rivers, but broader patterns across all rivers were less consistent.

Parameter estimates for the five Western Shore rivers, shown in Table 2, had k constants that ranged from 19.39 to 28.33, and a b coefficient range between 3.69 and 6.54. At first glance, therefore, the parameters are distinct from river to river. However, a test of significance (Paternoster et al. 1998) found this was not universally the case when taking account of the standard error. The Rappahannock is the most distinct river, with a statistically significant “k” compared to all other channels aside from the

Potomac. The James was also distinct from the Mainstem and the Potomac. Taking account of further differences in the “b” coefficient, we decided to proceed on the basis that each river merited its own model, the policy we adopted for the subsequent storm analysis. Developing a single model meant to encompass all water bodies will merely increase errors. As illustrated in Figure 3 for two locations on the

Rappahannock River, overlaid time-series of all MODIS-derived estimates and all CBP measurements in various years and at various stations followed similar trends.

Max Mean F crit b std b “t k std k “t Channel n r TSS Depth, m F stat value b error statistic” k error statistic” Mainstem 224 0.83 181 9.46 488.93 3.88 3.96 1.04 31.99 23.42 1.06 22.11 Rappa- hannock 81 0.85 106 12.57 211.01 3.96 4.24 1.11 13.85 28.33 1.95 14.53 York 78 0.71 57 13.2 76.24 3.97 6.54 1.14 13.96 21.02 2.41 8.73 Patuxent 140 0.64 75.5 9.46 95.27 3.91 4.33 1.1 15.36 22.32 2.29 9.76 James 101 0.74 58.18 12.57 117.74 3.94 5.57 1.13 14.38 19.39 1.79 10.85 Potomac 137 0.86 124.4 13.2 373.81 3.91 3.69 1.07 18.86 24.51 1.27 19.33 Table 2. Model parameters. Estimated model parameters for the four major Western Shore rivers and the Mainstem, and the significance (given by the t statistic) and standard error associated with each. The F statistic gives a measure of model suitability. All models pass close to (0,0); see Figure 2.

a) b)

Figure 3. Chesapeake TSS time series. All available MODIS-derived TSS estimates (solid line), and CBP TSS measurements (dots), for two different years at two different sites in the Rappahannock River, are here overlaid to provide an example of the extent to which each data source mirrors the other. Dashed lines denote the prediction intervals at 80% confidence.

Aside from depth, width, and the constraints of the available data, differences in channel relationships may be due to a divergence in sediment sources and composition. For example, although a river may discharge into the Mainstem, it does not hold necessarily that the two entities will share the same sediment content; in fact, not all Chesapeake Rivers routinely act as sediment sources to the

Mainstem. Sediments in the Potomac and other smaller Western Shore tributaries are believed to circulate solely in those rivers without escape to the Mainstem except during extremely high flows, in part due to the convergence at the limit of salt intrusion corresponding to the estuarine turbidity maximum which traps these particles (Schubel 1969). Particle flocs that reach the Mainstem during large storms can be

20 torn apart into their constituent sediments before their destination, changing the sediment composition and potentially affecting reflectance (Fugate & Friedrichs 2003). Mainstem sediment sources are in large part due to organic production or bank erosion (Schubel 1969). These sources exceed tributary-linked inflow of sediments, and several rivers are sediment sinks rather than sediment sources (Cerco et al. 2004). Each tributary has its own compound of land covers, and its own signature blend of forest, urban, agricultural and wastewater particles. The mixture of all these in the Mainstem will produce a spectrum of grain size, organic and mineral content, and color. Given these varying factors, different tributary channels might plausibly exhibit different TSS-reflectance relationships.

Error within channels

Having established channel differences, we then investigated whether individual stations within a single channel also have slight variations in reflectance-TSS relationships. Studies on the Irish Sea found that backscatter coefficients ranged widely depending on station (Binding et al. 2005). The Bay Mainstem had enough data per station to permit such an examination. Strikingly, the relationships are strongest in the Upper Bay (higher latitudes), and decline in strength towards the Lower Bay (Figure S3). It is a mirror contrast to relationships uncovered between MODIS-derived versus CBP-measured chlorophyll a concentrations for which estimates are best in the Lower Bay (Son & Wang 2012). These patterns may reflect the influence of freshwater versus seawater constituents in different parts of the Bay (from inorganic TSS in the Upper Bay to organic TSS/chlorophyll a in the lower Bay).

We analyzed the extent of such within-channel data dispersion for the major Western Shore rivers by removing 30 randomly-selected reflectance values from each river, not used in modeling the reflectance-TSS relationship. We applied the models to the segregated values, then graphed CBP TSS measurements opposite MODIS-derived estimates. The MODIS-derived estimates have a close to equal mix of over- and under-estimation, indicating that our exponential model choice has a useful predictive skill (Figure S4). Correlation coefficients between CBP and MODIS TSS estimates were above 0.75 for the major Western Shore rivers with the sole exception of the York, which is again the narrowest of these

21 rivers. These correlation coefficients are in range of those found in other estuaries (Table 3; Doxaran et al.

2009, Petus et al. 2010, Ondrusek et al. 2012).

Channel Correlation Percent of Mean percent Mean absolute Mean ratio coefficient error < 30% difference % difference Chesapeake Bay Mainstem 0.9141 43.33 5.217 46.21 1.052 Rappahannock 0.8811 63.33 9.866 36.86 1.099 James 0.77 43.33 12.91 56.19 1.129 York 0.4228 46.67 17.73 42.22 1.177 Potomac 0.8238 60 5.095 28.72 1.051 Patuxent 0.77 43.33 12.91 56.19 1.129 Table 3. Accuracy analysis. Comparison of CBP TSS concentration measurements and MODIS-derived estimates.

The mean ratio for each of the six main channels was quite close to 1, and mean absolute percent difference (defined as the absolute difference of MODIS-derived TSS estimates minus CBP TSS measurements, divided by the latter) was between 28-57% (Table 3). We also calculated the proportion of all “percent errors” which were less than 30%; our proportions are in harmony with similar figures calculated for the Adour River in France (Petus et al. 2010).

Error between modeled and measured TSS values is partly due to spatial dissonance between

CBP TSS measurements collected at a specific geographic coordinate, and MODIS light reflectance of a

250 meter pixel (Stumpf & Pennock 1989). Furthermore, other water constituents induce error. CDOM, chlorophyll a, and pheophytin were measured with varying frequency simultaneously with TSS by the

CBP, and it was possible to study how these constituents affected the TSS-reflectance relationship in the

Chesapeake Bay. CDOM originates on land and thus exists primarily in near-shore waters. CDOM absorbs highly in the blue wavelengths (Carder et al. 1989, Binding et al. 2005, Le et al. 2013b). In the

Chesapeake, the correlation between blue/ultra-violet absorption and red reflectance is very high; thus, presence or absence of CDOM should be indirectly visible in red reflectance (Tzortziou et al. 2007). We found the linear correlation between CBP CDOM measurements between 2000-2014, for all Chesapeake water bodies, and MODIS red reflectance was r = 0.60 based on all 112 coincident points (compared to r

= 0.50 for a linear regression with TSS concentrations, again for all water bodies). When regressing both

CDOM and TSS concentrations in concert as two independent variables against MODIS red reflectance

(reflectance = a*TSS + b*CDOM + ε), correlation improved to 0.69. Similar improvements did not occur when taking account of chlorophyll a or pheophytin. Thus, the primary driver of red light reflectance would seem to be a combination of TSS and CDOM. Based solely on MODIS reflectance data, it is not possible to distinguish between the two constituents, causing a major source of error, especially as

CDOM and TSS, based on 728 datapoints, do not covary in the Chesapeake (r = 0.18) nor in other coastal waters (Cannizzaro et al. 2013). This is likely due to different transport processes for sediments and dissolved substances (Gong & Shen 2010).

Chlorophyll a and pheophytin are both pigments that affect water color and reflectance, but often at a more complicated set of wavelengths rather than just the red wavelength used here for TSS (Le et al.

2013a). Previous work found that high levels of phytoplankton do increase absorption at the red wavelength (Stumpf & Pennock 1989). However, according to an F test of model comparison, removal of high chlorophyll a outliers (greater than 1.5 standard deviations from the mean) did not produce a model significantly different compared to the model including all data. High outliers of pheophytin were similarly ineffective at changing the relationship. As with CDOM, correlations between TSS and chlorophyll a (r = 0.124 based on 41,788 paired points) and pheophytin (r = 0.275 based on 41,151 points) were negligible.

Other limitations of this remote sensing model of water quality include that relationships are based, in this case, on surface CBP measures of TSS at 1 meter depth or less. There is a substantial difference between bottom TSS and surface TSS based on CBP data; bottom-water turbidity is higher due to the effects of sediment resuspension (Ward 1985, Olson et al. 2009), and perhaps better estimated through different sediment budget models (Cerco et al. 2004). Additionally, model results are also not transferrable to conditions under which MODIS datastate variables record high aerosols.

Storms in the Chesapeake

Cognizant of the errors in our reflectance-TSS models, we sought to determine if these models were nevertheless capable of capturing significant changes in TSS concentrations during storms. During the study period of 2000-2014, there were over 3516 rain events recorded separately by 12 NOAA rain gages in the lower Western Shore plus Washington D.C. (Figure 1) for which at least two MODIS- derived TSS estimates (one within a day of the storm) were available before the next storm hit. As a comparison, relying solely on CBP measures of TSS means only 51 rain events fulfilled these same criteria.

We began by plotting TSS estimates one day before and after the storm and found that even in situations with high rainfall, most pre- and post-TSS estimates lay on a 1:1 ratio, and only in rare cases were these estimates significantly different. The few cases in which pre- and post-storm TSS differed significantly occurred mainly during small rain events, strongly implying that it was not the rain itself behind the change (Figure 4). We also looked at high discharge data in lieu of rainfall (data procured from two USGS stations in Lower Western Shore study area with continuous records), and a similar result emerged. TSS estimates did not appear affected by high discharges, whether a single high flow or an average of several days’ worth of high flow, and this was true across all land cover areas (urban, agricultural, forested.)

24 a) b)

Figure 4. TSS estimates derived from MODIS. These estimates are derived a day before and after a storm strikes for the rain gage catchment in the graph title.

We then focused only on post-storm TSS estimates (Figure 5). Following a visual inspection, it was clear, first, that TSS estimates following small storms are quite variable. Second, post-storm TSS values do not increase with larger storms (the sole possible exception being urban Norfolk). In fact, in six of the watersheds, it appeared that more rainfall (greater than 125 millimeters) actually shifted the boxplot distribution of TSS concentrations to lower values compared to the TSS distribution following storms that dropped less than 50 millimeters of rain. In four of these six cases, the Kruskal-Wallis test showed the differences to be significant (at the 0.1 level) between the rainfall categories (Kruskal & Wallis 1952).

These stations include Williamsburg 2.0, West Point, Warsaw, and Quantico. The remaining two stations,

Smithfield and Hopewell, may have fallen short because their highest rain category has captured only a handful of datapoints. These observations suggest several possibilities. First, the variability following small storms hints that it is not the precipitation itself that explains TSS concentrations, but other factors.

Second, larger rainstorms may contribute more water than sediments to the Chesapeake, thus reducing

TSS concentrations by dilution. Though previous literature has been divided on this matter, our study suggests that declines in post-storm TSS are a common result in several Chesapeake sub-watersheds along the Western Shore. It is also interesting to note that except for Smithfield, the six watersheds

25 mentioned above feature high-forest, low-urban land covers (in contrast to urban Norfolk where more rain seems to produce higher TSS concentrations). We expected higher TSS values following large storms in agricultural areas due to erosion, especially during the post-harvest season when no standing crop holds soil in place (Stevenson et al. 1985, Gellis et al. 2009). However, we did not uncover such a pattern. We also add that our dataset comprises fewer 125+ millimeter-storms in proportion to the other categories

(number of events noted in Figure 5), so the data may not be fully representative. Large storms are rarer by nature and may be followed by lingering cloud cover, rendering Chesapeake light reflectance unobservable.

a) b)

c) d)

26 e) f)

g) h)

27 i) j)

k) l)

Figure 5. Box-plots of post-storm TSS estimates. These are categorized by rainfall amount, with a separate plot for all 12 gages. The time interval between the storm and the data is a day or less. Numbers indicate the total rainfall events counted for each category.

We also caution that TSS levels are usually highest along the shore, and the lowest in the middle of the channels (Ward & Twilley 1986). Most CBP stations suitable for this study are not directly by the shoreline, due to land contamination of the satellite pixel. Thus, we might expect that post-storm TSS levels at our CBP stations will not be exaggeratedly high. This also points to a further limitation of these

TSS-reflectance models, in that MODIS-derived estimates cannot give information about how near-shore areas respond to storms.

Our results indicate a very slight correspondence between land cover and rainfall effects on TSS concentrations. Previous research shows similar ambiguous land cover results. Landsat-based land cover changes studied for Maryland Eastern Shore including the Nanticoke and Pocomoke rivers did not find that a 20-year urban growth spurt that increased urban areas in excess of 120% resulted in increased nutrients, sediments, or chlorophyll a; rather, rainfall, wetland areas, and sewage plant discharges were the key drivers of sediment concentrations (Aighewi et al. 2013). Urban effects on water quality can be unreliable because urbanization can occur in small, haphazard suburbs, such that even in areas with 11-

20% impervious surfaces, it is not always possible to find a real ‘urban signature’ of street residue in river suspended sediments (Devereux et al. 2010). A study on the Rappahannock River, a mostly rural lower

Western Shore watershed, likewise found no strong correlation between population growth and nutrient concentrations, a dissonance attributed to land management and conservation efforts (Prasad et al. 2014).

Next, we examined if the rapidity with which TSS spikes decline back to normal levels depends on the peak TSS level (Figure 6). For this query, we did not restrict ourselves to storm events but rather to any event in which TSS was estimated at over 20 mg/L, and then fell significantly. We deemed TSS to have fallen significantly when the magnitude of the decrease was greater than the uncertainty bounds

(Bailey & Werdell 2006). This condition means that we take account of the satellite reflectance error, and can say with 80% confidence that we are only considering events in which there was a significant change in TSS. As a second condition, the TSS also had to drop below 15 mg/L, the generally accepted level at which submerged aquatic seagrasses can grow (Gurbisz & Kemp 2014) and so we could measure TSS decline to a common ending point. Finally, we removed all events for which we did not have at least 1 satellite image for every four days of the TSS stabilization period. At first glance, the results show that all

TSS peaks exceeding 75 mg/L decreased to 15 mg/L in 9 or fewer days while some smaller peaks lingered longer than this. The results seem to indicate that the greater the TSS peak, the greater the chance that it diminishes to below 15 mg/L quickly. However, we tested this same hypothesis under different

29 reflectance-TSS models (variations of the exponential model and a quadratic model) and each model changed the graph appearance. Thus, we hesitate to draw conclusions here, whereas our other storm/wind results were consistent between all models explored.

Figure 6. Peaks in the MODIS-derived TSS estimates. These are graphed against the number of days it took for TSS to fall significantly to below 15 mg/L.

Whether a storm was localized or widespread over several gages did not appear to be of great consequence to TSS concentrations. First, we noted that the local or widespread character of a storm did not appear to vary overmuch with the mean rainfall of the storm, except that more local storms were more likely to have high rainfall outliers (Figure S5). Second, re-drawing the boxplots in Figure 5 such that the categories were “number of gages with same-day rain” rather than rainfall amount hinted that median post-storm TSS levels were fairly even regardless of category.

Given the somewhat unremarkable influence of rain on MODIS-derived TSS estimates, we examined NOAA Climate Data Center wind data collected at four gages (Norfolk, Montross, St. Inigoes, and Reagan National). These gages include the more heavily urban watersheds. We did not find evidence that wind data (in 2 minute intervals or as average daily speeds) influence post-storm TSS levels, even though wind-driven bottom resuspension is a major contributor to TSS concentrations, particularly in shallow waters. Such resuspension will, however, mainly affect bottom waters, leaving unaffected the

30 surface waters emitting MODIS-measured reflectance (Zheng et al. 2015). We tried accounting for this by limiting stations to the 23 shallow CBP stations positioned where the Bay is 3 meters deep or less, but the best TSS-daily wind speed correlations were 0.10 and 0.12 at Montross and Reagan National. We then used hourly wind data collected at five NOAA “tides and currents” stations (Figure 1) and tested correlations between TSS levels at shallow CBP stations and 4-hour and 8-hour wind averages preceding the Terra overpass. We matched the NOAA wind station with the closest shallow stations. Most correlations did not surpass 0.1, and many were essentially at zero. The best correlation we found involved Washington D.C. 4-hour average wind data. If we considered only wind averages above the third quartile (to represent the strongest wind events), the correlation with MODIS-estimated TSS at shallow CBP stations near Quantico was 0.27. If we considered the strongest 8-hour wind speed averages, the correlation rose to 0.35. These results remained fairly constant even when derived with different models. The generally low wind-TSS correlations we uncovered mirror a previous study where the best correlation value between satellite-estimated TSS plumes and 8-hour average wind speeds was 0.45; it may be that daily satellite images are poor at exhibiting instantaneous wind effects (Zheng et al. 2015).

Ward (1985) reported that winds must surpass a velocity of 25 km/hr in order to resuspend bottom sediments significantly, and areas deeper than 2 meters required even stronger winds. The maximum daily wind speeds in our dataset are 22 km/hr.

TSS maxima driven by neither rain nor wind can include the estuarine turbidity maximum

(ETM), which occurs mid-river where saltwater and freshwater currents collide. ETMs are evidence of an increase in bottom shear stress and erosion, and the resulting TSS concentrations can thereafter by trapped by gravitational circulation (Schubel 1972, Stevenson et al. 1985, North et al. 2004). The York River, at the confluence of its Mattaponi and Pamunkey tributaries, consistently showed a slight TSS increase as one moves downstream and then a TSS decrease closer to the mouth; this may be evidence of the ETM zone (Figure 7a, b).

We returned once more to land cover and mapped TSS at various CBP stations following heavy rainfall (Figure 7). The maps included here are representative of TSS patterns at these stations following

31 every major storm. The maps show that as one approaches the mouth of the river, TSS levels fall, as is expected (York and Rappahannock rivers). The difference between upstream and downstream portions of the river can approach 20-30 mg/L. We also see that storms by the nation’s capital cause far higher TSS levels right in the city area compared to TSS levels in the more rural area at the head of the Patuxent

River (Figure 7d). This effect again can reflect increased streambank erosion driven by more impervious area (Pizzuto et al. 2000), as well as solids washed off urban streets, and was seen in all Potomac storms.

Land cover effects are crucial at the moment as the Chesapeake Bay is experiencing population growth and predicted increases in urban lands (Jantz et al. 2005, Ondrusek et al. 2012). The closing 16 years of the last century saw an increase of nearly 300 km2 of impervious surface in the Washington D.C.-

Baltimore corridor alone (Sexton et al. 2013).

32 a) b)

c) d)

Figure 7. MODIS-derived TSS estimates following high rainfall. Axis units give UTM coordinates. Note that TSS color scales differ from map to map. Note that TSS color scales are standardized for Figures 7a-7c, but differ for 7d.

In closing our discussion, we note that our results suggest that particular tributaries from a single region may carry particles that interact differently with natural light. This is a surprising result, one that should be tested in other field settings and by performing appropriate laboratory experiments. We also note that our work was possible only because the Chesapeake Bay is located in a densely populated

33 region of a wealthy country, where government was able and willing to support regular field monitoring. We encourage local managers and other researchers to test our model relationships against ground data going into the future, to use the relationships to compare TSS concentrations in the vicinity of major Chesapeake cities, and to expand the maps from isolated pixels to modeling all river pixel reflectance data as TSS measures. With continued growth of population and migration into urban and coastal environments, we expect growing TSS-contamination of coastal waters worldwide. Wherever feasible, local governments should be encouraged to support monitoring that will enable the productive use of satellite imagery in managing coastal environments. In ideal circumstances, the remote-sensing and field-monitoring communities would plan and execute joint investigations that could be designed to answer some of the questions raised by our work.

Section 4: Conclusions

Although they are imprecise, MODIS-derived estimates of TSS concentrations can be used to study water quality changes following storms within specified bounds of confidence. They provide a greatly increased coverage compared to existing data, particularly data that must be collected in a narrow window of time following storms. We recommend the use of predictions bounds to properly account for error. In other cases, we used an aggregate of storm data, and finding the same trend for every event also improves confidence. According to our literature review, this is the first time that the reflectance-TSS concentration relationship for a water body was constructed separately for the Chesapeake Mainstem and individual rivers. The results give evidence that such decentralized construction is necessary, and relying on a single relationship for the entire estuary will result in greater errors. In particular, our results indicate a lack of relationship in some Eastern Shore areas like the Nanticoke River and Tangier Sound to which

Mainstem models are perhaps best not applied. MODIS-derived TSS estimates indicate that rainfall is not a primary factor in causing TSS spikes in the Bay, and that higher rainfall levels lead to an overall slight decrease in TSS concentrations, presumably due to dilution or faster sediment velocity traveling out of the river. The evidence that urban watersheds lead to greater TSS levels following storms was apparent in

34 some analyses, but not in all. It especially appears to be a localized phenomenon that dissipates fairly quickly. We have presented evidence of remote sensing in an estuary as a useful tool for creating large datasets of TSS response following any estuarine disturbance, and as a useful way to look at land cover effects.

CHAPTER 2: HOW WAR, DROUGHT, AND DAM MANAGEMENT IMPACT WATER SUPPLY IN THE TIGRIS AND EUPHRATES RIVERS

The Tigris and Euphrates Rivers in the Middle East are the longest in Western Asia. In Arabic, the Euphrates is called Furaat and means ‘sweet water’, while the Tigris is called Dijla. These rivers have many important differences compared to the Chesapeake Bay. They lie in a region of the world with comparatively less precipitation, as well as less water quality ground data. While the CBP has established a large database with water quality measurements and other parameters, there is nothing similar in the

Tigris and Euphrates – and I was unable to access the data that exists. This chapter provides a contrast as to what sort of information one can extract about a location when the satellite data, and not the ground data, is the primary source of information.

The co-authors for this paper are Mejs Hasan, Aaron Moody, Larry Benninger, and Heloise

Hedlund. The paper is currently under review at Ambio.

Section 1: Introduction

The yearly supply of water that a river delivers is a result of many factors including climate, watershed, and water management. Engineering projects for flood control and hydropower alter ecology, flow, and sediment deposits. Water withdrawals for domestic, farming, and industrial use have consequences for water supply and quality (Al-Layla & Al-Rawi 1988; Vörösmarty et al. 2010).

One human activity of potentially great consequence for river systems is armed conflict. Damage to water systems was documented during conflicts in Kosovo, Afghanistan, and other places (UNEP

1999; UNEP 2003). Water infrastructure like dams can be directly targeted by airstrikes or battles

(MacQuarrie 2004). Displacement, explosions, and movement of heavy equipment increase dust that then settles on rivers and accumulates in reservoirs (Sissakian et al. 2013; Moridnejad et al. 2015). Conversely,

36 conflict sometimes leads to better water quality. For example, reservoir salinity can improve if agriculture is disrupted and reduces irrigation return flows (UN-ESCWA & BGR 2013; Eklund et al. 2016).

Abandoned farms can free water otherwise consumed by crops, increasing river flow (Müller et al. 2016).

Satellite data can shed light on such impacts. Satellite sensors capture optical signatures of water that stand in contrast to desert sand, vegetation, and cities, thereby delineating the water surface area (Gao

1996; McFeeters 1996). Different approaches have been tested for delineating inland water bodies, often with high accuracy (over 90%) both regionally and globally (Klein et al. 2014; Klein et al. 2017; Pekel et al. 2016). Depending on the satellite used, the resulting estimates of water surface area can have high spatial resolution, high temporal resolution, or a compromise between the two.

In this paper, we use satellite images to explore the effects of conflict and other factors on the

Tigris and Euphrates Rivers, which flow through Turkey, Syria, and Iraq. The Tigris and Euphrates allow countries like Iraq, comprised largely of desert, to enjoy abundant water resources compared to some neighboring countries (MacQuarrie 2004; UN-ESCWA & BGR 2013). Both rivers are fed by snowmelt from Turkey’s Taurus Mountains and Armenian highlands. The Tigris is also fed by the Zagros

Mountains in Iran. The normal snowmelt occurs between April and June (MacQuarrie 2004). Dams along the Tigris and Euphrates collect snowmelt in reservoirs to prevent flooding, generate hydropower, and supply irrigation canals (Table 4; Altinbilek 2004). During droughts, the water supply in these reservoirs can fall sharply (Cockburn 2009).

River Dam country Year Capacity completed (km3) Tigris Mosul Iraq 1985 11.1 Kralkizi Turkey 1997 1.92 Batman Turkey 1998 1.18 Dicle Turkey 1997 0.60 Devegecidi Turkey 1972 1.18 Euphrates Haditha Iraq 1984 8.2 Tishrine Syria 1999 1.9 Tabqa Syria 1975 11.7 Karkamis Turkey 1999 0.16 Birecik Turkey 2000 1.22 Ataturk Turkey 1992 48.70 Karakaya Turkey 1987 9.58 Keban Turkey 1975 31 Other Euphrates water diverters/storage areas Ramadi Iraq 1955 -- Barrage Lake Iraq 1948 3.3 Habbaniya Lake Razaza Iraq 1951 26 Lake Iraq 1954 72.8 Therthar Falluja Iraq 1985 -- Barrage Table 4. Dams and barrages along the Tigris and Euphrates. Dams ignored, due to small size, include Baath on the Euphrates (storage capacity of 0.09 km3) and Goksu (0.06 km3) on the Tigris. Capacities and years from Altinbilek 2004. The recent history of armed conflict around the Tigris and Euphrates Rivers includes the 1980-88

Iran-Iraq War; the 1990-91 Gulf War; the 2003-11 progression of violence marked by the United States invasion and occupation of Iraq, and sectarian fighting; sundry episodes of isolated airstrikes; and ongoing civil wars and militant insurgencies since around 2012. Obtaining complete information from a stressed war zone can be difficult. This paper applies a formal analysis of satellite data to approximate water supply and compare the different ways that natural, engineering, and conflict stressors affect the

Tigris and Euphrates. We focus on locations that have borne at least one direct episode of conflict: Mosul

Dam on the Tigris River, and Haditha Dam, Falluja Barrage, and Ramadi Barrage, all on the Euphrates

River. We evaluate how our analysis of changes in water supply compares to government and news

38 reports of conflict-driven change, as well as documentation of drought. We did not test these conflict- water relationships in a quantitative or statistical sense, or broaden the view to examine cooperation-water relationships. But this study may set the stage for a more statistically rigorous analysis of water-climate- cooperation events as possible future research.

Section 2: Materials and Methods

Study area

Mosul and Haditha Dams were built in 1985 and 1986 to provide hydropower and irrigation

(Table 4; Figure 8a; Altinbilek 2004). Approximately 140 km southeast from Haditha Dam stands a series of gates and locks (Ramadi Barrage and Falluja Barrage) through which authorities can manipulate how much water continues down the Euphrates, and how much is instead diverted to peripheral water routes

(Kassim et al. 2006; UN-ESCWA & BGR 2013). One of these routes, the Warar Canal, extends south from the Ramadi Barrage, and channels excess spring floods into lakes devoted to irrigation, recreation, and wildlife (Figure 8c). The first of these lakes, Lake Habbaniya, has two outflow canals: one diverts water back to the Euphrates, and the other carries water south into Lake Razaza (Kassim et al. 2006). A third lake, Therthar, lies between the Tigris and Euphrates and is connected to both by canals; like

Habbaniya and Razaza, this lake is also meant for flood control.

39 a.

c. Figure 8. Map of the Tigris/Euphrates study area. Map a shows the Tigris and Euphrates rivers as they wend their way through Iraq to the Persian Gulf, and the major dams, with an inset map showing the location on a world map. The satellite imagery is a most-recent-pixel-value composite of Landsat images between January 1 and September 15, 2014, timed so that the most recent pixels were cloud-free. The Tigris and Euphrates vector layers were obtained from a river and lake shapefile produced by Natural Earth. The watersheds were determined through maps (UN-ESCWA and BGR 2013) and shapefiles from the World Resource Institute. Map b shows a close-up of Mosul Dam, and the section of the Tigris River used to approximate discharge (burgundy polygon). Map c shows a close-up of the lakes, barrages, and canals between Ramadi and Falluja in the Anbar Province. These Landsat images were retrieved from Google Earth Engine.

Upstream of Mosul Dam, the Tigris is relatively undeveloped, with five smaller Turkish dams established. The flow of the Euphrates, on the other hand, is restricted by three Syrian dams and five

Turkish dams upstream of Haditha Dam in Iraq (Figure 8a). Most of the largest dams were inaugurated during the 1970s-80s. The turn of the millennium ushered in a period of small-sized dam building in

Turkey and Syria.

The Turkish headwaters of the rivers develop in areas that receive upwards of 800-1000 mm of precipitation per year (MacQuarrie 2004; UN-ESCWA & BGR 2013). Further south, four local rain gages

41 around the Tigris River near Mosul Dam recorded average rainfall of 325 mm/year between 1990 and

2009 (Zakaria et al. 2012). Southern Syria and Iraq receive about 75-150 mm/year (MacQuarrie 2004).

Precipitation along the Tigris and Euphrates has declined recently, to the point that some researchers consider the region under a prolonged drought since 1999 (UN-ESCWA & BGR 2013). The two most serious periods of low precipitation occurred during the hydrological years (October to September) of

1998-2000 and 2007-2009 (Trigo et al. 2010), with another period occurring 2013-14 (Figure S6c).

Among conflicts that have directly affected the stability of the Tigris/Euphrates water supply

(Table S1), a notable example is the militant capture of Mosul Dam in August 2014. The capture was accompanied by a threat to destroy the dam and flood downstream Baghdad, or the possibility of the dam’s accidental failure during airstrikes (BBC 2014). In fact, a visual inspection of the head of Mosul

Dam Lake pre- and post-battle shows dramatic changes in lake-surface area (Figure 9). However, quick visual inspections may be misleading. We therefore applied a formal analysis.

42 a) b)

c) d) Figure 9. The head of Mosul Dam illustrated in Landsat 8 imagery. This is shown on a) June 9, b) July 11, c) August 12, and d) September 13, 2014. Rapid changes in the water-surface area of the reservoir are evident. Militants captured the dam on August 8, and relinquished control on August 16. Data

The data we needed to complete this analysis included information on the amount of water in the reservoirs studied; discharge information along the Tigris and Euphrates, for insights into water management practices; precipitation data; and information on when conflicts and droughts occurred.

We used Landsat calibrated, ortho-rectified, top-of-the-atmosphere (TOA) reflectance images to measure water-surface area of rivers and reservoirs, because this was our indicator of water supply and how it changed in response to war, drought, or management (Table S2, Figure S7 for row/path information, image distribution over time). Landsat imagery dates from 1984, is set at 30x30 m2 pixel resolution, and involves 4 different U.S. satellite sensors (Thematic Mapper from Landsats 4 & 5,

Enhanced Thematic Mapper+ (ETM+) from Landsat 7, and Operational Land Imager (OLI) from Landsat

8). These sensors and their satellite orbits have been designed to provide continuity in the data products so

43 that imagery over the entire data archive can be readily combined and compared over time (e.g. Hui et al.

2008).

We validated our results using 10x10 m2 resolution multi-spectral Sentinel-2 images over Mosul and Haditha reservoirs, because of their finer spatial resolution compared to Landsat images. These

Sentinel-2 images come from eleven different months between 2015 and 2016. For each reservoir, we found 10 cloud-free Sentinel-2 images captured within two days of a Landsat image, or 20 images in all.

We used satellite-based altimeters to fill gaps in the Landsat record. Mosul reservoir has altimetry data collected by the Ocean Surface Topography Mission since 2008 (NASA 2008; Birkett & Beckley

2010). Haditha reservoir altimetry dating from 2002 was collected by TOPEX/Poseidon and ENVISAT

(Cretaux et al. 2011).

We used established studies to identify time of major droughts based on meteorological, hydrological, and agricultural evidence (Trigo et al. 2010), as well as estimates of rainfall from the

PERSIANN Climate Data Record. The PERSIANN record dates from 1982 and provides daily precipitation based on a combination of satellite data, modeled outputs, and rain gage data (Ashouri et al.

2015). We wanted this data to assess how drought affected reservoir water quantity.

We had access to archived monthly discharge records from several Iraqi river stations, though no data more recent than 2005 (Saleh 2010). This data granted insight into water management practices, as well as allowing us to extrapolate to more recent estimates of discharge.

Our approach focused on the surface area of various water bodies, rather than the total volume.

Most of the reservoirs were filled before satellite-based topographic information became globally available, and we also lacked cross-sections of the rivers, information essential for deriving volume. We did have access to 1983 and 2011 depth-height-surface area data from Mosul reservoir (Issa 2015).

The repeat period for Landsat data is sixteen days, and gaps in the data record due to clouds can extend for month-long stretches. Thus, floods or quick changes that occurred entirely within a repeat period cannot be captured (Yamakazi et al. 2015; Klein et al. 2017). We used all possible images, but

44 accept that a conspiracy of sensor absence or clouds would thwart complete data on quickly-occurring extreme events.

We used news articles, reports, and books depicting events back to 1990 to build our timeline of various conflict flashpoints.

Methods

We assembled a Landsat database in Google Earth Engine for each day on which high-quality images captured the entirety of each reservoir of interest between 1984 and 2016 (Table S2). Using the cloud.score functionality in Google Earth Engine, we masked out all pixels exceeding a cloud score of

0.25 (¼ of the pixel is cloud contaminated) and used images for which less than 1% of pixels were cloud- masked out.

We measured water-surface area by applying the normalized difference water index (NDWI)

(McFeeters 1996). NDWI is a normalized difference of the green and near infrared Landsat bands ([near- infrared – green]/[near-infrared + green]). Image pixels scoring above a threshold were classified as water

(Klein et al. 2014). A threshold was determined by first creating two polygons for each reservoir, one entirely over land and one over water. Average NDWI, and the standard deviation, was calculated for the water polygon and the land polygon. The land and water distributions were separated by a gap whose magnitude varied from image to image; we placed the land-water threshold within the gap (Figure S8).

Water in rivers and along shorelines, being shallower than reservoirs and susceptible to containing bottom reflectance, have NDWI values on the lower tail of the water distribution. Thus we adjusted the threshold for every image to include as much of the lower tail as possible without trespassing onto the land distribution. We applied the following formula to achieve this goal:

[푥̅ − (푥̅ +푛∗ 푆 )] 푇 = 푤 퐿 퐿 [Equation 1] 푆푤 −1 where xw, xL, Sw, and SL represent the mean values and standard deviations of the land and water distributions, T is the number of standard deviations below the water mean where the threshold is placed,

45 and n is an adjustable value that can be set to achieve the best visual match. The value of n changed depending on the brightness of the image, and ranged from 2 to 11.

We converted the resulting number of water pixels into square kilometers, which represented the lake-surface area on the given day of each image. Using this information, the rate of change per day was calculated between successive images (except those separated by years-long gaps). The number of datapoints per reservoir varied based on the number of cloud-free images available. For example, Mosul reservoir had 22 non-consecutive years with data, sixteen of which had at least five data-points, while

Haditha reservoir had 27 years with data, twenty of which had at least five data-points. We defined years- long gaps as being at least 1.5 years. The frequency of these gaps also varied by reservoir; Mosul had five, Haditha had one. With two exceptions, all remaining data intervals for Mosul and Haditha were less than one year. For intervals that lasted several months, we accepted that the rate of change in lake-surface area gives less specific information. We removed contaminated data, such as a July 2009 sandstorm that obscured Haditha reservoir. We converted the 1983 and 2011 Mosul reservoir depth-height-surface area data into regression equations and thus were able to estimate total water volume from that reservoir’s surface-area input.

We performed a correlation analysis between altimeter water level data and Landsat-based reservoir-surface area for each of Mosul and Haditha reservoirs. We then used linear regressions to estimate surface area from water level data when Landsat images were not available.

When validating our data, for the twenty Landsat images paired with a Sentinel-2 image, we randomly generated 22 geographical coordinates to use as our validation points. This meant a total of 440 validation points, which allowed for at least 50 validation points in each of our two land classifications

(water or not-water; Congalton and Green 2008). Twenty-four points were discarded because the corresponding Sentinel image did not extend to that particular coordinate. The remaining 416 validation points were randomly distributed over the reservoirs, desert, and shorelines (Figure S9). From these points, we calculated our accuracy in correctly classifying all real-world water as water on the satellite images, and our accuracy in not misclassifying real-world land as water on the satellite images (Table 5).

We also include the kappa coefficient which adjusts accuracy estimates to correct for the random chance of assigning the correct labels (water or land).

Sentinel Reference Data Landsat Water Land Total User’s classification accuracy Water 50 0 50 100% Land 1 365 366 99.73% Total 51 365 416 Producer’s 98.04% 100% Total: 99.75% accuracy Table 5. Tigris/Euphrates error matrix and accuracy results.

To get general discharge trends after the official records end in 2005, we substituted river-surface area. River-surface area is often a suitable indicator of discharge so we tested whether this was true on the

Tigris and Euphrates (Bjerklie et al. 2003). We compared the historically recorded discharges to same- month river-area derived from Landsat images at selected river reaches. We selected the river reaches by picking a section right next to the dam or barrage whose discharge we were trying to recreate. Each stretch of river was also chosen such that no canal or tributary joined or separated from the river in the middle of the stretch to keep the sourcing of the entire stretch consistent. We tested four sections of river downstream of Mosul dam, and four sections upstream, downstream, and in between Haditha Dam,

Ramadi Barrage, and Falluja Barrage.

Section 3: Results

Extreme events

Both Mosul and Haditha reservoirs, over the entire period of record, have a surface area generally ranging between 300 and 400 km2 (Figure 10). Haditha mean lake-area was 302 km2. Mosul mean lake- area before 2004 was 351 km2 (9.8 km3, based on 1983 Mosul bathymetry), while post-2004, it declined to

313 km2 (around 7.2 km3, based on 2011 Mosul bathymetry) (Figure 10b).

47 a)

b) Figure 10. Surface area of Haditha (a) and Mosul (b) reservoirs. Haditha estimates are based on Landsat and altimetry data; Mosul only on Landsat. Lines of significance are two standard deviations from the mean. Background colors indicate: red = conflict; gray = upstream dam building; tan = droughts. Close-ups of significant changes in water supply are provided in Figure S11.

We identified the lake-area values that were beyond two standard deviations from mean reservoir area. Lake-surface area fell to levels below this range twice for Haditha reservoir (multi-month stretches in 2009 and 2015), and thrice for Mosul (February 1991, March 2011, December 2015) (Figure 10; Table

6). Lake-surface area never exceeded the upper bound of this range for either reservoir. Likewise, we

48 categorized lake-area rates of change as substantial if they occurred beyond three standard deviations from mean reservoir change rates (Figure 11). Three deviations were chosen to eliminate high rates occurring repeatedly between May and July, which we assumed were caused by spring floods. Five occasions fit this standard: at Mosul reservoir in August 1990, February 1991, and March-April 2011, and at Haditha reservoir in February 1991 and June 2014 (Table 6).

Substantially high or low lake-areas Substantial rates of change in lake-area

Dam Year Most likely cause Dam Year Most likely cause

Haditha 2009 Drought Haditha February 1991 Conflict Haditha 2015 Drought Haditha June 2014 Conflict Mosul Feb 1991 Conflict Mosul August 1990 Conflict Mosul March 2011 Averting dam failure Mosul February 1991 Conflict

Mosul December 2015 Averting dam failure Mosul Spring 2011 Unknown

Table 6. Summary of substantial changes in water supply along the Tigris and Euphrates.

50 a)

b) Figure 11. Rates of change in reservoir lake-area of Haditha (a) and Mosul (b) reservoirs. Lines of significance are three standard deviations from the mean. Background colors indicate: red = conflict; gray = upstream dam building; tan = droughts. Close-ups of significant rates are provided in Figure S12.

Classification error

We calculated user’s and producer’s accuracy ranging from 98-100% for all classes (Table 5), and a kappa coefficient of 98.9%. We attribute the high accuracy to our formula that set the land-water threshold individually for each image.

Altimeters for gaps

The altimeter-Landsat correlation for Mosul reservoir was r2 = 0.90; for Haditha, it was r2 = 0.99

(Figure S10). We decided to therefore only use the altimetry data for Haditha reservoir, because we could

51 depend that the estimates produced from it would be nearly identical to estimates we would have derived from Landsat images, and the transitions between the two sets of estimates would not be noticeable. We used linear regression to convert height estimates during Landsat gap years into lake-surface area estimates.

River surface area for discharge

Some river stretches correlated to discharge better than others. On the Euphrates, the best correlations were found on the stretch of river just upstream of Ramadi Barrage (correlation of 0.73 with

Husayba station) and a stretch just downstream of Falluja Barrage (correlation of 0.64 with Hindiya station) (Figure 8c). The two sections in between Ramadi and Falluja Barrages did not perform well, likely because the water flow is too regulated between the two barrages. With Mosul discharge station, the most successful section of river had a correlation of 0.79. This stretch of river was located about 70 km downstream. This is likely because the Mosul discharge station (located right at Mosul dam) is actually upstream of a second regulatory dam (Figure 8b; Adamo & Al-Ansari 2016). There may exist extra regulatory measures on the river banks that prevent flooding in Mosul city. This extra layer of human management could be why the river extents we tested within the area of the regulatory dam and the city of Mosul itself did not correlate well with river discharge. We used river-surface area in the designated areas to look at general trends in discharge, rather than for generating specific estimates.

Droughts

Haditha reservoir underwent a two-year decline in surface area of 72% between May 2007 and

October 2009 (Figure S11d). This occurred during a major drought, and as the upstream Keban, Ataturk, and Tabqa reservoirs declined between 4-11%. During the 2013-2014 drought, those same three upstream reservoirs declined by 5-13%, but supplies were replenished during the following year until they were at

95-111% of their original pre-drought levels. This upstream replenishment coincided with a 62% drop in surface area at Haditha reservoir (Figure S11e). We did not have adequate data over the farthest

52 downstream lake on the Tigris, Lake Therthar, but other studies have also shown a substantial drop in its water height during drought (Trigo et al. 2010).

Conflicts

The First Gulf War, occurring between August 1990 and February 1991, saw rapid changes in lake-surface area. Between August 18-26, 1990, the Mosul reservoir lost an average of 3.3 km2 of surface area per day, in total falling from 372 km2 to 346 km2 (Figure S11a; Figure S12a; Figure S13a,b). In volume terms based on 1983 Mosul bathymetry, this represented a decline from 10.8 km3 to 9.6 km3.

During the same span of days, the downstream Lake Therthar increased at a rate of 2.73 km2/day. Over the next five months, Mosul reservoir continued losing surface area, at an average rate of 0.5 km2/day.

Then, came another plunge: between January 25 and February 10, 1991, the reservoir lost about 3.4 km2/day of lake-surface, for a final surface area of 215 km2, and a volume of 3.3 km3. At the same time

(January 17 – February 10, 1991), Haditha reservoir lost an average of 2.5 km2 of lake-surface per day, a loss of 21% in three weeks (Figure S12c; Figure S13c,d).

Between June 25 and July 11, 2014, a time consumed by both militant battles and drought,

Haditha reservoir lake-surface area declined at a rate of 2.0 km2/day, a substantial rate of loss only otherwise exceeded by the drainage of Haditha Dam in February, 1991 (Figure S12d). Around the same time, flooding was clearly visible between the towns of Falluja and Abu Ghraib. The land area affected had only 3.8 km2 water surface on February 26, but increased to 92.6 km2 by the next available non- cloudy data point, May 17 (Figure S14). By June 2, the wet area receded to 37.0 km2, then 11.9 km2 by

June 18, and remained below that point for the rest of the year.

Other instances of significance

Mosul reservoir experienced an increase of 4.0 km2/day in surface area between March 29 and

April 14, 2011, and then of 2.2 km2/day between April 14 and May 16, 2011 (Figure S12b). Mosul reservoir also reached extreme low levels in March 2011 and December 2015, at a time when wet-season management of the dam was changing to avoid its failure. (Figure S11b,c).

Relations to other reservoirs

By examining scatterplots of the major upstream reservoirs versus downstream Mosul and

Haditha reservoirs (Figure 12; Table 7), we found that same-day surface areas were typically positively correlated within individual years. These high correlations (r ≥ 0.85) suggest that within a single year, regional climate conditions drove water availability, and these conditions worked evenly over all the reservoirs: if one reservoir increased, the same driving factor caused an increase in the other reservoirs.

There are a few exceptions with poor or negative correlations (2002; 2011 and 2015 between Haditha and all upstream reservoirs, and especially so with Syrian reservoirs). From year to year, the slopes of the

Mosul-Turkish reservoir comparisons were remarkably similar, although after 2002, they became flatter, indicating that upstream dams started filling quicker relative to Mosul Dam (Figure 12a). Yearly

Euphrates correlations were more prone to high scatter.

Haditha + Haditha + Syrian + Haditha + all Mosul + Turkish Syrian Turkish dams, upstream Turkish upstream dams upstream dams Euphrates dams upstream dams year r n r n r n r n r n 2000 0.991 3 0.999 4 2001 0.948 4 1 1 2002 -0.592 5 0.750 5 -0.734 4 0.902 4 0.992 3 2006 1.00 3 0.898 4 0.993 3 1.00 3 0.968 4 2007 0.957 6 0.929 5 0.978 5 0.925 5 0.995 3 2009 0.974 4 2 2 0.961 6 2011 -0.503 3 -0.588 3 0.993 3 -0.517 3 0.564 6 2013 0.913 8 -0.408 8 -0.326 8 0.877 8 0.775 8 2014 0.892 5 0.122 10 -0.281 5 0.943 5 0.714 5 2015 0.162 6 -0.794 7 -0.635 6 -0.648 6 2016 0.976 5 0.176 7 0.324 4 0.954 4 0.972 6 Table 7. Correlations between dam lake sizes. Correlations after the Syrian Civil War began are outlined in red. R refers to the correlation; n refers to the number of dates for each year on which usable images were available for all reservoirs.

55 a) b)

c) Figure 12. Scatterplots of reservoir size. These scatterplots compare upstream reservoir sums versus Mosul (on the Tigris; a) and Haditha (on the Euphrates). We separated the Haditha graph by the Syrian (b) and Turkish (c) upstream dams. Only years with at least 3 points are shown. To maximize data-points, we included only the largest reservoirs on the Euphrates (Keban, Karakaya, Ataturk, and Tabqa). Using all Euphrates reservoirs resulted in identical plots, but with fewer points. Section 4: Discussion

Seasonality

In general, both Mosul and Haditha reservoirs follow a consistent pattern of spring flooding- induced peaks, followed by contraction over the summer and fall. The pattern is steadier for Mosul reservoir than Haditha, likely because there are fewer and smaller dams upstream from Mosul. Mosul reservoir reaches peak surface area most commonly in June, and occasionally as late as July or as early as

May, after snow in the uplands melts. These peaks diminish over the summer (due either to downstream releases, evaporation, or irrigation usage) until the lake reaches its smallest surface area generally in

October, sometimes in November or December. According to our results, there is no evidence that the timing of the June peak size has changed.

Yearly peaks at Haditha reservoir are not consistent, even during non-drought times. The peaks occur as early as February, and sometimes as late as November. The yearly timing of the smallest lake size is equally varied. It appears releases from upstream dams regulate lake-area as well as the seasonal precipitation.

Droughts and Conflicts

Conflict is the most strongly associated event with quick and substantial changes in lake-surface area; but is rarely the associated event for substantial changes happening over a long period of time

(Table 6). Instead, drought or dam management seemed to be the associated event.

For example, Haditha reservoir-surface area routinely fell substantially during the three drought periods. Lake Therthar has also been shown to reduce in water quantity during drought (Trigo et al.

2010). But the median rate of change in Haditha reservoir area during the three droughts ranges from -

0.35 to 0.06 km2/day. This is a much slower rate of decline than rates in both Haditha and Mosul reservoirs during the first Gulf War. The declines were so sudden they suggest a dam breach, but in fact, a

UN report confirmed that dam managers had partially drained the reservoirs as a precaution in case the dams failed during attack (Aga Khan 1991). Each drainage occurred just before the main periods of warfare: August 1990 when Iraq invaded Kuwait, and January-February 1991 when air and ground forces assaulted Iraq. The concurrent increase in Lake Therthar, downstream of Mosul reservoir, suggests at least some of the water released was funneled there. These were the most rapid declines in reservoir surface area we observed.

At the same time, not all conflicts produced rapid changes in lake-area. One example is the steadiness of lake-area during the U.S. invasion of Iraq between March 19 and May 1, 2003, the subsequent occupation, and the lingering tensions throughout the Anbar Province around Haditha Dam, punctuated by the destructive battle in nearby Falluja throughout 2004 (Traub 2006). Rates of change in

57 the Haditha lake-surface ranged from a loss of 0.80 km2/day to a gain of 0.99 km2/day between March

2003 and March 2006. Mosul reservoir also exhibited stable rates of lake-surface change throughout spring 2003, based on available data. The precautions taken prior to combat during the first Gulf War are not in evidence. We speculate that it was harder to make plans for a conflict driven by another power’s decisions, rather than one’s own instigations.

The same lack of rapid changes in water supply is true of the August 8, 2014 militant attack on

Mosul Dam, and the fight to retake it a week later (Adamo & Al-Ansari 2016; BBC 2014). Militants had earlier overtaken the city of Mosul, approximately 40 kilometers downstream, prompting many dam workers to flee the city starting around June 2014 (Adamo & Al-Ansari 2016). In spite of the bombardments, the chaotic availability of dam workers, and the conflict-fueled fires etched on Landsat imagery from this time (August 28, 2014), the rates of change in lake-surface area were not substantial.

The greatest rate of loss in lake-area, of 1.16 km2/day, occurred between August 12 and September 13.

Mosul reservoir has experienced sharper rates of decrease during many peaceful summers (Figure 11b).

The median rate of lake-area change, based on the six usable Landsat 8 images retrieved between June 9 and September 13, was -0.46 km2/day. Exactly 24 years before these battles, Mosul reservoir was experiencing its sharpest rate of surface area decline preparatory to the first Gulf War; yet the clashes of

2014 raged without any of the same preparations, perhaps pointing either to the surprise of the attack, or a change in strategy by dam managers.

Some recent flare-ups in Iraq featured the capture by militants, and then re-capture by government forces, of Falluja and Ramadi between January 2014 and June 2016 (Cockburn 2014; Milner

2014; BBC 2015; BBC 2016; BBC 2017; Figure 8c). Each of these cities has a barrage controlling water flow on the Euphrates. The various campaigns featured mass civilian displacement, airstrikes, military camps, and the exchange of barrage control multiple times, with the various consequences for working conditions and water regulation. Despite this, the actual military campaigns (January 2014, September

2014, May 2015, January 2016, May-June 2016) were not associated with any substantial changes in the river area upstream of Ramadi Barrage, downstream of Falluja barrage, or between the two barrages.

Militants are reported to have manipulated the Ramadi and Falluja Barrages so as to block water from continuing downstream on multiple occasions. For example, one report suggested that the Ramadi

Barrage was manipulated in June 2015 to send water through the Warar Canal into Lake Habbaniya, rather than continuing along the Euphrates (Gander 2015). However, no substantial increase in the water surface area of Warar Canal, nor of Lake Habbaniya, was detected at this time. The surface area of Lake

Habbaniya fell steadily throughout 2015 until September, broken by a slight uptick in June, but not enough to substantially or lastingly alter the overall trajectory. Other reports of barrage manipulations were met with similarly routine water-surface measurements. Perhaps the manipulations were short-lived enough to be contained between Landsat overpasses. Some are reported to have lasted mere days (von

Lossow 2016).

The sole evidence we found on Landsat imagery of militant water manipulations was the flooded land between Falluja and Abu Ghraib during spring, 2014. This coincided with closure of Falluja Barrage.

The surfeit of water removed from the Euphrates rushed into an irrigation canal, causing it to burst its banks (IRIN 2014).

Evidence suggests that not just militants but also the Iraqi government manipulated water for conflict. A substantial decline of Haditha reservoir occurred during June and July, 2014. There is no substantial increase, or hoarding of water, in the Syrian or Turkish reservoirs upstream, to explain the decrease at Haditha. Instead, we detected a substantial increase in river area downstream of Haditha Dam, indicating a sharp release from the dam. The mass release was a government act to swell the Euphrates

River and impede militants, fresh from victories across Iraq, from attacking Haditha (Rubin & Nordland

2014). To avoid further flooding near Falluja and Abu Ghraib, the excess water was presumably diverted at the Ramadi Barrage into Lake Habbaniya, and from there into Lake Razaza (Figure 8c). Lake Razaza increased in surface area from 329 km2 to 579 km2 between June 18 and August 5, 2014.

While we found that water-surface area in Therthar Lake varied, these variations were not clearly tied to conflict- or drought-provoked incidents at Haditha and Mosul reservoirs.

Upstream dams

The 1990s saw increases in dam building. Comparing pre-1992 to post-2000 lake-surface areas, we found that Mosul lake-area remained constant throughout the constructions (Figure 10b). Haditha reservoir likewise did not substantially decline in size directly following dam-building, but afterwards, droughts had a disproportionate effect on Haditha compared to upstream reservoirs. This implies that droughts combined with water management affect the farthest downstream dam the most. Iraqi marshes situated at the farthest downstream point of the Tigris and Euphrates also contract in size during droughts

(Al-Handal & Hu, 2014).

Reservoir sizes on the same river were generally correlated within a single year. Exceptions sometimes occurred when snowmelt expanded the upstream dams, while evading Haditha; or when a delayed pulse expanded Haditha late in the summer while the other dams contracted. Syrian reservoirs also often did not follow the patterns of other reservoirs, particularly after the Syrian Civil War began in

2011 (Figure 12b). Either Haditha/Syrian reservoirs were no longer correlated, or they were negatively correlated, meaning that as Syrian reservoirs expanded, Haditha contracted (Table 7). This perhaps reflects the weakening of Iraqi-Syrian agreements for sharing Euphrates water as other issues took precedence (UN-ESCWA & BGR 2013). As of February 2013, all three Syrian Euphrates dams were controlled by militants renowned for flouting international conventions (von Lossow 2016). Tabqa Dam, the largest on the Syrian Euphrates, remained in militant hands until May 2017 (Dearden 2017).

Managing for dam failure

The sustained decline in median Mosul reservoir levels after 2006 probably reflects the deliberate decrease in volume held by the reservoir to deal with a potential for dam failure (Adamo & Al-Ansari

2016, Filkins 2017). This threat is also behind dam managers releasing water over the winter in anticipation of incoming snowmelt (Filkins 2017). Perhaps this explains why in recent years the lake-area shrank to extreme low levels in March 2011 and December 2015 that Mosul reservoir otherwise had not

60 experienced. The reservoir area remained low for a brief span of time (unlike the prolonged depressions seen in Haditha during droughts) and the reservoir rebounded following snowmelt.

Uncertainties

There is one substantial change in lake-area which we could not account for: the sharp increase experienced by Mosul reservoir between March 29 and May 16, 2011. Snowmelt might account for some of the increase, but no other snowmelt period was similar. The temperatures in spring 2011 were not unusually warm in the Mosul Dam watershed compared to other years (Figure S15). We did not have evidence that downstream water release from Mosul Dam was cut off in 2011; nor that upstream reservoirs suddenly lost surface area; nor that conflict might have driven these high increases in the Mosul reservoir. There was no outright civil war occurring in Iraq at the time. The Syrian Civil War was mostly restrained to civil disobedience and protests until July 2011, and centered around Damascus and the south of the country – far from the Tigris River border that northern Iraq and Syria share.

Reverse pathway of water to conflict

While this paper dealt exclusively with how conflict affects water resources, other scholars have addressed the converse issue: the situation of when water resources – most often a scarcity of them – leads to conflict (Link et al. 2016). This has in fact occurred in the Tigris and Euphrates basin. The 1975 filling of the Tabqa Dam led to troop build-ups between Iraq and Syria, and conflicts between Turkey and

Syria have erupted not infrequently over the flow of the Euphrates River (Beschorner 1992, MacQuarrie

2004). Conflict can be either violent or non-violent; and since non-violent conflict is less likely to receive media attention and be recorded, data of which is more common is not complete. Water-linked disputes can also be a driver that brings the various actors involved together to cooperate and try to share the limited water resources in a way suitable for everyone. In fact, when looking at 17 different transboundary rivers, the bulk of them showed no link between the need for sharing the river water and an increased risk of conflict (Link et al. 2016). Another study based on a database of transboundary freshwater conflicts found that for a single standard deviation decrease in average precipitation, there was

61 an average relative increase in the probability of conflict of around 50% among 83 different river basins

(Bring and Sjöberg 2017).

Frameworks that describe the many pathways and outcomes by which water scarcity leads to conflict or cooperation are just now being proposed (Link et al. 2016). In the same way, we have here shown that mechanisms by which conflict affects water resources are also varied. Decisions to manipulate water resources as a result of conflict can come during actual military action, or pre-empt it. Many strategic considerations can underpin the decision. And just like not all cases of water scarcity or transboundary sharing of rivers leads to conflict, neither does all conflict come hand-in-hand with a clearly-marked imprint on water resources. The links between water and conflict, no matter which is affecting which, are many and often specific to the particular context where conflict or cooperation is occurring.

Section 5: Conclusions

Our results suggest that conflict was associated with more rapid fluctuations in reservoir-surface area, but not with the greatest absolute changes. Large changes wrought by drought occurred at a comparatively gentle pace. Dam management practices mean that droughts stress the downstream segments of river the most. This is evident at Haditha reservoir, and may well also come to affect Mosul reservoir if proposed upstream projects, such as Turkey’s Ilisu Dam, are completed (UN-ESCWA & BGR

2013). Reservoir-surface areas were generally correlated along a single river, with the exception of

Syria’s since the onset of the Syrian conflict. Broader climate factors probably affected all these lakes together, at least on a yearly level. We recommend further monitoring of the water situation on these rivers, as ongoing wars, population increases, and possible declines in future water supply due to climate change will likely increase scarcity. Water is poised to be the most critical resource in the arid Middle

East. Satellite monitoring could contribute to an international framework for peaceful adjudication of water resources, and for monitoring the effects of drought, conflict, and management on water resources.

CHAPTER 3: SEASONAL CHANGES IN THE MARSHES OF SOUTHERN IRAQ DUE TO DROUGHT, DEVELOPMENT, AND CONFLICT

In this chapter, I remain by the Tigris and Euphrates Rivers, merely moving southward to study the Ahhwar marshes. These marshes are the legendary home of the Garden of Eden. In this chapter, I used satellite images to study how the seasonal cycles of these marshes has changed over time. These marshes have faced many shocks over the past half century, both human and natural. I tried to explore where human fingerprints can be found on marsh changes, and where natural factors are more at play.

The co-authors of this paper are Mejs Hasan, Aaron Moody, Larry Benninger, and Colin West, and it will tentatively be submitted to the journal Earth perspectives.

Section 1: Introduction

Wetlands are among the most productive ecosystems on earth, providing habitat for unique organisms, benefits such as food and recreation for humans, and the filtration of pollutants and particles before they reach larger bodies of water (Hansson et al. 2005). When situated on an arid site affected by spring snowmelt, wetlands can create a unique ecosystem (Partow 2001; Jacobson and Galat 2006). Such an area will experience widespread flooding during the spring, followed by a hot summer during which the floods evaporate and the water withdraws to its permanent lakes. These temporary floods can be important habitats and entice unique organisms to the wetland ecosystem.

Iconic wetlands, known simply as Al-Ahhwar (Arabic for ‘the marshes’), lie where the Tigris and the Euphrates rivers – the two longest in Western Asia – meet in southern Iraq (Figure 13; Maulood 1981;

UN-ESCWA & BGR 2013). These marshes are believed to be more than 6000 years old (Fawzi et al.

2016). One of their hallmark features is the changes in their extent from season to season. The largest extent occurs in the spring (April through June) due to runoff from snowmelt in the mountains of Turkey

63 and Iran, several hundred kilometers upstream on the Tigris and Euphrates Rivers (Al-Handal and Hu

2014). Within a single year, the spring flood discharge on the Tigris River can be eighty times as great as minimum flows (Partow 2001). This seasonal nature of discharge is directly reflected in the seasonal nature of the marshes, and affects everything from phytoplankton, zooplankton, and the rest of the food chain, to the nutrient concentration (Al-Imarah et al. 2006; Hammadi et al. 2007; Salman et al. 2014).

While these marshes are contained almost entirely within Iraq (with a small segment shared across the border with Iran), studying the marshes has international implications as they serve as important habitat for migratory birds, as well as a nursery for the shrimp furnishing a part of Kuwaiti marine fish catch (Partow 2001; Abed 2008). They filter nutrients and sediments from river water before discharge into the Persian Gulf (Abaychi and DouAbul 1985; Hussain and Grabe 2009). Furthermore, the marsh ecosystem includes many archeological sites connected with ancient civilizations, and is proposed as the site of the legendary Garden of Eden (Bedair et al. 2006). This ecosystem enabled a subsistence lifestyle for roughly 5000 years by as many as half a million Marsh Arabs (Partow 2001; Stevens and

Ahmed 2011). The marshes were declared a Wetland of International Significance in 2008 by a conference of the Ramsar Convention (Stevens and Ahmed 2011).

In the middle of the last century, the Iraqi marshes may have covered as much as 20,000 square kilometers (Becker 2014; Chen et al. 2011). They consisted of both permanent and temporary pools and contained reeds (e.g., Phragmites australis), bulrushes (Schoenoplectus litoralis), cattails (Typha domengensis), sedges (Carex), eel grass (Vallisneria spiralis), pondweed (Potamogeton crispus), and all the associated fauna which found refuge in them (Al-Saadi et al. 1985; Bedair et al. 2006; Abed 2008).

However, starting in the 1950s, with the creation of large water diversions upstream in Iraq and other drainage projects, the marshes started to contract (Bedair et al. 2006). The earliest water diversions –

Lake Habbaniya built in 1948, Lake Razaza built in 1951, Lake Therthar built in 1954 – were intended to channel excess spring floods coming down the Tigris and Euphrates in order to prevent flooding in

Baghdad (Partow 2001; Altinbilek 2004). While this aim was met and discharge fell (Figure S16), it cost the marshes water and the marsh area declined (Bedair et al. 2006). By the 1970s, researchers reported

64 seeing far fewer migratory birds stopping at the marshes (Bedair et al. 2006). Since then, the situation has further deteriorated due to deliberate draining and marsh destruction during the Iran-Iraq War (1980-

1988) and following the first Gulf War (1990-1991) (Munro and Touron 1997; Partow 2001, Bedair et al.

2006; Hussain and Grabe 2009; Al-Handal and Hu 2014). Droughts in 1998-2000 and 2007-2009 led to further declines in marsh extent (Trigo et al. 2010; Becker 2014). The combination of precipitation runoff and human activities affecting marsh health represents an example of a coupled human-natural system

(Folke et al. 2002). A dynamic relationship exists between people and the environment in that the variables acting on the marshes include natural ones (rainfall, run-off) and also human activities

(damming, drainage, water withdrawals).

The change in size the marshes have experienced due to drought, conflict, and development has been well-documented. In this paper, we ask whether threshold changes have likewise occurred in the seasonal cycles of the marshes. We examine seasonal cycles over time by recording the timing of maximum and minimum marsh extent for each year throughout the period of record (1975-2017), the size of the marsh at minimum and maximum extent, and the ratios of marsh size comparing different seasons.

We use available satellite imagery, complementing and extending previous remote-sensing studies, which have not substantially examined seasonal cycles (Munro and Touron 1997; Partow 2001; Al-Handal and

Hu 2014; Becker 2014).

Other papers have studied ecology or hydrology of these marshes (Maulood et al. 1981;

Fitzpatrick 2004). In contrast, we approach the marshes as a coupled human-natural system and examine the marsh seasonal cycle in the context of the natural precipitation and discharge, and in context of damming and drainage projects. We try to find relationships between the seasonality of the marshes and these independent variables, and quantitatively determine the natural or human drivers behind anomalies.

The marshes of southern Iraq consist of three sections: Al Hammar, Central, and Hawiza marshes

(Figure 13a). We focus our research for the freshwater Hawiza marsh, because this is the only marsh that has remained at least somewhat intact during the last three decades, and so we are able to study both human and natural impacts simultaneously (Munro and Touron 1997; Al-Abbawy and Al-Mayah 2010).

Hawiza marsh is located on the Iraq-Iran border. Its main source of water is the Tigris River, and it receives additional water from Iran’s Karkheh River (Figure 13).

66 a)

Figure 13. A map and a conceptual diagram of the marshes in southern Iraq. The water sources for the marshes, the surrounding cities, upstream dams, and the countries involved are shown.

Section 2: Data and Methods

Study area

The marshes of southern Iraq formed from the freshwater and sediment drained from Turkey,

Iran, and Iraq by the Tigris and Euphrates rivers. Because the last 300 kilometers or so of their path towards the Persian Gulf is very flat, the Tigris and Euphrates rivers break off into multiple channels, becoming like braided rivers (Partow 2001). Geographical formations to the east and west direct these various channels into a compressed outlet in the Gulf, slowing the rivers down and causing them to deposit large amounts of sediment (Partow 2001). This sediment is the foundation for the formation of the marshes, and its annual arrival continually refreshes the marshes with new material and nutrients. The bulk of material arrives in spring with melting snow from Turkish and Iranian highlands flooding the rivers downstream. Hawiza marsh includes areas of open water (2-2.5 m), and shallower water in which grows dense vegetation (Al-Abbawy and Al-Mayah 2010).

Marsh Arabs shaped the marsh landscape in several ways. They dredged the marshes to form islands where settlements could be built, and fished, hunted, and farmed. Remnants of their activities and livestock waste are found in the soil.

Data

The marsh extent was measured entirely using Landsat surface reflectance satellite data, accessed using Google Earth Engine. Necessary atmospheric, geometric, and radiometric corrections had already been applied to the surface reflectance product (USGS 2017a; USGS 2017b). The Landsat record accessed was created by five satellites, each with a different sensor (Landsat 2 Multispectral Scanner,

Landsat 4 and 5 Thematic Mapper, Landsat 7 Enhanced Thematic Mapper Plus, and Landsat 8

Operational Land Imager). This record, which for our study area included data on 32 years and whose earliest images dated to the 1970s, nevertheless included many months and years without data, as well as images with missing or cloudy pixels. Our attempt to create the most complete library of quality images possible was approached through cloud-masked composites for each month. The composite for each

68 month was formed by any images available during that month, plus the two previous months, but gave primacy to the most recent pixel. This meant that we were able to preserve the most recent pixels for every month, but fill any gaps created by clouds or missing pixels, which would otherwise render the image unusable, with older data. Using this approach, we created a dataset for the 32 years in question, where eleven years had data for at least ten months, and a further seven years had data for over half the months of the year. We created similar libraries of Landsat images for the upstream Mosul and Karkheh reservoirs, whose areas we wanted to compare to the marsh area.

Discharge data was retrieved for the southern-most water gage stations on the Tigris River - the station by Kut, Iraq, approximately 250 km north of Hawiza marsh (USGS 2012). The purpose of this data was to look at how discharge affected marsh area.

Methods

We delineated our study area of Hawiza marsh by a polygon drawn around its extent as seen on various 1970s Landsat data. This was the largest extent to which we had access, and it encompassed the extent of the marsh in all subsequent years. Inside this study area, we classified pixels for each three- month composite as being either water or marsh vegetation. Pixels classified as water were determined by the decision tree-based CFMask algorithm performed by the USGS (Foga et al. 2017; USGS 2018). The extent of the marsh vegetation was determined using NDVI, or normalized difference vegetation index, a well-known algorithm based on the red and near-infrared wavelengths of light, or (NIR – Red)/(NIR +

Red) (Pettorelli et al. 2005). Plants have a unique spectral signature whereby the near infrared light reflected is much more than the red light; thus vegetated areas will produce high values of NDVI. The specific NDVI distributions of vegetated and not-vegetated land cover varied from image to image.

Therefore, we set up a dynamic threshold to act as the dividing classification line. The threshold was determined by first creating two polygons, one entirely over marsh vegetation and one over neighboring desert sand. Average NDVI, and the standard deviation, was calculated for the vegetated polygon and the desert polygon. The desert and vegetated distributions were separated by a gap whose magnitude varied

69 from image to image, and we placed the sand-vegetation threshold within the gap. Pixels whose NDVI values fell above the threshold were classified as vegetation.

The water and vegetated pixels falling within the confines of the Hawiza marsh polygon were converted into area measures (square kilometers). This produced a monthly timeline from which we were able to distinguish the yearly peak and minimum marsh extent, the timing of these points, and to compare the marsh area of different seasons.

Our approach rested on the assumption that the marsh area consists of open water and live vegetation. Therefore, the area of the marsh covered by dead reeds, especially in early spring before these have started growing again, were not included as the ‘marsh extent’ (Al-Imarah et al. 2006). This approach also allowed us to monitor the extent of rebirth in the marshes during the summer, and is similar to the definition of ‘marsh extent’ used by previous marsh studies (Becker 2014).

We created box-plot distributions of marsh extent by month. This allowed us to find the months that in general had the peak or the minimum distribution. We then looked at every year and located the month with the peak or minimum extent for each specific year. We defined a hydrological year in terms of these marshes as from November of one year, to October of the next. This allowed us to regard the commencement of winter rains during November, and the following spring, summer, and fall, as a single unit. We only included years in this analysis that had sufficient data across the expected peak months

(May through August) and the expected minimum months (November to April). For example, the year

1977 had data for the months of April to August, so we picked the maximum extent among that. But since the winter and early spring months were missing, we were unable to determine when the minimum extent occurred and how small it was. For the fourteen years in which there was sufficient data to determine both a minimum and peak marsh extent, we took the ratio of the marsh size at the two extremes. This ratio allowed us to gauge whether the difference in the yearly extremes of marsh extent has been changing as well. We compared the vegetated portion of the marshes to the open water surface, noting that vegetation may be hiding the shallow water in which it grows.

The Hawiza marsh depends on water from the Tigris and Karkheh Rivers, and the sediments and nutrients they deliver, for its existence. Therefore, we compared how the monthly marsh extent compared to monthly discharge data at the Kut discharge station (USGS 2010). We also compared the Hawiza marsh extent to the surface area of both Mosul and Karkheh reservoirs, which lie upstream on the rivers sourcing Hawiza marsh. Since Karkheh Dam began filling in early 2000, we only used data since that year (Partow 2001).

Section 3: Results

Error analysis

Due to the lack of accessible fieldwork data, it was difficult to quantify errors in our estimates of marsh surface area. One approach we took was to compare estimates from Landsat 7 and Landsat 8, during the years for which those two satellites overlapped in obtaining images of the marshes (Figure 14).

For water extent, the root mean square error (RMSE) between the two satellite estimates was 63, the correlation coefficient (r2) was 0.87, and the p-value was less than 0.001. The corresponding values for plant extent were a littler larger, with a RMSE = 168 and r2 = 0.84. The p-value remained less than 0.001.

71 a)

b) Figure 14. Marsh error analysis. Comparison of open water and vegetated marsh extent estimates from Landsat 7 and Landsat 8. Solid lines indicate a 45 degree line.

Size of the marsh

As other papers have reported, the size of Hawiza marsh has decreased since the 1970s (Munro and Touron 1997; Partow 2001; Al-Handal and Hu 2014; Becker 2014). Our data shows this with more time steps compared to previous papers (Figure 15). There appears to be several sudden shifts in the extent of the marsh (Table 8). The first occurred sometime between the late 1970s and 1985, and lasted until about 1995. During this time, the average extent of Hawiza marsh declined by 25% on average following the 1986 completion of the Mosul Dam. The second shift occurred around 1998-1999, a large drop that held for about four years. The large drop in marsh extent seen at this time (Figure 15) occurred during a period of multiple stressors to the marsh ecosystem. These include the deliberate measures taken to drain the dams as government-sponsored retaliation against the Marsh Arabs, by building an extensive system of dykes and canals (Partow 2001). In 2001, the Karkheh Dam completion further slowed inputs of fresh water and sediments. There was a final shift around 2004-2005, lasting till the present day during which the marsh extent increases. This occurred after at least some of the drainage dykes and canals were destroyed. All four stages have large variability (Table 8). Sometimes, as in the drop in marsh extent during 2007-2010 (Figure 15), these variations coincide with natural phenomena like drought (Trigo et al.

2010). The imagery representative of marsh size for each stage is also provided (Figure 16).

Years Event Average marsh total extent (square km) 1970s Pre-drainage and pre-damming 3081 ± 273 period 1980s-1995 Before deliberate drainage, but after impoundment of Mosul 2358 ± 508 period Dam 1998-2003 Drainage works are being implemented in Hawiza marsh. 1200 ± 542 period Karkheh Dam is introduced in 2001 (Partow 2001). Also coincides with a severe drought (Trigo et al. 2010).

2003- Some of the drainage canals are destroyed. The marsh extent 1663 ± 399 onwards varies and includes a large drop coinciding with a 2007-2009 drought. The marsh remains greater in extent compared to the 1998-2003 period.

Table 8. Different periods of marsh surface area size.

Figure 15. Timeline of the total marsh surface area. The four phases are separated by color.

74 a)

75 c)

76 e)

f) Figure 16. Landsat images of the marshes.These images come from the a) 1970s, b) 1988, c) 1995, d) 2000, e) 2006 and f) 2016. All three marshes are shown, with Hawiza marsh located in the northeast part of each image. The images are created based on August data, with previous months filling in in cases of cloud-masked or missing data. The composites feature two near-infrared bands and the red band.

Seasons

We examined the distributions of marsh extent by month (Figure 17). The month with the greatest median extent was July, and July is also the month in which the greatest marsh extent ever was recorded.

March was the month with the smallest median distribution. The range in distribution for almost all months is at least 2500 km2. Particular trends are apparent, including that the upper quartiles of the distributions are dominated by marsh size from the first decades of data.

We then looked more specifically at the peak and minimum extent by year. For this, we only included years in this analysis that had sufficient data across the expected peak months (May through

August) and the expected minimum months (November to April). We measured a hydrological year in terms of these marshes as from November of one year, to October of the next. This allowed us to regard the commencement of winter rains during November, and the following spring, summer, and fall, as a single unit. Our data showed that the maximum marsh surface area most commonly occurs in July, and if not in July, then usually in late spring or early summer (Figure 18a). Since 2000, the date of the maximum extent has become much more variable. Previously, the maximum always occurred prior to

August, but in the 2000s, maximum extents have occurred in almost all months. It is also obvious that the peak marsh extent has fallen since the 1970s (Figure 17). The minimum marsh surface area has most commonly occurred in March, though there was also variability here (Figure 18b). We saw that early spring or late fall have also seen the yearly minimum marsh extent. The size of the minimum has also fallen since the 1970s. In fact, the “minimum” marsh size of the earlier years is greater than the

“maximum” size of many of the more recent years.

Figure 17. The range of marsh extent by month. Plots based on all available data.

79 b) Figure 18. Timing of marsh peak and minimum extent. When the a) peaks and b) minimums of total marsh surface area occur for each year. Size of the label denotes the size of the marsh in square kilometers. Because Landsat images were more frequent during summer months (the expected ‘peak’ time) as opposed to the winter or spring (the expected ‘minimum’ time), there are more points for the peak graph (a).

The surface area of the marshes are not equally comprised of plants or water (Figure 19). For every month and across every stage of marsh size, the live vegetated portions of the marsh are greater than the portions with open water, except during the spring floods (Figure 19). Additionally, the peak open water extent of the marshes does not coincide with the normal summer peak of the entire marsh extent. In our data, the open water peak has occurred in March or April during 80% of the years (Figure

S17).

80 a) b)

c) d)

e) Figure 19. The monthly extent of vegetated and open water marsh area. Area in square kilometers.

The yearly ratio of the peak vegetated marsh area to the minimum vegetated marsh area has steadily fallen over time, meaning that the seasonal change in the vegetated area of the marsh is getting smaller (Figure 20). This trend is apparent regardless of the marsh peak size, and is indicative of a marsh that has smaller seasonal fluctuations.

Figure 20. The yearly ratio of the peak vegetated marsh area to minimum marsh area. The size of the dot refers to the peak size of the marsh during that year.

Human and natural factors

Correlations were weak between the monthly marsh extent and monthly discharge data at the Kut discharge station, both when it came to comparing the open water or the vegetated marsh areas. The strongest correlation we found was 0.44 between the discharge data, in cubic meters per second, and the open water area of marsh. Likewise, weak correlation was found between the yearly accumulation of precipitation in the upper Tigris watershed and the peak marsh extent, either in vegetated, open water, or total terms.

We also compared the Hawiza marsh extent to the surface area of both Mosul and Karkheh reservoirs, which lie upstream on the rivers sourcing Hawiza marsh. Since Karkheh Dam began filling in early 2000, we only used data since that year (Partow 2001). We found that for most years, the reservoirs and the total marsh extent were generally positively correlated, with a few dramatic departures in 2002,

2010, 2014, and 2016 (Table 9).

Year Number of data points Correlation Significant with 95% confidence 2000 8 0.81 Yes 2001 11 0.98 Yes 2002 8 0.48 No 2009 7 0.68 No 2010 6 -0.23 No 2011 5 0.91 Yes 2013 10 0.75 Yes 2014 10 0.32 No 2015 11 0.90 Yes 2016 10 0.57 No Table 9. Correlations between upstream reservoirs and Hawiza marsh area.

Section 4: Discussion

Differences in marsh size seem associated with the human or natural circumstances of the time.

For example, the introduction of Mosul dam in 1986 is associated with the reduction in size of Hawiza marsh on average by about one-fourth. The creation of the dam could conceivably have withheld both water and sediment from the marshes. Another cause could have been the rise of irrigated farm fields which the dam instigated, which would both have consumed water and returned lower-quality irrigation returns. The late 1990s/early 2000s brought a host of complications to the marshes. These include deliberate measures to drain the marsh, the advent of Karkheh Dam in April 2001, and a severe 1998-

2000 drought (Partow 2001; Trigo et al. 2010). Hawiza marsh shrank, on average, to less than half its

1970s size during this time, almost certainly as a result of these natural and human-induced events.

Following the collapse of some of the drainage canals in 2003, Hawiza marsh entered the fourth stage of areal extent. It recovered slightly, though average extent hovered just over half the 1970s extent. This final stage included the most severe drought in the region in 70 years, from 2007 to 2009 (Trigo et al.

2010), reflected in a decrease of Hawiza marsh area during the same time.

What accounts for the seasonal changes in the marshes? By November, the marshes have usually subsided from their peak growth from the summer, and plants, including the reeds, are dying or decomposing (Al-Imarah et al. 2006). This process continues until the marshes reach their minimum extent, usually in March (Figure 18b). As temperatures rise and the new swell of nutrients from the spring

83 flood arrives, production again increases, with the marshes growing in size and reaching their peak area sometime over the summer (most commonly July, according to our data). These peak times coincide with previously reported evidence of peaks in ecosystem activity during summer. For example, recent studies of Hawiza marsh show that macroinvertebrate abundance peaked in summer and spring (Ali et al. 2007) and that nitrates and other nutrients decline in summer due to ecosystem uptake (Al-Imarah et al. 2006).

On the other hand, most field surveys have found that foundational ecosystem production of phytoplankton and zooplankton both are greater in the spring, with counts falling during the hottest summer months (Al-Saadi et al. 1996; Ajeel et al. 2006). These do not correspond with the summer peak of marsh surface area. There may be multiple complex reasons for this. It is possible that the nutrients swept in during the spring floods keep marsh reeds fortified and nourished enough to continue expanding even during the hottest part of the year (July and August). Alternatively, because phytoplankton are fast reproducers, their population may increase rapidly when the first spring nutrient supply arrivers. As the season progresses and vascular plants emerge and start consuming nutrients, the phytoplankton would have access to less and decline. Zooplankton, in turn, would also decline since phytoplankton are their main food source. Other field studies of the estuaries around the marshes have also concluded that temperature has less of an influence on ecosystem activity than salinity and freshwater discharge (Ajeel et al. 2006). That means that peak marsh vegetation can occur in summer in spite of the hot temperatures of

40-50 degrees Celsius.

Sometimes, the maximum or minimum marsh extent did not occur when expected. In some years such as 2005, 2015, and 2016, the peak marsh size occurred in December or November. This seemed to occur when the marsh vegetation remaining at that time was greater than the maximum of the following year’s growth. The minimum marsh extent, which usually occurred in the spring, sometimes occurred in the fall if the marsh vegetation quickly died down following the summer.

Discharge plays a large role in nourishing the marshes. In fact, the decrease in the peak vegetated areas of Hawiza marsh mirrors what has happened with spring discharge from the Tigris (Figure S16).

Since the early 1980s, even before the completion of Mosul Dam in 1985, the yearly peak discharge at

Kut had declined compared to the 1960s and 1970s – which had already declined compared to the 1950s.

At Shatt Al-Arab, the river which carries marsh discharge to the Gulf, the flow dropped from 1000 m3/s in the 1970s to less than 100 m3/s during the height of the 2007-2009 drought (Stevens and Ahmed 2011).

This, combined with the correlations between the marsh water extent and the discharge at Kut, again points to the primacy of discharge in moderating the size of the marshes (Figure S18). The marsh vegetation, especially the foundational phragmites reed, is dependent on water depth and nutrients in water for growth; the deeper the water, the heavier and taller the reed stems can grow, and the more area they cover (Al-Abbawy and Al-Mayah 2009). During drought, of which we saw ample evidence in decreased marsh size, the reeds growing in the marshes stop growing or die (Stevens and Ahmed 2011).

Overall, fish, birds, mammals, and other species in the marshes have declined or are at risk (Partow 2001;

Stevens and Ahmed 2011).

We also addressed if the timing of the marsh peaks and minimums had changed over time. The peak of the entire marsh area during all periods occurs mostly in July while the peak open water area occurs a month or two (March and April) in advance (Figure 18; Figure S17). From 1977 to around 2010, our data showed how consistently peak marsh area occurred in July. At that point, there is more inconsistency. However, the peak open water area has been even less consistent. Reports from the 1980s claimed that the peak flooding occurred between April and June (Al-Saadi et al. 1996). Further reports from 1976 refer to June flooding which affected the chlorosity, pH, and dissolved oxygen in the canals and estuaries south of the marshes (Saad 1983). This and other 1970s reports also mention that the March water clarity was higher, and water turbulence lower, often during the 1970s compared to months later in the spring, because the floods and the related sediments and turbidity had not yet arrived (Huq et al.

1978). Our satellite-based data from 1977 shows that the open water area of the marshes peaked in May

(Figure S17). Thus, there is some evidence that the current occurrence of peak marsh open water in

March is a new phenomenon. This change could potentially be due to dams that sequester some part of the spring floods, in order to replenish the reservoirs with water needed for hydropower generation. It

85 could also be a general decline in snowmelt, meaning the spring flood is smaller and concludes earlier in the year.

An important dimension, but one we were not able to cover, is pollution, nutrients, and sediments in the marshes (Partow 2001). Concentrations of various particles in marsh water have been studied in various field surveys, but it is difficult putting together a comprehensive picture that both shows changes over space and over time (Adam et al. 2007; Awad et al. 2008; Hussain and Grabe 2009). Finer spatial and temporal resolution of the different pools, canals, and patches of vegetation would be welcome. We attempted to use satellite reflectance as a proxy for the chlorophyll content of marsh water (Le et al.

2013c). We were able to establish a strong relationship (correlation of 0.8) between the chlorophyll a measurements taken from Hawiza marsh between 2006 and 2007 and satellite reflectance in the red wavelength (AlMaarofi et al. 2014). Ultimately, because we only had two years’ worth of field data, we could not use this relationship to estimate chlorophyll a from the 1980s to the present. Without this problem, the satellite data would have provided stable, continuous way of monitoring chlorophyll a content by proxy when field data were lacking.

As the earliest dams and water infrastructure projects in Iraq were built starting in the 1950s, the satellite record cannot capture the effect of these on the seasonal pattern of the marshes. Should aerial photographs or other maps of the pre-1950s marshes, on a seasonal basis, exist, these would provide further insight. However, based on the data from 2000 onward, we found that marsh area and upstream reservoir area were strongly and positively correlated. This indicates overall climate conditions are affecting the entire watershed simultaneously in the same way. Human management likely played a role in the years where this was not the case (2002, 2010, 2014, 2016). There was also a low correlation between marsh extent and discharge. This may be because the southernmost discharge station on the

Tigris at Kut is still roughly 250 kilometers north of Hawiza marsh, and the land between the two sites is irrigated. Some water at Kut may be consumed by irrigation before reaching the marshes, and also, there still remain some canals through which water flow bypasses Hawiza marsh.

Although Hawiza marsh survived the shocks of the last four decades the most intact, its rehabilitation may be less prioritized in the future. Instead, Iraq has designated the Central Marshes as the first national park in Iraq, which means they may be directing water flow from the Tigris and Euphrates to the Central marshes rather than the two other marshes (Nature Iraq 2013). Hawiza marsh did not serve as the central location for Marsh Arab villages; its historically dense vegetation made it a harder place to live in, visit, and even in which to conduct scientific study, which may be why the Central marsh has been prioritized (Hussain and Taher 2007; Hussain and Grabe 2009; Fawzi et al. 2016).

Section 5: Conclusions

The Hawiza marshes are a coupled human-natural system. Human factors (dams and drainage) interfere with the natural cycles of the marshes. Timing of peak total marsh extent has become more variable compared to the 1980s (Figure 18), the peak extent is much smaller, and the ratio to the minimum extent is smaller (Figure 20). Also, the peak open water area, signifying the arrival of snowmelt, is potentially arriving earlier. The upstream dams have somewhat changed the natural rhythm of marsh behavior, as well as muted it. The overall decrease in the marsh area is made clear by the evidence that the minimum marsh size over the course of a single year three decades ago is greater than the maximum size during most years in the present day (i.e., the last decade).

The recovery of these marshes has a very strong international dimension. Not only does the health of the marshes affect internationally migrating birds and Persian Gulf fisheries, but the actual allocation of the water used for the marshes is not just a matter under of upstream control. For example, when building Karkheh Dam, part of the plans called for 200 million gallons of the stored water to be sent, via a

210 kilometer pipeline, to Kuwait, which has among the lowest access to fresh water worldwide (Partow

2001). The ease of water access to the marshes depends not just on Iraq and the upstream nations, but in fact on the water needs of the entire region. If the entire region was able to coordinate their water use, these marshes would have a greater chance of withstanding droughts and development.

In light of the post-1970s upstream dams in Turkey and Syria, perhaps it can be considered if Iraq still needs its original flood control measures implemented in the 1950s. These measures included the creation of the Razaza, Habanniya, and Therthar lakes; Partow 2001. The water in these lakes, since they are often stored for long periods and exposed to high temperatures and evaporation, becomes progressively saltier and less usable for humans and ecosystems over time (Rahi 2010). If such extensive flood control is no longer needed, some of those lakes can be decommissioned and the freed water can be fed to the marshes. A region-wide effort to distribute water fairly between all parties, and taking account of environmental needs, will likely be helpful to ensure the longevity of the marshes.

CONCLUSION

The efforts of the American government in providing free access to satellite images to the public have enabled a lot of research on environmental change on the surface of the earth, as well as how human activities or natural factors contribute to that change.

In this dissertation, I probed how satellite data and data obtained in the field can work in concert.

In the Chesapeake Bay, the combination of fieldwork TSS data and reflectance data produced a harmony that extended TSS estimates far beyond the scope of the fieldwork campaigns. I was able to use those additional estimates to tell very specific stories of how TSS concentrations in the Bay changed following major storms, which would not have been possible otherwise. I also explored how different land-uses

(city, forest, farms) along the shore of the major Bay rivers were associated with different patterns in TSS post-storms.

On the other hand, this approach was less successful when I tried it on the Tigris and Euphrates

Rivers. I was not able to produced long-term relationships between water quality and satellite reflectance data, either along the rivers or in the open pools of the marshes at the southern nape of the rivers.

Although some short-term relationships emerged, the record was too short to confidently extrapolate to different time periods. Nevertheless, the amount of information gleaned from satellites in the Tigris and

Euphrates was still plentiful. Water supply in reservoirs, marsh extent – I was able to study patterns in all of these.

It is tempting to believe that these satellite images can narrate the entire story of how and why water resources are changing. But the fact remains that when one is working out of an office in North

Carolina, there is only so much insight one has to offer about two rivers on the other side of the world.

Being aware of the limitations of satellite images ais important when ground data is lacking. When studying the effect of conflict on the Tigris and Euphrates, news stories and United Nations reports were

89 helpful in providing the important background context to explain patterns in the satellite imagery. Those reports provided information on dam management and motives for various changes. While studying the marshes, I had fewer news articles that could illuminate what was actually happening on the ground.

There were several instances where I was unable to decide if an anomaly was due to human intervention or a change in natural factors. Even if I concluded human acts were involved, I did not always know exactly what they were.

No matter the circumstances or surroundings, water is affected by the local happenings. This can be from a storm, as in the Chesapeake, or from droughts, as in Iraq. It can be in an urban setting or a rural setting. The impacts can come from events as mundane as storm pipes, to as dramatic as conflict. We see that nearly all human actions end up affecting our local water sources in some way; natural factors can sometimes exaggerate those effects, or produce unintended consequences. Special care should be taken by all interested parties when using or monitoring water supply and quality.

APPENDIX 1: EXTRA NOTES ON TSS

These notes pertain to the use of Chesapeake Bay TSS data in chapter 1. These are two different laboratory methods to test the amount of suspended material in a sample of water. With SSC, the researcher gathers a full bottle of the river water sample and then pours the entire sample through a filter, dries the sediment, and weighs it. With TSS, however, the researcher gathers a full bottle of the river water and then, pours or removes just a tiny portion of the full sample into a test tube. That sub-sample is the only portion of the water that flows through the filter, and has its sediment measured. Such a sub-sample, however, may not be very representative (Gray et al. 2000, Glysson et al.

2000). The studies cited in the papers above found that when the sand content is negligible, the TSS and

SSC laboratory methods will result in similar measures. However, when the sand content is high, then there is a discrepancy between the two. This is because in pouring the sample into the sub-sample, the sand does not pour representatively. Sand (and all grains that are sized like sand) settle quicker, so they are less likely to make it into the sub-sample tube. This is why, when TSS-SSC data pairs are plotted on a graph, the TSS value is usually a little bit less than the SSC value.

Therefore, TSS-derived data is not as reliable as SSC-derived data.

However, the CBP database measurements rely almost entirely on the TSS laboratory method.

They did try to circumvent the sand-settling-too-quickly issue by shaking the sample vigorously before pouring it into the sub-sample.

The CBP data that met our restrictions (sample taken at a depth less than or equal to 1 meter; at a station where the land is far enough away that a MODIS pixel over that station covers only water) included 43,798 TSS data points. The SSC equivalent was only 404 points. There were roughly 4000 SSC points in total, but most of these were upstream in the Susquehanna River, at stations where MODIS pixels don’t fit. It was not possible to find enough clear-day satellite images to create reflectance-SSC paired data points out of these 404 SSC measurements for the Susquehanna, let alone for dozens of other tributaries. Out of the 43,798 potential TSS data points, less than 2000 matched to a cloud-free MODIS

Terra pixel. For the sake of a spatial broad study, we accepted that our ground measures of suspended material also contain slightly higher errors compared to other laboratory methods.

APPENDIX 2: SUPPLEMENTARY FIGURES

Figure S1. Percent of various land covers for the 12 gage-station sites in the Chesapeake Bay; NLCD 2011 data.

93 a) b) Figure S2. Significance of the reflectance-TSS models. This is denoted by the F statistic and shows that many rivers had a reflectance-TSS concentration relationship that was not strongly significant. The strength of the F statistic was more closely linked to number of data points (a) than the maximum CBP measure of TSS in that river (b). This seems to suggest that some, but not all, rivers with weaker relationships could improve were more data points available.

S3a) S3b)

S3c) S3d)

S3e) S3f) Figure S3. Reflectance-TSS relationships for Mainstem stations. These are organized by latitude and relationship strength.

S4a) S4b)

S4c) S4d)

S4e) S4f) Figure S4. CBP measures of TSS concentrations versus MODIS-derived estimates. These are paired by the same day and same station, presented for randomly selected datapoints in some major Western Shore rivers.

Figure S5. Box-plots of the mean rainfall distribution. These are given for all gages and are plotted in categories according to how many gages recorded rain on the same day.

S6a) S6b)

S6c) S6d)

S6e) Figure S6. Cumulative precipitation based on PERSIANN estimates, 1983-2016. Black box-plots reflect the median precipitation in the various dam watersheds in the Tigris/Euphrates across all years. The super-imposed red plots are the precipitation distributions over the watersheds for a specific year. Figures a-c represent times of drought; d is a normal rainfall year; e represents a hydrological year which begins at lackluster levels, then slowly increases.

S7a)

S7b) Figure S7. Distribution of dates with satellite data by month and by year, for Mosul reservoir.

100

S8a)

S8b)

101

S8c)

S8d) Figure S8. NDWI distributions for land and reservoir polygons. These are based on 149 days of Landsat imagery over Mosul Dam Lake. All distributions are drawn in a. Figures 1b-d show distributions from individual dates; red lines mark the threshold set.

102

S9a)

S9b) Figure S9. Geographical position of points used in validation analysis. The points were distributed equally among images from the following days: August 14, 2015; December 22, 2015; May 30, 2015; June 29, 2016; July 16, 2016; August 18, 2016; September 4, 2016; September 17, 2016; October 4, 2016; November 6, 2016; December 23, 2016.

103

S10a)

S10b) Figure S10. Altimeter-Landsat scatterplots. These compare values of lake extent and water height measured on the same day.

104

S11a)

S11b)

S11c)

105

S11d)

S11e) Figure S11. Close-ups of substantial changes in lake surface area. For Mosul reservoir, this occurred in a) February 1991, b) March 2011, and c) December 2015. For Haditha reservoir, it was for several months in d) 2009 and in e) 2015. Lines of significance are two standard deviations from mean. Background colors indicate: red = conflict; gray = upstream dam building; tan = droughts.

106

S12a)

S12b)

S12c)

107

S12d) Figure S12. Close-ups of rapid changes of lake surface area. For Mosul reservoir, these occur during a) the first Gulf War period of August 1990-February 1991 and b) spring 2011. For Haditha reservoir, these also occurred during c) the first Gulf War, and then again d) during June 2014. Lines of significance are three standard deviations from mean. Background colors indicate: red = conflict; gray = upstream dam building; tan = droughts.

108

S13a)

S13b)

109

S13c)

S13d) Figure S13. Images representing changes in Mosul and Haditha reservoirs.These changes occurred before and after the first Gulf War, with a) Mosul reservoir on August 26, 1990, b) and then on February 10, 1991; c) Haditha reservoir on August 26, 1990, and d) then on February 10, 1991.

110

S14a)

S14b) Figure S14. Floods in near Abu Ghraib. The aerial view of the surrounding land on a) February 26, 2014 and on b) May 17, 2014.

111

Figure S15. Median monthly temperatures in the Tigris/Euphrates. The figures are based on 8-day MODIS Land Surface Temperature 8-day composite data. 2011 in red bold.

112

Figure S16. The discharge at Kut, Iraq.

113

Figure S17. The month during which Hawiza marsh has its peak area of open water, by year.

114

Figure S18. Scatterplot of same-month marsh water extent and discharge at Kut station.

115

APPENDIX 3: SUPPLEMENTARY TABLES

Conflict Years Impact on water systems/supply Citation

First Gulf 1990- Several dams destroyed or damaged in Iraq; Mosul Aga Khan 1991; War 91 Dam directly struck; dams drained pre-emptively to Beschorner 1992; minimize flooding if breached; water pipes, irrigation Filkins 2017 systems damaged; repairs delayed from lack of spare parts Mosul Dam August Dam captured by militants who threatened to both Adamo & Al- Battle 2014 breach the dam and flood those downstream; or close Ansari 2016; BBC the dam gates to deprive water to those downstream 2014a; von Lossow 2016 Ramadi and 2014- After capturing barrages, militants diverted water to Milner 2014; Falluja 2016 harm those living downstream; or to open new attack Gander 2015; BBC Barrages routes across empty riverbeds; or to render rivers 2016; von Lossow impenetrable to Iraqi army forces. 2016 Table S1. List of conflicts. Summaries of water impacts from recent major conflicts in the downstream portions of the Tigris and Euphrates rivers.

Reservoir Path, All days when reservoir Days used in analysis (free row fully covered of clouds) Ataturk 173, 34 225 115 Batman 172, 33 361 253 171, 33 171, 34 Birecik 173, 34 352 214 Devegecidi 172, 34 222 178 Dicle 172, 33 203 162 Falluja Barrage and 169, 37 231 204 surrounding river Habbaniya 169, 37 231 184 Haditha 170, 36 271 222 Karakaya 173, 33 240 141 Karkamis 173, 34 186 148 Keban 173, 33 239 116 Kralkizi 172, 33 188 128 Mosul 170, 34 249 151 170, 35 Ramadi Barrage and 169, 37 231 201 surrounding river Razaza (only 2014) 169, 37 22 14 Tabqa 173, 35 239 139 Therthar 169, 36 219 128 169, 37 Tishrine 173, 35 242 118 Table S2. Landsat path/row for each reservoir, and number of days with data.

116

REFERENCES

Abaychi, J. K., and A. A. Z. DouAbul. 1985. Trace metals in Shatt al-Arab River, Iraq. Water Research 19: 457–462. doi:10.1016/0043-1354(85)90037-5.

Abed, J. M. 2007. Status of water birds in restored Southern Iraqi Marshes. Marsh Bulletin 2: 64–79.

Abed, J. M. 2008. Survey of migratory ducks in West Hammar marsh. Marsh Bulletin 3: 162– 170.

Adam, R. S., I. J. M. Al-Shawi, and F. J. M. Al-Imarah. 2007. Distribution of some chemical elements in the Marsh Lands of Southern Iraq After Rehabilitation. Marsh Bulletin 2: 11– 17.

Adamo, N., and N. Al-Ansari. 2016. Mosul Dam full story: Safety evaluations of Mosul Dam. Journal of Earth Sciences and Geotechnical Engineering 6: 185–212.

Aga Khan, S. 1991. Report to the Secretary-General dated 15 July 1991 on humanitarian needs in Iraq prepared by a mission led by Sadruddin Aga Khan. United Nations, Document S/22799, New York, USA.

Aighewi, I. T., O. K. Nosakhare, and A. B. Ishaque. 2013. Land use–land cover changes and sewage loading in the Lower Eastern Shore watersheds and coastal bays of Maryland: Implications for surface water quality. Journal of Coastal Research 290: 1073–1082. doi.org/10.2112/JCOASTRES-D-11-00195.1

Ajeel, S. G., T. A. Khalaf, H. H. Mohammad, and M. Abbas. 2006. Distribution of zooplankton in the Al-Hawizah, Al-Hammar marshes and Al-Izze river South of Iraq. Marsh Bulletin 1: 140–153.

Al-Abbawy, D. A. H., and A. A. Al-Mayah. 2009. Influence of some ecological factors on plant tall, cover percent, and biomass of Phragmites australis in marshes of southern Iraq after inundation. Marsh Bulletin 4: 121–136.

Al-Abbawy, D. A. H., and A. A. Al-Mayah. 2010. Ecological survey of aquatic macrophytes in restored marshes of southern Iraq during 2006 and 2007. Marsh Bulletin 5: 177–196.

Al-Handal, A., and C. Hu. 2014. MODIS Observations of human-induced changes in the Mesopotamian Marshes in Iraq. Wetlands 35: 31–40. doi.org/10.1007/s13157-014-0590-6

Ali, A. H., N. M. Aziz, and H. A. Hamza. 2007. Abundance, occurrence, seasonal changes and species composition of Macroinvertebrates in the restored Iraqi southern marshes. Marsh Bulletin 2: 80–95.

117

Al-Imarah, F. J. M., I. J. . Al-Shawi, A. M. Issa, and M. G. Al-Badran. 2006. Seasonal variation for levels of nutrients in water from Southern Iraqi Marshlands after Rehabilitation 2003. Marsh Bulletin 1: 82–90.

AlMaarofi, S. S., A. A. Z. Douabul, B. G. Warner, and W. D. Taylor. 2014. Phosphorus and nitrogen budgets of the Al-Hawizeh marshland after re-flooding. Hydrobiologia 721: 155– 164. doi:10.1007/s10750-013-1657-8.

Al-Saadi, H. A., A. A. Al-Lami, and T. I. Kassim. 1996. Algal ecology and composition in the Garmat Ali River, Iraq. Regulated Rivers: Research and Management 12: 27–38.

Altinbilek, D. 2004. Development and management of the Euphrates–Tigris basin. International Journal of Water Resources Development 20: 15–33. doi:10.1080/07900620310001635584.

Alwan, A. R. A. 2006. Past and present status of the aquatic plants of the marshlands of Iraq. Marsh Bulletin 1: 160–172.

Ashouri, H., K. Hsu, S. Sorooshian, D. Braithwaite, K. Knapp, L. Cecil, B. Nelson, and O. Prat. 2015. PERSIANN-CDR: Daily precipitation climate data record from multisatellite observations for hydrological and climate studies. Bulletin of the American Meteorological Society 96: 69–83. doi:10.1175/BAMS-D-13-00068.1.

Alwan, A. R. A. 2006. Past and present status of the aquatic plants of the marshlands of Iraq. Marsh Bulletin 1: 160–172.

Bailey, S. W., and P. J. Werdell. 2006. A multi-sensor approach for the on-orbit validation of ocean color satellite data products. Remote Sensing of Environment 102: 12–23. doi:10.1016/j.rse.2006.01.015.

BBC. 2015. Ramadi battle: IS prepares to defend seized Iraqi city. Retrieved 1 August, 2016, from http://www.bbc.com/news/world-middle-east-32791448

BBC. 2016. IS conflict: Iraqi forces “retake most” of Falluja. Retrieved 21 June, 2017, from http://www.bbc.com/news/world-middle-east-36557092

BBC. 2017. Islamic state and the crisis in Iraq and Syria in maps. Retrieved 20 June, 2017, from http://www.bbc.com/news/world-middle-east-27838034

Becker, R. 2014. The Stalled Recovery of the Iraqi Marshes. Remote Sensing 6: 1260–1274. doi.org/10.3390/rs6021260

Bedair, H. M., H. T. Al Saad, and N. A. Salman. 2006. Iraq’s southern Marshes Something special to be conserved; A case study. Marsh Bulletin 1: 99–126.

Bedair, H. M., and H. T. Al-Saad. 2007. An introductory perspective to the ecotoxicology of the iraqi marshes anthropogenic problems. Marsh Bulletin 2: 171–179.

118

Beschorner, N. 1992. Water and instability in the Middle East. London: Adelphi Papers, Brassey’s Ltd. London.

Binding, C. E., D. G. Bowers, and E. G. Mitchelson-Jacob. 2005. Estimating suspended sediment concentrations from ocean colour measurements in moderately turbid waters; the impact of variable particle scattering properties. Remote Sensing of Environment 94: 373–383. doi:10.1016/j.rse.2004.11.002.

Birkett, C., and B. Beckley. 2010. Investigating the performance of the Jason-2/OSTM radar altimeter over lakes and reservoirs. Marine Geodesy 33: 204–238.

Bjerklie, D., S. Lawrence Dingman, C. Vorosmarty, C. Bolster, and R. Congalton. 2003. Evaluating the potential for measuring river discharge from space. Journal of Hydrology 278: 17–38. doi:10.1016/S0022-1694(03)00129-X.

Boomer, K. M. B., D. E. Weller, T. E. Jordan, L. Linker, Z.-J. Liu, J. Reilly, G. Shenk, and A. A. Voinov. 2013. Using multiple watershed models to predict water, nitrogen, and phosphorus discharges to the Patuxent Estuary. JAWRA Journal of the American Water Resources Association 49: 15–39. doi:10.1111/j.1752-1688.2012.00689.x.

Bowers, D. G., and C. E. Binding. 2006. The optical properties of mineral suspended particles: A review and synthesis. Estuarine, Coastal and Shelf Science 67: 219–230. doi:10.1016/j.ecss.2005.11.010.

Bring, A. and E. Sjöberg. 2017. Complex climate effects on cooperation and disputes in transboundary river basins. University of Utah, Department of Economics Working Paper No. 2017-2, Salt Lake City, Utah.

Cannizzaro, P., J., C. Hu, K. L. Carder, C. R. Kelble, N. Melo, E. M. Johns, G. A. Vargo, and C. A. Heil. 2013. On the accuracy of SeaWiFS ocean color data products on the West Florida Shelf. Journal of Coastal Research 292: 1257–1272. doi:10.2112/JCOASTRES-D-12- 00223.1.

Carder, K. L., R. G. Steward, G. R. Harvey, and P. B. Ortner. 1989. Marine humic and fulvic acids: their effects on remote sensing of ocean chlorophyll. Limnology & Oceanography.

Cerco, C. F., M. R. Noel, and L. Linker. 2004. Managing for water clarity in Chesapeake Bay. Journal of Environmental Engineering 130: 631–642. doi:10.1061/(ASCE)0733- 9372(2004)130:6(631).

Cerco, C. F. 2016. Conowingo Reservoir sedimentation and Chesapeake Bay: State of the science. Journal of Environmental Quality 45: 882–886. doi:10.2134/jeq2015.05.0230.

Chanat, J., D. Moyer, J. Blomquist, K. Hyer, and M. Langland. 2016. Application of a weighted regression model for reporting nutrient and sediment concentrations, fluxes, and trends in concentration and flux for the Chesapeake Bay nontidal water-quality monitoring network,

119

results through water year 2012. U.S. Geological Survey, Scientific Investigations Report 2015-5133, Reston, Virginia.

Chen, Z., C. Hu, and F. Muller-Karger. 2007. Monitoring turbidity in Tampa Bay using MODIS/Aqua 250-m imagery 109: 207–220. doi:10.1016/j.rse.2006.12.019.

Chen, Z. Q. R., M. L. Kavvas, N. Ohara, M. L. Anderson, and J. Yoon. 2011. Impact of water resources utilization on the hydrology of Mesopotamian Marshlands. Journal of Hydrologic Engineering 16: 1083–1092. doi:10.1061/(ASCE)HE.1943-5584.0000208.

Cockburn, P. 2014. The harm before the storm: army battles to expel resurgent al-Qa’ida from Iraq. Retrieved 21 June, 2017, from http://www.independent.co.uk/news/world/middleeast/the-harm-before-the-storm-army-battles-to-expel-resurgent-al-qa-ida-from-iraq- 9040155.html

Collard, R. 2014. Iraq’s battleground dams are key to saving the country from ISIS. Retrieved 6 October, 2016, from http://time.com/3303403/strikes-against-isis-in-iraq-dams/

Congalton, R., and K. Green. 2009. Assessing the accuracy of remotely sensed data: Principles and practices. 2nd ed. Boca Raton: Taylor & Francis Group.

Crétaux, J., W. Jelinski, S. Calmant, A. Kouraev, V. Vuglinski, M. Bergé-Nguyen, M. Gennero, F. Nino, et al. 2011. SOLS: A lake database to monitor in the Near Real Time water level and storage variations from remote sensing data. Advances in Space Research 47: 1497– 1507. doi:10.1016/j.asr.2011.01.004.

Devereux, O. H., K. L. Prestegaard, B. A. Needelman, and A. C. Gellis. 2010. Suspended- sediment sources in an urban watershed, Northeast Branch Anacostia River, Maryland. Hydrological Processes 24: 1391–1403. doi:10.1002/hyp.7604.

Doxaran, D., J. M. Froidefond, and P. Castaing. 2003. Remote-sensing reflectance of turbid sediment-dominated waters. Reduction of sediment type variations and changing illumination conditions effects by use of reflectance ratios. Applied Optics 42: 2623–2634.

Doxaran, D., J.-M. Froidefond, P. Castaing, and M. Babin. 2009. Dynamics of the turbidity maximum zone in a macrotidal estuary (the Gironde, France): Observations from field and MODIS satellite data. Estuarine, Coastal and Shelf Science 81: 321–332. doi:10.1016/j.ecss.2008.11.013.

Eklund, L., A. Persson, and P. Pilesjö. 2016. Cropland changes in times of conflict, reconstruction, and economic development in Iraqi Kurdistan. Ambio 45: 78–88. doi:10.1007/s13280-015-0686-0.

Fawzi, N. A.-M., K. Goodwin, M. Stevens, and B. Mehdi. 2016. Effects of Mesopotamian Marsh (Iraq) desiccation on the cultural knowledge and livelihood of Marsh Arab women. Ecosystem Health and Sustainability 2. doi:10.1002/ehs2.1207.

120

Filkins, D. 2017. Before the flood. Retrieved 20 January, 2017, from http://www.newyorker.com/magazine/2017/01/02/a-bigger-problem-than-isis

Fitzpatrick, R. W. 2004. Changes in soil and water characteristics of natural, drained and re- flooded soils in the Mesopotamian marshlands : Implications for land management planning. CSIRO Land and Water, Client Report, Canberra, Australia.

Focazio, M. J., L. N. Plummer, E. Busenberg, L. J. Bachman, and D. S. Powars. 1998. Preliminary estimates of residence times and apparent ages of groundwater in the Chesapeake Bay watershed, and water-quality data from a survey of springs. U.S. Geological Survey, Water-Resources Investigations Report 97-4225, Richmond, Virginia.

Folke, C., S. Carpenter, T. Elmqvist, L. Gunderson, C. S. Holling, and B. Walker. 2002. Resilience and sustainable development: Building adaptive capacity in a world of transformations. Ambio 31.

Foga, S., P. Scaramuzza, S. Guo, Z. Zhu, R. Dilley Jr, T. Beckmann, G. Schmidt, J. Dwyer, et al. 2017. Cloud detection algorithm comparison and validation for operational Landsat data products. Remote Sensing of Environment 194: 379–390.

Fugate, D. C., and C. T. Friedrichs. 2003. Controls on suspended aggregate size in partially mixed estuaries. Estuarine, Coastal and Shelf Science 58: 389–404. doi:10.1016/S0272- 7714(03)00107-0.

Gander, K. 2015. Isis use water as a weapon in Iraq, by shutting dam on the Euphrates River. Retrieved 2 August, 2016, from http://www.independent.co.uk/news/world/middle-east/isis- use-water-as-a-weapon-in-iraq-by-shutting-dam-on-the-euphrates-river-10295763.html

Gao, B. 1996. NDWI—A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment 58: 257–266. doi:10.1016/S0034- 4257(96)00067-3.

Gellis, A., C. Hupp, M. Pavich, J. Landwehr, W. Banks, B. Hubbard, M. Langland, J. Ritchie, et al. 2009. Sources, Transport, and Storage of Sediment at Selected Sites in the Chesapeake Bay Watershed. U.S. Geological Survey, Scientific Investigations Report 2008-5186, Reston, Virginia.

Ghobadi, Y., B. Pradhan, H. Z. M. Shafri, N. bin Ahmad, and K. Kabiri. 2015. Spatio-temporal remotely sensed data for analysis of the shrinkage and shifting in the Al Hawizeh wetland. Environmental monitoring and assessment 187: 4156. doi.org/10.1007/s10661-014-4156-0

Glysson, G. D., J. R. Gray, and L. M. Conge. 2000. Adjustment of total suspended solids data for use in sediment studies. U.S. Geological Survey Report, Reston, Virginia.

Gong, W., and J. Shen. 2010. A model diagnostic study of age of river-borne sediment transport in the tidal York River Estuary. Environmental Fluid Mechanics 10: 177–196. doi:10.1007/s10652-009-9144-5.

121

Gray, J. R., G. D. Glysson, L. M. Turcios, and G. E. Scharz. 2000. Comparability of suspended- sediment concentration and total suspended solids data. U.S. Geological Survey, Water- Resources Investigations Report, Reston, VA.

Gurbisz, C., and W. M. Kemp. 2014. Unexpected resurgence of a large submersed plant bed in Chesapeake Bay: Analysis of time series data. Limnology and Oceanography 59: 482–494. doi:10.4319/lo.2014.59.2.0482.

Hammadi, N. S., A. Q. Jassim, and H. M. Al-Sodani. 2007. Occurrence and seasonal variations of phytoplankton in the restored marshes of southern iraq. Marsh Bulletin 2: 96–109.

Hansson, L. A., C. Brönmark, P. A. Nilsson, and K. Åbjörnsson. 2005. Conflicting demands on wetland ecosystem services: Nutrient retention, biodiversity or both? Freshwater Biology 50: 705–714. doi:10.1111/j.1365-2427.2005.01352.x.

Hui, F., B. Xu, H. Huang, Q. Yu, and P. Gong. 2008. Modelling spatial-temporal change of Poyang Lake using multitemporal Landsat imagery. International Journal of Remote Sensing 29: 5767–5784. doi:10.1080/01431160802060912.

Hussain, N. A., and M. A. Taher. 2007. Effect of daily variations, diurnal fluctuations and tidal stage on water parameters of East Hammar marshland, southern iraq. Marsh Bulletin 2: 32– 42.

Hussain, N. A., H. A. Saoud, and E. J. Al Shami. 2008. Species composition and ecological indices of fishes in the restored marshes of Southern Mesopotamia. Marsh Bulletin 3: 17– 31.

Hussain, N. A., and S. A. Grabe. 2009. A review of the water quality of the Mesopotamian (Southern Iraq) Marshes prior to the massive desiccation of the early 1990s. Marsh Bulletin 4: 98–120.

Huq, M. F., H. A. Al-Saadi, and H. A. Hameed. 1978. Phytoplankton ecology of Shatt al-Arab reservoir at Basrah, Iraq. Verhandlungen Internationale Vereinigung fur Theoretische und Angewandte Limnologie 20: 1552–1556.

Issa, I. 2015. Sedimentological and hydrological investigation of Mosul Dam Reservoir. PhD Thesis. Luleå, Sweden: Luleå University of Technology.

Jacobson, R. B., and D. L. Galat. 2006. Flow and form in rehabilitation of large-river ecosystems: An example from the Lower Missouri River. Geomorphology 77: 249–269. doi:10.1016/j.geomorph.2006.01.014.

Jantz, P., S. Goetz, and C. Jantz. 2005. Urbanization and the loss of resource lands in the Chesapeake Bay watershed. Environmental management 36: 808–25. doi:10.1007/s00267- 004-0315-3.

122

Joodaki, G., J. Wahr, and S. Swenson. 2014. Estimating the human contribution to groundwater depletion in the Middle East, from GRACE data, land surface models, and well observations. Water Resources Research 50: 2679–2692. doi:10.1002/2013WR014633.

Jordan, T. E., D. L. Correll, and D. E. Weller. 1997. Relating nutrient discharges from watersheds to land use and streamflow variability. Water Resources Research 33: 2579– 2590.

Kassim, T., H. Al-Saadi, and R. Farhan. 2006. Vertical distribution of phytoplankton in Habbaniya Lake, Iraq. Marsh Bulletin 1: 19–31.

Kay, S., J. D. Hedley, and S. Lavender. 2009. Sun glint correction of high and low spatial resolution images of aquatic scenes: A review of methods for visible and near-infrared wavelengths. Remote Sensing 1: 697–730. doi:10.3390/rs1040697.

Klein, I., A. Dietz, U. Gessner, A. Galayeva, A. Myrzakhmetov, and C. Kuenzer. 2014. Evaluation of seasonal water body extents in Central Asia over the past 27 years derived from medium-resolution remote sensing data. International Journal of Applied Earth 26: 335–349. doi:10.1016/j.jag.2013.08.004.

Klein, I., U. Gessner, A. Dietz, and C. Kuenzer. 2017. Global WaterPack – A 250 m resolution dataset revealing the daily dynamics of global inland water bodies. Remote Sensing of Environment 198: 345–362. doi:10.1016/j.rse.2017.06.045.

Kruskal, W. H., and W. A. Wallis. 1952. Use of ranks in one-criterion variance analysis. Journal of the American Statistical Association 47: 583–621. doi:10.1080/01621459.1952.10483441.

Kelley, C., S. Mohtadi, M. Cane, R. Seager, and Y. Kushnir. 2015. Climate change in the Fertile Crescent and implications of the recent Syrian drought. Proceedings of the National Academy of Sciences 112: 3241–3246.

Le, C., C. Hu, J. Cannizzaro, and H. Duan. 2013a. Long-term distribution patterns of remotely sensed water quality parameters in Chesapeake Bay. Estuarine, Coastal and Shelf Science 128: 93–103. doi:10.1016/j.ecss.2013.05.004.

Le, C., C. Hu, J. P. Cannizzaro, D. English, F. Muller-Karger, and Z. Lee. 2013b. Evaluation of chlorophyll-a remote sensing algorithms for an optically complex estuary. Remote Sensing of Environment 129: 75–89. doi:doi:10.1016/j.rse.2012.11.001.

Le, C., C. Hu, D. English, J. Cannizzaro, Z. Chen, L. Feng, R. Boler, and C. Kovach. 2013c. Towards a long-term chlorophyll-a data record in a turbid estuary using MODIS observations. Progress in Oceanography 109: 90–103. doi:10.1016/j.pocean.2012.10.002.

123

Link, P., J. Scheffran, and T. Ide. 2016. Conflict and cooperation in the water-security nexus: a global comparative analysis of river basins under climate change. University of Hamburg, Working Paper CLISEC-31, Hamburg, Germany.

Liu, J., T. Dietz, S. R. Carpenter, M. Alberti, C. Folke, E. Moran, A. N. Pell, P. Deadman, et al. 2007. Complexity of coupled human and natural systems. Science 317: 1513–1516. doi:10.1126/science.1144004.

Liu, X., and M. Wang. 2014. River runoff effect on the suspended sediment property in the upper Chesapeake Bay using MODIS observations and ROMS simulations. Journal of Geophysical Research: Oceans 119. Blackwell Publishing Ltd: 8646–8661. doi:10.1002/2014JC010081.

MacQuarrie, P. 2004. Water security in the Middle East: Growing conflict over development in the Euphrates-Tigris Basin. PhD Thesis. Dublin, Ireland: Trinity College.

Mahamed, S. S. 2008. Phosphorus and nitrogen in the Al-Hawizeh marshes, southern Iraq. Ph.D. Thesis. Waterloo, Canada: University of Waterloo.

Maulood, B. K., G. C. F. Hinton, B. A. Whitton, and H. A. Al-Saadi. 1981. On the algal ecology of the lowland Iraqi marshes. Hydrobiologia 80: 269–276. doi:10.1007/BF00018367.

McFeeters, S. K. 1996. The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International Journal of Remote Sensing 17: 1425–1432. doi:10.1080/01431169608948714.

McGonigal, K. H. 2010. Nutrients and suspended sediment transported in the Susquehanna River Basin, 2009, and trends, January 1985 through December 2009. Susquehanna River Basin Commission, Publication 234, Harrisburg, Pennsylvania.

Millennium Ecosystem Assessment. 2005. Ecosystems and human well-being: Synthesis. Washington, D.C: Island Press.

Miller, R. L., and B. A. McKee. 2004. Using MODIS Terra 250 m imagery to map concentrations of total suspended matter in coastal waters. Remote Sensing of Environment 93: 259–266. doi:10.1016/j.rse.2004.07.012.

Milner, A. 2014. Mosul Dam: Why the battle for water matters in Iraq. Retrieved 17 June, 2017, from http://www.bbc.com/news/world-middle-east-28772478

Moridnejad, A., N. Karimi, and P. Ariya. 2015. Newly desertified regions in Iraq and its surrounding areas: Significant novel sources of global dust particles. Journal of Arid Environments 116: 1–10. doi:10.1016/j.jaridenv.2015.01.008.

124

Moyer, D., R. Hirsch, and K. Hyer. 2012. Comparison of two regression-based approaches for determining nutrient and sediment fluxes and trends in the Chesapeake Bay Watershed. U.S. Geological Survey, Scientific Investigations Report 2012-5244, Reston, Virginia.

Müller, M., J. Yoon, S. Gorelick, N. Avisse, and A. Tilmant. 2016. Impact of the Syrian refugee crisis on land use and transboundary freshwater resources. Proceedings of the National Academy of Sciences of the United States of America 113: 14932–14937. doi:10.1073/pnas.1614342113.

Munro, D. C., and H. Touron. 1997. The estimation of marshland degradation in southern Iraq using multitemporal Landsat TM images. International Journal of Remote Sensing 18: 1597–1606.

National Aeronautics and Space Agency. 2008. Ocean Surface Topography Mission/Jason-2 Mission Overview. Retrieved 29 June, 2017, from https://www.nasa.gov/mission_pages/ostm/overview/index.html

Nature Iraq. 2013. Press release on the declaration of Iraq's first National Park. Retrieved 7 July, 2016, from http://www.natureiraq.org/news/press-release-national-park1

North, E. W., S. Y. Chao, L. P. Sanford, and R. R. Hood. 2004. The influence of wind and river pulses on an estuarine turbidity maximum: Numerical studies and field observations in Chesapeake Bay. Estuaries 27: 132–146.

Novo, E. M. M., J. D. Hansom, and P. J. Curran. 1989. The effect of sediment type on the relationship between reflectance and suspended sediment concentration. International Journal of Remote Sensing. doi:10.1080/01431168908903967.

Olson, M., M. Malonee, and M. E. Ley. 2012. Guide to using Chesapeake Bay Program water quality monitoring data. Chesapeake Bay Program, Report 304-12, Annapolis, Maryland.

Ondrusek, M., E. Stengel, C. S. Kinkade, R. L. Vogel, P. Keegstra, C. Hunter, and C. Kim. 2012. The development of a new optical total suspended matter algorithm for the Chesapeake Bay. Remote Sensing of Environment 119: 243–254. doi:10.1016/j.rse.2011.12.018.

Partow, H. 2001. The Mesopotamian Marshlands : Demise of an ecosystem. United Nations Environment Programme, Report TR.01-3, Nairobi, Kenya.

Paternoster, R., R. Brame, P. Mazerolle, and A. Piquero. 1998. Using the correct statistical test for the equality of regression coefficients. Criminology 36: 859–866.

Pekel, J., A. Cottam, N. Gorelick, and A. Belward. 2016. High-resolution mapping of global surface water and its long-term changes. Nature 540: 418–422. doi:10.1038/nature20584.

125

Pettorelli, N., J. O. Vik, A. Mysterud, J. M. Gaillard, C. J. Tucker, and N. C. Stenseth. 2005. Using the satellite-derived NDVI to assess ecological responses to environmental change. Trends in Ecology and Evolution 20. doi:10.1016/j.tree.2005.05.011.

Petus, C., G. Chust, F. Gohin, D. Doxaran, J.-M. Froidefond, and Y. Sagarminaga. 2010. Estimating turbidity and total suspended matter in the Adour River plume (South Bay of Biscay) using MODIS 250-m imagery. Continental Shelf Research 30: 379–392. doi:10.1016/j.csr.2009.12.007.

Pizzuto, J. E., W. C. Hession, and M. McBride. 2000. Comparing gravel-bed rivers in paired urban and rural catchments of southeastern Pennsylvania. Geology 28: 79. doi:10.1130/0091-7613(2000)028<0079:CGRIPU>2.0.CO;2.

Prasad, M. B. K., M. C. Maddox, A. Sood, S. Kaushal, and R. Murtugudde. 2014. Nutrients, chlorophyll and biotic metrics in the Rappahannock River estuary: implications of urbanisation in the Chesapeake Bay watershed, USA. Marine and Freshwater Research 65. CSIRO: 475. doi:10.1071/MF12351.

Rahi, K. A., and T. Halihan. 2010. Changes in the salinity of the Euphrates River system in Iraq. Regional Environmental Change 10: 27–35. doi:10.1007/s10113-009-0083-y.

Richardson, C. J., and N. A. Hussain. 2006. Restoring the Garden of Eden: An ecological assessment of the Marshes of Iraq. BioScience 56: 477. doi:10.1641/0006- 3568(2006)56[477:RTGOEA]2.0.CO;2.

Rubin, A. and R. Nordland. 2014. Sunni militants advance toward large Iraqi dam. Retrieved 4 July, 2017, from https://www.nytimes.com/2014/06/26/world/middleeast/isis-iraq.html

Ruffin, K. K. 1998. The persistence of anthropogenic turbidity plumes in a shallow water estuary. Estuarine, Coastal and Shelf Science 47: 579–592. doi:10.1006/ecss.1998.0366.

Saad, M. A. H. 1978. Seasonal variations of some physicochemical conditions of Shatt al-Arab estuary, Iraq. Estuarine and Coastal Marine Science 6: 503–513. doi:10.1016/0302- 3524(78)90027-0.

Saad, M. A. H. 1983. Effect of pollution on phytoplankton in the Ashar Canal, a highly polluted canal of the Shatt al-Arab Estuary at Basrah, Iraq. Hydrobiologia 99: 189–196.

Saleh, D. 2010. Stream gage descriptions and streamflow statistics for sites in the Tigris River and Euphrates River Basins, Iraq. U.S. Geological Survey, Data Series 540, Reston, Virginia.

Salman, S. D., M. F. Abbas, A.-H. M. Ghazi, H. K. Ahmed, A. N. Akash, A. A. Z. Douabul, B. G. Warner, and T. Asada. 2014. Seasonal changes in zooplankton communities in the re- flooded Mesopotamian wetlands, Iraq. Journal of Freshwater Ecology 29: 397–412. doi:10.1080/02705060.2014.907547.

126

Schubel, J. R. 1969. Size distributions of the suspended particles of the Chesapeake Bay turbidity maximum. Netherlands Journal of Sea Research 4: 283–309.

Schubel, J. R. 1972. Environmental framework of coastal plain estuaries. Vol. 133. Geological Society of America Memoirs. Geological Society of America. doi:10.1130/MEM133.

Sexton, J. O., X.-P. Song, C. Huang, S. Channan, M. E. Baker, and J. R. Townshend. 2013. Urban growth of the Washington, D.C.–Baltimore, MD metropolitan region from 1984 to 2010 by annual, Landsat-based estimates of impervious cover. Remote Sensing of Environment 129: 42–53. doi:10.1016/j.rse.2012.10.025.

Son, S., and M. Wang. 2012. Water properties in Chesapeake Bay from MODIS-Aqua measurements. Remote Sensing of Environment 123: 163–174. doi:10.1016/j.rse.2012.03.009.

Steinmetz, F., P.-Y. Deschamps, and D. Ramon. 2011. Atmospheric correction in presence of sun glint: application to MERIS. Optics express 19. Optical Society of American (OSA): 9783–800. doi:10.1364/OE.19.009783.

Stevens, M., and H. Ahmed. 2011. Eco-cultural restoration of the Mesopotamian Marshes, Southern Iraq. In Human Dimensions of Ecological Restoration, ed. D. Egen, E. E. Hjerpe, and J. Abrams, 289–298. Washington, DC: Island Press.

Stevenson, J. C., M. S. Kearney, and E. C. Pendleton. 1985. Sedimentation and erosion in a Chesapeake Bay brackish marsh system. Marine Geology 67: 213–235. doi:10.1016/0025- 3227(85)90093-3.

Stumpf, R. P., and J. R. Pennock. 1989. Calibration of a general optical equation for remote sensing of suspended sediments in a moderately turbid estuary. Journal of Geophysical Research. doi:10.1029/JC094iC10p14363.

Susquehanna River Basin Commission (SRBC). Sediment and Nutrient Assessment Program (SNAP). Retrieved 15 February, 2016, from http://www.srbc.net/programs/cbp/nutrientprogram.

Sutton, A. J., T. R. Fisher, and A. B. Gustafson. 2009. Effects of restored stream buffers on water quality in non-tidal streams in the Choptank River Basin. Water, Air, and Soil Pollution 208: 101–118. doi:10.1007/s11270-009-0152-3.

Tourian, M. J., O. Elmi, Q. Chen, B. Devaraju, S. Roohi, and N. Sneeuw. 2015. A spaceborne multisensor approach to monitor the desiccation of Lake Urmia in Iran. Remote Sensing of Environment 156. Elsevier Inc.: 349–360. doi:10.1016/j.rse.2014.10.006.

Traub, J. 2006. The best intentions: Kofi Annan and the UN in the era of American world power. 1st ed. New York: Farrar, Straus and Giroux.

127

Trigo, R. M., C. M. Gouveia, and D. Barriopedro. 2010. The intense 2007–2009 drought in the Fertile Crescent: Impacts and associated atmospheric circulation. Agricultural and Forest Meteorology 150: 1245–1257. doi:10.1016/j.agrformet.2010.05.006.

Tzortziou, M., A. Subramaniam, J. R. Herman, C. L. Gallegos, P. J. Neale, and L. W. Harding. 2007. Remote sensing reflectance and inherent optical properties in the mid Chesapeake Bay. Estuarine, Coastal and Shelf Science 72: 16–32. doi:10.1016/j.ecss.2006.09.018.

UN-ESCWA and BGR (United Nations Economic and Social Commission for Western Asia, Bundesanstalt fur Geowissenschaften und Rohstoffe). 2013. Inventory of Shared Water Resources in Western Asia. Beirut, Lebanon.

United Nations Environment Programme. 1999. The Kosovo conflict – Consequences for the environment and human settlements. Nairobi, Kenya.

United Nations Environment Programme, 2003. Afghanistan: Post-conflict environmental assessment. Nairobi, Kenya.

USGS. 2017. Product guide: Landsat 4-7 surface reflectance (LEDAPS) product version 8.1. U.S. Geological Survey, Product Guide, Reston, Virginia.

USGS. 2017. Product guide: Landsat 8 surface reflectance code (LASRC) product. U.S. Geological Survey, Product Guide, Reston Virginia.

Villa, P., M. Boschetti, A. Scozzari, and S. Vignudelli. 2014. Analysis of vegetation dynamics in Middle East area during 2002-2013 in relation to the 2007-2009 drought episode. In International Geoscience and Remote Sensing Symposium (GARSS), 4260–4263. Institute of Electrical and Electronics Engineers Inc. doi:10.1109/IGARSS.2014.6947430. von Lossow, T. 2016. Water as weapon: IS on the Euphrates and Tigris. Stiftung Wissenschaft und Politik, Comments 3, Berlin, Germany.

Voss, K. A., J. S. Famiglietti, M. Lo, C. Linage, M. Rodell, and S. C. Swenson. 2013. Groundwater depletion in the Middle East from GRACE with implications for transboundary water management in the Tigris-Euphrates-Western Iran region. Water resources research 49: 904–914. doi:10.1002/wrcr.20078.

Vermote, E. F., J. C. Roger, and J. P. Ray. 2015. MODIS surface reflectance user’s guide: Collection 6. MODIS Land Surface Reflectance Science Computing Facility, Report, Greenbelt, Maryland.

Vörösmarty, C., P. McIntyre, M. Gessner, D. Dudgeon, A. Prusevich, P. Green, S. Glidden, S. Bunn, et al. 2010. Global threats to human water security and river biodiversity. Nature 468: 334–334. doi:10.1038/nature09549.

128

Ward, L. G., W. Michael Kemp, and W. R. Boynton. 1984. The influence of waves and seagrass communities on suspended particulates in an estuarine embayment. Marine Geology 59: 85–103. doi:10.1016/0025-3227(84)90089-6.

Ward, L. G., and R. R. Twilley. 1986. Seasonal distributions of suspended particulate material and dissolved nutrients in a Coastal Plain Estuary. Estuaries 9: 156. doi:10.2307/1352127.

Yamazaki, D., M. Trigg, and D. Ikeshima. 2015. Development of a global ~90m water body map using multi-temporal Landsat images. Remote Sensing of Environment 171: 337–351. doi:10.1016/j.rse.2015.10.014.

Zakaria, S., N. Al-Ansari, M. Ezz-Aldeen, and S. Knutsson. 2012. Rain water harvesting and supplemental irrigation at Northern Sinjar Mountain, Iraq. Geoscience Research 3: 100– 108.

Zhang, Q., D. C. Brady, W. R. Boynton, and W. P. Ball. 2015. Long-term trends of nutrients and sediment from the nontidal Chesapeake Watershed: An assessment of progress by river and season. JAWRA Journal of the American Water Resources Association 51. Blackwell Publishing Inc.: 1534–1555. doi:10.1111/1752-1688.12327.

Zhang, Q., W. P. Ball, and D. L. Moyer. 2016. Decadal-scale export of nitrogen, phosphorus, and sediment from the Susquehanna River basin, USA: Analysis and synthesis of temporal and spatial patterns. Science of The Total Environment 563-564: 1016–1029.

Zheng, G., P. M. DiGiacomo, S. S. Kaushal, M. a. Yuen-Murphy, and S. Duan. 2015. Evolution of sediment plumes in the Chesapeake Bay and implications of climate variability. Environmental Science & Technology 49: 6494–6503. doi:10.1021/es506361p.

129