INVESTIGATING THE IMPACT OF MANMAD E RESERVOIRS ON LARG E - SCALE HYDROLOGY AND WATER RESOURCES USING HIGH - RESOLUTION MODELING By Sanghoon Shin A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the de gree of Civil Engineering Doctor of Philosophy 2019 ABSTRACT INVES T IGATING THE IMPACT OF MANMAD E RESERVOIRS ON LARG E - SCALE HYDROLOGY AND WATER RESOURCES USING HIGH - RESOLUTION MODELING By Sanghoon Shin Manmade reservoir s are important component s of the te rrestrial hydrolog ic system . Dam installments fragment river systems, and reservoir operations alter flow regimes. The total storage capacity of existing global reservoirs is large enough to hold one sixth of a nnual continental discharge to global ocean s . Due to growing energy demands, hundreds of large dams are being built and planned around the world , especially in the developing countries . Therefore, there is an urgent need to develop a better understanding of the impact of the existing and new dams on h ydrologic al , ecological, agricultural, and socio - economic systems . Owing to increasing computational power and needs to understand and sim ulate processes in small - scale , hydrological models are advancing towards hyper - resolution global hydrological models. One of benefits of the increased spatial resolution is that the dynamics of surface water inundation over natural river - floodplain system s and manmade reservoir s can be explicitly represented; however, existing global models are not capable of simulating the river - floodplain - reservoir inundation dynamics in an integrated manner . This dissertation addresses th is important standing issue by d eveloping a high - resolution, continental - scale model to simulate the spatial and temporal dynamics of reservoir storag e and release, thus paving pathways toward hyper - resolution surface water modeling in continental - to global - scale hydrological and climat e models. The newly developed model is applied to simulate reservoirs within the contiguous United S t ates (CONUS) and the Mekong River Basin (MRB) in Southeast Asia. With respect to the model development, the following advances are made over the previous global reservoir modeling studies: (1) an existing algorithm for reservoir operation is improved by conducting analytic al analysis and numerical experiments and by introducing new calibra tion features for reservoir operation; (2) the spatial extent and its seasonal dynamics of reservoirs are explicitly simulated and reservoirs are treated as an integra l part of river - flood plain routing, thus reservoir storage is no longer isolated from riv er and floodplain storages; and (3) a novel approach for processing and integrating high - resolution digital elevation models ( DEMs ) in river - floodplain - reservoir routing is introduced. The newly developed reservoir scheme is integrated within the river - flo odplain routing scheme of a continental hydrological model, LEAF - Hydro - Flood, which is set for the CONUS , where abundant data are available for model validation . T hen, the reservoir scheme is integrated into a global hydrodynamics model, CaMa - Flood, to inv estigate the historical impact of manmade reservoirs in the MRB that is experiencing an unprecedented boom in hydropower dam construction . Using the new scheme, the role of flood dynamics in modulating the hydrology of the MRB and the potential impact of f low regulation by the dams on the inundation dynamics are investigated. The significance of hydrologic effect of increasing dams is compared with that of climate variability. The fully cou pled river - reservoir - floodplain storage simulation approach presente d in this dissertation provides an advancement in hydrological modeling in terms of the representation of surface water dynamics, which is indispensable for better attribution of the obser ved changes in the water cycle, prediction of changes in water resou rces, and the understanding of the continually changing environmental and ecological system s . iv ACKNOWLEDGEMENTS I would like to express my gratitude to my advisor, Dr. Yadu Pokhrel, f or guiding me to think big and critically, identify and focus on important issues, and properly organize and present the research results. I also thank to the committees, Drs. Shu - Guang Li, Mantha Phanikumar , and Lifeng Luo, for giving me many great and he lpful advices. My parents have sacrificed their lives and have always prayed for me; I will never be able to pay back to them. Jiyeong is the best supporter in my life; She is always happier than me for my accomplishments, and she gives me courage me whe n I am discouraged. I also thank to my brothers and sister who share many valuable memories with me. I send many thanks to lab mates, Farshid, Suyog, Mateo, and Tamanna, who helped each other and discussed about various topics. I also thank to Eunsang for he lping me settle in East Lansing. This study was supported by NASA (Award#: 80NSSC17K0259), National Science Foundation (Award#: 1752729), and WaterCUBE project from Michigan State University (Award#: GR100096). Simulations were conducted using Cheyenne (d oi:10.5065/D6RX99HX) provided by NCAR's Computational and Information Systems Laboratory, sponsored by the National Science Foundation. Lastly, I dedicate this thesis to the Lord, who is always bigger than I thought. v TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ........................ vii LIST OF FIGURES ................................ ................................ ................................ ..................... viii Chapter 1. Introduction ................................ ................................ ................................ ................... 1 1.1. Research Motivation ................................ ................................ ................................ .......... 1 1.1.1. Manmade Reservoirs in Global Hydrology ................................ ............................ 1 1.1.2. High - resolution Modeling of Reservoir Dynamics at the Continental Scale ......... 3 1.1.3. Application of a New High - resolution Reservoir Scheme: Mekong River Basin .. 7 1.2. Research Goal, Objectives, and Questions ................................ ................................ ........ 9 1.3. Dissertation Outline ................................ ................................ ................................ ......... 11 Chapter 2. High - resolution Modeling of Storag e Dynamics at the Continental Scale ................. 12 2.1. Introduction ................................ ................................ ................................ ..................... 12 2.2. Materials and Methods ................................ ................................ ................................ .... 13 2.2.1. The Hydrological Model LEAF - Hydro - Flood - Dam (LHFD) .............................. 13 2.2.2. Data and Preprocessing ................................ ................................ ......................... 20 2.3. Results and D iscussion ................................ ................................ ................................ .... 34 2.4. Summary and Conclusion ................................ ................................ ............................... 39 Chapter 3. Development of an Improved Reservoir Operation Scheme ................................ ...... 41 3.1. Introduction ................................ ................................ ................................ ..................... 41 3.2. Improvements on Reservoir Operation Scheme ................................ .............................. 42 3.2.1. The Improve d Reservoir Operation Scheme ................................ ......................... 42 3.2.2. Ana lytical Comparison with the Existing Reservoir Operation Schemes ............ 44 3.2.3. The Calibration of R ................................ ................................ ............................. 49 3.3. Experimental Settings ................................ ................................ ................................ ...... 51 3.4. Results and Discussion ................................ ................................ ................................ .... 53 3.4.1. LHFD Model Validation for the CONUS ................................ ............................. 53 3.4.2. Improvements in Reservoir Release and Storage ................................ ................. 55 3.4.3. The R ole of R in Releas e and Storage Simulations ................................ .............. 61 3.4.4. Stability of Reservoir Simulation over the CONUS ................................ ............. 63 3.4.5. Flow Regime Change due to Reservoirs ................................ ............................... 65 3.5. Summary and Conclusion ................................ ................................ ............................... 68 Chapter 4. Sensitivity Analysis for the Effect of Upstream Flow Regulation on Flood Dynamics in th e Lower Mekong River Basin and Tonle Sap Lake ................................ ............................... 70 4.1. Introduction ................................ ................................ ................................ ..................... 70 4.2. Materials and Methods ................................ ................................ ................................ .... 72 4.2.1. HiGW - MAT ................................ ................................ ................................ .......... 72 4.2.2. CaMa - Flood ................................ ................................ ................................ .......... 72 4.2.3. Simulation Settings ................................ ................................ ............................... 73 4.2.4. Terrestrial Water Storage (TWS) and its Estimation ................................ ............ 76 4.2.5. Data ................................ ................................ ................................ ................... 78 vi 4.3. Results and Discussi on ................................ ................................ ................................ .... 79 4.3.1. Model Evaluation ................................ ................................ ................................ .. 79 4.3.2. Role of River - Floodplain Storage on TWS dynamics and Historical Variability 86 4.3.3. Potential Effects of Flow Regulation on Flood Dynamics in the LMRB ............. 90 4.4. Summary and Conclusion ................................ ................................ ............................. 111 Chapter 5. Impact of Manmade R eservoirs on Mekong River Basin Hydrology over the Past Years . ................................ ................................ ................................ ................................ .. 112 5.1. Introduction ................................ ................................ ................................ ................... 112 5.2. Materials and Methods ................................ ................................ ................................ .. 115 5.2.1. CaMa - Flood and HiGW - MAT models ................................ ............................... 115 5.2.2. Dams and Rese rvoirs database ................................ ................................ ............ 116 5.2.3. Observed Data for Model Validation ................................ ................................ .. 119 5.2.4. Incorporation of Reservoirs into CaMa - Flood ................................ .................... 120 5.2.5. Reservoir Operation Scheme ................................ ................................ .............. 123 5.2.6. Simulation Settings ................................ ................................ ............................. 128 5.3. Results and Discussion ................................ ................................ ................................ .. 130 5.3.1. River Discharge and Water Level ................................ ................................ ....... 130 5.3.2. Flood Occurrence ................................ ................................ ................................ 136 5.3 .3. Flooded Area and Surface Water Storage ................................ ........................... 143 5.4. Results and Discussion ................................ ................................ ................................ .. 150 Chapter 6. Summary and Conclusion ................................ ................................ ......................... 152 REFERENCES ................................ ................................ ................................ ........................... 154 vii LIST OF TABLES Table 3 - 1 Summary of experimental settings ................................ ................................ ........... 51 Table 3 - 2 The attributes of reservoirs selected for calibration (Source: GRanD database) 52 Table 3 - 3 Summary of performance measures for different simulations .............................. 57 Table 4 - 1 Geographic location of stations in MRB ................................ ................................ .. 74 Table 4 - 2 Comparison of flooded areas with Arias et al. (2012) for the major flood regions around Tonle Sap Lake indicated by thick black line in Figure 4 - 4a. ................................ ... 85 Ta ble 4 - 3 Changes in major flood characteristics (e.g., onset, magnitude, duration, and amount) compared to the baseline s imulation at different stations analyzed in Figure 4 - 8. 94 Table 4 - 4 A summary of the results of potential changes in flooded area around Tonle Sap Lake under different flow regulation scenarios from this study and those from Arias, Piman, et al. (2014) . ................................ ................................ ................................ ................... 102 Table 4 - 5 Same as Table 4 - 3 but for dry year (1998). ................................ ........................... 109 Table 4 - 6 Same as Table 4 - 3 but for wet year (2000). ................................ ........................... 110 Table 5 - 1 Observation stations of the Mekong River Commission ................................ ...... 120 Table 5 - 2 Simulatio n settings ................................ ................................ ................................ ... 129 viii LIST OF FIGURES Figure 2 - 1 An example of the storage simulation problem by storage buffer effect (SBE) for Lake Mead. ................................ ................................ ................................ ................................ .. 19 Figure 2 - 2 Reservoirs locations and an example of imported surface extent. ....................... 22 Figure 2 - 3 Comparison of drainage area at dam locations based on GRanD database and LHF model (a) before and (b) after the a pplication of coordinates correction. .................... 23 Figure 2 - 4 Temporal disaggregation of USGS water use data to monthly time scale for Butte county in California. ................................ ................................ ................................ ......... 25 Figure 2 - 5 River and reservoir elevation parameterizations. ................................ ................. 27 Figure 2 - 6 Lo ngitudinal bed profiles o f Colorado river derived from 90 m HydroSHEDS void - filled (green), HydroSHEDS conditioned (blue), and MERIT (red) DEMs. ................ 30 Figure 2 - 7 Simulated river - reservoir - floodplain storages averaged over 1985 - 201 0 period 36 Figure 2 - 8 Same as in Figure 2 - 7 but for additional regions ................................ .................. 38 Figure 3 - 1 Demonstration of the need to use . ................................ ................ 47 Figure 3 - 2 Model va lidation for annual mean flow over the contiguous US. ........................ 53 Figure 3 - 3 Validation of seasonal river discharge at the major gauging stations over the contiguous US (unit: 10 3 m 3 /s). ................................ ................................ ................................ .. 54 Figure 3 - 4 Evalua tion of simulated release and storage for the six selected reservoirs in th e Columbia, Colorado, San Joaquin, Sacramento, and Missouri river basins. ....................... 56 Figure 3 - 5 Same as in Figure 3 - 4 but for additional reservoirs ................................ .............. 58 Figure 3 - 6 Boxplot showing stability of simulated reservoir sto rages for 1,889 GRanD reservoirs estimated from frequency analysis for the occurrence of (a) overfilling and (b) underfilling. ................................ ................................ ................................ ................................ . 65 Figure 3 - 7 Eff ect of reservoir operation on long - term average daily river discharge with the exceedan ce probability of (a) 90% (low flow; Q 90 ), (b) 50% (median flow; Q 50 ), and (c) 10% (high flow; Q 10 ), shown as relative change between Q DAM (i.e., R cal ) and Q NAT (i.e., N AT simulation). ................................ ................................ ................................ ................................ .. 67 Figure 4 - 1 Long - term (1981 - 2010) mean river discharge (m 3 /s) simulated by CaMa - Flood at 10km spatial resolution over the entire Mekong River Basin (MRB). .............................. 80 ix Figure 4 - 2 Evaluation of simulated river discharge with observations obtained from the Mekong River Commission (MRC) at locations indicated by red circles in Figure 4 - 1. ...... 81 Figure 4 - 3 Simulated annual mean flood depth downscaled to 500m spatial reso lution using high resolution SRTM topography data for (a) average year (mean of 1981 - 2010), (b) dry year (1998), and wet year (2000). ................................ ................................ ............................... 82 Figure 4 - 4 Monthly flood occurrence and daily water surface elevation. ............................. 83 Figure 4 - 5 Role of river - floodplain storage on TWS dynamics over the MRB. .................... 87 Figure 4 - 6 Same as in Figure 4 - 5 but only for the Lower Mekong reg ion. ........................... 88 Figure 4 - 7 Relationship between ri ver - floodplain storage from CaMa - Flood (monthly) and climate variability (annual mean precipitation and temperature) over the Lower Mekong domain . ................................ ................................ ................................ ................................ ......... 89 Figure 4 - 8 Dai ly river discharge at Pakse (PA) station (location shown in Fig ure 4 - 1 and Table 4 - 1) simulated by HiGW - MAT model with and without considering the existing dams. ................................ ................................ ................................ ................................ ............ 90 Figure 4 - 9 Potential changes in daily river discharge by flow regu lation. ............................ 92 Figure 4 - 10 Same as in Figure 4 - 9 but for simulated water level ................................ ........... 97 Figure 4 - 11 Flood occurrence and the effects of flow regulation on it. ................................ .. 99 Figure 4 - 12 Flooded areas hav ing different flood occurrence estimated from t he baseline simulation results presented in Figure 4 - 11 and their changes under different flow regulation scenarios ................................ ................................ ................................ .................. 101 Figure 4 - 13 Same as in Figure 4 - 11 but for dry (1998) and wet (20 00) years. .................... 105 Figure 4 - 14 Simulated flood occurrence in dry and wet years for different flow regulation scenarios ................................ ................................ ................................ ................................ ..... 106 Figure 4 - 15 Same as in Figure 4 - 9 but for 1998 (dry year). ................................ .................. 107 Fi gure 4 - 16 Same as in Figure 4 - 9 but for 2000 (wet year). ................................ .................. 108 Figure 5 - 1 The spatial distribution of river discharge and commissioned dams (as of 2016) over the MRB. ................................ ................................ ................................ ........................... 118 Figure 5 - 2 Deviation at 3 - arcsec grids from dam crest le vel to water level to achieve recorded storage capacity for (a) the MERIT and (b) HydroSHEDS DEMs. .................... 121 Figure 5 - 3 Relationship between dam height in the database and adjusted dam height at 3 - arcmin CaMa - Flood modeling grids bas ed on the MERIT DEM ................................ ........ 123 x Figure 5 - 4 The profile of Lam Ta Khong P.S. dam. ................................ .............................. 126 Figure 5 - 5 Comparison of turbine design flow and 20% (Q 20 ), 30% (Q 30 ), and 40% (Q 40 ) stream flow exceedances for 12 reser voirs. ................................ ................................ ............. 126 Figure 5 - 6 Mean discharge and stream flow exceedances for 86 reservoirs. ...................... 127 Figure 5 - 7 Validation of simulated river discharge. ................................ .............................. 132 Figure 5 - 8 Validation of simulated water level. ................................ ................................ ..... 133 Figure 5 - 9 Potential changes in river discharge by the existing 86 dams estimated from the ALL simulations. ................................ ................................ ................................ ....................... 134 Figure 5 - 10 Pote ntial changes in water level by the existin g 86 dams estimated from the ALL simulations. ................................ ................................ ................................ ....................... 135 Figure 5 - 11 Simulated flood occurrence in 3 - arcsec (90 m) over the MRB. ....................... 137 Figure 5 - 12 Spatial validation of simul ated inundation dynamics with 0.0 0100º resolution GSW flood occurrence data. ................................ ................................ ................................ .... 138 Figure 5 - 13 Spatial validation of simulated inundation dynamics with 0.00025º resolution GSW flood occurrence data. ................................ ................................ ................................ .... 139 Figure 5 - 14 Spatial validation of simulated inundation dynamics with the Sentinel - 1 product. ................................ ................................ ................................ ................................ ...... 141 Figure 5 - 15 Historical flooded area and surface water storage dynamics over the MRB. 144 Figure 5 - 16 Historical flooded area and su rface water storage dynamics over the MRB for 1979 - 2016. ................................ ................................ ................................ ................................ .. 145 Figure 5 - 17 The monthly GSW flooded area for 1984 - 2016. ................................ ................ 146 Figure 5 - 18 The potential impacts of existing 86 dams on surface water dynami cs estimated from the ALL simulations for selected periods and years. ................................ ................... 148 Figure 5 - 19 The potential impacts of existing 86 dams on surface water dynamics estimated fr om the ALL simulations for the entire simulation period. ................................ ................ 149 1 Chapter 1. I ntroduct ion 1.1. Research Motivation 1.1.1. Manmade Reservoirs in Global Hydrology Water impoundment in reservoirs and flow regulation by dams have exe r ted a profound influence on the terrestrial water cycle (Lehner et al., 20 11b) . The total impoundment capacity of over 50,000 large and thousands of additional small dams built globally during the last century (Chao et al. , 2008; Vörösmarty et al ., 2003) has been estimated to be in between 7,000 - 10,000 km 3 (Chao et al., 2008; Graf, 1999; Lehner et al., 2011b; Lehner & D ö ll, 2004; McCully, 2001; Renwick et al., 2005; Wisser et a l., 2010) , which repres ents about one - sixth of the annual continental discharge to global oceans (T. Oki & Kanae, 2006) . There are growi n g evidences that the large dams have fragmented river sy stems globally (Dynesius & Nilsson, 1994; Graf, 1999; Nilsson et al., 2005; Postel et al., 1996) by changi ng the magnitude, timing, and duration of flows (Haddeland et al., 2006; Hanasaki et al., 2006; Y. Pokhrel, Hanasaki, Koirala, et al., 2012; Veldkamp et al., 2017; Zajac et al., 2017) and altering natural flow regimes (Poff et al., 19 97) . Studies have shown that the adverse effects of dams extend far beyond physical alteration of rivers and changes in downstream hydrology because the hydrologic alterations can have severe ecological and environmental conseque n ces. Many dams have threa tened the ecological integrity of terrestrial and river - floodplain ecosystems (Bunn & Arthington, 2002; Vörösmarty et al., 2010) by altering seasonal flood pulse (Arias, Cochrane, et al., 2014) , impeding species movement (Stone, 2016) , causing river channel incision (Ligon et al., 1995) and delta erosion (S. L. Yang et al., 2011) , blocking sediment flux (Gupta et al., 2012; Syvitski et al., 2005; Vörösmarty et al., 1997; Wisser et al., 2010; Xue et al., 2011) , and altering the transport of dissolved nutrients ( Eiriksdottir et al., 2017) . 2 Evidences suggest that some global mega deltas are sinking at an alarming rate owing to the reduction in sediment delivery by large dams (Bohannon, 2010; Schmidt, 2015; Syvitski et al., 2009) , which has increased the vulnerability of coastal communities to flooding under climate change (Alex Smajgl et al., 2015) . Large reservoirs also affect the terrestrial carbon cycle by changing nitrate remov al (Shuai et al., 2017) , carbon gases ebullition , and greenhouse gas emissions, especially methane in tropical reservoirs (Fearnside & Pueyo, 2012; Kemenes et al., 2007) . In the highly regulated river basins, reservoirs alter terrestrial water storage dynamics and storage in terminal lakes (Chaudhari et al., 2018; Felfelani et al., 2017; Y. Pokhrel et al., 2017; T. Zhou et al., 2016) and also affect groundwater systems (Zhao et al., 2 0 12) , which can have important implications on global sea level change (Chao et al., 2008; Y. Pokhrel, Hanasaki, Yeh, et al., 2012; Wada, Reager, et al., 2 0 16) . Further, surface energy budget is also impacted by reservoirs through water temperature cooling effect (Buccola et al., 2016) , thermal stratification (S. Wang et al., 2012) , and increase in evaporation from open water surfaces which has been linked to intensified extreme precipitation through regional climate feedback (Degu et al., 2011; Hossain, 2010; Hossain et al., 2010, 2012; Woldemichael et al., 2012) The growing recognition and consensus about the impairment of river ecosystems by dams (Babbitt, 2002) has resulted in an increase in dam removal in the US and other regions with aging dams (Doyle et al., 2005; Null et al., 2014; Pohl, 2002) . At the same time, there is an ongoing proliferation in large - scale dam c onstruction to fulfill growi ng energy needs in the developing world; dozens of large dams are being built and hundreds of others are planned in the Amazon and Mekong River basins among other regions (Grumbine & Xu, 2011; Y. Pokhrel et al., 2018; Sabo et al., 2017; Stone, 2016; Timpe & Kaplan, 2017; Winemiller et al., 2016; Zarfl 3 et al., 2015) . There fore, there is an urgent nee d to develop a better understanding of the hydrological and ecological impacts of the existing as well as the planned dams. It is even more critical to understand how the impacts of large dams will interfere with the impacts of climate change, especially i n regions that are likely to be heavily impacted by climate change such as the southwestern US (Cook et al., 2015; Rajagopalan et al., 2009) , and the Amazon (Y. P okhrel et al., 2014) and Mekong (Lauri et al., 2012) river basins. As discussed above, hist orical observations for pre - and post - dam periods can be used to examine the hydrological, geomorphic, and ecological changes caused by dams where such data exist (e.g., Räsänen et al., 2017; Timpe & Kaplan, 2017) ; however, the observed data alone is not sufficient to isolate the changes caused by natural climate variability and human activities. Models ar e indispensable tools that can be used for such isolation of natural and human - induced changes (Y. Pokhrel et al., 2017) as well as for future projections. Hence, continued efforts are indispensable to advance hydrological models for better attribution of th e observed changes in the wa ter cycle, prediction of changes in water resources, and understanding of Earth environmental system. 1.1.2. High - resolution Modeling of Reservoir Dynamics at the Continental Scale Modeling reservoirs involves determination of storage and release by using the in formation on inflow, storage capacity, and downstream demands. Owing to the difficulty in representing individual operation rules in large - scale models, studies have used generic operation schemes to simulate reservoir operation within continental and glob al scale hydrological models. Early works simulated reservoir releases by using a rectangular weir equation or retarding flow velocity (Coe, 2000; Döll et al., 2003; Meigh et al., 1999) . Pioneering works on grid - based, explicit simulation of reservoirs in global model s were presented by Hanasaki et al. (2006) and Haddeland et al. (2006) . A number of subsequent studies have improved and incorporated these 4 schemes into various other global hydrological models (GHMs) and land surface model s (LSMs) (e.g., Adam et al., 2007; van Beek et al., 2011; Biemans et al., 2011; Pokhrel et al., 2015; Pok hrel, Hanasaki, Koirala, et al., 2012; Voisin et al., 2013; Wada et al., 2014) , and other studies have modified them for reservoir - specific applications by including operation rules for the individual reservoirs (Mateo et al., 2014; Zhao et al., 2016) . There are also more recent studies that have used slightly di fferent appr oaches. For example, Solander et al. (2016) developed a generalized reservoir model by employing temperature as a primary factor to modulate seasonal changes in reservoir management, and Ehsani et al. (2016) developed a general reservoir operation scheme (GROS) using artificial neural networks. The aforementioned studies h ave made gre at strides in simulating the effects of dams on river discharge over large global basins; however, significant standing issues still remain in terms of better simulating the spatio - temporal dynamics of reservoir storage and release and making t he schemes c ompatible with hyper - resolution hydrological models (Benedict et al., 2017; Beven et al., 2015; Bierkens, 2015; Bierkens et al., 2015; Mccabe et al., 2 017; Wood et al., 2011) . Here four issues are identified related to reservoir schemes to be used in hyper - resolution hydrological models . There are more issues related to reservoir modeling, but these are of primary interest of this study . Fir st , most sch emes have been developed for macro - scale hydrological models with a typical grid size of 50 - 100 km, in which the storage capacities of one or more reservoirs located within a grid cell are lumped into a single storage component. Such lumped tre atment of mu ltiple reservoirs poses a major challenge in incorporating the exiting schemes in future models with an improved treatment of hydrological states and fluxes at relatively high spatial and temporal resolutions (Clark et al., 2017; Fatichi et al., 2016) . Second , the existing schemes do not explicitly represent reservoir surface extent dyn amics that i s a critically important process to 5 simulate reservoir evaporation under climate change using coupled atmospheric - hydrological models and examine the potential climate impacts of large reservoirs such as those suggested by Degu et al. ( 2011) . Th ird , the existing reservoir schemes are not designed for an integrated simulation of river, reservoirs, and floodplain which is essential when applying the models in regions such as the Amazon, Mekong, and Mississippi river basins (e.g., Allison et al., 2012; Gran et al., 2009; Grumbine & Xu, 2011; Kesel et al., 1974; Timpe & Kaplan, 2017) wher e r iver - floodplain - reservoir dynamics needs to be simulated as a coupled process. Significant advancements are therefore necessary to make existing reservoir schemes suitable for examining the impacts of dam regulation on the seasonally inundated floodplai ns, wetlands, and other flood pulse - dependent ecosystems in the upstream and downstream of reservoirs. Forth , it is essential to employ an improved treatment of topography data, especially to better represent river and reservoir bed elevations consistent w ith the proposed hyper - resolution models (D. Yamazaki et al., 2017) . The direct use of digital elevation models (DEMs) assuming that they represent bear - earth elevations is reasonable for modeling natural rivers and reservoirs tha t a re built after the production of DEMs or will be built in the future (e.g., Gernaat et al., 2017) . For existing reservoirs, however, the DEMs provide water surface elevation which can be tens to hundreds of meters above reservoir bed; the direct use of DEMs can thus cause numerical instability or yield unrealistic results (see Chapter 2. 2.2 ). The issues enumerated above have not yet been addressed even in recent studies. For example, Solander et al. (2016 ) intended to develop a generalized reservoir operation scheme for possible integration into global LSMs for long - term climate impact simulations, but they devised 6 the model with minimal complexity and without considering high resolution spatial and temporal dynamics of reservoir extent and release. Ehsani et al. (2016) developed a reservoir model which was designed to simulate flow regulation by multiple reservoirs in a basin, but the scheme aggregates reservoirs capacities into a large hypothetical reservoir, and hence does not explicitly simu late the spatial extent. Some recent enhancements in large - scale models employ relativ ely high - resolution grids (1 - 10 km) with reservoir modules (Voisin et al., 2017; Wada, de Graaf, et al., 2016) , but the spatial representation of reservoirs in these st u dies is identical to that in the global models, i.e., the reservoir storage is either aggregated into a single model grid where a dam is located or in a predefined reservoir area that is composed of multiple model grid cells ( i.e., level - pool) . Further, c o ncerted community efforts have been made to develop the national water model (NWM, http://water.noaa.gov/about/nwm ), which provides a platform for a detailed and high - resolution hydrological simulations incl u ding lakes and reservoirs; however, the issues identified above have not yet been res olved in the reservoir scheme of NWM. As of summer 2019, the NWM uses Muskingum - Cunge method for river water routing, level - pool method to represent reservoir storage, an d storage - proportional release equation for reservoir release (i.e., passive reservoir s; weir - like operation). Therefore, advancements are needed to make reservoir schemes commensurate with an increase in spatial resolution. Answering the questions relate d to the impacts of reservoirs on the terrestrial water cycle (e.g., how does the seas onally inundated floodplains, wetlands, and other flood pulse - dependent ecosystems change by manmade reservoirs ? ) is possible only through the use of such advanced models that explicitly simulate reservoir dynamics over large scales. In addition, consideri ng the increased use of reservoir operation schemes in continental to global scale studies on both historical analysis (e.g., Pokhrel et al., 20 1 7; Voisin et al., 2017) and future 7 predictions (e.g., Hejazi et al., 2015; Yamagata et al., 2018) , the continued improvement of reservoir schemes both by adding new capabilities and improving the existin g operation/release schemes becomes even more important. 1.1.3. Application of a New High - r esolution Reservoir Scheme: Mekong River Basin The Mekong River is one of the few large and complex global river systems that still remain mostly undammed (Gru mbine & Xu, 2011) , but the rapid socio - economic growth, increasing regional energy demands, and geopolitical opportunities have led to a recent rise in basin - wide construction of large hydropower dams (Y. Pokhrel et al., 2018; Winemiller et al., 2016) . In the Lancang River, which drains the upper portion of the Mekong River Basin (MRB), China is building dozens of mega dams (Y. Pokhrel et al., 2018) ; in the Lower Mekong River Basin ( LMRB), some large main stem dams are being built and about a dozen main stem and over hundred tributary dams are plann ed (Grumbine & Xu, 2011; Y. Pokhrel et al., 2018; Stone, 2016) . The new dams and reservoirs are expected to fulf ill the rapidly growing energy needs and provide other societal benefits; however, the positive benefits come with unprecedented negative social - environmental consequences (Sabo et al., 2017; Schmitt et al., 2018; Stone, 2016) . In t he MRB, of particular concern are the effects of flow regulation on the seasonal hydrological regime characterized by a strong unimodal flow pattern, known as the flood pulse (Junk et al., 1989) . The monsoon - driven flood pulse delivers a timely su pply of water and nutrient - rich sediments for flood - recession agriculture, inland fisheries, and extensive instream and wetland ecosystems , thus serving as a driving force for life and major ecosystems in the LMRB (Arias et al., 2013; Kummu & Sarkkula, 2008) . The flood pulse is also the primary driver of the unique flow reversal in the Tonle Sap River (T SR) that discharges water and sediments 8 from the Mekong River into Tonle Sap Lake (TSL) during wet season and drains w ater from the lake into the Mekong during dry season. The seasonal river - lake inundation dynamics around argest and most productive freshwater fishery (Baran & Myschowod a, 2009; Bonheur & Lane, 2002; Mekong River Commission, 2005) and provides dry - season flow for critical ecosystems and agriculture in t he Mekong Delta (Frappart et al., 2006) . Any alterations in the duration, amplitude, timing, and rapidity of the Mekong flood pulse and the resulting changes in flo odplain dynamics in the LMRB can thus severely impact a wide range of ecosystems and undermine regional food security (Kummu & Sarkkula, 2008) . A n understanding of the surface hydrology of the MRB, LMR B , and TSL has been improved by the studies (1) based on observations for identify ing the historical changes in hydrology of MRB (Arias et al., 2013; Inomata & F ukami, 2008) , (2) based on large - scale modeling and scenario analysis for analyzing and predict ing overall hydr ology of MRB (Costa - Cabral et al., 2008; Meko ng River Commission, 2010; Xue et al., 2011) , and (3) based on hydrodynamic model ing for understanding flood inundations in TSL (Arias et al., 2012; Arias, Piman, et al., 2014; Kummu & Sarkkula, 2008) . While su ch many pieces of knowledge ha ve been accrued , an integrated view that puts all the pieces in place to provide a holistic view for the entire MRB has not been yet presented. To address this gap, this study applies the newly developed high - resolution reservoir scheme in large - scale hydrological model over the entire MRB. The new modeling framework is indispensable for MRB where the flo w is characterized by a highly pronounced seasonal dynamics and unprecedent boom in the construction hy dropower dams is underway since reservoirs are to be modeled as an integral part of river - floodplain system. 9 1.2. Research Goal, Objectives, and Q uestions Th e aforementioned importance of understanding the role of manmade reservoirs in global hydrology (Chapte r 1.1.1) and the need for improving high - resolution reservoir scheme in large - scale models (Chapter 1.1.2) lead me to pursue the overarching goal t o impr ove our understand ing of the compounded impacts of manmade reservoirs and climate change on river flow and flood inundation dynamics by advanc ing our capability to better represent reservoirs in high - resolution large - scale models . The newly developed model ing tool is expected to be usefu l for the broad er hydrological modeling communit y to ward addressin g the increasing issues related to the sustainability of food , energy , and water systems under changing Earth environments. Toward achi e ving this goal, this d issertation is driven by the following specific science questions , which are categorized into two parts . Part 1. Development of a reservoir scheme for high - resolution global hydrological model : Q1. What are the technical challenges and oppo rtunities in bette r representing river - reservoir - floodplain storage at high - resolution in continental and global scales ? Q2. How can we better simulate reservoir release in large - scale models using the limitedly available information (e.g., lack of reservoir - specific rule curv es)? Part 2. Investigation of the impact s of dams on land hydrology and water resources in the MRB : Q3. What are the implications of potential flow regulation by new dams on downstream flood inundation dynamics? Q4. What role does the flood dynam ics play in modula ting the overall hydrology of the basin? Q5. How have the flood dynamics and surface water storage in the MRB changed over the past four decades? 10 Q6. Are the effects of dams significant compared to that of climate variability? Q7. What will be the role of existing res ervoirs in modulating surface water storage and inundation dynamics over the MRB in the future? The specific objective s are (1) to develop a high - resolution reservoir scheme that presents the dynamics of reservoir storage as an integral part of river - flood plain routing to be used in Global Hydrological Models (GHMs) , Land Surface Models (LSMs), and Earth System Models (ESMs) and (2) to investigate the impact of proliferating dams on land hydrology and water resources under climate change . The new reservoir scheme is firstly developed and validated for the Contiguous United States (CONUS) where abundant observation data is available. Then, the new model is applied to the Mekong River Basin (MRB) where the flow is characterized by a highly pronounced seasonal dynamics and unprecedent hydropower dam construction boom is underway . While the model is applied to the MRB in the present study, the new scheme can b e seamlessly incorporated using global datasets into any high - resolution river - floodplain routing schemes in GHMs and ESMs and applied over other regions or globally. 11 1.3. Dissertation Outline The abovementioned research question s are tackle d in individual c hapters (from C hapte r 2 through Chapter 6) . The following provides a brief summary of the remaining chap ters. Chapter 2. High - resolution Modeling of Storage Dynamics at the Continental Scale : Technical challenges and solutions in representing river - reservoir - floodplain storage at high - resolution are presented. Chapter 3. Development of an Improved Reservo ir Operation Scheme : A new reservoir operation scheme is developed. Problems of the existing reservoir operation schemes are elucidated, the parameterizations in the existing schemes are improved, and a calibration feature is newly introduced. Chapter 4. Sensitivity Analysis for the Effect of Upstream Flow Regulation on Flood Dynamics in the Lower Mekong River Basin and Tonle Sap Lake : As a surrogate of potential flow regulations by future dams, scenarios are set up by gradually changing timing (i.e., 1 - month early and delayed peak) and magnitude (up to 50% reduction) of flood peak from the upper Mekong river. For those scenarios, the changes in patterns of river flow and inundation i n the lower Mekong river basin (LMRB) and Tonle Sap Lake (TSL) are exami ned. Chapter 5. Impact of Manmade Reservoirs on Mekong River Basin Hydrology : Historical flood dynamics over the entire Mekong river basin (MRB) is simulated using the models with an d without reservoirs. The historical impact s of reservoirs on the flood dynamics of the MRB are investigated for various aspects (e.g., discharge, water depth, inundation extent , and surface water storage ). In addition, the definite future impact of reserv oirs is estimated. Chapter 6 . Summary and Conclusion 12 Chapter 2. High - resolution Mo deling of Storage Dynamics at the Continental Scale 2.1. Introduction A new high - resolution reservoir scheme is developed that addresses t he four issues of reservoir modeling identified in Chapter 1. 1. 2 . The new reservoir scheme is incorporated within a river - f loodplain routing scheme in a continental scale hydrological model LEAF - Hydro - Flood (Miguez - Macho & Fan, 2012a) (hereafter LH F), resulting in LEAF - Hydro - Flood - Dam (LHFD). The following advances over the previous reservoir schemes are made : (1) the spatial extent and its se asonal dynamics of reservoirs are explicitly simulated, and reservoirs are treated as an integrated part of river - floodplain routing, thus reservoir storage is no longer isolated from river and floodplain storages, and (2) a novel approach for processing a nd integrating high - resolution DEMs in river - floodplain - reservoir routing is introduced. LHFD model is teste d over the contiguous United States (CONUS) at 5 km grids using the abundant data of river flow, reservoir storage, and water use . It is noteworthy that new scheme and data processing algorithm can be seamlessly incorporated into any high - resolution river - floodplain routing models and applied over other regions or globally. The remainder of the chapter is organized as follows. The LHFD model, approac h for pre - processing of DEM, river - reservoir - floodplain routing, and the calibration method are described in Chapter 2. 2 . R esults of high - resolution river - reservoir - floodplain storages are provided in Chapter 2. 3 . S ummary and concluding remarks are presented in Chapter 2. 4. 13 2.2. Materials and Methods 2.2.1. The Hydrological Model LEAF - Hydro - Flood - Dam (LHFD) LEAF - Hydro - Flo od ( LHF ) is a continental - scale land hydrology model that resolves various hydrologic al processes in the re alm from canopy to groundwater aquifers on a physical basis (e.g., radiative energy transfer, turbulent exchange, heat conduction, snow covering and snow water melting, evapotranspiration, throughfall , runoff, river - floodplain flow, infiltration, soil mois ture and heat diffusion, lateral groundwater flow , etc.). The energy and water storages are simulated largely for four entities, i.e., (1) canopy air and vegetation, (2) bare soil surface, (3) 14 soil layers, and (4) river - floodplain . T he energy and water fluxes between the interfaces of these entities are also simulated. In this study, LHF is further developed by incorporating a new reservoir operati on scheme to form LEAF - Hydro - Flood - Dam (LHFD). LHF has been continuously developed since its original version of the Land Ecosystem - Atmosphere Feedback (LEAF) model (Walko et al., 2000) , whic h is a land surface scheme of the Regional Atmosphere Modeling Sy stem (RAMS) . T he original LEA F was extensively improved and enhanced to develop LEAF - Hydro for North America (Fan et al., 2007; Miguez - Macho et al., 2007) by allowing (1) the water table to rise and fall or the vadose zone to s hrink or grow, (2) the water table, recharged by soil drainage, to relax through discharge into rivers, and lateral groundwater flow, leading to convergence to low valleys, (3) two - way exchange between groundwater and rivers, representing both losing and g aining streams, (4) river routing to the ocean as kinematic waves, and (5) setting sea level as groundwater head boundary condition. LEAF - Hydro was further enhanced to develop LEAF - Hydro - Flood ( LHF ) (Miguez - Macho & Fan, 2012a) by introducing a river - floodplain routing scheme that solves the full momentum equation of open channel flow, taking into account the 14 backwater effect (t h e diffusion term) (D. Yamazaki et al., 2011) and the inertia of large water mass of deep flow (acceleratio n term) (Bates et al., 2010) . LHF was applied over the Amazon river basin us ing 2 km grids with 4 - minute time step for land hydrology and 30 - second for surface water routing, where it was extensively validated against observed water table, river discharge, and flooding (Miguez - Macho & Fan, 2012a) , soil moisture and evapotranspiration (Miguez - Macho & Fan, 2012b) , and terrestrial w ater storage (TWS) change (Y. Pokhrel et a l., 2013) using satellite data from Gravity Recovery and Cl imate Experiment (GRACE), and used for future projections (Y. Pokhrel et al., 2014) . The objective of Chapter 2 is on advancing the river - reservoir routing scheme. Hence, i n the remain der of C hapter 2.2.1 , a description for the existing river - reservoir routing scheme of LHF is provided , and an approach to implement reservoirs into LHF to develop LHFD is presented. 2.2.1.1. Existing Surface Water Routing Scheme in LHF T he 1 - D continuity eq uation is given as : (2 - 7) where, Q is discharge, A is flow cross section, and q is lateral flow. The 1 - D momentum equation is given as : (2 - 8 ) 15 where, v is mean flow velocity, g is gravitational acc el eration, d is flow depth, and z is bed elevation. Here, represents the local inertia (or acceleration), represents the advective inertia, represents water (pressure and pote nt ial) slope, and represents friction slope. The friction slope is calculated by Manning as: (2 - 9 ) where, R is hydraulic radius. By combining equations (2 - 8 ) and (2 - 9 ) and a ss uming R is equal to d , the following finite difference equation is obtained. (2 - 10 ) where, h is water surface elevation. Water d epth and level at the time step (t+ t) can be obtained according to mass balance using the flow v el ocity at t . Then, equation (2 - 10 ) can be im plicitly solved. (2 - 11 ) To enhance numerical stability, a second - order Runge - Kutta method is employed. Sea water level is used as a boundary condition at the ends of rivers. 16 Strict ly speaking, the assumption of R = d for Equation (2 - 9) is valid only if the river width ( b ) is suffi ciently greater than water depth ( d ). If such assumption is relaxed, in Equations (2 - 10) and (2 - 11) should be substituted to R (i. e. , cross section area divided by wetted perimeter; in case of rectangular channel that LHFD model employs) . In case of LHFD, more than 95% grid cells have 10 times greater channel width than mean water depth. The other grid cells ( less than 5%) even have small river discharge, hence the effects of the assumption of R = d is not considerable. 2.2.1.2. S patial R esolution of Routing Grids C omputational cost increases as the finer spatial resolution of modeling is used owing to (1) the increase of routing reaches and (2) the increase of temporal resolution to ensure numerical stabil ity. The number of routing reaches is proportional to the number of grid cells; hence it increase s quadratically with the grid cell size. The maximum temporal resoluti on , , is limited to satisfy the Courant - Friedrichs - Lew (CFL) co ndition for the numerical stability. (2 - 1 2 ) where, to enhance stability, is the flow depth at time step t. arger than 1) for explicit schemes and implicit scheme, respectively. CFL condition is a necessary , yet not sufficient , condition for convergence, hence the modeling time step is usually set smaller than used to calculate . 17 Considering the balance between the spatial resolution and available computational resources, the LHFD model is developed at 5 - km resolution. For 1 - year simulation, it takes 35 8 core - hours (2.3 - GHz Intel Xeon E 5 - 2697V4 processors ; NCAR Cheyenne ) . In total, ~60,000 core - hours are consumed for 6 sets of 28 - year simulation s, which is equivalent to 10 days when 252 cores are used. In the setting of 5 - km resolution, the default 1 - year simulation output s for daily hyd rological states and fluxes require ~ 119 GB (i.e., 40 variables × 1450 columns × 1510 rows × 4 bytes/cell × 2 - 30 GB/byte × 365 days) of storage space . For a single set of 28 - year simulation, the output files require 3.25 TB; here we have 6 set s of simulati ons (see Chapter 3 for details), hence the total size of output files is 19.5 TB. V arious spatial resolutions are set in other large - scale studies with consideration of computational cost as well . For example, 25 - km (15 - arcmin) resolution is used for glob al scale (D. Yamazaki et al., 2011) and 10 - km (5 - or 6 - arcmin) for regional scale studies (Mateo et al., 2014; D. Yamazaki et al., 2014) . Predec essors of LHFD model also have been set up in various spatial resolutions. The 5 - km routing grids are finer than those of LEAF - Hydro for North America (Fan et al., 2007; Miguez - Macho et al., 2007) , which is originally developed to have 12.5 k m of river water routing grids while the other land surface modules are set for 1.25 km resolution. LEAF - Hydro - Flood for South America (Miguez - Macho & Fan, 2012a, 2012b; Y. Pokhrel et al., 2013) has 1.5 - arcmin (2 - km) resolution with 1.83 - times bigger domain size (i.e., 1780×2250). The use of finer grid cell for LHFD model is possible , but it will require more co mputational resources and storage space. 18 2.2.1.3. A Challenge in Incorporating Reservoirs Storage Buffer Effect (SBE) T o allow water impoundment behind a dam and prevent flooded water brimming over neighboring basins, two constraints are added: (1) river discharge at the dam location is controlled by the reservoir operation rule, and (2) the downstream of a reservoir is hydraulically disconnected from the reservoir except for the one - way reservoir release. When the rive r - reservoir - floodplain elevations are accurate ly parameterized, solving the full momentum equation with the two constraints enables water impoundment within the reservoir and backwater flow to the upstream; however, we find these constraints to be insuffic ient because the appropriateness of simulated reservoir storage is highly dependent on the accuracy of DEMs. Specifically, we find that excessively accumulated water storages on non - reservoir upstream cells in the vicinity of reservoirs can cause critical problems in simulating the dynamics of reserv buffering contribution of non - reservoir upstream storages to reservoir storage dynamics as the storage buffering effect ( SBE). SBE does not cause water balance problem , but reservoir storage can be inappropriately simulated when SBE is excessive; the excessive non - reservoir upstream storages can significantly dampen the change in water level (or storage) within the reservoir . An example of erroneously simulated reservoi r storage is provided below. 19 Figure 2 - 1 An example of the storage simulation problem by storage buffer effect (SBE) for Lake Mead. When SBE is properly removed (red line), the storage fluctuates over time. When SBE is unremoved (green), the reservoir storage increases slowly and monotonously, and the seasonal fluctuation of release reduces significantly after the reservoir is fully filled. The simulation setting of the case where SBE is removed (red line) i s identical to H06 in Chapter 3.2.1. The simulation setting of the case where SBE is unre moved (green line) is identical to H06 in Chapter 3.2.1 except for the initial condition obtained from additional 30 - years of spin - up. Figure 2 - 1 shows an example for Lake Mead in the Colorado river for which reservoir storage is erroneously simulated when SBE is not eliminated. The simulation setting of Figure 2 - 1 is identical to H06 (see Chapter 3.2.1 for details) except for the initial condition obtained from ad ditional 30 - years of spin - up. Since water continues to back up to large upstream non - rese rvoir cells, it takes ~8 years to increase the storage by ~17% (i.e., from 55% to 72% of storage during 1983 - 1991 period), which is a very slow filling rate compared t o the 2.6 year hydraulic residence time of Lake Mead (Holdren & Turner, 2010) . Further, although the non - reservoir upstream storages are hydraulically connected to main reservoir storage, it is n ot considered in determining the release. Consequently, the release is persistently under estimated, and the reservoir storage can slowly increase near to its full capacity (e.g., from 1983 to 1999 in Figure 2 - 1). When completely filled, the non - reservoir u pstream storages unrealistically dampen the seasonal variation of reservoir storage (e.g. , from 1999 to 2010 in Figure 2 - 1). 20 It is noted that SBE is a virtual phenomenon which would not occur if a perfect DEM is used. For example, reservoir where the amount of its upstream river storage capacity is comparable to that o f reservoir storage capacity due to wrongly parameterized bathymetry that has too flat profile from the reservoir to its upstream . In this case, the excessive SBE beco mes obvious as the significant amount of water stored in the non - reservoir upstream cells becomes comparable to reservoir storage itself exist especially for large regions; hence, a proper treatment of SBE becomes essential for modelling the existing reservoir s in LHFD. Here, we eliminate the unintended behavior in storage simulations due to excessive SBE by first improving the parameterization of river - reservoir beds and floodplain elevations ( Chapter 2.2. 3 . 3) and then by imposing a constraint on reservoir bo undaries. At the boundaries, the slope between water level at non - reservoir cell and bed level of reservoir cell is used as the potential energy gradient at the interface between the reservoir cell and non - rese rvoir cell for surface water routing. Such tre atment of water surface gradient is identical to continually making the bed elevation of the upstream non - reservoir cells equal to the water level of the reservoir. 2.2.2. Data and Preprocessing 2.2.2.1. Dams and Reservoir Inf ormation W e use the dams and reservoir informa tion from the Global Reservoir and Dam (GRanD) database (Lehner et al., 2011a, 20 11b) . A total of 1,889 reservoirs within the CONUS domain, including some reservoirs in the Canadian portion of the Columbia and Missouri river basins, are imported (Figure 2 - 2a). The GRanD database provides the location and height of dams and the areal extent of reservoirs as well as other details such as the storage capacity, construction year, 21 and purpose of reservoir. Using the latitude and longitude information, dams can be located on the river network o f raster model grids derived from HydroSHEDS f low direction map (Lehner et al., 2008) . However, as spatial resolution becomes finer, such a simple method, when automated, ofte n yields significant inaccuracies causing disl ocations of main stem dams to the tributaries and vice versa. Such wrongful assignment of dam locations can cause severe problems in hydrologic simulations. For example, if a tributary dam is located in the mai n stem, a significantly larger drainage area c ould be erroneously assigned for that reservoir, which results in frequent over - filling and spilling of the reservoir. On the contrary, when a smaller basin area is assigned to a reservoir which should have had a larger basin area, under - filling and excess ively underestimated release can be caused. These are inevitable problems in high - resolution surface water modeling using raster grid flow direction maps that determine the direction of flow from a grid cell to its downstream among eight directions (i.e., 4 in cardinal and 4 in diagonal directions) even though, in reality, surface water can flow in any direction (see Shin & Paik, 2017 and refer ences therein) . In this study, we resolve the se issues by comparing the drainage area at a reservoir location estimated by the model with that obtained from GRanD database. If discrepancies in the two drainage areas are found for any dam, the dam location is adjusted to match the model estimated drai nage area with that from GRanD. Among the neighboring cells of the cell located based on the given latitude and longitude, the nearest cell whose estimated drainage area differs from GRanD database drainage are a by less than 20% (an arbitrarily set thresho ld) is determined as a dam location (Figure 2 - 3). 22 Figure 2 - 2 R eservoir s location s and an example of imported surface extent. (a) The 1,889 reservoirs simulated in LHFD model (blue circles) and 27 calibrated reservoirs (red circles), and (b) the rasterized maximum reservoir extent of Lake Sakakawea in the Missouri river shown by a cyan box in (a). The color coding in (b) represe nts the fraction of GRanD reservoir extent within LHFD grid cells. River network is shown by blue lines with the width scaled using the simulated river discharge. The inset in (b) shows Landsat imagery derived using Google Earth with the vector - form reserv oir extent from the GRanD database shown in white. 23 Figure 2 - 3 Comparison of drainage area at dam locations based on GRanD database and LHF model (a) before and (b) after the application of coordinates corr ection. NSE and LNSE are Nash Sutc liffe efficiency coefficient and Log Nash - Sutcliffe efficiency coefficient, respectively. In the newly developed scheme, a reservoir storage is no longer aggregated to a dam location but is spread over multiple upstream c ells from the dam location. Hence, the maximum reservoir extent has to be set to diagnose the total amount of reservoir storage for a given surface water profile. Considering seasonally varying reservoir surface area, the maximum reservoir extent should be sufficiently large enough to cover the reservoir when it is completely filled. For this purpose, we convert each reservoir polygon of GRanD database to raster grids to define the maximum reservoir extent. When the polygon is converted to raster grids, all grid cells having any overlaps wit h the polygon (Figure 2 - 2b, all color - filled cells) are preliminary classified as reservoir cells. In many cases, not all grid cells within the preliminary reservoir extent are hydraulically connected to the main reservoi r body (Figure 2 - 2b, cells with red - filled circles). The disconnected cells are small in number and account for small portion of reservoir extent in GRanD database (see small background values and the number of red circles in Figure 2 - 2b), 24 thus we discard those disconnected cells and import the remaining cells as the maximum reservoir extent (cells with white - filled circles in Figure 2 - 2b). This process is repeated for all 1,889 reservoirs. 2.2.2.2. Water Use Data D ownstream water use data is required for simulatin g reservoir release at each dam loc ation. We use the data from the US Geological Survey (USGS) that is available for 1981 - 2010 period (Maupin et al., 2014 ) . USGS provides the averaged wa ter use at 5 - year intervals but the model needs daily to monthly water use. Therefore, we temporally disaggregate the USGS data to the monthly scale by imposing the monthly irrigation water use patterns simulated by our 1 ° global model HiGW - MAT (Y. Pokhrel et al., 2015) as shown in the example of Butte County, California in Figure 2 - 4. Total irrigation water use (Figure 2 - 4a, red boxes) is disaggregated following the seasonal variability of monthly irrigation (F igure 2 - 4b). Here, the monthly irrigation demand is imported from HiGW - MAT model that incorporates the information on irrigated areas, crop types, and crop calendar (Y. Pokhrel et al., 2015) . To consider both i nter - and intra - annual variability, the monthly irrigation fraction is calculated for every 5 - years, i.e., dividing monthly irrigation demand by 5 - year average deman d. Water uses in the sectors other than irrigation (Figure 2 - 4a, green boxes) are assumed t o remain constant throughout the year since irrigation dominates water use seasonality while the others do not vary significantly. The monthly water use (Figure 2 - 4c ) is derived by adding the irrigation water use disaggregated by the monthly irrigation fra ction to the other water uses. This method generates a monthly time series of water use with seasonal and inter - annual variability while preserving the total amount of water use for each 5 - year intervals. 25 Figure 2 - 4 Temporal disaggregation of USGS water use data to monthly time scale for Butte county in California. (a) The averaged USGS water use in 5 - year intervals, (b) monthly irrigation fraction to total irrigation amount, and (c) total monthly water us e. To allocate water uses to each reservoir, the county - based values are then regridded to 5 km model grids. Each reservoir fulfills the downstream demands in the re gion within a given distance from the reservoir, located at lower elevation than the reserv oir. There is a varying range of downstream extent used in previous studies: 100 km for high - resolution simulations over the contiguous US (Voisin e t al., 2017) , and 250 km (Biemans et a l., 2011; Haddeland et al., 2006) and 1000 km (Hanasaki et al., 2006) over the globe. In this study, we set the downstream extent to 200 km, which is larger than the value of Voisin et al. (2017) but smaller than those used in global studies. Our rationale is that a sufficiently large downstream extent needs to be co nsidered in modeling large river basins such as the Colorado and Columbia, but 1000 km coul d be too large which is in the order of the entire length of such large rivers. To prevent excessive demand allocation, an upper limit of 0.8 is set for DPI (demand per inflow; see 26 Chapter 3.2.1 for detail), which is equivalent to further reducing the down stream extent if necessary. 2.2.2.3. River - reservoir Bed and Floodplain Elevations I ntegrated river - floodplain - reservoir routing requires reliable terrain data that represent bare - earth elevations. Acquiring such data for existing reservoirs is challenging, if not impossible, because the available DEMs provide only water surface elevatio ns over reservoirs which can sometimes be in tens to hundreds of meters above the actual reservoir bed elevation (e.g., red lines in Figure 2 - 5a). Alternatively field survey and remo tely sensed bathymetry data (e.g., Gao, 20 15) can be used but (1) field surveys are sparse, and (2) remotely sensed data are available only for a limited number of reservoirs and cover only the non - permanent parts of the water body. To overcome these data limitations, here we derive reservoir b ed elevations assuming that (1) an abrupt elevation drop on the longitudinal river profile occurs at reservoir locations, which closely corresponds to dam height and (2) most rivers have a concave upward profile resulting from geologic, hydrologic, and cli mate conditions (Figure 2 - 5a). 27 Figure 2 - 5 River and reservoir elevation parameterizations. (a) The longitudinal profiles of mean, preliminary, and refined bed elevations for Colorado, Columbia, Tennessee, and White rivers. The blue horizontal bars show the location and longitudinal extent of the major dams. For Tennessee and White rivers only the middle reach is shown, where dams exi st. (b) Storage - depth parameterization using theoretical inverted - triangle reservoir formulation for Lakes Mead, Shasta, and Havasu (red dots). Black diagonal lines represent 1:1 lines. 28 To derive river - reservoir bed and floodplain elevations, we use MERIT DEM (Multi - Error - Removed Improved - Terrain DEM; Yamazaki et al., 2017) , which is based on SRTM DEM (Shuttle Radar Topography Mission DEM) but include s multiple errors corrections made by separating absolute bias, stripe noise, speckle noise, and tree height bias using multiple satellite datasets and filtering techniques. In particular, distortions in topography slopes in SRTM DEM and other inconsistenc ies in error removal method have been improved in MERIT DEM, hence MER IT DEM is suitable especially for terrain - dependent hydrologic applications such as reservoir - floodplain simulations (D. Yamazaki et al., 2017) . In principle, t he flow direction results from one DEM is not applicable to another DE M, hence the use of MERIT DEM requires the retrieval of flow directions that can be conducted by various automated methods (e.g., Shin & Paik, 2017) , however manual corrections are essen tial which involve tedious and laborious tasks for large scale modeling. Hence, instead of retrieving new flow directions from MERIT DEM, we use the readily available flow direction data from HydroSHEDS (Lehner et al., 2008) which is already manually corrected, is also based on SRTM DEM, and has been widely used (e.g., Fan & Miguez - Macho, 2011; Gong et al., 20 11; Wada, de Graaf, et al., 2016) . To verify the consistency of MERIT DEM and HydroSHEDS - based flow directions, we compared the longitudinal river - reservoir bed profiles for selected ri ver basins; it is found that HydroSHEDS flow direction becomes consist ent with MERIT DEM when the spatial resolution is upscaled to the current model grid size of 5 km (Figure 2 - 6). The HydroSHEDS flow direction map is used after upscaling it to 5 km model grids in LHFD using a similar approach as in Miguez - Macho and Fan (20 12a), which is based on Yamazaki et al. (2009) . Detailed description of flow direction upscal ing is provided below. 29 To utilize flow directions from high - resolution DEM for coarse - resolution DEM, we employ the method of Yamazaki et al. (2009) with some modifications. While the original method of Yamazaki et al. (2009) allows designating the downstr eam grid cell among any grid cells in the domain, a common convention of designating downstream cell is to choose one among 8 - neighboring grid cells. Here, we follow the common convention, which makes our method slightly different than the method of Yamaza ki et al. (2009) in that the flow direction is chosen among 8 directions (i.e., 4 cardinal and 4 diagonal directions). In short, the upscaling procedure can be summarized as follows. Among fine - resolution pixels within a coarse - resolution grid cell of inte rest, the pixel having the largest upstream area is chosen as the outlet pixel. While tracing down the pixels along the fine - resolution flow path from the outl et pixel, the nearest outlet pixel of other grid cells is identified as the tentative next outlet pixel. If the grid cell containing the tentative next outlet pixel is one of 8 - neighboring grid cells from the grid cell of interest, the given grid cell is d etermined as the downstream grid cell. Otherwise, among the 8 - neighboring grid cells from the gri d cell of interest, the grid cell containing the nearest pixel on the fine - resolution flow path is determined as the downstream grid cell. 30 Figure 2 - 6 Longitudinal bed profiles of Colorado river derived fro m 90 m HydroSHEDS void - filled (green), HydroSHEDS conditioned (blue), and MERIT (red) DEMs. The elevations of (a) the cell on the flow path, and the lowest ce ll among the (b) 10 th ( 1.8km) , (c) 20 th ( 3.6k m) , and (d) 30 th ( 5.4km) - order neighboring cells from the flow path are used to derive the profiles. Dotted black circles represent t he reservoir locations in the order of Fontenelle, Flaming Gorge, Lake Powell, Lake Mead, Lake Mo have, and Lake Havasu (from upstream to downstream). The longitudinal profiles are extracted based on HydroSHEDS flow directions. As higher order neighboring c ells are used, the noise on MERIT DEM profile by the discrepancy between HydroSHEDS flow directio ns and MERIT DEM decreases. It is found that the void - filled HydroSHEDS has large noises and the conditioned HydroSHEDS has artificially lowered and stepwise l ongitudinal bed profiles even on the sections where dams do not exist. MERIT DEM contains less no ises and has stepwise longitudinal profiles only where dams exist. Since MERIT DEM does not provide river - reservoir bed elevations, especially where reservoirs exist, we devise a new algorithm to derive reservoir bed elevations using the information of sur face extent of reservoirs (Chapter 2.2. 3 .1). We apply the following procedure for the entire study domain; example results for the Colorado, Columbia, and Miss issippi river 31 basins are shown in Figure 2 - 5a. First, to derive the first - order estimate of river - reservoir bed elevations, the minimum (Figure 2 - 5a, preliminary bed elevation, red line) values of 90 m DEM cells within a 5 km model grid cell are selected, assuming that the cell with the lowest elevation is generally the river mouth of each model grid cell. Then, we obtain an elevation profile by fitting the elevations between upstream and downstream cells of the reservoir using an exponential func tion (Tanner, 1971) that is widely used f or describing longitudinal bed profiles. The elevation of cells along the fitted profile are lowered to the fitted profile. By doing so, large water body elevations in sections marked blue in Figure 2 - 5a are removed. Finally, to remove spikes and pits with minimal distortion of river - reservoir bed profiles, depressions are filled after applying the locally weighted scatterplot smoothing (LOWESS) filter (Cleveland, 1979) (Figure 2 - 5a, refined bed elevation, black line). As t his procedure is repeated for every flow reaches in the model domain (e.g., a flow reach starting from the middle of the main reach of Colorado river), every model grid cell is treated to have smoothly refined bed elevations. This process results in the re moval of tens to hundreds of meters of reservoir water depth to obtain the bed elevations. In LHF, each gri d cell is considered to have a rectangular river channel cross - section. Flood water (i.e., overbank flow) that overtops the river channel is simulate d to be evenly spread over the flat floodplain (e.g., Fan et al., 2017; Neal et al., 2012) . Current version of LHFD employs the same river cross - section parameterization method for reservoir parameterization: each cell has a river - reservoir bed elevation (Figure 2 - 5a, black line) and a floodplain elevation (described next) regardless of whether it is a non - reservoir or a reservoir cell; bed and floodplain elevations are assumed to be flat within a grid cell. We note that reservoir and non - reservoir cells are identically modeled and th e information on whether a cell is a reservoir cell ( Chapter 2.2. 3 .1; 32 Figure 2 - 2b) is used only for diagnosing the reservoir storage, which is the sum of surface water stored in all cells within the reservoir. In LHF, the floodplain elevations were determi n ed from high - resolution DEM using the climatological equilibrium water table (EWT) as a reference (Miguez - Macho & Fan, 2012a) ; that is, when a 5 km model grid cell corresponds to multiple high - resolution DEM pixels, assuming that the pixels lower than EWT represent perennial rivers, the floodplain elevation is determined to the average elevation of the pixels higher than EWT. T h is method could be reasonable when natural rivers are of interest, but as reservoirs are additionally considered in LHFD, the river - reservoir bed and floodplain elevations are additionally required to be parameterized to reasonably represent the reservoir bathymetry and the storage - depth relationship. For example, the reservoir storage when a reservoir is completely filled (i.e., water level reaches near dam crest) should be equal to the storage capacity; however, the storage at that water level can be cal c ulated to be far more (or less) than the storage capacity due to prevailing large errors in DEMs, for which a proper treatment of topography data is unavoidable. Owing to the lack of observations, we use the inverted triangle storage - depth relationship (Liebe et al., 2005 ; and where V , A , h , and a are storage, surface area, depth, and shape factor, respectively), which is one of the widely used relationships in global scale studies (Adam et al., 2007; van Beek et al., 2011; Wada et al., 2014) . First, for reservoir cells ( Chapter 2.2. 3 .1; Figure 2 - 2b), the mean values of 90 m DEM cells within a 5 km model grid cell (Figure 2 - 5a, green line) are regarded as preliminary floodplain elevations. Second, assuming the increasing water level that is flat within a reservoir, we adjust floodplain elevations of reservoir cells to satisfy the inverted tria n gle storage - depth relationship. Most 33 reservoirs can be well parameterized to represent the given relationship as shown as a straight profile or a partly uneven profile in comparison of theoretical and modeled storages (Figure 2 - 5b). A deviating profile of some reservoirs is attributed to the use of flat floodplain geometry, which inevitably incurs sudden large storage increments at water levels where the water overtops the floodplains. All aforementioned procedures are systemically automated, hence enablin g an easy inclusion of any available site - specific data to better represent bathymetry and storage - depth relationship. 2.2.2.4. Atmospheric Forcing and Other Parameterizations W e use the meteorological forcing data from North American Regional Reanalysis (NARR) (Mesinger et al., 2006) , which fully assimilates the observations from multiple sources, making it suited for driving continental scale hydrological models to mimic the obse r ved dynamics of water flows and storages. NARR produces the full atmospheric fields from 1979 to present and is available at 3 - hourly step; the original data at 32km grids are spatially interpolated using bilinear interpolation method to model grid resolu t ion as done in our previous studies (Miguez - Macho & Fan, 2012a; Y. Pokhrel et al., 2013, 2014) . All other model parameters are identical to those used in Fan et al. (2007) , Miguez - Macho et al. (2007) , and Miguez - Macho and Fan (2012a) . 34 2.3. Results and Discussion Figure 2 - 7 s h ows the spatial patterns of the integrated simulation of river - floodplain - reservoir storage at 5km model grids over the entire CONUS. This figure demonstrates that the broad spatial patterns of storages in rivers, reservoirs, and floodplains are clearly c a ptured by the model for large river basins. Validating these results over the whole study domain is not possible due to the lack of spatially - varying data of reservoir storage, depth, and extent, therefore we focus on selected river basins and reservoirs. The comparison of the reservoir surface extent with a satellite - based data for a portion of Colorado, Columbia, and Mississippi river basins (shown by rectangles in the top panel of Figure 2 - 7 ) is presented in the bottom panel of Figure 2 - 7 . The compariso n s for Yellowstone, Missouri, Ohio, and Tennessee river basins are provided in Figure 2 - 8 . The left column of the bottom panel in Figure 2 - 7 and Figure 2 - 8 shows the surface water occurrence data from Pekel et al. (2016) and the middle column presents the w ater storage depth from this study (i.e., a zoomed - in view of the results shown in the top panel). The surface water from 1984 to 2015 (e.g., the grids with pe r manent ground and permanent surface water have 0% and 100% values, respectively). Note that for a consistent comparison of results at the model grid scale of 5km, the 30m data from Pekel et al. (2016) is spatially averaged to 0.04° grids following Yamazak i et al. (2015) . Even though a direct comparison a nd evaluation of the simulated storage depth cannot be made because the data from Pekel et al. (2016) provides only the extent of surface water occurrence, these comparisons evidently suggest that the overall spatial patterns of river - reservoir - floodplain storages, especially the water storage extent in the upstream of the major reservoirs (reservoirs are shown as red dots in the left column of bottom panel in Figure 2 - 35 7 ), are well captured by the model. In the Colorado river basin, the cascade reservoirs a re accurately captured by the model and in the Mississippi both the flood extent along the main stem as well as the surface water extent in many small and large reservoirs can be readily discerned. A larger flood extent in the model can be seen along the M ississippi valley alluvial plain and its downstream, which could be because of an actual model overestimation of flood, the effect of vegetation over water bodies not captured in the open water data, or the inconsistency in temporal aggregation between th e model results and the data from Pekel et al. (2016) . Also shown in Figure 2 - 7 (right column of the bottom panel) are the results of river - reservoir storage from our global model (Y. Pokhrel et al., 2015) at 1 ° grids that used the original H06 reservoir scheme. We present these comparisons with a global model to demonstrate the major improvements made in the present study over the previous global studies in which multiple reservoirs are lumped into a single gr i d cell and reservoir storages are spatially spread across a large grid cell with relatively small storage depth. Note the large storage depth simulated by the new model around the dam locations in the Colorado river (Figure 2 - 7 , lower panel middle column) . The above described spatial features of river - reservoir - floodplain storage in Colorado, Columbia, and Mississippi river basins are similarly found in the Yellowstone, Missouri, Ohio, and Tennessee river basins (Figure 2 - 8 ). Some recent studies (e.g., Voisin et al., 2017; Wada, de Graaf, et al., 2016) have used the global models at a relatively finer grid (i.e., 10 - 15 km), but they use the similar approaches as in the global studies of Pokhrel et al. (2015 ) . 36 Figure 2 - 7 Simulated river - reservoir - floodplain storages averaged over 1985 - 2010 period (top; background shows shaded topographic relief). The bottom panels show the comparison of simulated storage fro m this study (middle) with the surface water occurrence data from Pekel et al. (2016) (left), and 1° grid global model results from Pokhrel et al. (2015) (right) for (a) Columbia, (b) Colorado, and (c) Mississippi river basins shown by black rectangles in the top 37 panel. The lower - right region in the Mississippi subplot, shown as dark blue color, is a part of the ocean. Red circles on the left column indicate dam locations. Simulated storages are grid - averaged water depths. The color coding for the middl e co lumn is same as that for the right column. 38 Figure 2 - 8 Same as in Figure 2 - 7 but for additional regions : (a) Yellowstone and upper Missouri, (b) lower Missouri, and (c) Ohio and Tennessee river basins. 39 2.4. S ummary and Conclusion This study presents the first results of a fully coupled river - reservoir - floodplain storage simulations at 5km grids over the CONUS. First, a new approach for processing DEM to derive reservoir bed elevation is presented, wh ich is cri tical for spatially explicit representation of reservoir storage dynamics in high - resolution simulations. Second, 5 - yearly water use data in the downstream of dams, which determines reservoir release, is temporally disaggregated to monthly values by using the temporal variations in irrigation water use obtained from global model simulations. Third, the issues of Storage Buffer Effect (SBE) is identified, and a new approach is proposed to eliminate SBE in the upstream of reservoirs. The comparison of simulat ed spatial distribution of integrated river - reservoir - floodplain storage with remote - sensing based water extent data demonstrates a promising capability of the model to simulate the spatial extent over and around large reservoirs and floodplains. The river - reservoir - floodplain parameterization can also be enhanced further by utilizing the data from satellite altimetry (e.g., Envisat and GLAS/ICESat) and surface water extent from remote sensing (e.g., MODIS and Landsat). The use of finer grids (Pokhrel et al., 2013) or sub - grid topography (Yamazaki et al., 2011) could enable a more detailed simulation of surface water storage dynamics. Despite some limitations, the present study presents a framework for explicit simulation of reservoir storage and surface extent dynamics that can be used within hyper - resolution hydrological models, thus providing a major advance over previous studies on large - scale reservoir simulations. Whi le the model i s tested over the CONUS at 5km grids in the present study, the scheme and data processing algorithms can be seamlessly incorporated into any high - resolution river - floodplain routing models and applied over other regions at scales ranging from a river basin to the entire globe using site - specific and global datasets (e.g., GRanD 40 database, MERIT DEM, and satellite - based products). According to a recent study (Fleischmann et al., 2019) , the continental scale hydrodynamics models are found to yield satisfactory results when the modeling reach length is small (<15km; 1 - 5km are preferable). The current grid size of 5km is in their recomm ended range, b ut it is possible to reduce the grid size . The only problem in reducing the grid size is computational cost, hence it would add more value on the newly developed model when the numerical scheme is optimized to reduc e the computational cost . 41 Chapter 3. Development of an Improved Reservoir Operation Scheme 3.1. Introduction Modeling reservoirs involves determination of storage and release by using the information on inflow, storage capacity, and downstream demands. Owing to the difficulty in representing indiv idual operatio n rules in large - scale models, studies have used generic operation schemes to simulate reservoir operation within continental and global scale hydrological models. Early works simulated reservoir releases by using a rectangular weir equation or retarding f low velocity (Coe, 2000; Döll et al., 200 3; Meigh et al., 1999) . Pioneering works on grid - based, explicit simulation of reservoirs in global models were presented by Hanasaki et al. (2006) and Haddeland et al. (2006) . A number of subsequent studies have improved and incorporated these schemes into various other global hydrological models (GHM s) and land surface models (LSMs) (e.g., Adam et al., 2007; van Beek et al., 20 11; Biemans et al., 2011; Pokhrel et al., 2015; Pokhrel, Hanasaki, Koirala, et al., 2012; Voisin et al., 2013; Wada et al., 2014) , and other studies have modified them for reservoir - specific applications by including operation rules for the individual r eservoirs (Mateo et al., 2014; G. Zhao et al., 2016) . In this Chapter, an improved reser voir operation scheme based on the original scheme of Hanasaki et al. (2006) is presented with analytical comparisons of existing and new operation schemes , and a new calibrat ion method that is computationally efficient is proposed in Chapter 3.2. The new reservoir scheme incorporated within LHFD model is applied to the CONUS with the various simulations settings (Chapter 3.3), and the simulations results are presented and disc ussed in Chapter 3.4. Lastly, summary and conclusion are given in Chapter 3.5. 42 3.2. I mprove ments on Reservoir Operation Scheme 3.2.1. The Improved Reservoir Operation Scheme Building upon the schemes of Hanasaki et al. (2006) (hereafter, H06) and Biemans et al. (2011) (hereafter, B11) we develop a n enhanced reservoir operation scheme by improving the parameterizations in these existing schemes and adding new capabilities to better simulate reservoir storage and release dynamics. For each operational year, starting with the first month when monthly mean inflow changes from above to below the annual mean flow (Hanasaki et al., 20 06), these schemes determine the total amount of annual release based on the initial reservoir storage and impose the variability of monthly release considering the seasonalit y of water use. There are two common issues in these parameterizations that we ad dress in this study: (1) excessive release in high demand season and (2) unstable storage simulation for small reservoirs. In the following, we introduce the revised scheme fi rst and provide the comparisons with H06 and B11 as necessary. The inter - annual v ariability of release is determined by the release coefficient ( [ - ]), which is the ratio between the initial storage at the beginning of the operatio nal year ( [L 3 ]) and the long - term target storage ( [L 3 - dimensional constant set to 0.85 (Hanasaki et al., 2006) and [L 3 ] is the reservoir storage capacity tak en from GRanD database. The release coefficient is calculated as: (3 - 1) Then, the provisional monthly release ( [L 3 /T]) is calculated as: 43 where and are long - term monthly and annual mean in flow [L 3 /T], respectively, and are monthly and annual mean demand [L 3 /T], respectively, DPI is the ratio between annual mean demand and annual mean inflow ( ) [ - ], and is the ratio between the minimum monthly release and long - term annual mean inflow [ - ] that assures . Regardless of the value of , which can be arbitrarily set to 0.1 (B11), 0.5 (H06), or any other values, the expected value of over the operational year is mathematically equal to . Note that can comprise various demands including domestic, industrial, irrigation, power generation, and others depending on the purpose of reservoirs. By considering the seasonality of various demand s , the reservoir scheme is devised to mimic the generic be havior of reservoirs with respect to their purpose. The targeted monthly release, , can then be calculated as: where (3 - 3) where R is the demand - controlled release rat io [ - ], c is the ratio between capacity and mean annual inflow ( ) [ - ]. As R varies from 0 to 1, the reservoir release changes from run - of - the - river flow to demand - controlled release. 44 In additi on to , all of the excess water is released when the reservoir is full. Conversely, release is prevented when the storage reaches to the minimum level. Finally, reservoir release is calculated under these two constraints of spilli ng and minimum storage level. 3.2.2. Analyti cal Comparison with the Existing Reservoir Operation Schemes We propose the following two major improvements to address the issues identified DPI <1 - M 3 - 2) to prevent excessive release in high demand season; DPI - 2a) and (3 - 2b) aim to fully satisfy water needs in low - demand regions and partially satisfy the needs in high - demand regions, respectively (Hanasaki et al., 2 006). Equation (3 - 2b) is comparable with the hedg ing rule that preserves some water to meet the future demands because high DPI makes a reservoir susceptible to drought conditions. Even if M is adjusted by user preference, equation (3 - 2b) should produce a dampened release pattern compared to the release pattern from equation (3 - 2a). However, unless proper modification of the criterion (i.e., DPI <1 - M ) is followed the release variability gets amplified instead of being dampened. The analytical derivation of D PI <1 - M is provided below. Using equation (3 - 2) we can write: (3 - 4) (3 - 5) (3 - 6) (3 - 7) 45 Where , , , and are the maximum and minimum provisional monthly release calculated from equations (3 - 2a) and (3 - 2b), respectively, and are the maximum and minimum of , respectively, is the minimum release ratio, is the ratio between annual mean demand and annual mean inflow, and is the annual mean flow. Then, the differences of maximum and minimum values between equations (3 - 2a) and (3 - 2b), and , are: (3 - 8) (3 - 9) Compared to the use of equation (3 - 2a), if the us e of equation (3 - 2b) enables less variability in release, and should be positive and negative, respectively. Since , , , the equation (3 - 2a) should be used when (3 - 10) Second, we propose using instead of (H06; B11). Here we provide an analytical derivation of a new equation for R . The target s torage, , can be calculated from a simple mass balance as follows: (3 - 11) After the rearrangement of equation (3 - 11), ( 3 - 12 ) 46 The relationship between initial storage and t a rget storage for varying c of equation (3 - 12) is shown in Figure 3 - 1 a, which demonstrates the need for new parameterization of R in e quation (3 - 3) . The equation (3 - 3) is intended to prevent overflow and storag e depletion when reservoir storage capacity is small compared to annual flow (i.e., lower case ) by allowing some portion of run - of - the - river flow (Hanasaki et al., 2006). The need of equation ( 3 - 3) comes from the assumption in eq uation ( 3 - 1) where the initial storage scaled t o the reservoir storage capacity (i.e., ) serves as a surrogate for potential outflow scaled to the expected inflow (i.e., ) to achieve a long - term target st orage (i.e., ). Henc e , the rate of outflow variation (e.g., 20 % reduction/increment in outflow) contributes to the rate of storage change by times of it (e.g., 20/ c % reduction/increment in storage). As a result, when is l arge, equation (3 - 1) allows reservoir storage to be gradually guided to rather than abruptly filling or releasing the storage up to within a year; conversely, when c is small, unless a proper reduction of R is followed, the storage simulation is expected to be unstable since the target storage ( ) is repeatedly set to either or that causes frequent over - filling and under - filli n g; hence the reduction of R is needed for those cases. 47 Figure 3 - 1 Demonstration of the need to use . (a) The relationship between initial storage and target storage for varying c when R is not reduced ( R =1), and (b) the comparison of various equations for R for four groups of c values. Therefore, f or the stability of reservoir storage, the slope term of should be positive (Figure 3 - 1 ). Otherwise, the t r ansition between over - fill ing and under - filling of reservoir storage will be repeated. The unstable reservoir storage can be simulated for reservoirs having the following c value. ( 3 - 13 ) For reservoirs satisfying equation ( 3 - 13), the stability of reservoir storage simulation can be enhanced by setting the value of R to less than 1. The principle is that the portion of run - of - the - river has an effect of decreasing inflow and outflow terms in the equation (3 - 11) to and , respectively. When the effect of R is considered, the slope term between initial storage and target storage can be rewritten as . Hence, the stabil ity of reservoir storage simulation can be enhanced when R is set as follows. (3 - 14) 48 Hence, the necessary condition of R for the stability of storage simulation is: (3 - 15) T he reservoirs having a re found to require a reduction of R as shown in the relationship between and (Figure 3 - 1 a). In contrast, the existing equation (i.e., ) reduces R values only for some of thos e reservoirs (i.e., ), hence can result in unstable storage simulation, specifically for reservoirs having (Fig ure 3 - 1 a, dotted lines). For example, for a reservoir having , when the initial storage is 30% and 95% of capacity, is set to 159% and 71% of capacity, respectively (Figure 3 - 1 a, pink dotted line). Amon g the feasible cases of reducing R , the necessary condition to en s ure reservoir storage stability is when , which can be provided by the new equation: . To effectively compare with , 4 groups of c values are defined a s follows: Groups I, II, III, and IV for , , , and , respec tively (Figure 3 - 1 b). Compared to , can potentially cause a reduced stability for reservoirs having (Group I; Figure 3b ), however, those reservoirs already have more than 82% of outflow as run - of - the - river flow (i.e., a relatively small reservoir influence on flows). I n addition, the difference in outflow portion of run - of - the - river flow between the old and new equations i s only less than 5%. Meanwhile, the necessary condition ensures a higher stability for reservoirs having (Group II; Figure 3b ), and newly introduces stability for reservoirs having which are not 49 covered by the original equation (Group III ; Figure 3 - 1 b). The new equation can also be compared with other univariate power functions of c , i.e., , which have a shape parameter of n . The equations with can results in an increased (when ) or reduced (when ) stability compared to the necessary condition (F i gure 3 - 1 b, dotted lines). In this study, to ensure a desired storage stability while satisfying the downstream demand, R is parameterized using . 3.2.3. The Calibration of R provides the necessary condition f o r R and c (see Chapter 3.2.2); however, R is not necessarily a p ower function or a univariate function of c because additional variables can be incorporated to define R in the form of polynomial or conditional equations. In addition, the current formulations using i m , d m , and R to impose seasonal variation of reservoir operation can be further improved in different ways; here we propose doing so by calibrating R . Specifically, we calibrate R for reservoirs having release and storage observations for our simulation period of 1983 - 2010; for other reservoirs we use a general function of . The specific objectives of calibrating R are to (1) improve the simulated release and storage of individual reservoir s, (2) examine the appropriateness of newly developed equation for R , i.e., , and (3) identify potential improvements on the reservoir operation scheme. The first objective is achieved by determining optimal R values, and the latt er two are accomplished by comparing R values from and calibration. Due to the interdependence between upstream and downstream reservoirs (Taeb et al., 2017) , iterative model si mulations are necessary for a concurrent calibration of multiple 50 reservoirs. Such iterative simulations involve excessive computational costs for high - resolution modeling at continental to global scales . To overcome this difficulty, we propose a calibratio n approach that utilizes the time series of river discharge simulated without considering dams and sequentially calibrates R from upstream to downstream. The rationale is that the inflow to a reservoir is determined by the releases from its immediate upstr eam reservoirs and the unregulated upstream river flows. The inflow to a reservoir having K of immediate upstream reservoirs is estimated as: (3 - 16) where is the inflow to the reservoir at time t , is the outflow of the k - th immediate upstream reservoir, is the river discharge from the without - dam simulation at the reservoir location and time t , K is the number of immediate upstream reservoirs, and T is the number of time steps. When K =0, is identical to . Using the inflow from equation (4) and the prescribed operation rule ( Chapter 3 ), release and storage are sequentially calculated from the upstream to downstream for all reservoirs. Notably, if observed discharge (i.e., release) is available, R is calibrated to minimize Root Mean Squared Error (RMSE) of release. For reservoirs where observatio ns are unavailable, the new equation is used to determine R . By doing so, while the entire hydrological model is not required to be iteratively run, the calibrated results of upstream reservoirs are reflected in the cal ibration o f downstream reservoirs. To pursue the objectives enlisted above under the uncertainties in large scale 51 modeling, we use the entire period for the calibration to find the optimal R values that explain the given release data best rather than disti nguishing the calibration and validation periods. 3.3. Experimental Settings Altogether six simulations for NAT (without dams), H06 (Hanasaki et al., 2006) , B11 (Biemans et al., 2011) , R old , R new , and R cal are conducted as summarized in Table 3 - 1. Note that all six simulations with different reservoir operation rules are conducted using the same settings for model parameters and forcing. While M can be set as a spatially varying parameter, we use it as a constant number either 0.1 (B11) or 0.5 (H06) (see Table 3 - 1) to make our resul ts comparable to the existing schemes that used a constant M . Considering the availability of USGS water use data, s imulations are conducted for the period of 1983 - 2010. A 20 - year spinup simulation is first conducted without considering reservoirs. Further , the first two years of simulations are discarded as spinup for reservoirs, thus 26 years of results (1985 - 2010) ar e analyzed . Of the 1,889 reservoirs imported from the GRanD database for the CONUS domain, 27 reservoirs are selected for calibration (Figur e 2 - 2a and Table 3 - 2), for which the long - term storage and release data are available from USGS , California Data Exc hange Center, US army corps of engineers, and US Bureau of Reclamation. Table 3 - 1 Summary of e xperimental s ettings Simulation 1 M Criterion in equation (3 - 2) 2 Equation for R Calibration of R NAT - - - - H06 0.5 - B11 0.1 - R old 0.1 - R new 0.1 - R cal 0.1 1 NAT: without dam simulation, H06: Hanasaki et al. (2006), B11: Biemans et al. (2011) 2 The criterion of H06 can also be written to sinc e =0.5 52 Table 3 - 2 The attributes of reservoirs selected for calibration (Source: GRanD database) Groups 1 ID 2 Reservoir Name Dam Name River (Lat, Lon) Storage Capacity (Mm 3 ) Main Purpose Data Source 3 I 31 0 Franklin D. Roosevelt Grand Coulee Columbia (47.95, - 118.98) 6395.6 Irrigation USBR 391 Lucky Peak Lake Lucky Peak Columbia (43.53, - 116.05) 378.7 Flood control USBR 423 Fontenelle Fontenelle Colorado (42.03, - 110.07) 185.6 Hydroelectricity USBR 88 4 Lake Sharpe Big Bend Dam Missouri (44.04, - 99.45) 2343.6 Flood control US Army 895 Lewis and Clark Lake Gavins Point Dam Missouri (42.85, - 97.49) 666.1 Flood control US Army II 101 Wickiup Reservoir Wickiup Reservoir Columbia (43.68, - 121.69) 267.0 Ir ri gation USBR 182 Folsom Lake Folsom Sacramento (38.71, - 121.16) 1102.7 Irrigation CDEC 396 Palisades Palisades Columbia (43.33, - 111.20) 1480.2 Irrigation USBR 411 American Falls American Falls Columbia (42.78, - 112.87) 2061.5 Irrigation USBR III 6 3 Rimrock Tieton Colorado (46.66, - 121.13) 244.2 Irrigation USBR 132 Shasta Lake Shasta Sacramento (40.72, - 122.42) 4890.7 Irrigation CDEC 148 Oroville Sacramento (39.54, - 121.48) 4366.5 Flood control CDEC 370 Lake Cascade Cascade Columbia (44.52, - 11 6.05) 805.5 Irrigation USBR 384 Jackson Lake Jackson Lake Columbia (43.86, - 110.59) 1076.8 Irrigation USBR 394 Anderson Ranch Anderson Ranch Columbia (43.36, - 115.45) 521.8 Irrigation USBR 541 Blue Mesa Blue Mesa Colorado (38.45, - 107.33) 923.2 Hyd ro electricity USBR IV 131 Clair Engle Lake Trinity Klamath (40.80, - 122.76) 2633.5 Irrigation CDEC 198 New Melones New Melones San Joaquin (37.95, - 120.52) 2985.0 Irrigation CDEC 210 Don Pedro San Joaquin (37.70, - 120.42) 2504.0 Irrigation CDEC 30 7 Fort Peck Lake Fort Peck Dam Missouri (48.00, - 106.41) 23560.0 Flood control USAC 386 Lake Owyhee Owyhee Columbia (43.64, - 117.24) 881.9 Irrigation USBR 451 Flaming Gorge Flaming Gorge Colorado (40.92, - 109.42) 4336.3 Water supply USBR 597 Lake Pow el l Glen Canyon Colorado (36.94, - 111.49) 25070.0 Hydroelectricity USBR 601 Navajo Navajo Colorado (36.80, - 107.61) 1278.0 Irrigation USBR 610 Lake Mead Hoover Colorado (36.02, - 114.73) 36700.0 Water supply CDEC 753 Lake Sakakawea Garrison Dam Missou ri (47.51, - 101.43) 30220.0 Flood control US Army 870 Lake Oahe Oahe Dam Missouri (44.46, - 100.40) 29110.0 Flood control US Army 1 Groups are classified by storage capacity and annual mean flow ( Chapter 3). 2 Global Reservoir and Dam (GRanD) Database I D 3 USBR: United States Bureau of Reclamation, CDEC: California Data Exchange Center, and US Army: United States Army Corps of Engineers 53 3.4. Results and Discussion 3.4.1. LHFD Model Validation for the CONUS The main focus of C hapter 3 is on simulating reservoir sto ra ge and release, but we first briefly discuss the evaluation of river discharge simulations in the CONUS scale. Because the model has been extensively evaluated over the North America as well as the Amazon in pr evious studies (Fan et al., 2007; Miguez - Macho et al., 2007; Miguez - Macho & Fan, 2012a, 2012b; Y. Pokhrel et al., 2014) , we revisit this briefly for annual strea mflow and seasonality of flow. Overall, annu al streamflow is accurately simulated in the major river basins (Figure 3 - 2). The seasonality of flow is also reproduced well in many large river basins, but further improvements are needed especially for small b asins (Figure 3 - 3 ). Since the site - specific ca libration is not conducted yet for the current version of LHFD except for the calibration of R values for 27 reservoirs, the hydrographs are expected to be improved by calibrations. The impact of reservoir oper ation on river flows is found to be large in s ome rivers (e.g., Colombia, Colorado, and Missouri rivers) and to be small in other rivers (e.g., Mississippi and Ohio rivers). Figure 3 - 2 Model validation for annual mean flow over the contiguous US. ( a) Comparison of simulated and observed annual mean flows averaged over the period 1983 - 2010 for (b) 96 USGS streamflow gauge stations. The 18 USGS stations loc ated near the confluences (indicated in blue) are chosen to show their hydrographs in Figure 3 - 3 . The other stations are indicated in red. 54 Figure 3 - 3 Validation of seasonal river discharge at the major ga uging stations over the contiguous US (unit: 10 3 m 3 /s). The locations of USGS gauge stations ar e in dicated in Figure 3 - 2 (b). 55 3.4.2. Improvements in Reservoir Release and Storage We present the evaluation of simulated reservoir release and storage from NAT, H06, B11, R old , R new , and R cal simulations (see Table 3 - 1 for experiment settings) with the observed d at a obtained from multiple sources (see Chapter 3.3). As discussed in Chapter 3.2.2, the results of release and reservoir storage can be categorized into four groups by c values defined in Figure 3 - 1. Hence, among 27 reservoirs selected for calibration o f R, we present the evaluation for six reservoirs, which are located in different geographic regions (i.e., Columbia, Colorado, San Joaquin, Sacramento, and Missouri river basins) and provide a good coverage of different group s (i.e., Groups I, II, III, IV ), i n Figure 3 - 4 with a summary of performance measures in terms of correlation and Root Mean Square Error (RMSE) in Table 3 - 3. The results for additional reservoirs are provided in Figure 3 - 5 . 56 Figure 3 - 4 Ev a luation of simulated release and storage for the six selected reservoirs in the Columbia, Colorado, San Joaquin, Sacramento, and Missouri river basins. NAT, H06, B11, R old , R new , and R cal denote different simulation settings ( Chapter 3.3). The panels o n t he right show the seasonal cycle. 57 Table 3 - 3 Summary of p erformance m easures for d ifferent s imulations River flow (Reservoir Release) Reservoir Storage NAT H06 B11 R old R new R cal H06 B11 R old R new R cal ( a) Big Bend Dam (Lake Sharpe, Missouri river, Group I) R - value - 0.04 0.04 0.04 0.08 1.00 R - value 0.04 0.04 0.04 0.08 1.00 CORR 1 - 0.11 - 0.05 - 0.05 - 0.05 - 0.07 0.10 CORR - 0.02 - 0.03 - 0.02 - 0.02 0.07 RMSE 2 1040 461 464 455 460 378 RMSE 1087 1085 1086 10 79 1 216 (b) Folsom (Folsom Lake, Sacramento river, Group II) R - value - 0.96 0.96 0.96 0.42 0.56 R - value 0.96 0.96 0.96 0.42 0.56 CORR 0.77 0.64 0.58 0.59 0.63 0.61 CORR 0.45 0.42 0.42 0.47 0.49 RMSE 69 83 91 90 84 85 RMSE 399 416 417 359 337 (c) Jacks on L ake (Jackson Lake, Columbia river, Group III) R - value - 1.00 1.00 1.00 0.80 0.70 R - value 1.00 1.00 1.00 0.80 0.70 CORR 0.03 0.26 0.29 0.28 0.31 0.31 CORR 0.32 0.33 0.33 0.45 0.48 RMSE 87 61 64 65 56 52 RMSE 398 386 392 322 299 (d) New Melones (New M el ones, San Joaquin river, Group IV) R - value - 1.00 1.00 1.00 1.00 0.50 1.00 1.00 1.00 1.00 0.50 CORR 0.34 0.30 0.25 0.26 0.27 0.40 CORR 0.87 0.86 0.86 0.86 0.89 RMSE 37 34 42 39 39 30 RMSE 383 394 391 401 345 (e) Glen Canyon (Lake Powell, Colorado ri ve r, Group IV) R - value - 1.00 1.00 1.00 1.00 0.87 1.00 1.00 1.00 1.00 0.87 CORR - 0.01 0.24 0.25 0.23 0.23 0.25 CORR 0.23 0.24 0.23 0.22 0.29 RMSE 889 239 239 243 244 253 RMSE 6438 6420 6476 6532 6322 (f) Garrison Dam (Lake Sakakawea, Missouri river, G ro up IV) R - value - 1.00 1.00 1.00 1.00 1.00 R - value 1.00 1.00 1.00 1.00 1.00 CORR - 0.05 0.22 0.21 0.22 0.20 0.21 CORR 0.28 0.27 0.26 0.26 0.26 RMSE 982 303 304 305 311 303 RMSE 5302 5339 5318 5215 5197 1 Correlation 2 Root Mean Square Error 58 Figure 3 - 5 Same as in Figure 3 - 4 but for additional reservoirs 59 Figure 3 - 5 (continued) Results of individual reservoir release and storage suggest that (1) the large seasonal variation of river flow in the simula ti on without reservoirs is dampened by the inclusion of reservoirs (comparison of NAT and others), (2) the new criterion in equation ( 3 - 2) (i.e., DPI >1 - M ) prevents excessive release in high - demand season (comparison of B11, R old , R new, and R cal ), (3) the u se o f new equation of R mitigates the issues of storage depletion, particularly for 60 reservoirs in Groups II and III (comparison of H06, B11, R old , R new, and R cal ), (4) the proper adjustment of R value improves seasonal dynamics of the release and/or storag e in most cases (comparison of all simulations), and (5) all reservoir schemes are found to exhibit poor performance for small reservoirs (Group I), suggesting t he need of further improvements (comparison of all). As the relative size of storage capacity to m ean flow (i.e., c ) increases (i.e., a transition from Group I toward Group IV in Figure 3 - 1 b), the reduction in peak becomes more pronounced as the reservoir s can retain a large portion of the excess inflow during wet season and release it later during d ry ing seasons (Figure 3 - 4 def). On the contrary, for reservoirs having small c (e.g., Figure 3 - 4 abc), the magnitude of variability in release may not decrease a s substantially since the excessive inflows easily fill the reservoir storages up to their capa ci ti es, beyond which the remainder is released as spillway overflow. When c is very small (i.e., Group I), adjusting R may not necessarily influence reservoir st orage stabilization (Figure 3 - 4 a). However, with moderate c values (i.e., Groups II and III), r ed uc ing R becomes crucial, and the use of new equation for R effectively enhances the release and storage simulations (Figure 3 - 4 bc). Specifically, for the reser voirs in Group III, the equation for R from H06 (i.e., ) always sets R = 1, and hence the amplified release patterns and the consequent over - and under - filling storage patterns are simulated due to the periodic excessive inflows (Fi gure 3 - 4 c). Similarly, the overestimated R values by for reservoirs in G roup II can also have frequent over - and under - filling storage patterns (Figure 3 - 4 b). After the proper adjustment of R value by using the newly proposed equ ation (i.e., ), the issues of amplified release and unstable storage ar e mitigated, and the releases and storages of reservoirs in Groups II and III are significantly 61 improved (Figure 3 - 4 bc). The comparable R values from (i.e., R new ) and the calibration (i.e., R cal ) demonstrate the appropriateness o f new equation, specifically for reservoirs in Groups II and III. 3.4.3. The Role of R in Release and Storage Simulations The R values of the reservoirs in Group IV (e.g., New Melones, Lake Powell, and Lake Sakakawea; Figure 3 - 4def) are calculated to be 1 by bo th and equations. Consequently, many reservoirs in Group IV show sim ilar storage - release patterns for H06, B11, R old , and R new simulations. For R cal simulation, depending on the calibration effect, the ma gn itudes of changes in release and storage can vary substantially. For example, the release and storage of New Melones ( Figure 3 - 4d) with R =1 and R =0.5, before and after calibration, respectively, show significantly different trends and patterns. Meanwhi le , the calibrated R for Lake Powell (Figure 3 - 4e) is rather high (i.e., R =0.87; Table 3 - 3), which imposes a relatively sm all interannual seasonality in release. Further, for Lake Sakakawea (Figure 3 - 4f), calibration yields R =1 (i.e., identical to that fro m bo th old and new equations), making the changes in release and storage insignificant. However, the release and storage f or New Melones are significantly improved when R is calibrated, demonstrating that results could be potentially improved by further im pr ov ing the R equation which is currently a univariate power function of c . As shown for Lake Sakakawea, certain advanceme nts beyond improving R could also be necessary since seasonal variation in natural and actual release can differ from each other. Dis ti nc t results are derived from the calibration of R for cascade of reservoirs such as the Lake Sharpe (Figure 3 - 4a) located in the main stem of Missouri river that has three large upstream reservoirs: Lake Oahe, Lake Sakakawea (Figure 3 - 4f), and Fort Peck La ke . Due to high 62 of mai n river stem and small , the R value of Lake Sharpe is set to 0.04 and 0.08 by and equations, respectively. With an uncalibra te d (small) R , r m is determined to be similar to (Figure 3 - 4a, green line in right panel ) which results in a highly variable release patterns in H06, B11, R old , and R new simulations; for example, the maximum amount of monthly tar ge t release (i.e., ) is ~ 3 times of storage capacity that inevitably causes an oscillating storage simulation. Conversely, the calibration of R reflects the steady patterns of observed release and storage (Figure 5a, black lines) to d etermine r m ; r m is determined to be steady when R is set to 1 (by calibration) and d m is small (as given). In R cal simulation, the inflow to Lake Sharpe is simulated to be steady except for the peaks during 1997 - 1998 due to the large upstream reservoir s in cascade (see Figure 3 - 4f). While the oscillation is not entirely eliminated despite steady r m and stable inflow, its frequency is reduced. It is noteworthy to recall that the reservoirs having low (i.e., Group I) hardly reco ve r the storage stability once deviates significantly from (see discussions in Chapter 3.2.2). In addition, the calibration increases R (i.e., boosting the instability) instead of decreasing it. Since t he c urrent simulations start from natural river storages with small for the 1 st operational year, is ove re stimated and results in large for the 2 nd year; for the 2 nd year is underestimated that results in small for the 3 rd year and vice versa. Even if is set exactly equal to , a given and several consequent operation years w il l have steady reservoir storage at most, but even with small changes in year - to - year inflow from the upstream (Figure 3 - 4f) or occasional large inflows (e.g., flooding pulses during 1997 - 1998) reservoir storage can readily begin to oscillate. 63 Even if th e inter - annual inflow is stable and a steady intra - annual release is planned to Lake Sharpe in R cal simulation, storage is not stabilized even after 28 years. Based on the reasoning above, the reservoir storage of Lake Sharpe will remain unstable regardl es s of an extended simulation period. Considering a common rationale in hydrological modeling that ex tended model runs with steady input will result in equilibrium states, the results of Lake Sharpe are counter - intuitive. Such modeling issues have not been e xp lored in previous studies because their focus has been on reproducing the large - scale patterns, e specially of reservoir release. When the grid resolution is increased and smaller reservoirs are considered, the issues encountered in Lake Sharpe can unin te nt ionally and ubiquitously spread over a large domain, which can adversely affect the magnitude and timing of simulated surface water fluxes and cannot be addressed by an extended spinup. This calls for caution in using the proposed scheme when applied f or s mall reservoirs, and the need for further improvements by developing generic reservoir operation rule that can provide potentially stable results for small reservoirs and through incorporation of reservoir - specific operation rules. 3.4.4. Stability of Reserv oi r Simulation over the CONUS As discussed above, the stability of reservoir storage is significantly improved in the enhanced scheme, but the stability varies largely among reservoirs in the four different groups. The improvements in stability are examine d by using a frequency analysis for over - filled and under - filled days for all 1,889 reservoirs (Figur e 3 - 6). Over - filled and under - filled days are defined as days when a reservoir storage is >95% of the maximum capacity and <105% of the minimum storage, re sp ec tively. While overfilling and under - filling can happen in real - world operation, too frequent occu rrences in simulations arise from the deficiencies in reservoir release parameterizations described in Chapter 3.2.2. When the over - filled and under - filled d ay s of H06, 64 B11, and R old are compared for each group of reservoirs using Kruskal - 0.05), none of the groups shows statistically significant differences for both over - filled and under - filled days. When H06, B11, R old , R new , and R cal are t es te d, statistically significant differences are found for the over - filled days in group II (p=1e - 7) and group III (p=5e - 17) and for the under - filled days in group III (p=2e - 5). From the results of decreasing quantiles and average of over - filled and under - fi ll ed days, it is concluded that the new reservoir operation scheme mitigates the problem of unstabl e reservoir storage simulation, especially for reservoirs in Groups II and III. In terms of enhancing the storage stability, the contribution of the new R eq ua tion (i.e., ) (compared to R old and R new ) is found to be larger than t he contribution of the new criterion of equation (3 - 2) (i.e., DPI <1 - M ) (in comparison of H06, B11, and R old ). The reservoirs in Group I and Group IV are le ss i nfluenced by different reservoir operation schemes due to inherent and large variability (as disc ussed with Lake Sharpe in Figure 3 - 4a) and the identical R =1 values, respectively. 65 Figure 3 - 6 Boxplot sh ow in g stability of simulated reservoir storages for 1,889 GRanD reservoirs estimated from frequency analysis for the occurrence of (a) overfilling and (b) underfilling. 3.4.5. Flow Regime Change due to Reservoirs Finally, we examine the effects of rese rvoir opera ti on on river flows having different exceedance probabilities to ensure that the reservoir - induced changes in extreme flows are well simulated over the entire CONUS domain. Figure 3 - 7 presents the difference between Q DAM (R cal simulation) and Q N AT (NAT sim ul at ion) presented as percentage change relative to Q NAT for exceedance probabilities of 90% (low flow; Q 90 ), 50% (median flow; Q 50 ), and 10% (high flow; Q 10 ) (Figure 3 - 7). Expectedly, the Q 90 is increased and Q 10 is decreased in much of the stu dy domain ( Fi gu res 3 - 7a and 3 - 7c). Further, an increase in the median flow can be seen in most regions (Figure 3 - 7b). Meanwhile, sporadic decreases in Q 90 and Q 50 and increase in Q 10 can also be seen, similar to that found in previous studies that examined the impact s of dams on extreme flows. This is likely due to the following two reasons. First, frequent over - and under - filling of 66 relatively small reservoirs (Figure 3 - 6) potentially cause a reduction in low flow and an increase in high flow as can be see n in the lo we r portion of the Mississippi river basin where there are large number of such small reservoirs. Second, in the river - floodplain parameterizations in LHF model, the channel geometry and elevations of tributaries are sensitive to the water level in the mai n st em of the river, which can cause a magnified flow pattern in the tributaries, especially when flows are modified through reservoir regulation. For example, if river - reservoir bed elevations of tributaries are parameterized to be rather flat, when the w at er level at the main stem, which acts as the lower boundary condition for tributaries, is modified by reservoir operation, the variability of flow and water level in the tributary could increase. Despite some unintended outcomes due to the int erference b et we en river - floodplain and reservoir parameterizations, the overall patterns of increased low flow and decreased high flows are reproduced by the model. This is particularity true for large rivers having reservoirs with large storage capacity. 67 Figure 3 - 7 Effect of reservoir operation on long - term average daily river discharge with the exceedance probability of (a) 90% (low flow; Q 90 ), (b) 50% (median flow; Q 50 ), and (c) 10% (high flow; Q 10 ), sh own as rela ti ve change between Q DAM (i.e., R cal ) and Q NAT (i.e., NAT simulation). To avoid overly exaggerated relative change, results for grid cells having very small river discharge (i.e., Q 99 < 1 m 3 /sec) are represented to have 10% change (light yellow color). 68 3.5. S um ma ry and Conclusion The temporal dynamics of reservoir storage and release is improved by a novel parameterization that enables stabilized storage and release simulations and by employing a computationally - efficient calibration method that does not requi re i terative reservoir simulations over the entire domain. S imulated results of reservoir release and storage for the selected reservoirs are evaluated with observations, and the importance o f resolving certain unintended behaviors in the previously used r es er voir operation schemes is discussed. The proposed model improvements result in a better simulation of the seasonal dynamics of reservoir storage; however, further improvements are needed to account for diverse features of real - world reservoir operation r ul es. From the calibration of R , the limitations of generic reservoir operation rules that imposing release seasonality using demand seasonality are identified , and t he potential further im provements on the formulation of R are found . S pecifically, furth er i mprovements are necessary to enhance simulation stability for small reservoirs (i.e., Group I) and better simulate the seasonality in storage and release for large reservoirs (i.e., Group IV). In addition, t he unintended effects of reservoir operation on e xtreme flows due to main stem - tributary interference in regions with small reservoirs could be improved by enhancing the proposed reservoir release scheme. There are many other important issues in reservoir modeling which should be addressed in the fol lo wi ng studies . For example, a novel approach is needed to implement the optimal operation of multi - objective cascade reservoirs (e.g., Huang et al., 2002; Yang et al., 2017) in l ar ge - scale mod (Bellman, 1961; Labadie, 2004) and high uncertainties. Future studies shoul d al so conside r incorporating long - distance water transfer from reservoirs, which can be a significant water balance component in 69 some regions and has been increasingly discussed in recent literature (Hanasaki et al., 2018; Y. Pokhrel et al., 201 6; W ada et al., 2017) . 70 Chapter 4. Sensitivity Analysis for the Effect of Upstream Flow Regulation on Flood Dynamics in the Lower Mekong River Basin and Tonle Sap Lake 4.1. Introduction In Chapter 4, the question s of Q 3 . What are the implications of potential flow regu la ti on by new dams on downstream flood inundation dynamics ? is answered by examining different scenarios of flow regulation designed by altering the magnitude and timing of flood pulse near Stung Treng, a location in the vicinity of the proposed site for th e massive Sa mbor dam with 18km barrier, which, if built, is feared to severely fragment the river dolphin population and block fish migration (Brownell et al., 2017; Fawthrop, 2018) . This approach is novel in that we use the altered flow patterns over scenari os o f flow regulation by the existing and planned dams because the existing dams have caused little hydrologic alterations in mainstream flows (Arias, Piman, et al., 2014) , and the number, construction time, and size of dams to be built remains highly uncertain. It is note d that the flow alteration patterns in Chapter 4 are not designed to perfectly capture the actual flow regulati on b y any specific future dam but to provide a framework for sensitivity analysis under different degree of flow regulation by a single of multiple dams in the upstream of the location where we alter the flows. Further, while future dams could be operated fo r different purposes, our analysis is based on flow regulation patterns of hydropower and flood control dams whereby a reduced flood peak is accompanied by an increased low flow. Thu s, the approach enables a mechanistic understanding of the changes in do wn st ream flooding under different levels of potential flow regulation by any upstream dams. Further, while previous studies have mostly focused on the TSL, a relatively complete pictur e of flood dynamics and its consequent TWS variations across the entire MR B is provided. By doing 71 Q 4 . What role does the flood dynamics play in modulating the overall hydrology of the basin? also addressed . 72 4.2. Materials and Methods 4.2.1. HiGW - MAT HiGW - MAT (Y. Pokh re l et al., 2015) is a global hydrological model based on the global land surface model (LSM) MATSIRO (Takata et al., 2003) coupled with the river routing model TRIP (T ai kan Oki & Sud, 1998) . HiGW - MAT simulates both the natural water cycle and human activities such as irrigation, flow regulation, and groundwater pumping, but the human impact schemes are turned off b ecause the objective here is to examine the effects o f po tential flow regulation by future dams, not limited to the existing ones; note that the existing dams have caused little impact on the Mekong flow (Arias, Piman, et al., 2014) . MATSIRO simulates key vegetation, surface hydrological, soil moisture, and groundwater processes on a full physical basis. A complete description can be found in Takata et al. (2003) , with further details on its recent impro vements in the studies of Pokhrel and coauthors (Y. Pokhrel et al., 2015, 2016; Y. Pokhrel, Hanasaki, Koirala, et al., 2012) . 4.2.2. CaMa - Flood CaMa - Flood (D . Yamazaki et al., 2011, 2014) is a hydrodynamic model, which computes river hydrodynamics (i.e., river discharge, flow velocity, water level, and inundated area) by solving shallow water equation of open channel flow, explicitly accounting for backwa te r effects using the local inertial approximation (D. Yamazaki et al., 2013) . The physics of CaMa - Flood is similar to the river - floodplain routing module of LHF (see Chapter 2.2.1). In this chapter, CaMa - Flood version - 3.6 is used with regional settings at 10km resolution fo r the MRB (D. Yamazaki et al., 2014) , w hich includes the capability for downscaling output to 500m grids; version - 3.6 accounts for channel bifurcation, a critically important process to realistically simulate river - floodpl ain dynamics in the Mekong Delta. In CaMa - Flood, w ater level and inun da te d area s are 73 diagnosed from water storage in each unit catchment; river discharge from each unit catchment is calculated using the shallow water equation; water storage at each unit catchment is updated by a mass conservation equation considering discha rg e input from the upstream unit catchment(s), discharge output to the downstream unit cat chment, and local runoff input from HiGW - MAT. The 10km river network map is generated by u pscaling the 3 arc - second (90m) HydroSHEDS flow direction map (Lehner et al., 2008) and digital elevation model from SRTM3 DEM (D. Yamazaki et al., 2014) s is set basin wide at 0.03 and 0.10, respectively, following Yamazaki et al . (2011, 2012, 2014) ; sensitivity of the coefficient to model results is discussed in Yamaz aki et al. (2011) . All other model parameters including river width are identical to those in Yamazaki et al. (2014) . 4.2.3. Simulation S ettings First, the HiGW - MAT model is used to simulate runoff and all TWS components (i.e., soil moisture, snow, river sto ra ge , and groundwater) for 1979 - 2010 period using identical settings, parameters, and forcing data as in Pokhrel et al. (2015) ; the first two years a re discarded as spinup; results for 1981 - 2010 are analyzed. Since HiGW - MAT i s a global model, results for the MRB (90 - 110ºE, 5 - 35ºN) are extracted from global simulations at 1º × 1º grids. Runoff is used to drive the CaMa - Flood model and the storage componen ts are used for TWS analysis. Note that the HiGW - MAT is used for a single si mu la tion without considering flow alterations. Then, similar ly to the previous studies (D. Yamazaki et al., 2014; F. Zhao et al., 2017) , daily runoff from HiGW - MAT is used in CaMa - Flood to simulate r iv er - floodplain hydrodynamics at 10km over the MRB. The 10km resolution flood depth is then downscaled to 500m grids using SRTM3 high - resolution DEM assuming that the water levels of 500m grids 74 within a 10km grid is identical (D. Yamazaki et al., 2014) . A series of simulations are conducted f or : (1) 1981 - 2010 period using continuous HiGW - MAT runoff and with no flow alterations ; (2) an average year, using the climatological mean daily runoff for 1981 - 2010 period; (3) a historical dry year (1998); and (4) a historical wet year (2000). For (2), ( 3) , and (4), a baseline simulation is firstly conducted without flow alterations. Then , simulations with different degree of dampened flood peak (i.e., by 10, 20, 30, 40, and 50%) and early and delayed arrival of the peak by one month (see details below) a re conducted. Table 4 - 1 Geographic lo cation of stations in MRB Station Name Latitude Longitude Classification LP Luang Prabang 19.89 102.14 Mainstream Mekong PA Pakse 15.12 105.80 Mainstream Mekong ST S tu ng Treng 13.53 105.95 Mainstream Mekong Dam Dam 13.55 105.95 Mainstream Mekong KT Kratie 12.49 106.02 Mainstream Mekong KC Kampong Cham 12.00 105.47 Mainstream Mekong KL Kompong Luong 12.58 104.22 Tonle Sap R i ver LO Lake Outlet 12.52 104.47 Tonle Sa p R i ver PK Prek Kdam 11.81 104.80 Tonle Sap R i ver PP Phnom Penh Port 11.58 104.92 Mainstream Mekong NL Neak Luong 11.26 105.28 Mainstream Mekong KK Koh Khel 11.27 105.02 Bassac River CD Chau Doc 10.70 105.13 Bassac River VN Vam Nao 10.58 105.36 Mains tr ea m Mekong MTu My Thuan 10.27 105.92 Mekong Delta MTo My Tho 10.36 106.37 Mekong Delta CT Can Tho 10.03 105.79 Bassac River/Song Hau To generate the altered flood pulse patterns as a surrogate of flow regulations by future dams, we change the timing a nd magnitude of flood peak near Stung Treng gauging station (13.53°N, 105.9 5 °E ; Table 4 - 1 ) in the Mekong river, immediate downstream of the confluence of the 3S river systems, a location near the proposed site for the massive Sambor dam (Fawthrop, 2018) . This approach enables us to mechanistically examine the changes in flood magnitude, 75 timing, duration, and extent under different levels of dam regulations or altered flow patterns due to climate change. Although the m aj ority of the proposed large dams are likely to be used for hydropower generation, no information is available on how these dams will be operated. However, as most dams do, the new dams will alter the magnitude and timing of river flow by attenuating th e pe ak and increasing low flow. Thus, to capture these altered flow patterns, we generate a proxy of dam release using a release equation modified from the reservoir operation rule proposed in Chapter 3 , which can be written as: (4 - 1) Where, Q i,DAM is the altered flow, Q i,NAT is the simulated natural flow at the dam location, and Q mean is the mean annual natural flow for each operational year. M is a calibration parameter that determines the release; e.g., when M is u ni ty , the equation represents a constant release throug hout the year, and when M is zero, release is equal to the natural flow, representing no reservoir effect. Here, we calibrate M to attenuate peak by 10, 20, 30, 40, and 50% from the baseline (i.e., ave ra ge year) flow. Because Q mean is different for each ye ar, M values are differently calibrated among years, i.e., average (1981 - 2010 mean), dry (1998), and wet (2000), to maintain the same degree of peak flow attenuation. Once Q mean and M are determined, Q i, DA M is generated using equation ( 4 - 1) that produces e nhanced low - flow to compensate for peak flow reduction, preserving water balance. For the scenarios with altered timing of peak, Q i,DAM is derived by shifting the peak of the hydrograph one month earli er o r later. Note that the scenarios of altered timing are analyzed only for 10, 30, and 50% peak flow attenuation scenarios. These scenarios are designed to reflect the compounded impacts of flow regulation and climate change to flood dynamics under the u nc er tainties in climate change as well as in number, 76 sp ecifications, and operation rules of future dams. Here, t he reduction in magnitude and a delay in timing of the flood peak are typically caused by hydropower and flood - control dams. The timing of peaks c an change as timing s of precipitation and snow melt in g are altered under climate change. While these scenarios may not capture the actual flow regulations by future dams, they represent the plausible scenarios of the cumulative effects of upstream dams, si mi lar to those observed in other large river basins s uch as the Colorado (Y. Po kh re l et al., 2016) , and also climate changes on flow regime. Hence, t his approach enables a mechanistic understanding of the changes in flood dynamics in the LMRB including Mekong Delta region by different levels of flow regulations under climate change. 4.2.4. Te rrestrial Water Storage (TWS) and its Estimation TWS is composed of water stored over and underneath the land surface; thus, it is estimated by vertically integrating snow water, canopy water, river and floodplain water, soil water, and groundwater st or age s over a given spatial domain, typical ly a river basin. Mathematically, this can be expressed as (Y . P okhrel et al., 2013) : TWS = Surface water + Subsurface water Surface water = FW + RW + SW + CW (4 - 2) Subsurface water = VW + GW Where, FW = water on the floodplains RW = water in the river channels SW = snow water CW = water stored in cano py su rfaces VW = soil water in the vadose zone (unsaturated store) GW = groundwater (below the water table, saturated store) The TWS derived from the measurements made by the Gravity Recovery and Climate Experiment (GRACE) satellite mission (Tapley, 2004 ) provides the vertically - integrated TWS and thus includes all components listed in Equation (4 - 2). In hydrological models such as 77 HiGW - MAT, however, each of the components is typically simulated on an individual basis. T hus , vertically integrated TWS f or comparison with GRACE - based TWS is estimated by adding all components using Equation (4 - 2). Because GRACE measures the TWS variations over large regions, the GRACE data and model results are typically compared as basin a ver ages (Fe lfelani et al., 2017; Y. Pokhrel et al., 2013; Syed et al., 2009) over river basins having an area larger than the GRACE footprint of ~200,000 km 2 (Yeh et al., 2006) . In the present study, we estimate the basin - averaged TWS from both GRACE and HiGW - MAT model by taking an area - weighted av era ge: (4 - 3) where s is the LSM or GRACE estimate, a i is the cell area, S i is th e w eighted estimate for each cell inside the basin, n is the number of cells in a basin, A is the total area of the basin, and H ( x , t ) represents the estimate of water storage for basin at time t . Following Felfelani et al. (2017) , simulated TWS components from HiGW - MAT are vertically integrated to derive TWS anomal ies averaged over the MRB for 2 00 2 - 2 010 period , an overlapping period between GRACE and simulations. Two sets of basin - averaged TWS are derived from the model results. In the first set, the flood water (FW) component in Equation (4 - 1) river - floodplain storage is lu mp ed in the river water (RW) component of the TRIP routing model used in HiGW - MAT . Then, another set of TWS time series is derived by replacing the river storage in HiGW - MAT - based TWS by FW based on the explicit simulation by the CaMa - Flood model without a lt eri ng the other TWS components. The two sets of TWS are then compared with the TWS from GRACE to examine the role of river - floodplain storage in modulating TWS variations. Note that the river storage in HiGW - MAT and river - f loodplain 78 storage in CaMa - Flood a re simulated using the same runoff from HiGW - MAT, thus the mass balance in TWS computations is preserved. The component contribution of river - floodplain water to the total TWS is calculated as the ratio of seasonal amplitud e of river - floodplain storage t o the seasonal amplitude in the simulated total TWS (Y. Pokhrel et al., 2 013) . For uniformity, CaMa - F lo od results at 10km grids are first upscaled to the grid resolution of GRACE data and HiGW - MAT model (i.e., 1º). 4.2.5. Data Historical observations of river discharge and water level are obtained from the MRC. For the analysis of TWS variations, we use both th e Sph erical Harmonics (SH) and mascon - based GRACE products. The level - 3 SH - based products are obtained from three processing centers: (i) the Center for Space Research (CSR), (ii) the Jet Propulsion Laboratory (JPL), and (iii ) the German Research Center fo r Geo science (G F Z), available at: https://grace.jpl.nasa.gov/data/get - data/ . The mascon products are obtained from Scanlon et al. (2016) . 79 4.3. Results and Discussion 4.3.1. Model Evaluation Over the MRB (Figure 4 - 1), river discharge is found to be reasonably reproduced (Figure 4 - 2). Meanwhil e, th e oscillating hydrographs with high frequencies i n the Mekong delta regions ( CT and M Tu stations ) are not reproduced by CaMa - Flood because the boundary condition near ocean is assumed to be steady in CaMa - Flood version 3.6 . Some of deviating hydrograp hs in those regions are attributed by the errors in DEMs in flat delta regions and high uncertainties in parameterization of river geometries in Mekong Delta region that cause the biases in bifurcation scheme (D. Yamazaki et al., 2 01 7) . Otherwise, river discharges on main stems (LP, PA, KT, and PP stations) and flow reversal in TS R (PK station) are well simulated. The related discussions are given in the model evaluation on flooded area below. 80 Figure 4 - 1 Long - term (1981 - 2010) mean river discharge (m 3 /s) simulated by CaMa - Flood at 1 0km spatial resolution over the entire Mekong River Basin (MRB). The upper right and lower left insets show river discharge for the Lower Mekong during p ea k f low season in 1998 (dry year) and 2000 (wet year), respectively. Red circles show the locations f or river discharge validation presented in Figure 4 - 2 ; station names are: LP (Luang Prabang), PA (Pakse), ST (Stung Treng), KT (Kratie), PP (Phnom Penh Po rt ), PK (Prek Kdam), CD (Chau Doc), CT (Can Tho), and MTu (My Thuan). 81 Figure 4 - 2 Evaluation of simulated river discharge with observations obtained from the Mekong River Commission (MRC) at locations indica te d b y red circles in Figure 4 - 1 . While the simulated results are shown for the period of 1985 - 201 0, observed data are shown for the period available. The right panels show the daily climatological mean over the period for which observations were availabl e. Fo r PK station, limited data were available only for year 2004. 82 Next, we evaluate flood occurr ence (i.e., number of flooded months per year) and water surface elevation. Modeled flood occurrence, derived from the flood depth downscaled to 500m grids ( Fi gur e 4 - 3 ), is compared with the satellite - based 30m global data (upscaled to 500m) of historical water occurrence (Pekel et al., 2016) (Figure 4 - 4 a b ). A good agreement can be observed in terms of the broad patterns of flooded areas, but discrep ancies are evident in flood occurrence itself. Notable differences can be seen around the northwest portion of the TSL, where numerous previous studies (Arias, Piman, et al., 2014; Kummu & Sarkkula, 2008; Sakamoto et al., 2007) as well as our results suggest a permanent water occurrence (i.e., 12 - month flood occurrence) but the satellite data indicat e non existence of such permanent water. This possible underestimation of permanent water occurrence in the satellite data results from underestimated water occurrence during April - July, likely due to the presence of relatively shallow and turbid water . F igure 4 - 3 Simulated annual mean flood depth downscaled to 500m spatial resolution using high resolution SRTM topography data for (a) average year (mean of 1981 - 2010), (b) dry year (1998), and wet year (2000 ). The region enclosed by magenta lines shows the areas of major flood around Tonle Sap Lake (TSL) and Lower Mekong within Cambodia (source: https://data.humdata.org/ ), and the thick black outline marks the floode d areas around TSL used in previous studies (Arias et al., 2012) . 83 Figure 4 - 4 Monthly flood occurrence and daily water s ur face elevation. (a, b) Comparison of simulated flood occurrence (number of months) with sat ellite - based flood occurrence of Pekel et al. (2016) . The region enclosed by magenta lines shows the areas of major flood around TSL and Lower Mekong within Cambo di a (so urce: https://data.humdata.org/ ), and the thick black outline marks the flooded areas around TSL used in Arias et al. (2012) . (c - g) C omparison of simulated (CaMa - Flood) and observed (obtained from MRC) wate r surface elevation at five stations indicated by red circles in (b). Observations are shown only for the period available. 84 Further, the model simulates high flood occurrence a ro und t he main body of the lake, along river channels, and the flat floodplains in the Mekong Delta, expected but not seen in the satellite data. This could be a possible model overestimation caused by the uncertainties in topographic and climate data, or an unde restimation in the satellite product, which represents areas such as in the gallery forests and flood - recession agriculture around the lake and in the delta r eg ion (Arias, Cochrane, et al., 2014) . Mor eover, while the model provides a continuous simulation of monthly flood occurrence, only limitedly available cloud - free images are used in the satellite product (Pekel et al., 2016) . Note that the stripes in the Mekong Delta region (Figure 4 - 4 b) result from small but inherent errors in the digital elevation model (DEM) in low - lying ar eas (D. Yamazaki et al., 2017) . For TSL region, the model clearly captures the areas of major flood (magenta lines in Figure 4 - 4 a,b). Further, comparison of simulated flooded areas with the estimates from a previous study (Arias et al., 2012) for the major flooded regions around TSL (black outline in Figure 4 - 4 a) suggests that the model well captures the total flooded areas both during dry and w et seas ons (Table 4 - 2 ). 85 Table 4 - 2 Comparison of flooded areas with Arias et al. (2012) for the major flood regions around Ton le Sap Lake indicated by thick black line in Figure 4 - 4 a. Dry Season Flooded Area (km 2 ) Difference (%) b MODIS a GIS a CaMa - Flood (this study) MODIS GIS 5/8/2000 2,841 3,072 3,442 17 11 4/15/2001 2,751 3,096 3,671 25 16 5/25/2002 2,580 2,433 3,144 18 23 6 /2/20 03 2,605 3,003 3,173 18 5 5/16/2004 2,579 2,281 3,100 17 26 5/1/2005 2,841 3,177 3,036 6 - 5 5/1/2006 2,667 2,442 3,143 15 22 5/17/2007 2,626 3,029 3,405 23 11 Wet Season Flooded Area (km 2 ) Difference (%) MODIS GIS CaMa - Flood (this study) MODI S GIS 10/23/2000 14,763 14,030 14,521 - 2 3 10/8/2001 14,392 13,792 13,038 - 10 - 6 10/16/2002 14,264 13,103 12,517 - 14 - 5 10/24/2003 12,037 10,863 9,330 - 29 - 16 10/23/2004 12,264 10,894 10,669 - 15 - 2 10/16/2005 13,026 12,665 10,480 - 24 - 21 10/24/2006 13 ,180 12,624 13,635 3 7 10/16/2007 12,404 12,300 10,668 - 16 - 15 a MODIS and GIS data are from Arias et al. (2012) b - - 86 Since the satellite data could contain uncertainties, we evaluate the modeled water surface elevation a primary determinant of flood extent, depth, and occurrence with the ground - based observation to add further confidence to our flood simulations ( Fi gure 4 - 4 c - g). Evidently, both the seasonal magnitude and temporal variability of water elevation are well captured by the model, especially at the Kompong Luong (KL) in the TSL and Phnom Penh Port (PP). Simulated water levels are not as accurate in the lo wer p ortion of the delta (e.g., Can Tho; Fig ure 4 - 4 g), which could be due to the uncertainties in DEM, river width, and channel bathymetry represented in our river bifurcation scheme (D. Yamazaki et al., 2014) . At Can Tho, discrepancies could also be attributed to tide effects, not consider ed in t he current model. Given the scale of the model domain, uncertainties in data, and the diffic ulty in accurately representing channel bifurcation, we consider these results to be reasonable for this study. 4.3.2. Role of River - Floodplain Storage on TWS dyna mi cs an d Historical Variability Next, we examine the role of river - floodplain water storage in modu lating the TWS dynamics in the MRB using TWS variations simulated by the models and from GRACE satellites. By comparing the TWS solely from HiGW - MAT, the com bi ned T WS from HiGW - MAT and CaMa - Flood, and TWS from GRACE, we find that river - floodplain storage p lays a critical role in modulating the total TWS variations, and hence the hydrology of the MRB (Fig ure 4 - 5 ). First, the variations in river - floodplain stora ge from CaMa - Flood (solid blue line) exhibit substantially larger seasonal amplitude than the river storage (dashed blue line) in HiGW - MAT that does not consider TSR flow reversal and lacks floodwater storage. While certain inter - annual variations are obvi ou s, th e differences can be clearly discerned from the seasonal cycle ( Figure 4 - 5 b). Second, a one - month delay in the peak can be seen in river - floodplain storage in CaMa - Flood as compared to the river storage in HiGW - MAT, which e xp ected ly results from 87 larger flood plain storage in CaMa - Flood during wet season partly due to TSR flow reversal and a subsequent release in the dry season. Figure 4 - 5 Role of river - floodplain storage on T WS dyna mics over the MRB. (a) Black line shows the mean of TWS anomaly from spherical harmonics - and mascon - based GRACE products with the range between different products indicated by grey shading. For modeled TWS anomalies, four sets of result are shown: c ombin ed total TWS from H iGW - MAT and CaMa - Flood (solid red), total TWS only from HiGW - MAT (dashed red), river - floodwater storage from CaMa - Flood (solid blue), and river water storage from HiGW - MAT (dashed blue) which lumps the flood water storage. (b) The s eason al cycle. Results a re averaged for the entire MRB. Third, as a consequence of larger seasonal swing and delayed peak in CaMa - Flood, the combined TWS from HiGW - MAT and CaMa - Flood provides a better agreement with GRACE compared to the TWS from HiGW - MA T alone . The better agreem ent is reflected not only graphically, but also statistically; while the already - high R 2 (0.99) does not change, the root mean squared error (RMSE) reduces from ~32 to ~18mm ( Figure 4 - 5 b). Because GRACE p ro vides the vertically integrated total TWS, not its components (e.g., Pokhrel et al., 201 3) , ri ver - floodplain storage from CaMa - Flood could not be separately validated, but an independent evaluations of water level ( Figure 4 - 4 c - g) and river discharge ( Figure 4 - 1 and Figure 4 - 2 ) suggest that CaMa - Flood well simulates the overall hydrodynamics. Fourth, comparison of the 88 seasonal amplitude in the combined TWS (Fig ure 4 - 5 ; solid red line) and the river - floodplain storage from CaMa - Flood (Fig ure 4 - 5 b; solid blue line) su gg ests that river - floodplain storage explains ~27% of the seasonal amplitude of total TWS variations averaged over the entire MRB as opposed to only ~13% by river storage in HiGW - MAT (Fig ure 4 - 5 ; dashed blue line); for the LMRB, w hile CaMa - Flood river - floo dp lain storage contributes to ~49% of total TWS, HiGW - MAT river storage only accounts for ~12% (Fig ure 4 - 6 ). These findings imply that the potential alterations in the Mekong flood pulse and TSR flow reversal will affect not only the dynamics of flood patt er ns bu t also the overall basin hydrology because changes in surface water storage can alter other components of the basin water balance. Figure 4 - 6 Same as in Figure 4 - 5 but only for the Lower Mekong region. Note that GRACE data are not included here because of the reduced reliability of the data when averaged over small r egions. 89 Figure 4 - 7 Relationship between riv er - fl oodplain storage from CaMa - Flood (monthly) and climate variability (annual mean precipitation and temperature) over the Lower Mekong domain . Precipitation and temperature data are same as those used as input to HiGW - MAT model. Figure 4 - 7 shows the historical variations in river - floodplain storage from CaMa - Flood over the Lower Mekong domain under varying climate conditions (i.e., annual precipitation and temperature). Evident ly, annual storage variations are largely dictated by t he variab ilities in inter - annual precipitation (correlation of 0.67) and temperature (correlation of - 0.31). High precipitation, often combined with low basin - wide temperatures, lead to wet years and v ice versa; however, no significant trend in river - flood plain sto rage (Mann - Kendall test, p=0.958, - year period. The lowest and highest storages clearly stand out in years 1998 and 2000, which are among the driest and wettest yea rs, respectively in past few decades (Hung et al., 2012; Mekong River Commission, 2005) ; flood occ urrence in these years is discussed further in the next chapter . Note that these results do not account for the effects of existing dams, but such effects are relatively small compared to the flow volume in the main stem of the Mekong ( Figure 4 - 8 ). 90 Figure 4 - 8 Daily river discharge at Pakse (PA) station (location shown in Figure 4 - 1 and Table 4 - 1 ) s imulated by HiGW - MAT model with and without considering the existing dams. Observed data from MRC are also shown. The right panel shows the monthly seasonal cycle. Model results are taken from Pokhrel et al. (2015) . 4.3.3. Potentia l Effects of Flow Regulation on Flood Dynamics in the LMRB 4.3.3.1. Potential Effect s of Flow Regulation on Mean River Discharge in the LMRB Figure 4 - 9 presents the potential effects of upstream flow regulation (altered magnitude and timin g of peak ne ar Stung Treng, Table 4 - 1 ; marked by a star in F igure 4 - 11 ) on downstream flow dynamics in the mainstream Mekong, TSR, and some distributaries in the delta region (locations shown in F igure 4 - 11 ; Ta ble 4 - 1 ). Note that only the magnitude and timing of peak is altered, and mass balance is preser ved in all flow regulation scenarios. Results in Figure 4 - 9 represent a surrogate of an average year, defined as the mean for 1981 - 201 0 period; typical dry and wet yea rs are discussed next. Up to the PP station, highly similar altered flow patterns are observed to that at the dam location ( Figure 4 - 9 a - c) but interesting features e m erge in the downstream of PP an d in the TSR. Most notably, upstream flow alterations are found to severely impact the magnitude, timing, and direction of discharge into TSL (LO and PK stations; Figure 4 - 9 d,e), which could potenti ally disrupt the natural flood d ynamics in the TSR and cause a 91 regime shift in TSL water balance. Results suggest that, for different flow regulation scenarios, the peak of flow from the TSL to the Mekong would reduce by 7 - 37% and 7 - 34% ( Table 4 - 3 a) and that of the reversed flow from the Mekong into TSL by 11 - 80% and 15 - 88% ( Table 4 - 3 b) at LO and P K stations, respectively. Together, the changes in the peaks of the bi - directional flow in the TSR would dampe n the seasonal amplitude (i.e., maximum - minimum) of the hydrograph at LO and PK stations by 8 - 51% and 10 - 60% ( Table 4 - 3 c), respectively, under different upstream flow regulation scenarios. These changes in flood dynamics could sig nificantly alter the onset, dura tion, and amount of flow reversal in the TSR. We fi nd that the onset could be delayed by 1 - 38 days and 1 - 40 days ( Table 4 - 3 d), respectively, at LO and PK stations, with a reduction in the total dura tion of reversed fl ow by 2 - 51 da ys and 2 - 55 days ( Table 4 - 3 e). As a result, the total volume of water entering the TSL due to flow reversal could reduce from 12,202 (23,646) million m 3 by 14 - 87% (15 - 92%) at the LO (PK) station for different flow alteration scena rios ( Table 4 - 3 f). 92 Figure 4 - 9 Potential changes in daily river discharge by flow regulation. The thick black line shows the baseline flow (1981 - 2010 average); other colors repr esent different scenarios of 93 change in peak flow magnitude (solid lines). Das hed and dotted lines represent the scenarios of early and delayed peak timing, respectively, by one month for different degree of peak flow alterat ion represented by the color cod ing. Results of altered timing are shown only for 10, 30, and 50% peak flow r eduction scenarios. Station names are: KC (Kampong Cham), PP (Phnom Penh Port), LO (Lake Outlet), PK (Prek Kdam); NL (Neak Luong), KK (Koh Khel), C D (Chau Doc), VN (Vam Nao), MTu (My Thuan), MTo (My Tho), and CT (Can Tho); latitudes and longitudes are prov ided in Table 4 - 1 . 94 Table 4 - 3 Changes in major flood characteristics (e.g. , onset, magnitude, duration, an d amount) compared to the baseline simulation at different stations analyzed in Figure 4 - 8 . The Lake Outlet (LO) and Prek Kdam (PK) stations are indicated by grey shading. Numbers e nclosed in boxes are those noted in the text. Fo r 10, 30, and 50% peak flow alteration scenarios, results for the scenarios with one - month early and delayed peak ti ming are also provided. (a) Q max in baseline (m 3 /s) and Q max (%) (b) Q min in baseline (m 3 /s) a nd Q min (m 3 /s) Station Baseline (m 3 /sec) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early Station Baseline (m 3 /sec) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early Dam 36,329 - 10 - 10 - 10 - 2 0 - 30 - 30 - 30 - 40 - 50 - 50 - 50 Dam 4,318 1,280 1, 280 1,280 2,646 3,927 3,927 3,927 5,293 6,573 6,573 6,573 KC 35,747 - 8 - 8 - 10 - 17 - 27 - 26 - 29 - 37 - 46 - 46 - 48 KC 4,645 1,264 1,235 1,311 2,625 3,907 3,875 3,967 5,287 6,547 6,528 6,587 PP 22,450 - 6 - 5 - 7 - 1 2 - 19 - 17 - 20 - 27 - 36 - 34 - 38 PP 4,434 980 948 1 ,026 1,972 2,879 2,858 2,933 3,839 4,721 4,694 4,764 LO 4,479 - 7 - 16 - 4 - 14 - 22 - 29 - 19 - 29 - 37 - 42 - 36 LO - 2,210 240 ( - 11%) 458 ( - 21%) 146 ( - 7%) 551 ( - 25%) 887 ( - 40%) 1,105 ( - 50%) 798 ( - 36%) 1,293 ( - 58%) 1, 762 ( - 80%) 1,967 ( - 89%) 1,682 ( - 76%) PK 5,423 - 7 - 17 - 1 - 13 - 20 - 27 - 16 - 27 - 34 - 37 - 31 PK - 5,085 742 ( - 15%) 1,394 ( - 27%) 444 ( - 9%) 1,508 ( - 30%) 2,414 ( - 47%) 3,076 ( - 60%) 2,076 ( - 41%) 3,376 ( - 66%) 4,483 ( - 88%) 5,002 ( - 98%) 4,120 ( - 81%) NL 21,105 - 5 - 2 - 6 - 9 - 13 - 10 - 15 - 17 - 23 - 20 - 25 NL 5,749 1,244 1 ,131 1,509 2,709 3,865 3,754 4,062 4,965 5,921 5,843 6,038 KK 4,049 - 14 - 2 - 22 - 29 - 40 - 30 - 50 - 58 - 65 - 62 - 71 KK 25 46 40 60 134 221 210 237 313 403 394 416 CD 4,931 - 13 - 1 - 23 - 28 - 35 - 29 - 46 - 52 - 61 - 56 - 67 CD 69 36 31 45 123 210 203 223 300 396 387 4 15 VN 15,578 - 2 0 - 4 - 5 - 9 - 6 - 12 - 13 - 16 - 14 - 19 VN 4,938 1,046 933 1,288 2,266 3,168 3,065 3,305 3,976 4,654 4,600 4,749 MTu 5,408 - 4 0 - 11 - 10 - 16 - 11 - 23 - 24 - 28 - 26 - 32 MTu 930 275 244 342 641 933 899 980 1,219 1,472 1,453 1,507 MTo 1,028 - 10 - 5 - 2 2 - 20 - 32 - 30 - 65 - 67 - 70 - 70 - 71 MTo 2 16 12 26 85 149 142 160 216 253 253 252 CT 16,704 - 3 - 1 - 6 - 7 - 12 - 8 - 16 - 17 - 20 - 18 - 22 CT 4,948 1,049 934 1,288 2,268 3,170 3,067 3,307 3,968 4,638 4,587 4,727 (c) Q max - Q min in baseline (m 3 /s) and (Q max - Q mi n ) (%) (d) Onset of reversed flow in baseline (day of year) and its change (day) Station Baseline (m 3 /sec) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early Station Baseline (DO Y) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early Dam 32,012 - 15 - 15 - 15 - 31 - 46 - 46 - 46 - 62 - 77 - 77 - 77 LO 159 1 32 - 29 6 8 37 - 20 16 38 74 7 KC 31,102 - 13 - 13 - 15 - 28 - 43 - 42 - 46 - 59 - 74 - 73 - 76 PK 160 1 32 - 29 7 9 42 - 18 21 40 83 8 PP 18,016 - 12 - 12 - 14 - 25 - 39 - 38 - 41 - 55 - 71 - 68 - 73 (e) Duration of reversed flow in baseline (day) and its change (day) LO 6,690 - 8 - 17 - 5 - 18 - 28 - 36 - 25 - 39 - 51 - 57 - 49 Station Baseline (Days) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early PK 10,508 - 10 - 22 - 5 - 21 - 33 - 43 - 28 - 46 - 60 - 67 - 55 NL 15,356 - 15 - 10 - 18 - 30 - 43 - 38 - 47 - 56 - 70 - 66 - 73 LO 112 - 2 - 7 0 - 9 - 13 - 16 - 13 - 24 - 51 - 63 - 46 KK 4,024 - 15 - 3 - 23 - 32 - 46 - 36 - 57 - 66 - 75 - 72 - 82 PK 109 - 2 - 13 1 - 10 - 15 - 23 - 14 - 31 - 55 - 100 - 48 CD 4,862 - 14 - 1 - 24 - 31 - 40 - 34 - 52 - 59 - 70 - 65 - 77 (f) Volume o f reversed flow in baseline (10 6 m 3 ) and its change (%) VN 10,641 - 13 - 9 - 18 - 28 - 43 - 37 - 48 - 56 - 68 - 64 - 72 Station Baseline (Mm 3 ) 10% 10 % Delay 10% Early 20% 30% 30% Delay 30% Early 40 % 50% 50% Delay 50% Early MTu 4,478 - 11 - 6 - 20 - 26 - 41 - 33 - 50 - 57 - 67 - 64 - 72 MTo 1,026 - 12 - 7 - 24 - 28 - 47 - 44 - 81 - 88 - 94 - 95 - 95 LO 12,202 - 14 - 25 - 9 - 29 - 47 - 56 - 43 - 68 - 87 - 94 - 84 CT 11,7 57 - 13 - 9 - 19 - 29 - 44 - 37 - 51 - 57 - 69 - 65 - 72 PK 23,646 - 15 - 3 0 - 7 - 31 - 50 - 64 - 43 - 72 - 92 - 100 - 86 95 In the Mekong Delta region, flow regulations cause re latively predictable changes in river flow dynamics along the mainstream channels, but flood dynamics becomes highly unpredictable in the distributar ies and bifurcated channels. In the main channels (e.g., NL, VN, and CT stations; Figure 4 - 9 f,i,l), the alterations in the magnitude of seasonal amplitude (i.e., dampened peak and en hanced low flow) show a similar pattern to the u pstream flow alterations with a more pronounced increa se in low flow than the reduction in flood peak. This interesting phenomenon in the downstream of the confluence of the TSR results partly from the weaken ed flow reversal in the TSR; that is, because le ss water flows into the TSR during wet season, the mai nstream flow in the downstream of PP station is not significantly affected by TSR flow reversal. Consequently, in dry season, most of the enhanced low flo w from the upper main stem is directly discharge d to the downstream with considerably less contributio n from TSR flow. In other words, while the TSL plays a role of a detention reservoir by dampening flood peak and enhancing low flow in the downstream, suc h role becomes less significant when the mainstr eam flow is regulated by upstream reservoirs, reducing the wet - season flow into the lake. At the KK and CD stations ( Figure 4 - 9 g,h) which are on the Bassac River, a distributary of the Mekong originating near PP, flood peak is su bstantially reduced due to the dampened upstream flood peak; however, the increase in low flow is relatively small but distributed over a longer period than at the mainstream stations. Understanding these pot ential changes in high and low flows i s crucial because the Bassac River is a critical transportation corridor between Cambodia and Vietnam. Moreover, in these downstream locations likely increase in water levels due to sea level rise and tides could inter fere with the changes brought by upstr eam dam re gulation, causing unpredictable flow and water level patterns. Further downstream, highly unpredictable flows are observed at MTo (near My Tho in Vietnam) station ( Figure 4 - 9 k), 96 beca use of unpredictable changes in channel bifurcat ion dynamics; here, flood seasonality potentially ceases under high upstream flow regulation and delayed peak scenarios ( Table 4 - 3 a,b). Note that some of these changes could have res ulted from the uncertainties in channel bifurcat ion simulations in our model in the low - lying areas as discussed earlier. Figure 4 - 9 also includes the results from simulations with one month early and delayed peak at the dam loca tion. Overall, the changed timing of peak result s in correspondingly shifted hydrograph in the immediat e downstream of dam location; however, the altered timing is found to affect also the magnitude of peak flow in the TSR and delta region. At LO and PK st ations ( Figure 4 - 9 d - e ), the compounded effects of reduced peak and altered timing cause an even larger impact on the timing, duration, and amount of reversed flow than caused only by reduced peak (also see Table 4 - 3 d - f). Notably, results suggest that TSR flow reversal at PK station almost ceases if the flood peak reduced by 50% arrives with a one month delay relative to the baseline flow ( Figure 4 - 9 e, Table 4 - 3 b). In the delta region (e.g., KK and CD ; Figure 4 - 9 g - h), delayed (early) timing is found to increase (reduce) the flood peak magnitude, suggesting that there is an optimum timing for flood patterns to be maintained at th e base level; any changes in timing can causes a significant increase or decrease in the flood peak magnitude. The changes in water surface elevation are found to follow similar patters to the changes in discharge ( Figure 4 - 10 ) . 97 Figure 4 - 10 Same as in Figure 4 - 9 but for simulated water level (i.e., water surface elevation). 98 4.3.3.2. Potential Effects of Flow Regulation on Mean Flood Occurrence in the LMRB F igure 4 - 11 presents the change s in flood occurrence within the same spatial domain shown in Figure 4 - 4 . Evidently, downstream impacts vary among different scenarios of flow regulation. First, th e changes are relatively small for 10% and 20% peak reduction. Sec ond, increased flood occurrence can be seen at the vicinity of TSL and along the main river channe ls because of increased water retention during dry season. Away from the lake and in the flo oded agricultural areas, flood occurrence decreases significantly (by up to 5 months or more) because of large decline in flood water entering the TSL from the Meko ng during wet season. Overall, flooded areas in the TSL region (thick black line within the red rectangle in F igure 4 - 11 f) averaged for the high flood season (August - October) decrease by 413 km 2 (4.6%), 774 (8.6%), 1122 (12.5%), 1602 (17.8%), and 2075 (23.1%) for 10, 20, 30, 40, and 50% scenarios, respectively. Similarly , flooded areas during dry season (April - June) increase by 93 km 2 (2.9%), 311 (9.7%), 580 (18.1%), 8 62 (26.9%), and 1144 (35.7%). Results suggest that the flooded areas in the floodplains upstream of PP (green rectangle in F igure 4 - 11 f) also decrease significantly u nder all flow alteration scena rios. Third, no significant change in water occurrence can be observed within the main body of TSL, suggesting the presence of permanent water under all flow alteration scenarios, which coul d partly be attributed to a simple t reatment of lake bed elevation s owing to the inherent limitations in DEM data that provide only the water surface elevation over water bodies and the lack of spatially - explicit bathymetry data. Thus, these results should be interpreted with caution. 99 F igure 4 - 11 Flood occurrence and the effects of flow regulation on it. (a) Same as in Figure 4 - 4 but based on the baseline simulation with 1981 - 2010 mean runoff. (b - f) Chan ge in flood occurrence (number of months) for change in peak flow by different degree as indicated with reference to the baseline flood occurrence (a). Th e star in (b) shows the Stung Treng station where flow is altered. Rectangles in (f) show regions in t he LMRB discussed in the text. For magenta and black lines, see de scription in Figure 4 - 4 caption. Fourth, in the Mekong Delta region, a large increase in flood occurrence can be seen in the middle reach (post - flooding agricultur al areas; cyan rectangle in F igure 4 - 11 f ) for >30% flow regulation. Again, this results from a relatively small impact on the mainstream flow during flood season as less water enters the TSL and an increase in low because of dam r elease ( Figure 4 - 9 f,i,l). Farther from t he mainstream channels in the lower portion of the delta, flood occurrence mostly reduces because lowered water levels ( Figure 4 - 10 j,k) in the mainstream channels p revent frequent overtopping to the floodplains. Fifth, no changes in flood occurrence can be seen in mainstream Mekong and Bassac Rivers as well as other dis tributaries near the river 100 mouth. In these regions, the magnitude and timing of water levels are si mulated differently under different scenarios, but the water occur rence remains unchanged because of permanent water occurrence in the model. Finally, our re sults indicate that while the areas flooded for ~6 months are least impacted by flow regulation dur ing average years, the areas flooded for ~1 (~12) months could dec rease (increase) significantly under all flow alteration scenarios and both in the TSL regi on as well as the entire Lower Mekong domain ( Figure 4 - 12 ). This is a dir ect consequence of the reduced flood peak and increased low flow u nder all flow regulation sce narios. In terms of the impacts of changed flood peak timing, the effects tend to become smaller with increased duration of flood occurrence; in general, regions that are flooded over nine months are minimally impacted both in t he Lower Mekong and TSL regi ons ( Figure 4 - 12 ). Similar patterns were reported in a previous study (Arias, Pima n , et al., 2014) in that the reductions in peak and increases in low flows are amplified for higher degrees of flow regulation. Similarities are found also in terms of the least impa cted flood occurrence (in general, 40 - 60%, which is similar to ~6 months in a year). These comparisons are summarized in Table 4 - 4 , but it is noted that the results are not directly comparable because of the differences in simulat ion sett ings (see footnotes in Table 4 - 4 ), whic h results in considerably different baseline simulations (Two - sample K - S test, p=0.11 ). 101 Figure 4 - 12 Flooded areas having diff erent flood occurrence estimated from the baseline simulation results pre sented in F igure 4 - 11 and their changes und er different flow regulation scenarios for (a) the entire domain shown in F igure 4 - 11 , and (b) the TSL region marked by thick black line within the red rectang le in F igure 4 - 11 f. Plus signs and open cir cles show the results from the early and delayed peak flood timing by one month, respectively. 102 Table 4 - 4 A summary of the results o f potential changes in flooded area around Tonle Sap Lake under differ ent flow regulation scenarios from this study and those from Arias, Piman, et al. (2014) . Flooded days with t he smallest change are marked by grey shading. As noted in the text, the results are not directly comparable due to differences in simulation settings between t he two studies. a Spatial domain of Tonle Sap Lake region is exactly same as in Arias, Piman, et al. (2014) b DF: Water infrastr u cture development plan up to 2015; DF+3S: Cumulative impact of the DF 42 dams in the main tributaries and 3S rivers (see Arias, Piman, et al. (2014) for details). c To calculate flood occurrence, Arias, Piman, et al. (2 0 14) used 15 - years of simulations for 1986 - 2000 period, but this study uses 1 - year of s imulations driven by the climatological mean daily runoff for 1981 - 2010 period. 4.3.3.3. and Flo o d Occurrence in the LMRB B ecause the effects of flow regulation on downstream flood patterns can vary significantly during dry and wet years, we examine the results fo r 1998 and 2000 ( Figure 4 - 13 and Figure 4 - 14 ), which represent the historica l dry and wet years, respectively (Arias et al., 2012; Kummu & Sarkkula, 2008) ( Figure 4 - 4 , Figure 4 - 7 ). Although the broad spatial patterns of changes in flood occurrence during dry and wet years appear similar to those during the Flood occu rrence c (%) Flooded area (km 2 ) and its change from baseline simulation s in km 2 (% in parentheses) a Arias, Pim an, et al. (2014) This study Baseline DF b DF+3S b Baseline 10% 10% D elay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early 16670 - 94 ( - 0.6) - 120 ( - 0.7) 11129 - 729 ( - 4.4) 74 (0.4) - 1404 ( - 8.4) - 1206 ( - 7.2) - 1627 ( - 9.8) - 1258 ( - 7.5) - 2180 ( - 13.1) - 2117 ( - 12.7) - 2891 ( - 17.3) - 2770 ( - 16.6) - 3147 ( - 18.9) 14912 - 271 ( - 1.8) - 404 ( - 2.7) 10637 - 433 ( - 2.9) - 249 ( - 1.7) - 1048 ( - 7.0) - 888 ( - 6.0) - 1434 ( - 9.6) - 1230 ( - 8.2) - 2191 ( - 14.7) - 2229 ( - 14.9) - 2759 (18.5) - 2669 ( - 17.9) - 2987 ( - 20.0) 13495 - 322 ( - 2.4) - 569 ( - 4.2) 9729 - 382 ( - 2.8) - 538 ( - 4.0) - 477 ( - 3.5) - 784 ( - 5.8) - 1408 ( - 10.4) - 1538 ( - 11.4) - 1543 ( - 11.4) - 1888 ( - 14.0) - 2411 ( - 17.9) - 2451 ( - 18 .2) - 2563 ( - 19.0) 12074 - 262 ( - 2.2) - 423 ( - 3.5) 7905 - 208 ( - 1.7) - 352 ( - 2.9) - 12 ( - 0.1) - 448 ( - 3.7) - 692 ( - 5.7) - 814 (6.7) - 478 ( - 4.0) - 918 ( - 7.6) - 1148 ( - 9.5) - 1566 ( - 13.0) - 1101 ( - 9.1) 10407 - 149 ( - 1.4) - 236 ( - 2.3) 6187 - 63 ( - 0.6) - 124 ( - 1.2) 149 (1.4) - 147 ( - 1.4) - 236 ( - 2.3) - 323 ( - 3.1) - 18 ( - 0.2) - 301 ( - 2.9) - 427 ( - 4.1) - 470 ( - 4.5) - 251 ( - 2.4) 60 8874 - 36 ( - 0.4) - 52 ( - 0.6) 4827 53 (0.6) 6 (0. 1) 361 (4.1) 292 (3.3) 325 (3.7) 163 (1.8) 468 (5.3) 376 (4.2) 421 ( - 4.7) 399 (4.5) 527 (5.9) 0 7483 126 (1.7) 321 (4.3) 4186 74 (1.0) 91 (1.2) 171 (2.3) 171 (2.3) 270 (3.6) 261 (3.5 ) 365 (4.9) 383 (5.1) 515 (6.9) 491 (6.6) 619 (8.3) 6552 191 (2. 9) 290 (4.4) 3631 125 (1.9) 143 (2.2) 186 (2.8) 248 (3.8) 506 (7.7) 456 (7.0) 584 (8.9) 685 (10.5) 839 (12.8) 833 (12.7) 895 (13.7) 5603 199 (3.6) 354 (6.3) 3244 57 (1.0) 69 (1.2) 242 (4.3) 410 (7.3) 569 (10.1) 543 (9.7) 634 (11.3) 911 (16.3) 1097 (1 9.6) 1096 (19.6) 1151 (20.5) 4910 278 (5.7) 424 (8.6) 3103 73 (1.5) 61 (1.2) 67 (1.4) 240 ( 4.9) 587 (11.9) 575 (11.7) 619 (12.6) 843 (17.2) 1171 (23.9) 1153 (23.5) 1211 (24.7) 103 average year ( F igure 4 - 11 ), magnitudes vary, and some interesting features emerge. Substantially smaller (larger) flooded areas and occurrence can be seen during dry (wet) years ( Figure 4 - 13 a,e) compared to that in an average year ( F igure 4 - 11 a). Specifically, during the high flood season (Augu st - October), 51.3% (36.9%) more (less) areas are flooded in wet (dry) years compared to the average year. Similarly, during the dry season (April - June), 17.1% (1.4%) more ( less) areas are flooded in wet (dry) years. Further, for the 10% flow alteration sce nario, marked differences are not found in downstream flood occurrence between dry, normal, and wet years. However, varying patterns of change in flood occurrence become r e adily discernable between dry and wet years for the 30%, and even more so for the 50 % scenario. In the wet year, substantial areas in the western vicinity of the TSL experience an increase in flood occurrence by up to 6 months for 50% scenario, but the sa m e region experiences a notable decline in flood occurrence during the dry year ( Figure 4 - 13 d,h). As in normal year ( F igure 4 - 11 f), a marked reduction in flood oc currence is seen in the outer extents of t h e major flooded areas around TSL (shown by magenta line) in both dry and wet years ( Figure 4 - 13 d,h). No change in flood occ urrence that can be seen northwest of the TSL flooded areas in 50% scenario for 1998 ( Figure 4 - 13 d) is in fact due to no flood occurrence in all scenarios including the baseline ( Figure 4 - 13 a ) . The e ffects of flow regulation on the seasonal flood dynamics (similar to Figure 4 - 8 ) dur i ng dry and wet years are presented in the Figure 4 - 15 and Figure 4 - 16 , along with the changes in different flood characteristics ( Table 4 - 5 , Table 4 - 6 ) for all flow alteration scenarios. Evidently, the peak flood magnitude decreases and the low flow increases, resulting in significantly dampened flood pulse amplitude in all Lower Mekong sta tions ( Table 4 - 6 ). Notably, the TSR flo w reversal tends to completely cease under 50% peak reduction scenario 104 during dry year ( Figure 4 - 15 d - e, Table 4 - 5 d - e), which is likely also during the wet year if the flow under 50% peak reduction scenar i o is delayed by one month ( Figure 4 - 16 d - e, Table 4 - 6 d - e). The onset of TSR flow reversal tends to shift in the same direction and by a comparable duration to the duration of altered peak timing (i.e., on e month) for 10% scenario but the effects are varying for 30% and 50% scenarios ( Table 4 - 3 , Table 4 - 5 , Table 4 - 6 ) . The other flood characteristics (e.g., duration and volume of r e versed flow) are also affected to a considerably varying extent due to the alteration of timing under different scenarios of peak flow reduction ( Table 4 - 3 , Table 4 - 5 , Table 4 - 6 ) . In the downstream of PP station (cyan rectangle in Figure 4 - 13 h), as in the average year, substantial increase in flood occurrence is seen during the wet year (especially for >30% peak alteration scenarios) which is primarily d ue to a longer retention of water in these flat areas caused by increased low flow ( Figure 4 - 8 g,h) and higher dry - season water levels ( Figure 4 - 10 g,h). On the contrary, a significant reduction in flood o c currence can be seen in the upper portion of this region in the dry year because the relatively small increase in baseflow does not lead to a sustained floo d water during the low flow season. In the areas upstream of PP station (green rectangle in Figure 4 - 13 h), flooding is primarily caused by water overtopping river bank s during wet season, thus the reduced flood peak causes a marked decline in flood extents and occurrence under all scenarios during both dry and wet years. Finall y , within the region shown by a black rectangle ( Figure 4 - 13 h), a large decline in flood occurrence seen during a normal year ( F igure 4 - 11 b - f) is not evident during the wet year ( Figure 4 - 13 f - h) because of significantly larger flows in the wet year even in the 50% regulation scenario (see Figure 4 - 8 and Figure 4 - 16 ); in the dry year no change can be seen because this region is r a rely flooded ( Figure 4 - 13 a). 105 Figure 4 - 13 Same as in Figure 4 - 11 but for dry (1998) and wet (2000) years. For the altered flow scenarios, results for only 10, 30, and 50% altera t ions are shown. Rectangles in (h) show regions in the LMRB discussed in the text. For magenta and black lines, see description in Figure 4 - 4 caption. 106 Figure 4 - 14 Simulated flood occurrence in dry and wet y e ars for different flow regulation scenarios , i.e., the actual flood occurrence corresponding to the difference with the baseline shown in Figure 4 - 13 b - d (1998) and Figure 4 - 13 f - h (2000). 107 Figure 4 - 15 Same as in Figure 4 - 9 but for 1998 (dry year). 108 Figure 4 - 16 Same as in Figure 4 - 9 but for 2000 (wet year). 109 Table 4 - 5 Same as Table 4 - 3 but for dry year (1998). Blank fields in (d) and (e) denote no flow reversal. (a) Q max in baseline (m 3 /s) and Q max (%) (b) Q min in baseline (m 3 /s) and Q min (m 3 /s) Station Baseline (m 3 /sec) 10% 10% Delay 10% Early 20% 30% 3 0% Delay 30% Early 40% 50% 50% Delay 50% Early Station Baseline (m 3 /sec) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early Dam 21,327 - 10 - 10 - 10 - 20 - 30 - 30 - 30 - 40 - 50 - 50 - 50 Dam 3,694 980 980 980 1,961 2,941 2,941 2,941 3 ,921 4,901 4,901 4,901 KC 20,775 - 10 - 8 - 10 - 19 - 29 - 27 - 29 - 38 - 40 - 36 - 42 KC 4,005 955 933 996 1,916 2,886 2,860 2,921 3,848 4,826 4,800 4,876 PP 15,249 - 8 - 7 - 8 - 16 - 25 - 23 - 25 - 33 - 29 - 24 - 31 PP 3,915 764 741 771 1,478 2,173 2,163 2,212 2,864 3,559 3 ,536 3,618 LO 4,597 - 7 2 - 16 - 14 - 21 - 12 - 28 - 26 - 31 - 25 - 36 LO - 1,429 220 ( - 15%) 286 ( - 20%) 197 ( - 14%) 455 ( - 32%) 708 ( - 50%) 777 ( - 54%) 691 ( - 48%) 1,017 ( - 71%) 1,529 ( - 107%) 1,591 ( - 111%) 1,560 ( - 109%) PK 5,318 - 7 3 - 16 - 13 - 20 - 10 - 27 - 25 - 29 - 24 - 34 P K - 2,351 407 ( - 17%) 506 ( - 22%) 317 ( - 14%) 827 ( - 35%) 1,270 ( - 54%) 1,384 ( - 59%) 1,257 ( - 53%) 1,785 ( - 76%) 2,552 ( - 109%) 2,663 ( - 113%) 2,598 ( - 111%) NL 15,904 - 6 - 3 - 7 - 11 - 13 - 5 - 18 - 10 - 9 - 4 - 11 NL 5,045 933 898 1,184 1,840 2,785 2,677 3,219 3,858 4,601 4 ,479 4,743 KK 1,008 - 19 - 1 - 21 - 29 - 27 0 - 40 - 22 - 19 - 4 - 24 KK 16 13 12 24 49 101 93 128 174 228 218 240 CD 1,457 - 16 14 - 25 - 26 - 21 8 - 39 - 18 - 17 0 - 22 CD 29 40 36 53 78 122 106 150 189 244 241 269 VN 12,402 - 4 2 - 6 - 7 - 6 1 - 12 - 4 - 3 0 - 6 VN 4,359 764 735 990 1,544 2,339 2,233 2,649 3,176 3,707 3,615 3,829 MTu 3,571 - 7 3 - 10 - 11 - 9 2 - 17 - 7 - 5 0 - 9 MTu 792 185 178 242 390 622 591 719 887 1,061 1,029 1,102 MTo 278 0 13 - 1 0 0 6 0 0 0 - 1 0 MTo 2 0 0 0 12 52 45 70 108 146 140 156 CT 12,575 - 4 3 - 7 - 7 - 6 2 - 11 - 4 - 3 1 - 6 CT 4,367 762 734 988 1,541 2,342 2,232 2,647 3,179 3,706 3,615 3,833 (c) Q max - Q min in baseline (m 3 /s) and (Q max - Q min ) (%) (d) Onset of reversed flow in baseline (day of year) and its change (day) Station Baseline (m 3 /sec) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early Station Baseline (DOY) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early Dam 17,634 - 18 - 18 - 18 - 35 - 53 - 53 - 53 - 71 - 88 - 88 - 88 LO 159 2 32 - 27 4 30 62 0 32 - - - KC 16,770 - 18 - 16 - 18 - 35 - 53 - 51 - 53 - 71 - 78 - 73 - 81 PK 162 0 30 - 30 2 28 59 - 1 30 - - - PP 11,334 - 17 - 15 - 18 - 35 - 52 - 50 - 53 - 69 - 70 - 64 - 74 (e) Duration of reversed flow in baseline (day) and its change (day) LO 6,026 - 9 - 3 - 16 - 18 - 28 - 2 2 - 33 - 37 - 49 - 46 - 53 Station Baseline (Days) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early PK 7,669 - 10 - 4 - 15 - 20 - 30 - 25 - 35 - 40 - 54 - 51 - 58 NL 10,858 - 17 - 12 - 20 - 32 - 45 - 32 - 56 - 51 - 55 - 47 - 60 LO 103 - 6 - 9 - 7 - 15 - 31 - 32 - 29 - 48 - - - KK 992 - 20 - 2 - 24 - 34 - 37 - 10 - 53 - 39 - 42 - 26 - 49 PK 96 - 5 - 8 - 3 - 13 - 27 - 33 - 25 - 45 - - - CD 1,428 - 19 12 - 29 - 32 - 30 1 - 51 - 31 - 34 - 17 - 41 (f) Volume of reversed flow in baseline (10 6 m 3 ) and its change (%) VN 8,043 - 1 6 - 7 - 22 - 30 - 38 - 26 - 51 - 46 - 51 - 44 - 56 Station Baseline (Mm 3 ) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early MTu 2,779 - 16 - 3 - 21 - 29 - 34 - 19 - 48 - 41 - 45 - 37 - 51 MTo 277 0 13 - 1 - 5 - 19 - 10 - 26 - 39 - 53 - 51 - 5 7 LO 7,245 - 22 - 27 - 16 - 43 - 65 - 69 - 60 - 86 - 100 - 100 - 100 CT 8,208 - 16 - 4 - 22 - 30 - 37 - 24 - 50 - 45 - 50 - 43 - 56 PK 12,180 - 23 - 30 - 18 - 46 - 68 - 74 - 63 - 89 - 100 - 100 - 100 110 Table 4 - 6 Same as Table 4 - 3 but for w et year (2000). Blank fields in (d) and (e) denote no flow reversal. (a) Q max in baseline (m 3 /s) and Q max (%) (b) Q min in baseline (m 3 /s) and Q min (m 3 /s) Station Baseline (m 3 /sec) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early Station Baseline (m 3 /sec) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early Dam 51,826 - 10 - 10 - 10 - 20 - 30 - 30 - 30 - 40 - 50 - 50 - 50 Dam 4,733 2,266 2,266 2,266 4,532 6,798 6,798 6,798 9,065 11,331 11,331 11,331 KC 5 0,684 - 10 - 12 - 12 - 19 - 28 - 29 - 29 - 36 - 44 - 46 - 47 KC 5,107 2,247 2,258 2,287 4,569 6,817 6,814 6,627 9,055 10,160 11,286 9,451 PP 29,477 - 5 - 6 - 7 - 10 - 17 - 17 - 20 - 24 - 32 - 30 - 34 PP 5,042 1,610 1,555 1,776 3,233 4,794 4,760 4,898 6,338 7,835 7,806 7,313 L O 7,396 - 6 - 11 - 6 - 12 - 18 - 24 - 17 - 24 - 31 - 35 - 31 LO - 2,935 410 ( - 14%) 931 ( - 30%) 226 ( - 4%) 850 ( - 29%) 1,349 ( - 46%) 1,787 ( - 61%) 1,174 ( - 40%) 1,852 ( - 63%) 2,461 ( - 84%) 3,007 ( - 102%) 2,247 ( - 77%) PK 9,135 - 6 - 10 - 3 - 12 - 18 - 24 - 15 - 23 - 30 - 33 - 29 PK - 6,90 2 1,090 ( - 16%) 2,052 ( - 30%) 248 ( - 4%) 2,294 ( - 33%) 3,701 ( - 54%) 4,510 ( - 65%) 2,702 ( - 39%) 5,167 ( - 75%) 6,497 ( - 94%) 7,444 ( - 108%) 5,481 ( - 79%) NL 27,471 - 4 - 3 - 7 - 9 - 14 - 12 - 16 - 19 - 25 - 22 - 26 NL 7,473 2,051 1,960 2,264 3,772 5,306 5,274 5,449 6,670 7,985 7,981 8,006 KK 8,222 - 10 - 5 - 15 - 18 - 28 - 24 - 33 - 40 - 51 - 45 - 54 KK 99 143 135 163 291 457 450 475 634 788 782 802 CD 10,441 - 10 - 4 - 16 - 19 - 30 - 24 - 35 - 41 - 50 - 45 - 56 CD 138 139 116 166 286 502 489 533 804 1,083 1,063 1,136 VN 18,999 - 3 - 1 - 5 - 6 - 10 - 8 - 12 - 14 - 17 - 15 - 19 VN 6,466 1,652 1,546 1,862 2,914 4,047 3,980 4,153 4,889 5,685 5,668 5,756 MTu 7,140 - 3 - 2 - 7 - 8 - 13 - 10 - 16 - 18 - 22 - 20 - 25 MTu 1,342 522 487 590 982 1,406 1,379 1,452 1,763 2,123 2,113 2,155 MTo 1,737 - 6 7 - 17 - 16 - 19 - 13 - 27 - 26 - 3 0 - 27 - 35 MTo 41 111 102 125 213 213 213 214 214 213 213 213 CT 21,128 - 3 - 1 - 6 - 7 - 11 - 9 - 13 - 16 - 20 - 17 - 22 CT 6,484 1,654 1,542 1,857 2,904 4,032 3,959 4,146 4,932 5,858 5,837 5,900 (c) Q max - Q min in baseline (m 3 /s) and (Q max - Q min ) (%) (d) Onset of reversed flow in baseline (day of year) and its change (day) Station Baseline (m 3 /sec) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early Station Baseline (DOY) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 5 0 % Delay 50% Early Dam 47,094 - 16 - 16 - 16 - 32 - 47 - 47 - 47 - 63 - 79 - 79 - 79 LO 141 6 32 - 21 7 11 35 - 17 14 34 - 6 KC 45,577 - 16 - 18 - 18 - 31 - 46 - 48 - 47 - 60 - 72 - 76 - 73 PK 143 5 33 - 21 7 10 41 - 18 14 36 - 5 PP 24,436 - 12 - 14 - 15 - 25 - 40 - 40 - 45 - 55 - 70 - 68 - 71 (e) Duration of reversed flow in baseline (day) and its change (day) LO 10,332 - 8 - 17 - 7 - 17 - 26 - 35 - 24 - 35 - 46 - 54 - 44 Station Baseline (Days) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early PK 16,037 - 10 - 19 - 3 - 21 - 33 - 42 - 25 - 45 - 58 - 66 - 51 NL 19,998 - 16 - 14 - 21 - 31 - 46 - 44 - 50 - 60 - 74 - 70 - 76 LO 114 - 8 - 7 - 6 - 12 - 21 - 32 - 14 - 28 - 66 - - 45 KK 8,123 - 12 - 7 - 17 - 22 - 34 - 29 - 40 - 48 - 61 - 55 - 64 PK 111 - 7 - 13 - 7 - 13 - 23 - 57 - 14 - 53 - 80 - - 60 CD 10,303 - 1 1 - 5 - 18 - 22 - 35 - 29 - 41 - 50 - 61 - 56 - 67 (f) Volume of reversed flow in baseline (10 6 m 3 ) and its change (%) VN 12,532 - 18 - 15 - 23 - 33 - 48 - 44 - 51 - 60 - 71 - 68 - 74 Station Baseline (Mm 3 ) 10% 10% Delay 10% Early 20% 30% 30% Delay 30% Early 40% 50% 50% Delay 50% Early MTu 5,798 - 13 - 11 - 19 - 26 - 40 - 36 - 44 - 52 - 64 - 61 - 69 MTo 1,696 - 12 1 - 25 - 29 - 32 - 26 - 41 - 39 - 43 - 40 - 49 LO 15,181 - 18 - 40 - 7 - 38 - 57 - 75 - 46 - 75 - 92 - 100 - 86 CT 14,644 - 16 - 12 - 21 - 30 - 43 - 40 - 47 - 56 - 68 - 65 - 72 PK 31,642 - 20 - 4 9 - 3 - 42 - 63 - 85 - 46 - 82 - 99 - 100 - 89 111 4.4. Summary and Conclusion R iver - floodplain water storage is found to play a crucial role in modulating the hydrology of the MR B . P otential upstream flow regulations could disrupt the natural flood dynamics in the T S L region and Mekong Delta. Results indicate that the river - floodplain water explains ~26% of the total storage dynamics in the MRB and ~49% in the LMRB, suggesting tha t the potential flow alterations can largely modify the natural regime of the Lower Meko n g hydrology. It is found that the reduction in the peak of flood pulse by more than 20% near Stung Treng gauging station could cause a significant alteration on the wa ter balance of the TSL, potentially ceasing the flow reversal in the TSR and disrupting t he lake flood dynamics, if the flood peak at the same location is dampened by 50% and delayed by one - month. During average and wet years, flood occurrence is likely to increase at the outer fringe of the permanent water in the TSL and post - flooding agricu l tural regions in the middle reach of the Mekong Delta; however, during dry years flood occurrence could reduce by up to 5 months or more around the outer edge of the f looded areas in the TSL region, in the flood - recession agricultural region at the vicini t y of the Mekong upstream of Phnom Penh, and downstream portion of the Mekong delta. Further, while areas flooded for less than five months and over six months are like ly to be impacted significantly by flow regulations, areas flooded for 5 - 6 months could b e impacted the least. These results provide new insights about how the downstream flood dynamics could change under different levels of upstream flow regulation by pro posed dams, which have important implications for sustainable hydropower development to e nsure food security and ecological integrity in the Mekong region. 112 Chapter 5. Impact of Manmade Reservoirs on Mekong River Basin Hydrology over the Past Years 5.1. Introduction In this chapter , historical flood dynamics of the entire Mekong R iver B asin (MRB) is simulat e d at fine resolution grids (i.e., ~5km) with explicit representation of inundation dynamics of individual reservoirs. While Chapt er 4 mainly dealt with the investigati o n of the cumulative effect of upstream flow regulation on the flood dynamics in the spe c ific regions, i.e., the Mekong Delta and Tonle Sap Lake (TSL), the goal in this chapter is to model the operation of individual r eservoirs and the consequent changes in flood dynamics over the entire MRB. T he understanding of flood inundation dynamics in the entire MRB is currently limited. S tudies based on observations have identified that there are shifts in flow patterns after some dams have been built ( e.g., Arias, Cochrane, et al., 2014; Räsänen et al., 20 1 2; Sabo et al., 2017; W. Wang et al., 2017) ; however, these studies have not considered the inundation extent dynamics due to data limitations. There has been an increase in model - based studies in the past decade; however, these studies have focused on s o me parts of the MRB such as the UMRB (e.g., Räsänen et al., 2012) , LMRB ( e.g., D a ng et al., 2018; Trung et al., 2018) , 3S - river basin (e.g., Wild & Loucks, 2014) , Mun river basin (e.g., Akter & Babel, 2012) , and TSL and the Mekong Delta region (Minh et al., 2019; A. Smajgl et al., 2015) . There are some modeling studies for the entire MRB ( e.g., Lauri et al., 2012; Piman et al., 2013; Sridhar et al., 2019; W. W ang et al., 2016) of both natural river - floodplain and manmade - reservoirs over the entire MRB. A recent and notable study is by Bonnema and Hossain (2017) that used a 0.1° VIC model over the MRB to derive the time series inflow into some reservoirs in the MRB. The inundation of individual 113 reservoirs was of interest in their study, hence the inundation over t he other parts of natural river - floodplain was not simulated. As such, while there is an increasing body of literature on the hydrologic changes in different parts of the MRB, there is a lack of an integrated study that holistically and explicitly simula t es the natural and human - induced changes in flood dynamics over the entire basin. The goal of this study is to fill this gap by using a newly integrated river - floodplain - reservoir hydrodynamics model for the entire MRB. We address the following research q u estions: . How have the flood dynamics and surface water storage in the MRB changed over the past four decades? Are the effects of dams significant compared to that of climate variability? What will be the role of existing reservoirs in mod ulating surface water storage and inundation dynamics over the MRB in the future? We answer these questions by simulating the historical dynamics of river - floodplain - reservoir storages over the entire M RB for the period of 1979 - 2016 and at a spatial reso lution of 3 - arcmin (5 km). The simulated flood extent is downscaled to a further finer resolution of 3 - arcsec (90 m) to investigate the fine - scale inundation extent and patterns. Such high - resolution and large - scale modeling is accomplished by integrating a newly developed reservoir scheme (Chapter 3) into a global river - floodplain hydrodynamics model CaMa - Flood (D. Yamazaki et al., 2013) over the MRB. Simulations of river discharge and water level are vali dated with ground - based observations and those of ri ver - floodplain - reservoir inundation dynamics with state - of - the - art remote sensing products. The newly integrated model simulates reservoirs as an integral part of river - floodplain systems in high - resoluti on and over large domains, making it 114 suitable for in vestigating the changing hydrology of the Mekong river where natural hydrologic processes currently dominate river - floodplain hydrodynamics but the influence of dams is increasing at an unprecedented rate . 115 5.2. Materials and Methods 5.2.1. CaMa - Flood and HiGW - MAT mo dels In the Chapter 5 , we use CaMa - Flood version - 3.94 which includes major updates to the previous version used in Chapter 4 . The spatial resolution is set to 3 - arcmin (5km) with the capability to downsca le simulated flood depth to 3 - arcsec (90m) resolutio n. For an improved representation of channel bifurcation and the processes therein (D. Yamazaki et al., 2014) which are critically important in the LMRB the maximum number of bifurcation channels is increased from 6 to 10 com pared to that in Chapter 4 . The previously used SRTM DEM (Shuttle Radar Topography Mission DEM) has been suggested to have multiple errors including absolute bias, stripe noise, speckle noise, and tree height bias (D. Yamazaki et al., 2017) . Here we use MERIT DEM (M ulti - Error - Removed Improved - Terrain DEM; Yamazaki et al., 2017 ) and MERIT Hydro (D. Yamazaki et al., 2019) in which those errors have been resolved such that the stripe - like artifacts in simulated flood extents found in low - relief areas, specifically in the Mekong D e lta region (Chapter 4) , are eliminated. These advancements are essential for a more realistic simulation of river - floodplain dynamics in the MRB. In addition, a new reservoir inundation and release scheme has been incorporated into the modeling framework ( details in Chapter 5. 2.4 and 5. 2.5). CaMa - Flood is fed by the runoff simulated by HiGW - MAT (Y. N. Pokhrel et al., 2015) , which is a global hydrological model based on the land surface model (LSM) called the MATSIRO (Takata et al., 2003) . HiGW - MAT simulates both the natural water cycle and human activities from canopy to bedrock including evapotranspiration, infiltration, irrigation, flow regulation, and groundwater pumping on a full physical basis. In this study, we use HiGW - MAT in the natural setting (i.e., the human impact schemes are turned o ff) as done in Chapter 4 116 because the objective is to provide runoff as input to CaMa - Flood and reservoirs are simulated within CaMa - Flood (see Chapter 5. 2. 6 ). 5.2.2. Dams and Reservoir s database The specifications (e.g., location, dam purpose, commissioned year, dam dimensions, storage capacity, power generation capacity, etc.) of existing, under cons truction, and planned dams (455 in total) within the MRB are obtained from the Research Program on Water, Land, and Ecosystem (WLE; https://wle - mekong.cgiar.org/ ). However, the WLE database contains significan t omissions (e.g., missing attributes), and sometimes, errors. For example, of the 455 dams, only 128 and 173 include information on storage capacity and height, res pectively. Since the direct employment of erroneous dam specifications results in erroneous hydrodynamics modeling, all missing and erroneous specifications should be carefully curated to yield reasonable model results for validation against satellite - base d datasets such as the Global Surface Water (GSW) data (Pekel et al., 2016) . In this study, the dams that existed as of 2016 (end of simulation period constrained by the availability of WFDEI forcing data; see Chapter 5. 2.6) are imported to the CaMa - Flood modeling framework when they satisfy at least one of the following criteria: 15m, (2) storage capacity > 1 million m 3 (Mm 3 ), and (3) installed hydropower capacity > 100 Mega Watts (MW). The first criterion is commonly used to classify large dams (e.g., International Commission on Large Dams (ICOLD); (Binnie, 1987; Greathouse et al., 2006; Räsänen et al., 2017) . The second criterion is set since small inundation extents are fou nd in satellite images for dams having a height of <15 m and a storage capacity < 1 Mm 3 . When the first and second criteria are applied, most of hydropower dams, even those with installed 117 hydropower capacity in the single - digit (in MW), are included. Howev er, there were some large hydropower dams (in terms of installed capacity) for which dam height and storage values were missing, hence we additionally set the third criterion of 100MW. We fill the missing values from various resources including published r eports from the Mekong River Commission (MRC; http://www.mrcmekong.org ), Project Design Documents (PDD) provided by Clean Development Mechanism (CDM) of United Nations Framework Convention on Climate Change (UNFCCC ) ( https://cdm.unfccc.int ), documents from construction and design companies, and other peer - reviewed literature. These resources are also used to correct any erroneous records in the dam specification database . As a result, 86 dams are selected ( Figure 5 - 1 ). 118 Figure 5 - 1 The spatial distribution of river discharge and commissioned dams (as of 201 6 ) over the MRB. The background s hows the simulated long - term mean river discharge (1979 - 201 6 ; 5 - km grids). The filled circles show the height and storage capacity of 86 dams with varying color and size, respect ively. The cumulative reservoir storage capacity from 1960 through 2020 is s hown as a subplot on the lower - left corner with the highlights on the values of 2010 and 2016. The hydrological gauging stations are displayed as red hollow circles. Station names are: LP (Luang Prabang), PM (Pak Mun), PA (Pakse), ST (Stung Treng), KT (Kr atie), KC (Kampong Cham), PK (Prek Kdam), PP (Phnom Penh Port), CD (Chau Doc), CT (Can Tho), KL (Kompong Luong), and MT (My Tho). 119 5.2.3. Observed Data for Model Validation The MRC provides the observation data for river discharge and water level. Among the observation stations, 10 stations for each variable either having several years of record or those located near river confluences are selected ( Table 5 - 1 ). For the validation of inundation extent, we use two remote sensing product s: Global Surface Water (GSW) data (Pekel et al., 2016) based on Landsat imagery and our own water body detection products from Sentinel - 1. For the selected regions that include t he TSL, natural river - floodplain regions, and top 16 reservoirs (selected by considering top 10 reservoirs in height, storage capacity, and surface inundation area , respectively ), the remote - sensing based flood occurrence is used to validate the simulated food occurrence. The GSW flood occurrence data are provided for the 1984 - 2018 period, which is different from CaMa - Flood simulation period, i.e., 1979 - 2016; however, the discrepancy is expected to be acceptable since 38 - and 35 - years timespans of CaMa - Floo d simulation and GSW data overlap for 33 years. For the Sentienal - 1 based products, the comparison is done only for the overlapping period (i.e., 2014 - 2016). To validate the simulated flooded areas over the entire basin, we also utilize the GSW monthly wat er extent; specifically, the time series of total flooded areas over the entire MRB from GSW and CaMa - Flood are compared. 120 Table 5 - 1 Observation stations of the Mekong Ri ver Commission Names Use for valida tion Location Abbreviation Full Name Discharge Elevation Latitude Longitude LP Luang Prabang O O 19.892 102.137 PM Pak Mun O O 15.282 105.468 PA Pakse O O 15.117 105.800 ST Stung Treng O O 13.533 105.945 KT Kratie O O 12.487 106.024 KC Kompong Cham O O 11.997 105.470 KL Kompong Luong O 12.575 104.215 PK Prek Kdam O 11.813 104.804 PP Phnom Penh Port O O 11.575 104.923 CD Chau Doc O 10.705 105.133 CT Can Tho O O 10.033 105.790 MTo My Tho O 10.345 106.347 5.2.4. I ncorporation of Reservoirs into CaMa - Flood When it comes to importing dams into a gridded hydrodynamic model, there are two important issues to be considered. First, the maximum water depth can be lower than the dam height provided in the dam database. It can be because of prevalent errors in the d atabase or due to the mixed definition and usage of dam height (i.e., dam structure height or the maximum water storage depth). Second, as discussed in detail in Chapter 2 , DEM grid elevations represent the water surface level, not the river - reservoir bed elevation. When the pre - existing water depth is large, the deviations between the bed elevation and the DEM - based elevation can be too large to be ignored (Chapter 2.2.2) . Specifically, when large dams are constr ucted before the DEM is produced, such discr epancies can be relatively large. The above - mentioned issues exist even when a high - resolution DEM (e.g., MERIT DEM) is used. Our examination of the dams in the MRB using two 3 - arcsec global DEMs, namely the MERI T and HydroSHEDS, reveal s that most of reser voirs need considerably lower level of water impoundment than their dam heights ( Figure 5 - 2 ). The MERIT DEM, which is employed in this study, shows less deviations from the recorded dam h eight than the HydroSHEDS DEM does (vertical lines in Figure 5 - 2 ), likely 121 owing to more realistic representations of ground elevations by multiple er rors corrections in MERIT DEM. Figure 5 - 2 Deviation at 3 - arcsec grids from dam crest level to water level to achieve recorded storage capacity for (a) the MERIT and (b) HydroSHEDS DEMs. Vertical lines indicate the deviations. 122 When dams are located on the river network of raster mode l grids, a simple conversion of the latitude and longitude information into the model grid coordinates can cause a dislocation of dams that can lead to erroneous mod eling results; for example, some dams in the tributaries can be wrongly located in the main stem which may cause an unusually high dam inflow and outflow, and vice versa (Chapter 2.2.2) . Accurately locating dams is also important for realistic representatio n of inundation patterns since the upstream inundation starts from dam locations. The appro ach in Chapter 2.2.2, i.e., fine - tun ing the reservoir location in the model grid to the grid cell having the most similar upstream basin area to the known value , can not be used since To overcome the limitation, dam locations in 3 - arcsec river network are first determined as the nearest river cells from the given latitude and longitude. Then, the upstream basin area in 3 - arcsec grid DEM is used as a reference for locating the dams into 3 - arcmin model grid. Of the 86 selected dams, dam height and storage capacity information was missing for 9 and 17 dams, respectively. For the other 60 dams having both attributes, we find the ratio between database dam height and the maximum water level (i.e., water depth at the dam location to achieve recorded storage capacity) to be ~70% ( Figure 5 - 3 ). Based on this finding, the maximum water level for the 9 dams is set at 70% of the dam height reported in the database. The stora ge capacities of the 17 dams are estimated as the storage when the water level at the dam location reaches 70% of the reported dam height. The inundation extents at those levels are found to be reasonable in the visual inspections with GSW data. 123 Figure 5 - 3 Relationship between dam height in the database and adjusted dam height at 3 - arcmin CaMa - Flood modeling grids based on the MERIT DEM 5.2.5. Reservoir Operation Scheme The 86 selected dams in the MRB can be cla ssified into three general categories based on their purpose reported in the database: i rrigation (22), hydropower (62), and multi - purpose (2). We use the reservoir operation scheme in Chapter 3.2.1 for irrigation dams, which determines the seasonality of reservoir release by utilizing the water demand in a region and an optimization scheme t hat maximizes hydropower generation for hydropower and multi - purpose dams. Details about the demand - driven scheme can be found in Chapter 3.2.1 . The reservoir operation scheme in Chapter 3.2.1 can also be applied for reservoirs with any other purposes than irrigation if the seasonal water demands for that purpose are available. However, since the seasonal water demand data or prox ies are not available for hydropower 124 prod uction in the MRB, we use an optimization approach for hydropower reservoirs. Hydropower reservoir operation can be formulated as an optimization problem that maximizes hydropower benefit, F [$] as: ( 1 ) where P ( t ) is electricity price [$/Watts - hour], W ( t ) is the generated electrical energy [Watts] during unit time span of [hr], is efficiency [ - ], is specific weight of water [kg/m 3 ], Q ( t ) is the reservoir release (m 3 /s), is the turbine design flow (m 3 /s), and H ( t ) is turbi ne head [m]. Since no data are available on P ( t ), which is a rather complicated function of demand and supply and various other variables (Aggarwal et al., 2009; Weron, 2014) , we assume constant P ( t ) over time. Consequently, the optimization problem is simplified to maximize the total energy p roduction. Such simplific ation can ignore the inundation variability in small time scales (e.g., diurnal and weekly), specifically near local reservoir areas; however, such small - scale variations are expected to be averaged out as the size of the domain of interest (i.e., the enti re MRB) and the time scale (i.e., monthly) become greater . In addition to the maximization of F , we also consider a common practice in hydropower management that stores as much as water in low - demand and wet seasons and releases wa ter gradually to be prepared for high - demand and dry seasons. Such practice can be formulated to minimize the variation of reservoir storage ( ). In achieving two objectives of hydropower reservoir operation, the maximization of F is assumed to preced e the minimization of . In other words, after estimating the total discharge amount through turbine (i.e., 125 ) that maximizes F , the identical amount is redistributed without additional spillway discharge t o make the minimum under the constraints of reservoir storage capacity and time - varying inflow In ( t ). To avoid the excessive computational cost in the optimization from the iterative hydrodynamic model running , we optimize the re servoir operation using the approach in Chapter 3.2.3 that employs simulated river discharge without considering dams in optimization (or calibration) of parameters in reservoir operation scheme . Using the approach in Chapter 3.2.3, t he optimized reservoir releases are sequentially calculated from the uppermost to lowermost reservoirs , firstly to maximize F , and secondly to minimize . From the aforementioned data sources ( Chapter 5. 2.3), for 12 dams are obtained. The WLE database does not provide information on , but the installed energy capacity W max is available. For a given W max , can be calculated as , wh ere H max is the maximum available head. The only available proxy of H max is the dam height, however, equating the dam height to H max could yield too large since, in many cases, turbines are located at further downstream (i.e. , lo wer) locations from dam locations to obtain high water heads ( Figure 5 - 4 ). For this reason, we employ the streamflow with 30% probability of exceedance (i.e., Q 30 ) a s , which has been widely employed in th e previous global studies (Gernaat et al., 2017; Hoes et al., 2017; Y. Zhou et al., 2015) . To examine the uncertainty caused by the choice of exceedance probability, the simulations using Q 20 and Q 40 are also set up ( Chapter 5. 2.6) for a sensitivity analysis purpose . Those three flow exceedances are found to 126 reasonably approxima te ( Figure 5 - 5 ) and are also closely related t o the long - term average flow ( Figure 5 - 6 ). Figure 5 - 4 The profile of Lam Ta Khong P. S. dam. The dam height and maximum water depth are 42 m and 40 m, respectively, and the head difference before and after turbine is 360 m. Figure 5 - 5 Comparison of turbine design flow and 20% (Q 20 ) , 30% (Q 3 0 ) , and 40% (Q 4 0 ) stream flow exceedances for 12 reservoirs. Red circles indicate Q 3 0 , and the upper and lower bounds indicate Q 20 and Q 4 0 , respectively. A diagonal dashed line represents 1:1 line. 127 Figure 5 - 6 Mean discharge and stream flow exceedances for 86 reservoirs. Red circles indicate Q 30 , and the upper and lower bounds indicate Q 20 and Q 40 , respectively. While the real - world reservoir operation rule can be complicated, no information is publicly available for any of the reservoirs in the MRB; hence any model development in the MRB has to rely on a generic reservoir scheme that may not fully account for the complex dynamics in the actual operation rules. Anot her challenge is the validatio n of reservoir release and storage because observed river discharge is generally not available at locations immediate downstream of many of the reservoirs, and storage measurements are not available for any of the reservoirs i n the MRB. Thus, to consider t he uncertainties in reservoir operation scheme s and data, we additionally consider two hypothetical hydropower reservoir operation modes, following a similar approach employed by Piman et al. (2013): full - level and low - level. Full - level operation maintains the water level up to the maximum level and releases the surplus water through turbines or spillway. Low - level operation releases , which can be Q 20 , Q 30 , or Q 40 , 128 so long as the reservoir water level is above the minimum level. F rom the contrasting operation policies of these two extremes, we intend to simulate the probable maximum changes in inundation extent (full - level operation) and river discharge (low - level operation). Conversely, the probable m inimum changes in inundation e xtent and river discharge are simulated by low - level and full - level operations, respectively. 5.2.6. Simulation Settings HiGW - MAT is run for 1979 - 2016 period for which the WFDEI forcing data are available, and the runoff from HiGW - M AT is used as the input for Ca Ma - Flood model. With different settings for reservoirs, 15 simulations are conducted that are categorized into three groups: NAT, HIST, and ALL ( Table 5 - 2 ). NAT is a simulation with natural river - floodplain settings that simulates surface water dynamics without considering reservoirs. To assess the impact of reservoirs on MRB hydrology, NAT is compared with the historical simulations of river - reservoir - floo dplain dynamics (HIST). HIST simulations use the reservoir operation scheme o f Chapter 3.2.1 for irrigation reservoirs and optimized (opt), full - level (full), and low - level (low) operation scheme s (details in Chapter 5. 2.5) for hydropower reservoirs. The u se of different hydropower operation schemes is intended to provide the upper and lower bounds of changes under the uncertainty in reservoir operation. HIST simulations are grouped into HIST - opt, HIST - full, and HIST - low, and are further detailed according to the setting of Q turbine using Q 20 , Q 30 , and Q 40 . In addition, to estimate the probable impact of the existing 86 dams, we set up ALL simulations , which is designed to investigate the difference in hydrodynamic responses of the Mekong river with and with out the newly built reservoirs. To make ALL simulations comparable to NAT and HIST simulations, the same WFDEI forcing data and HiGW - MAT runoff are used for ALL simulations as well. The main difference is the start timing of reservoir 129 operation in which al l 86 dams are assumed to exist from 1979. That is, in ALL simulations the ope ration of all 86 dams begins from 1979 regardless of the year each dam is historically commissioned. Other than the start year of reservoir operation, the modeling settings of ALL simulations (e.g., runoff time series ; reservoir operation modes; Q turbine ) are identical to those of HIST simulations. Table 5 - 2 Simulation s ettings Reservoir operation Irrigation reservoir operation scheme Start year of reservoir operation Hydropower reservoir operation scheme Q tur bine NAT X - - - - HIST - opt - Q20 O Chapter 3.2.1 Individual historical commissioned year Optimized level Q 20 HIST - opt - Q30 Q 30 HIST - opt - Q40 Q 40 HIST - low - Q20 Low level Q 20 HIST - low - Q30 Q 30 HIST - low - Q40 Q 40 HIST - full Full le vel - ALL - opt - Q20 The start year of simulation (1979) Optimized level Q 20 ALL - opt - Q30 Q 30 ALL - opt - Q40 Q 40 ALL - low - Q20 Low level Q 20 ALL - low - Q30 Q 30 ALL - low - Q40 Q 40 ALL - full Full level - 130 5.3. Results and Discussion 5.3.1. River Discha rge and Water Level Figure 5 - 7 and Figure 5 - 8 present the evaluation of simulated river discharge and water level , respectively, at the selected stations ( Table 5 - 2 ) in the mainste m and tributaries of the Mekong r iver. The high fluctuations in the observations in the delta region (e.g., Can Tho) are due to the tidal effect, which is not yet considered in the current version of the model. In terms of the monthly variability of inunda tion extents that this study main ly focuses on, the tidal effect in a sub - monthly time scale is expected to be averaged out in a monthly time scale. T he total reservoir storage capacities have been doubled in 2010 and 2016, respectively, compared to the to tal capacities of their preceding decades ( Figure 5 - 1 ), hence the impacts of reservoirs are investigated for two periods, i.e., before and after 2010. Overall, the reservoir operation dampens peak flow (level) and increases low flo w (level), and the impact of reservoir operation on river discharge and water level is smaller than that of inter - annual climate variability. The reservoir operation has exerted a limited influence on the variations of discharg e and water level during 1979 - 2009, except for the Mun river basin (Pak Mun station) where many reservoirs were already built before or early in the simulation period. As more dams are built in 2010, especially the mega - sized dams built after 2014, the imp acts on reservoir operation o n river discharge and water level become more pronounced such that the difference between NAT and HIST simulations is noticeable in the 2014 - 2016 period. From these results, we conclude that climate variability has dominated th e flood dynamics over the res ervoir operation in the past. Nonetheless, the potential impact of existing 86 dams is found to be considerable as found in the ALL simulations where the river discharge and water level are significantly dampened by the 86 dams that are assumed to be in op eration from 1979 ( Figure 5 - 9 and Figure 5 - 10 ). Sinc e the 131 number of dams and their storage capacities are increasing in a rapid pace ( Figure 5 - 1 ), the impact of reservoirs on the fu ture MRB hydrology is expected to be accelerated. 132 Figure 5 - 7 Validation of simulated river discharge. Th e orange shading indicates the level of uncertainty in the optimized hydropower operation . Daily tim e series, seasonal cycle for the time span when the observations are available, and seasonal cycle for 2010 - 2016 are presented in three columns from left to right, respectively. The locations of stations are shown in Figure 5 - 1 . 133 Figure 5 - 8 Validation of simulated water level. The orange shading indicates the level of uncertainty in the optimized hydropower operation. Daily time series, seasonal cycle for the time span when the o bservations are available, and seasonal cycle for 2010 - 2016 are presented in three columns from left to right, respectively. The locations of stations are shown in Figure 5 - 1 . 134 Figure 5 - 9 Potent ial changes in river discharge by the existing 86 dams estimated from the ALL simulations . In the ALL simulations, the 86 dams are modeled to be operated from the beginning (i.e., 1979) through the ending (i.e., 2016) years for the given cli mate conditions, which are identical to those given to the HIST (historical) simulations. 135 Figure 5 - 10 Potential changes in water level by the existing 86 dams estimated from the ALL simulations. In the ALL s imulations, the 86 dams are modeled to be operated from the beginning (i.e., 1979) through the ending (i.e., 2016) years for the given climate conditions, which are identical to those given to the HIST (historical) simulations. 136 5.3.2. Flood Occurrence Figure 5 - 11 presents the 90 - m (3 - arcsec) grid flood occurrence (i.e., flood frequency in percentile) for the period of 1979 - 2016 over the entire MRB, which is derived by downsc aling the monthly - average flood depth simulated by CaMa - F lood. Flood occurrence in the major lakes, natural river channels, and reservoirs ( Figure 5 - 1 ) is broadly reproduced by the model. Because no ground - based observations exist to evaluate the flooded extents and storages in the Mekon g, we evaluate the results with remote sensing - based pr oducts from Landsat (GSW data; Pekel et al., 2016) and Sentinel - 1. It is worth noting that the GSW data has a spatial resolution of 0.00025º, hence the data is upscaled data to four - times coarser resol ution (i.e., 0.00100º) for a consistent comparison with CaMa - Flood results at 0.00083º resolution Figure 5 - 12 . The results at the original resolution of 0.00025º are also provided in Figure 5 - 13 . A reprojec tion of GSW data to the identical resolution of model results is avoided since it adds distortions, specifically for the number of pixels having small non - zero flood occurrence that delin eates the maximum inundation extent. 137 Figure 5 - 11 Simulated flood occurrence in 3 - arcsec (90 m) over the MRB. Black boxes indicate the regions used to validate the simulated spatial inundation dynamics. 138 Figure 5 - 12 Spati al validation of simulated inundation dynamics with 0.00100 º resolution GSW flood occurrence data . Simulated flood occurrence (left; CaMa - Flood) for 1979 - 2016 is compared with the GSW flood occurrence (right; Pekel et al. 2016) for 1984 - 2018. The location s of sub - panels are shown in Figure 5 - 11 . The resolution of CaMa - Flood is 0.00083º . 139 Figure 5 - 13 Spatial validation of simulated inundation dynamics with 0.00025 º resolution GSW flood occurrence data . Simulated flood occurrence (left; CaMa - Flood) for 1979 - 2016 is compared with the GSW flood occurrence (right; Pekel et al. 2016) for 1984 - 2018. The locations of sub - panels are shown in Figure 5 - 11 . The resolution of CaMa - Flood is 0.00083º . In general, the patterns of simulated and satellite - based flood occurrences are similar for both natural river - floodplain - lake and manmade reservoir systems ( Figure 5 - 12 and Figure 5 - 13 ; hereafter only Figure 5 - 12 is mentioned ). First, the main body of TSL is accurately simulated to be always flooded (dark blue; 100% flood occurrence) in the CaMa - Flood, but it is sometimes wrongly represented n ot to be flooded (light blue; <100%) in the GSW data. Second, the 140 variations from the center of TSL to the maximum extents are smooth and continuous in the CaMa - Flood result while they are abrupt and sometimes intermittent (i.e., no flood areas in white) i n the GSW data. Those differences are similarly found in other natural river - floodplain regions ( Figure 5 - 12 FG), which can be attributed to either a model overestimation or sate llite product underestimation. The model can overestim ate the flood occurrence due to uncertainties in meteorological (e.g., errors in the magnitude and spatio - temporal distribution of precipitation), topographic data (e.g., overrepresentation of flat area s; underrepresentation of natural and artificial river bank), and model parameterization (e.g., under - and over - estimations of evapotranspiration and runoff). The underestimation in satellite products can be the case since Landsat imagery cannot penetrate t he interfering objects over the water body, e.g., clou ds, debris, and vegetations. Due to such limitation, the number of cloud - free images is limited especially in monsoon season, which can lead to the underrepresentation of flood occurrence in satellite p roducts. Such limitation can be complemented by using the advanced remote sensing data. For example, Sentinel - 1 in short - wavelength (5.5 cm) can penetrate clouds so that the flood extent in flooding season can be better represented, but the penetration of Sentinel - 1 is still limited, specifically for very den se vegetation canopy at high biomass ( Figure 5 - 14 ). For the vegetation dense regions, L - band remote sensing products can be useful (Urbazaev et al., 2018) . The stripes - like noises in Sentinel - 1 product are due to the processing issues from IPF of ESA , which should be removed for further quantitative analys is. 141 Figure 5 - 14 Spatial validation of simulated inundation dynamics with the Sentinel - 1 product. Simulated flood occurrence (left; CaMa - Flood) for 2014 - 2016 is compared with the Sentinel - 1 occurrence (ri ght) for 2014 - 2016 . The locations of sub - panels are shown in Figure 5 - 11 . The resolutions of CaMa - Flood and GSW data are identical to 0.00083º . For the reservoir inundation ( Figure 5 - 12 B - E;H - S), the sim ilarity in the inundation extent and patterns between the simulation and remote sensing product demonstrates that the issues of high uncertainties in the da m and reservoir dataset ( Chapter 5. 2.2) are reasonably curated; if not, e.g., the erroneous dam spec ifications are employed, the simulated inundation extents can be too smaller or larger than the satellite - based inundation extents. The core - part of 142 reservo irs built before 1979 and 1984 for CaMa - Flood and GSW data, respectively, have large permanent water - Flood results, those reservoirs show smaller seasonal variation of inundation near the reservoir boundaries. It is attributed to th e flat areas within reservoirs in the SRTM data, which is the main source of MERIT DEM that we emplo yed. When the SRTM was launched on February 2000, if the reservoirs already impounded water in their upstream, the water surface elevation that is flat over the reservoir are measured. The water body portion is not yet removed while many other errors and n oises are removed (D. Yamazaki et al., 2017, 2019) . The grid cells within the flat area are simulated to be inundated all at once, hence the CaMa - Flood simulates the permanent reservoir water body port ions of the reservoirs built before 2000 to be larger and less seasonally varying compared to the GSW data represents. On the contrary, the bathymetries of reservoirs built after 2000 can be parameterized to have more spatially varying elevations, henc e fl ood occurrences change within reservoir extents. There are some reservoirs whose new permanent water area (i.e., yellow - to - green color) is shown in a different color in different products and results of CaMa - Flood, GSW data, and Sentinel - 1 data (e.g., Xiao wan, Nuozadu, and Nam Ngum 2 in Figure 5 - 12 B, C, E, respectively) . It could be because the time period of the remote sensing product is different from the period of the model (GSW data), the limited images availability due to the a tmosphere condition (GSW data), or the time period is set to too early from the launching date to regularly and stably acquire the remote sensing products (Sentinel - 1). While t he inundatio n pattern starts from the location of a dam , the dam location can b e located at any positions within a modeling grid. Hence, the downscaled flood extent can contain the errors in the inundation pattern starting point in the degree of less than 5 - km scale. For example, the inundations near the dam locations initiate slight ly at downstream of dams (e.g., 143 Nuozadu, Miaowei, and Nam Khan 2 in Figure 5C, K, and O, respectively) or at upstream of dams (e.g., Theun - Hinboun exp., Yali, and Manwan in Figure 5H, J, an d L, respectively). Such issues cannot be completely eliminated, bu t they can be alleviated by increasing the spatial resolution. Additionally, to better simulate spatial variation of inundation extent of the reservoirs built before the SRTM lunch, the mor e realistic reservoir bathymetry can be produced by utilizing other credible data or local data (van Bemmelen et al., 2016; Busker et al., 2018; Li et al., 2019) . Overall, the maximum inundation extent s are found to be reasonably well simulated, and the recent reservoirs show more spatially varying flood depth within the reservoirs. As such, the modeling approach used in this study is pr omising in terms of modeling future dams as well as existing dams. 5.3.3. Flooded Area and Surface Water Storage Figure 5 - 15 shows the historical surface water storage and extent dynamics over the MRB simulated by CaMa - Flood in terms of the changes in seasonal variation for entire period for 1979 - 2016 ( Figure 5 - 15 a) and at dec adal intervals ( Figure 5 - 15 b - e) and intra - annual variability of the selected years (1998 and 2015 for dry years; 2000 for wet years; Figure 5 - 15 f - h), along with the comparison of GSW monthly flooded area. The daily time series for the entire period of 1979 - 2016 are presented in Figure 5 - 16 . Overall, the flooded area accounts for 3 - 8% (24,000 - 64,000 km 2 ) and less than 1% (<800km 2 ) of the MRB area in wet and dry seasons, respectively, according to CaMa - Flood results. Evidently, large deviations are fou nd between the flooded area of CaMa - Flood and GSW data, specifically in the years before 2000 and flooding seasons ( Figure 5 - 15 and Figure 5 - 16 ). The deviations are partly attributed to the model 144 overestimation as discussed in Chapter 5. 3.2; however, considering many of Landsat image 2000 and flooding seasons ( Figure 5 - 17 ), the large deviations are likely attributed to the underesti mation of GSW data. This assertion is partly supported by the decreased deviations in Figure 5 - 16 ; Figure 5 - 17 ). Figure 5 - 15 Historical flooded area and surface water storage dynamics over the MRB. The seasonal average and daily time series are presented for (a - e) the selected periods and (f - h) dry - and wet - years, respectively. The orange and pink shadi ng s indicate the level o f uncertainty in the optimized hydropower operation for the flooded area and surface water storage, respectively. The monthly GSW flooded area for 1984 - 2016 is presented for the periods and years wherever available. 145 Figure 5 - 16 Historical flooded area and surface water storage dynamics over the MRB for 1979 - 2016 . The orange and pink shadings indicate the level of uncertainty in the optimized hydropower operation for the flooded a rea and surface water st orage, respectively. The monthly GSW flooded area for 1984 - 2016 is presented for the periods and years wherever available. 146 Figure 5 - 17 The monthly GSW flooded area for 1984 - 2016 . Pixels are classified to either records of a month are masked out (i.e., no lines or broken lines) when the entire pixels in the Whil e the inter - annual variability is evident in the surface water storage and extent dynamics, no significant inter - annual trend is found in the NAT simulation for the minimum, maximum, mean, standard deviation, and the maximum amplitude o f flooded are a (Mann - Kendall test, p>0.31 for all variables, - value is provided) and surface water storage (p>0.42), respectively. For the HIST simulations, except for HIST - low - Q20 that is designed to provide the lowest bound in floo ded area and storage water change by dams ( Chapt er 5. 2.6), all HIST simulations show increasing trends in the minimum value of 147 flooded area and minimum surface water storage, respectively (p<0.01). Meanwhile, none of HIST simulations shows significant tren ds in standard deviations and the maximum amplit udes (p>0.39). Results suggest that the reservoirs are likely to have exerted influence on the surface water storage and extent dynamics over the MRB in dry season while the inter - annual climate variability h as dominated the inter - annual variability of sur face water storage and extent dynamics, specifically for wet season. As the number and storage capacities of reservoirs increase, the response of surface water storage and extent dynamics can be considerably different from the past even if the identical c limate condition is given. In HIST simulations, two dry - year of 1998 and 2015 are similarly dry, but the variation of change by reservoirs is evidently greater in 2015 than that in 1998. The continuing impact s of existing 86 dams (as of 2016) on the future MRB hydrology can be found in ALL simulations, where the all existing dams are simulated to be operated since 1979 ( Chapter 5. 2.6). Compared to HIST simulations, ALL simulations show higher variability in th e surface water storage and extent dynamics over the entire simulation period ( Figure 5 - 18 ; Figure 5 - 19 ). It is noted that the ALL simulations show no significant trends in both flooded area and surface storage for any of the attributes of the minimum, m aximum, mean, standard deviation, and the maximum amplitude (p>0.4) since the meteorological forcing of HIST simulation is also used for ALL simulation. The result should be interpreted with caution since different results can be obtained when other climat e projections having wetter or drier trend are used. 148 Figure 5 - 18 The potential impacts of existing 86 dams on surface water dynamics estimated from the ALL simulations for selected periods and years . In the ALL simulations, the 86 dams are mode led to be operated from the beginning (i.e., 1979) through the ending (i.e., 2016) years for the given climate conditions, which are identical to those given to the HIST (historical) simulations. 149 Figure 5 - 19 The potential impacts of existing 86 dams on surface water dynamics estimated from the ALL simulations for the entire simulation period . In the ALL simulations, the 86 dams are modeled to be operated from th e beginning (i. e., 1979) through the ending (i.e., 2016) years for the given climate conditions, which are identical to those given to the HIST (historical) simulations. 150 5.4. Results and Discussion In this chapter, the historical dynamics of surface water storage and extent dynamics over the entire MRB are simulated for the period of 1979 - 2016 (HIST simulations). Through the modeling of the natural river - floodplain - lake system and manmade reservoirs in an integral manner, the natural and human - induced changes in flood dynamics over the entire basin are investigated altogether. To do so, the newly developed reservoir scheme (Chapter s 2 and 3) is incorporated into a global river - floodplain hydrodynamics model (CaMa - Flood; Yamazaki et al. 2013 ). F or reser voir operation schemes, the demand - driven operation scheme ( Chapter 3.2.1 ) and optimized operation scheme are used for irrigation and hydropower reservoirs, respectively. By using wide range of turbine designed flow Q turbine ( Q 20 , Q 30 , and Q 40 ) and various operation modes (opt, low, and full), the uncertainties in reservoir operations are considered. The results are validated with the ground - based observations for discharge and water level and with the remote sensing products (Landsat and Senti nel - 1 derived products) for spatial inundation dynamics. Along with the historical (HIST) simulations, the probable future impact of the existing 86 dams (as of 2016) is estimated from ALL simulations, where the same meteorological data and reservoir opera tion schemes are used yet the all existing dams are assumed to be operated from the beginning of the modeling period (i.e., 1979). Results suggest that the inter - annual flood dynamics of surface water storage in the MRB have been mainly controlled by clima te variabilit y over the past four decades . Compared to the climate variability, the reservoir impacts are small in the past, specifically before 2010; however, the surface water dynamics is found to be already significantly altered by the reservoirs, and r esults also s uggest that the surface water storage dynamics in the future would be likely different from that in the past even without additional dam constructions. As a greater number of mega - sized are 151 proposed, planned, and under construction, the impact s of reservoi rs on the MRB hydrology is expected to be accelerated. As the change of surface water storage and extent dynamics is (Sabo et al., 2017) of hydropower reservoir op eration rule would be desirable for the sustainable hydro - ecological system of the MRB. 152 Chapter 6. Summary and Conclusion Manmade reservoirs are important components of the terrestrial water cycle. Considering the importance of manmade reservoirs in med iating proces ses in both local and global scales, improving reservoir modeling is a n indispensable effort in hydrologic and climate modeling ; however, reservoirs have been poorly modeled by treating them as separated entities from natural river - floodplain system not on ly in course - resolution models but also in recently developed high - resolution models. In Chapters 2 and 3 , the new high - resolution continental - scale reservoir storage dynamics and release scheme is presented by enhancing exist ing schemes and adding critica l novel parameterizations to improve reservoir storage and release simulations. The new scheme simulates river - floodplain - reservoir storages in an integrated manner considering their spatial and temporal variations. A new cali bration scheme is also incorpo rated to better simulate reservoir dynamics considering cascade - reservoir effects. Further, since no reservoir bathymetry data are available over large domains, a state - of - the - art digital elevation model and reservoir extent d ata are used to derive reservo ir bed elevation. The new scheme can be used as a standard - alone surface water model (e.g., CaMa - Flood) that digests runoff outputs from Land Surface Models (e.g., HiGW - MAT) or can be integrated within the river - floodplain rou ting scheme of hydrological mo dels (e.g., LEAF - Hydro - Flood and Community Land Model). The new modeling framework for integrated simulation of river, reservoirs, and floodplain is first tested for the Contiguous US having abundant data. Comparison of result water inundation extents , and the results of reservoir release and storage are also found to be improved. 153 In Chapter s 4, the po tential disruption of flood dy namics in the Tonle Sap River (TSR) and Mekong Delta by upstream flow regulation is investigated in the form of sensitivity analysis. It is found that the effects of flow regulation on downstream river - floodplain dynamics are relatively predictable along t he mainstem of Mekong river, but flow regulations could potentially disrupt the flood dynamics in the TSR and small distributaries in the Mekong Delta. Modeling results suggest that TSR flow reversal could cease if the Mekong flood pulse is dampened by 50% and delayed by one - month. As the upstream regulation is intensified, flood occurrence near Tonle Sap Lake and Mekong river reaches is increased while other regions are less flooded. In Chapter 5, the historical dynamics of s urface water storage and exten t dynamics over the entire MRB are simulated for the period of 1979 - 2016 . The optimized reservoir operation scheme is newly incorporated to the modeling framework with the consideration of uncertainties in reservoir operations . Results suggest that the int er - annual flood dynamics of surface water storage in the MRB have been mainly controlled by climate variability over the past four decades . Compared to the climate variability, the reservoir impacts are small in the past, specifically before 2010; however, the surface water dynamics is found to be already significantly altered by the reservoirs, and results also suggest that the surface wate r storage dynamics in the future would be likely different from that in the past even without additional dam construct ions. It is expected that the new modeling framework presented in this study will inform the incorporation of reservoirs in hyper - resoluti on models to improve simulations of terrestrial water storage and flow and examine reservoir - atmosphere interactions o ver large domains. 154 REFERENCES 155 REFERENCES Adam, J. C., Haddeland, I., Su, F., & Lettenmaier, D. P. (2007). Simulation of reservoir influences on annual and seasonal streamflow changes for t rivers. Journal of Geophysical Research Atmospheres , 112 (24), 1 22. https://doi.org/10.1029/2007JD008525 Aggarwal, S. K., Saini, L. M., & Kumar, A. (2009). Electricity price forecasting in deregulated mark ets: A review and evaluat ion. International Journal of Electrical Power & Energy Systems , 31 (1), 13 22. https://doi.org/10.1016/j.ijepes.2008.09.003 Akter, A., & Babel, M. S. (2012). Hydrological modeling of the Mun River basin in Thailand. Journal of Hydr ology , 452 453 , 232 246. https://doi.org/10.1016/j.jhydrol.2012.05.059 Allison, M. A., Demas, C. R., Ebersole, B. A., Kleiss, B. A., Little, C. D., M eselhe, E. A., et al. (2012). A water and sediment budget for the lower Mississippi - Atchafalaya River in fl ood years 2008 - 2010: Impl ications for sediment discharge to the oceans and coastal restoration in Louisiana. Journal of Hydrology , 432 433 , 84 97. ht tps://doi.org/10.1016/j.jhydrol.2012.02.020 Arias, M. E., Cochrane, T. A., Piman, T., Kummu, M., Caruso, B. S., & Killeen, T. J. (20 12). Quantifying changes in flooding and habitats in the Tonle Sap Lake (Cambodia) caused by water infrastructure developmen t and climate change in the Mekong Basin. Journal of Environmental Management , 112 , 53 66. https://doi.org/ 10.1016/j.jenvman.2012.07 .003 Arias, M. E., Cochrane, T. A., Norton, D., Killeen, T. J., & Khon, P. (2013). The Flood Pulse as the Underlying Driver of Vegetation in the Largest Wetland and Fishery of the Mekong Basin. AMBIO , 42 (7), 864 876. https://doi.or g/10.1007/s13280 - 013 - 0424 - 4 Arias, M. E., Piman, T., Lauri, H., Cochrane, T. A., & Kummu, M. (2014). Dams on Mekong tributaries as significant contri butors of hydrological alterations to the Tonle Sap Floodplain in Cambodia. Hydrology and Earth System Scie nces , 18 (12), 5303 5315. https://doi.org/10.5194/hess - 18 - 5303 - 2014 Arias, M. E., Cochrane, T. A., Kummu, M., Lauri, H., Holtgrieve, G. W., Koponen, J ., & Piman, T. (2014). Impacts of hydropower and climate change on drivers of ecological productivity of So tant wetland. Ecological Modelling , 272 , 252 263. https://doi.org/10.1016/j.ecolmodel.2013.10.015 Babbitt, B. (2002). What G oes Up, May Come Down Learning from our experiences with dam construction in the past can guide and improve dam removal in the futur e. BioScience , 52 (8), 656 658. https://doi.org/10.1641/0006 - 3568 Baran, E., & Myschowoda, C. (2009). Dams and fisheries in t he Mekong Basin. Aquatic Ecosystem Health & Management , 12 (3), 227 234. https://doi.org/10.1080/14634980903 149902 156 Bates, P. D., Horr itt, M. S., & Fewtrell, T. J. (2010). A simple inertial formulation of the shallow water equations for efficient two - dimensi onal flood inundation modelling. Journal of Hydrology , 387 (1 2), 33 45. https://doi.org/10.1016/j.jhydrol.2 010.03.027 van Beek, L. P . H., Wada, Y., & Bierkens, M. F. P. (2011). Global monthly water stress: 1. Water balance and water availability. Water Res ources Research , 47 (7), 1 25. https://doi.org/10.1029/2010WR009791 Bellman, R. (1961). Adaptive control pro cesses: a guided tour . Pr inceton University Press. van Bemmelen, C. W. T., Mann, M., de Ridder, M. P., Rutten, M. M., & van de Giesen, N. C. (2016). Determining water reservoir characteristics with global elevation data. Geophysical Research Letters , 43 (21 ), 11,278 11,286. https:/ /doi.org/10.1002/2016GL069816 Benedict, I., Van Heerwaarden, C. C., Weerts, A. H., Hazeleger, W., & Benedict, I. (2017). An evaluation of the importance of spatial resolution in a global climate and hydrological model based on the Rhine and Mississippi bas in. Hydrol. Earth Syst. Sci. Discuss. , 2017 (September), 1 28. https://doi.org/10.5194/hess - 2017 - 473 Beven, K., Cloke, H., Pa ppenberger, F., Lamb, R., & Hunter, N. (2015). Hyperresolution information and hyperresolution ignorance in modelling the hydrology of the land surface. Science China Earth Sciences , 58 (1). https://doi.org/10.1007/s11430 - 014 - 5003 - 4 Biemans, H., Haddeland, I., Kabat, P., Ludwig, F., Hutjes, R. W. A., Heinke, J., et al. (2011). Impact of reservoirs on river disch arge and irrigation water supply during the 20th century. Water Resources Research , 47 (3). https://doi.org/10.1029/2009WR008929 Bierkens, M. F. P. (2015). Global hydrology 2015: State, trends, and directions. Water Resources Research , 51 (7). https://doi.or g/10.1002/2015WR017173 Bi erkens, M. F. P., Bell, V. A., Burek, P., Chaney, N., Condon, L. E ., David, C. H., et al. (2015). Hyper - Hydrological Processes , 29 (2), 310 320. https://doi.org/10.1 002/hyp.10391 Binnie, G. M. (1987). Masonry and Concrete Dams 1880 - 1941. Industrial Archaeology Review , 10 (1), 41 58. https://doi.org/10.1179/iar.1987.10.1.41 Science , 327 (5972) , 1444 1447. https://doi. org/10.1126/science.327.5972.1444 Bonheur, N., & Lane, B. D. (2002 ). Natural resources management for human security in Environmental Science and Policy , 5 (1), 33 41. Bonnema, M., & Hossain, F. (2017). Inferring rese rvoir operating patterns across the Mekong Basin using only space observations. Water Resources Research , 53 (5), 3791 3810. https://doi.org/10.1002/2016WR019978 157 Brownell, R. L., Reeves, R. R., Thomas, P. O., Smith, B. D., & Ryan, G . E. (2017). Dams threate n rare Mekong dolphins. Science , 355 (6327), 805.1 805. https://doi .org/10.1126/science.aam6406 Buccola, N. L., Turner, D. F., & Rounds, S. A. (2016). Water temperature effects from simulated dam operations and structures in the Mid dle Fork Willamette River , western Oregon (p. 49). Reston, VA. https://doi.org/10.3133/ofr2 0161159 Bunn, S. E., & Arthington, A. H. (2002). Basic principles and ecological consequences of altered flow regimes for aquatic biodiversity. Environmental Managem ent , 30 (4), 492 507. http s://doi.org/10.1007/s00267 - 002 - 2737 - 0 Busker, T., de Roo, A., Gela ti, E., Schwatke, C., Adamovic, M., Bisselink, B., et al. (2018). A global lake and reservoir volume analysis using a surface water dataset and satellite altimetry. Hydrology and Earth Syste m Sciences Discussions , 1 32. https://doi.org/10.5194/hess - 2018 - 21 Chao, B. F., Wu, Y. H., & Li, Y. S. (2008). Impact of artificial reservoir water impoundment on global sea level. Science , 320 (5873), 212 LP 214. Retrieved from h ttp://science.sciencemag. org/content/320/5873/212.abstract Chaudhari, S., Felfelani, F., Sh in, S., & Pokhrel, Y. (2018). Climate and anthropogenic contributions to the desiccation of the second largest saline lake in the twentieth century. Journal of Hydro logy , 560 , 342 353. https ://doi.org/10.1016/j.jhydrol.2018.03.034 Clark, M. P., Bierkens, M . F. P., Samaniego, L., Woods, R. A., Uijlenhoet, R., Bennett, K. E., et al. (2017). The evolution of process - based hydrologic models: Historical challenges and the collective quest for phys ical realism. Hydrology and Earth System Sciences , 21 (7), 3427 344 0. https://doi.org/10.5194/hess - 21 - 3427 - 2017 Cleveland, W. S. (1979). Robust Locally Weighted Regression and Smoothing Scatterplots. Journal of the American Statisti cal Association , 74 (368), 829 836. https://doi.org/10.1080/01621459.1979.10481038 Coe, M. T. (2000). Modeling terrestrial hydrological systems at the continental scale: Testing the accuracy of an atmospheric GCM. Journal of Climate , 13 (4), 686 704. https:/ /doi.org/10.1175/1520 - 044 2(2000)013 <0686:MTHSAT>2.0.CO;2 Cook, B. I., Ault, T. R., & Smerdon, J. E. (2015). Unprecedented 21st century drought risk in the American Southwest and Central Plains. Science Advances , 1 (1). Retrieved from http://advances.science mag.org/content/1/1/e1400 082.abstra ct Costa - Cabral, M. C., Richey, J. E., Goteti, G., Lettenmaier, D. P., Feldkötter, C., & Snidvongs, A. (2008). Landscape structure and use, climate, and water movement in the Mekong River basin. Hydrological Processes , 22 (12), 1731 1746. https:// doi.org/10 .1002/hyp.6740 158 Dang, T. D., Cochrane, T. A., Arias, M. E., & Tri, V. P. D. (2018). Future hydrological alterations in the Mekong Delta under the impact of water resources development, land subsidence and sea level rise. J ournal of Hydrology: Regi onal Studi es , 15 , 119 133. https://doi.org/10.1016/j.ejrh.2017.12.002 Degu, A. M., Hossain, F., Niyogi, D., Pielke, R., Shepherd, J. M., Voisin, N., & Chronis, T. (2011). The influence of large dams on surrounding climate and preci pitation patterns. Geophy sical Rese arch Letters , 38 (4). https://doi.org/10.1029/2010GL046482 Döll, P., Kaspar, F., & Lehner, B. (2003). A global hydrological model for deriving water availability indicators: Model tuning and validation. Journal of Hydrolog y , 270 (1 2), 105 134. htt ps://doi.o rg/10.1016/S0022 - 1694(02)00283 - 4 Doyle, M. W., Stanley, E. H., Orr, C. H., Selle, A. R., Sethi, S. A., & Harbor, J. M. (2005). Stream ecosystem response to small dam removal: Lessons from the Heartland. Geomorphology , 71 ( 1 2), 227 244. https://do i.org/10.1 016/j.geomorph.2004.04.011 Dynesius, M., & Nilsson, C. (1994). Fragmentation and flow regulation of river systems in the northern third of the world. Science , 266 . https://doi.org/10.1126/science.266.5186.753 Ehsani, N., Fekete, B. M., Vörösmarty , C. J., & Tessler, Z. D. (2016). A neural network based general reservoir operation scheme. Stochastic Environmental Research and Risk Assessment , 30 (4), 1151 1166. https://doi.org/10.1007/s00477 - 015 - 1147 - 9 Eiriksdottir, E. S., Oe lkers, E. H., Hardardotti r, J., & G islason, S. R. (2017). The impact of damming on riverine fluxes to the ocean: A case study from Eastern Iceland. Water Research , 113 , 124 138. https://doi.org/10.1016/j.watres.2016.12.029 Fan, Y., & Miguez - Macho, G. (2011 ). A simple hydrologic fr amework fo r simulating wetlands in climate and earth system models. Climate Dynamics , 37 (1), 253 278. https://doi.org/10.1007/s00382 - 010 - 0829 - 8 Fan, Y., Miguez - Macho, G., Weaver, C. P., Walko, R., & Robock, A. (2007). Incorporating water table dynamics in climate modeling: 1. Water table observations and equilibrium water table simulations. Journal of Geophysical Research Atmospheres , 112 (10), D10 125. https://doi.org/10.1029/2006JD008111 Fan, Y., Miguez - Macho, G., Jobbágy, E. G., Ja ckson, R. B., & Otero - Cas al, C. (2017). Hydrologic regulation of plant rooting depth. Proceedings of the National Academy of Sciences , 114 (40), 10572 LP 10577. Retriev ed from http://www.pnas.org/content/114/40/10572.abstract Fatichi, S., Vivoni, E. R., O gden, F. L., Ivanov, V. Y ., Mirus, B., Gochis, D., et al. (2016). An overview of current applications, challenges, and future trends in distributed process - based models in hydrology, 537 , 45 60. https://doi.org/10.1016/j.jhydrol.2016.03.026 Fawthrop, T. (2 018). Leaked report warns Mekong river . London. Retrieved from 159 https://www.theguardian.com/environment/2018/may/16/leaked - r eport - warns - cambodias - biggest - dam - could - literally - kill - mekong - river Fearnside, P. M., & Pueyo, S. (2012). Greenh ouse - gas emissions from tropical dams. Nature Climate Change , 2 (6), 382 384. https://doi.org/10.1038/nclimate1540 Felfelani, F., Wada, Y., Longu evergne, L., & Pokhrel, Y. (2017). Natural and human - induced terrestrial water storage change: A global analysis using hydrological models and GRACE. Journal of Hydrology , 553 , 105 118. https://doi.org/10.1016/j.jhydrol.2017.07.048 Fleischmann, A., Paiva, R., & Collischonn, W. (2019). Can regional to continental river hydrodynamic models be locally relevant? A cross - scale comparison. Journal of Hydrology X , 3 , 100027. https://doi.org/10.1016/j.hydroa.2019.100027 zenave, A., Ramillien, G., Le Toan, T., & Mognard - Campbell, N. (2006). Water volume cha nge in the lower Mekong f rom satellite altimetry and imagery data. Geophysical Journal International , 167 (2), 570 584. https://doi.org/10.1111/j.1365 - 246X.2006.03184.x G ao, H. (2015). Satellite remote sensing of large lakes and reservoirs: from elevation a nd area to storage. Wiley Interdisciplinary Reviews: Water , 2 (2), 147 157. https://doi.org/10.1002/wat2.1065 Gernaat, D. E. H. J., Bogaart, P. W., Vuuren, D. P. van, Bie mans, H., & Niessink, R. (2017). High - resolution assessment of global technical and eco nomic hydropower potentia l. Nature Energy , 2 (10), 821 828. https://doi.org/10.1038/s41560 - 017 - 0006 - y Gong, L., Halldin, S., & Xu, C. Y. (2011). Global - scale river routin g - an efficient time - delay algorithm based on HydroSHEDS high - resolution hydrography. Hy drological Processes , 25 ( 7), 1114 1128. https://doi.org/10.1002/hyp.7795 Graf, W. L. (1999). Dam nation: A geographic census of american dams and their large - scale hydro logic impacts. Water Resources Research , 35 (4), 1305 1311. https://doi.org/10.1029/1999 WR900016 Gran, K. B., Bel mont, P., Day, S. S., Jennings, C., Johnson, A., Perg, L., & Wilcock, P. R. (2009). Geomorphic evolution of the Le Sueur River, Minnesota, USA, and implications for current sediment loading. In L. A. James, S. L. Rathburn, & G. R. Whittecar (Eds.), Managem ent and Restoration of Fluvial Systems with Broad Historical Changes and Human Impacts . Geological Society of America. Retrieved from https://do i.org/10.1130/2009.2451(08) Greathouse, E. A., Pringle, C. M., & Holmquist, J. G. (2006 ). Conservation and manag ement of migratory fauna: dams in tropical streams of Puerto Rico. Aquatic Conservation: Marine and Freshwater Ecosystems , 16 (7), 695 712. https ://doi.org/10.1002/aqc.804 160 Grumbine, R. E., & Xu, J. (2011). Mekong Hydropower Developm ent. Science , 332 (6026), 178 LP 179. Retrieved from http://science.sciencemag.org/content/332/6026/178.abstract Gupta, H., Kao, S. J., & Dai, M. (2012). The role of me ga dams in reducing sediment fluxes: A case study of large Asian rivers. Journal of Hyd rology , 464 465 , 447 458. https://doi.org/10.1016/j.jhydrol.2012.07.038 Haddeland, I., Skaugen, T., & Lettenmaier, D. P. (2006). Anthropogenic impacts on continental surface water fluxes. Geophysical Research Letters , 33 (8), L08406. https://doi.org/10.1029 /2006GL026047 Hanasaki, N ., Kanae, S., & Oki, T. (2006). A reservoir ope ration scheme for global river routing models. Journal of Hydrology , 327 (1 2), 22 41. https://doi.org/10.1016/j.jhydrol.2005.11.011 Hanasaki, N., Yoshikawa, S., Pokhrel, Y., & Kanae, S . (2018). A global hydrol ogical simulation to specify the sources of wat er used by humans. Hydrol. Earth Syst. Sci. , 22 (1), 789 817. https://doi.org/10.5194/hess - 22 - 789 - 2018 Hejazi, M. I., Voisin, N., Liu, L., Bramer, L. M., Fortin, D. C., Hathaway, J. E., et al. (2015). 21st cent ury United States emissions mitigation could in crease water stress more than the climate change it is mitigating. Proceedings of the National Academy of Sciences , 112 (34), 10635 10640. https://doi.org/10.1073/pnas.1421675112 Hoes, O. A. C., Meijer, L. J. J., van der Ent, R. J., & van de Giesen, N. C. ( 2017). Systematic high - resolution assessment of global hydropower potential. PLOS ONE , 12 (2), e0171844. https://doi.org/10.1371/journal.pone.0171844 Holdren, G. C., & Turner, K. (2010) . Characteristics of Lake Mead, Arizona - Nevada. Lake and Reservoir Manag ement , 26 (4), 230 239. https://doi.org/10.1080/07438141.2010.540699 Hossain, F. (2010). Empirical Relationship between Large Dams and the Alteration in Extreme Precipitation. Wor ld , 11 (3), 97 101. https://doi.org/10.1061/(asce)nh.1527 - 6996.0000013 Hossain, F., Jeyachandran, I., & Pielke, R. (2010). Dam safety effects due to human alteration of extreme precipitation. Water Resources Research , 46 (3). https://doi.org/10.1029/2009WR00 7704 H ossain, F., Degu, A. M., Yigzaw, W., Burian, S., Niyogi, D., Shepherd, J . M., & Pielke, R. (2012). Climate Feedback Based Provisions for Dam Design, Operations, and Water Management in the 21st Century. Journal of Hydrologic Engineering , 17 (8), 837 8 50. ht tps://doi.org/10.1061/(ASCE)HE.1943 - 5584.0000541 Huang, W. - C., Yuan, L. - C., & Lee, C. - M. (2002). Linking genetic algorithms with stochastic dynamic programming to the long - term operation of a multireservoir system. Water Resources Research , 38 (12), 4 0 1 40 9. https://doi.org/10.1029/2001WR001122 161 Hung, N. N., Delgado, J. M., Tri, V. K., Hung, L. M., Merz, B., Bárdossy, A., & Apel, H. (2012). Floodplain hydrology of the Mekong Delta, Vietnam. Hydrological Processes , 26 (5), 674 686. https://doi.org/10.10 02/hyp .818 3 Inomata, H., & Fukami, K. (2008). Restoration of historical hydrological data of Tonle Sap Lake and its surrounding areas. Hydrological Processes , 22 (9), 1337 1350. https://doi.org/10.1002/hyp.6943 Junk, W. J., Bayley, P. B., & Sparks, R. E. (1 989). The flood - pulse concept in river - floodplain systems. In D. P. Dodge (Ed.), Proceedings of the International Large River Symposium (LARS) (pp. 110 127). Canadian Special Publication of Fisheries and Aquatic Sciences 106. Kemenes, A., Forsberg, B. R., & Mela ck, J. M. (2007). Methane release below a tropical hydroelectric dam. Geophysical Research Letters , 34 (12), 1 5. https://doi.org/10.1029/2007GL029479 Kesel, R. H., Dunne, K. C., McDonald, R. C., Allison, K. R., & Spicer, B. E. (1974). Lateral erosion and o verb ank deposition on the Mississippi River in Louisiana caused by 1973 flooding. Geology , 2 (9), 461 464. https://doi.org/10.1130/0091 - 7613(1974)2<461:LEAODO>2.0.CO;2 Kummu, M., & Sarkkula, J. (2008). Impact of the Mekong River flow alteration on the Tonle Sap flood pulse. Ambio , 37 (3), 185 192. https://doi.org/10.1579/0044 - 7447(2008)37[185:IOTMRF]2.0.CO;2 Labadie, J. W. (2004). Optimal Operation of Multireservoir Systems: State - of - the - Art Review. Journal of Water Resources Planning and Management , 13 0 (2), 93 1 11. https://doi.org/10.1061/(ASCE)0733 - 9496(2004)130:2(93) Lauri, H., de Moel, H., Ward, P. J., Räsänen, T. A., Keskinen, M., & Kummu, M. (2012). Future changes in Mekong River hydrology: impact of climate change and reservoir operation on discha rge. H ydro logy and Earth System Sciences , 16 (12), 4603 4619. https://doi.org/10.5194/hess - 16 - 4603 - 2012 Lehner, B., & Döll, P. (2004). Development and validation of a global database of lakes, reservoirs and wetlands. Journal of Hydrology , 296 (1 4), 1 22. h ttps:/ /doi .org/10.1016/j.jhydrol.2004.03.028 Lehner, B., Verdin, K., & Jarvis, A. (2008). New Global Hydrography Derived From Spaceborne Elevation Data. Eos, Transactions American Geophysical Union , 89 (10), 93 94. https://doi.org/10.1029/2008EO100001 Lehne r, B., Rei dy Liermann, C., Revenga, C., Vorosmarty, C., Fekete, B., Crouzet, P., et al. (2011a). Global Reservoir and Dam Database, Version 1 (GRanDv1): Dams, Revision 01. NASA Socioeconomic Data and Applications Center (SEDAC). Retrieved from http://dx.do i.org/ 10.7 927/H4N877QK 162 Lehner, B., Reidy Liermann, C., Revenga, C., Vorosmarty, C., Fekete, B., Crouzet, P., et al. (2011b). High - River - Flow Management. Frontiers in Ecology and the Envi ronmen t , 9 , 494 502. Retrieved from http://dx.doi.org/10.1890/100125 Li, Y., Gao, H., Jasinski, M. F., Zhang, S., & Stoll, J. D. (2019). Deriving High - Resolution Reservoir Bathymetry From ICESat - 2 Prototype Photon - Counting Lidar and Landsat Imagery. IEEE T ransac tion s on Geoscience and Remote Sensing , 1 11. https://doi.org/10.1109/TGRS.2019.2917012 Liebe, J., van de Giesen, N., & Andreini, M. (2005). Estimation of small reservoir storage capacities in a semi - arid environment: A case study in the Upper East R egion of G hana. Integrated Water Resource Assessment , 30 (6), 448 454. https://doi.org/10.1016/j.pce.2005.06.011 Ligon, F. K., Dietrich, W. E., & Trush, W. J. (1995). Downstream Ecological Effects of Dams. BioScience , 45 (3), 183 192. https://doi.org/10.2307 /13125 57 M ateo, C. M., Hanasaki, N., Komori, D., Tanaka, K., Kiguchi, M., Champathong, A., et al. (2014). Assessing the impacts of reservoir operation to floodplain inundation by combining hydrological, reservoir management, and hydrodynamic models. Water Resour ces Research , 50 (9), 7245 7266. https://doi.org/10.1002/2013WR014845 Maupin, M. A., Kenny, J. F., Hutson, S. S., Lovelace, J. K., Barber, N. L., & Linsey, K. S. (2014). Estimated use of water in the United States in 2010: U.S. Geological Survey Circu lar 14 05 ( p. 56). Retrieved from https://dx.doi.org/10.3133/cir1405 Mccabe, M. F., Aragon, B., Houborg, R., & Mascaro, J. (2017). CubeSats in Hydrology: Ultrahigh - Resolution Insights Into Vegetation Dynamics and Terrestrial Evaporation. Water Resources Res earch , in review . https://doi.org/10.1002/2017WR022240 McCully, P. (2001). Silenced rivers: the ecology and politics of large dams (Enl. and u). New York;London; Zed Books. Meigh, J. R., McKenzie, A. A., & Sene, K. J. (1999). A grid - based approach to water scarc ity estimates for eastern and southern Africa. Water Resources Management , 13 (2), 85 115. https://doi.org/10.1023/A:1008025703712 Mekong River Commission. (2005). Over view of the Hydrology of the Mekong Basin . Vientiane. Mekong River Commission. (201 0). Im pacts on the Tonle Sap Ecosystem . Vientaine. Mesinger, F., DiMego, G., Kalnay, E., Mitchell, K., Shafran, P. C., Ebisuzaki, W., et al. (2006). North American Regional Reanalysis. Bulletin of the American Meteorological Society , 87 (3), 343 360. https: //doi. org/10.1175/BAMS - 87 - 3 - 343 163 Miguez - Macho, G., & Fan, Y. (2012a). The role of groundwater in the Amazon water cycle: 1. Influence on seasonal streamflow, flooding and wet lands. Journal of Geophysical Research Atmospheres , 117 (15). https://doi.org/10.102 9/2012 JD017539 Miguez - Macho, G., & Fan, Y. (2012b). The role of groundwater in the Amazon water cycle: 2. Influence on seasonal soil moisture and evapotranspiration. Journal of Geophysical Research Atmospheres , 117 (15), D15114. https://doi.org/10.1029/2012 JD0175 40 Miguez - Macho, G., Fan, Y., Weaver, C. P., Walko, R., & Robock, A. (2007). Incorporating water table dynamics in climate modeling: 2. Formulation, validation, and so il moisture simulation. Journal of Geophysical Research Atmospheres , 112 (13), 1 16. https ://doi.org/10.1029/2006JD008112 Minh, H. V. T., Kurasaki, M., Ty, T. V., Tran, D. Q., Le, K. N., Avtar, R., et al. (2019). Effects of Multi - Dike Protection Systems on Surface Water Quality in the Vietnamese Mekong Delta. Water , 11 (5), 1010. https://d oi.org /10.3390/w11051010 Neal, J., Schumann, G., & Bates, P. (2012). A subgrid channel model for simulating river hydraulics and floodplain inundation over large and data sp arse areas. Water Resources Research , 48 (11). https://doi.org/10.1029/2012WR012514 Nilsso n, C., Reidy, C. A., Dynesius, M., & Revenga, C. (2005). Fragmentation and flow Science , 308 . https://doi.org/10.1126/science.1107887 Null, S. E., Medellín - Azuara, J., Escriva - Bou, A., Lent, M., & Lund, J. R. (2014). Optimizing the dammed: Water supply losses and fish habitat gains from dam removal in California. Journal of Environmental Management , 136 , 121 131. https://doi.org/10.1016/j.jenvman.2014.01.024 Oki, T., & Kanae, S. (2006). Global Hydrologica l Cycl es and World Water Resources. Science , 313 (5790), 1068 1072. https://doi.org/10.1126/science.1128845 Oki, Taikan, & Sud, Y. C. (1998). Design of Total Runoff Integrating Pathways (TRIP) A Global River Channel Network. Earth Interactions , 2 (1), 1 37. https: //doi.org/10.1175/1087 - 3562(1998)002<0001:DOTRIP>2.3.CO;2 Pekel, J. - F., Cottam, A., Gorelick, N., & Belward, A. S. (2016). High - resolution mapping of global surface water and its long - term changes. Nature , 540 (7633), 418 422. https://doi.org/10.1038/ nature 20584 Piman, T., Lennaerts, T., & Southalack, P. (2013). Assessment of hydrological changes in the lower Mekong Basin from Basin - Wide development scenarios. Hydrological Processes , 27 (15), 2115 2125. https://doi.org/10.1002/hyp.9764 Poff, N. L., Alla n, J. D., Bain, M. B., Karr, J. R., Prestegaard, K. L., Richter, B. D., et al. (1997). The natural flow regime. BioScience , 47 (11). https://doi.org/10.2307/1313099 164 Pohl, M. M. (2002). Bringing Down Our Dams: Trends in American Dam Removal Rationales. Journ al of the American Water Resources Association , 38 (6), 1511 1519. https://doi.org/10.1111/j.1752 - 1688.2002.tb04361.x Pokhrel, Y., Hanasaki, N., Koirala, S., Cho, J., Yeh, P. J. - F., Kim, H., et al. (2012). Incorporating Anthropogenic Water Regulation Module s into a Land Surface Model. Journal of Hydrometeorology , 13 (1), 255 269. https://doi.org/10.1175/JHM - D - 11 - 013.1 Pokhrel, Y., Hanasaki, N., Yeh, P. J. F., Yamada, T. J., Kanae, S., & Oki, T. (2012). Model estimates of sea - level change due to anthropogenic impact s on terrestrial water storage. Nature Geoscience , 5 (6), 389 392. https://doi.org/10.1038/ngeo1476 Pokhrel, Y., Fan, Y., Miguez - Macho, G., Yeh, P. J. F., & Han, S. C. (2013). The role of groundwater in the Amazon water cycle: 3. Influence on terrestr ial wa ter storage computations and comparison with GRACE. Journal of Geophysical Research Atmospheres , 118 (8), 3233 3244. https://doi.org/10.1002/jgrd.50335 Pokhrel, Y., Fan, Y., & Miguez - Macho, G. (2014). Potential hydrologic changes in the Amazon by the end of the 21st century and the groundwater buffer. Environmental Research Letters , 9 (8), 084004. https://doi.org/10.1088/1748 - 9326/9/8/084004 Pokhrel, Y., Koirala, S., Yeh, P. J. - F., Hanasaki, N., Longuevergne, L., Kanae, S., & Oki, T. (2015). Incorporati on of groundwater pumping in a global Land Surface Model with the representation of human impacts. Water Resources Research , 51 (1), 78 96. https://doi.org/10.1002/2014WR015602 Pokhrel, Y., Hanasaki, N., Wada, Y., & Kim, H. (2016). Recent progresses in inco rporat ing human land water management into global land surface models toward their integration into Earth system models. Wiley Interdisciplinary Reviews: Water , 3 (4), 548 574. https://doi.org/10.1002/wat2.1150 Pokhrel, Y., Felfelani, F., Shin, S., Yamada, T. J., & Satoh, Y. (2017). Modeling large - scale human alteration of land surface hydrology and climate. Geoscience Letters , 4 (1), 10. https://doi.org/10.1186/s40562 - 017 - 0076 - 5 Pokhrel, Y., Burbano, M., Roush, J., Kang, H., Sridhar, V., & Hyndman, D. (2018) . A Re view of the Integrated Effects of Changing Climate, Land Use, and Dams on Mekong River Hydrology. Water , 10 (3), 266. https://doi.org/10.3390/w10030266 Pokhrel, Y. N., Koirala, S., Yeh, P. J. - F., Hanasaki, N., Longuevergne, L., Kanae, S., & Oki, T. (2 015). Incorporation of groundwater pumping in a global Land Surface Model with the representation of human impacts. Water Resources Research , 51 (1), 78 96. https://doi.org/10.1002/2014WR015602 Postel, S. L., Daily, G. C., & Ehrlich, P. R. (1996). Human app ropria tion of renewable fresh water. Science , 271 (5250), 785 788. https://doi.org/10.1126/science.271.5250.785 165 Rajagopalan, B., Nowak, K., Prairie, J., Hoerling, M., Harding, B., Barsugli, J., et al. (2009). Water supply risk on the Colorado River: Can man agemen t mitigate? Water Resources Research , 45 (8), 1 7. https://doi.org/10.1029/2008WR007652 Räsänen, T. A., Ko ponen, J., Lauri, H., & Kummu, M. (2012). Downstream Hydrological Impacts of Hydropower Development in the Upper Mekong Basin. Water Resources Ma nageme nt , 26 (12), 3495 3513. https://doi.org/10.1007/s11269 - 012 - 0087 - 0 Räsänen, T. A., Someth, P., Lauri, H., K oponen, J., Sarkkula, J., & Kummu, M. (2017). Observed river discharge changes due to hydropower operations in the Upper Mekong Basin. Journal of Hydro logy , 545 , 28 41. https://doi.org/10.1016/j.jhydrol.2016.12.023 Renwick, W. H., Smith, S. V., Bartley, J. D., & Buddemeier, R. W. (2005). The role of impoundments in the sediment budget of the conterminous United States. Geomorphology , 71 (1 2), 99 11 1. htt ps://doi.org/10.1016/j.geomorph.2004.01.010 Sabo, J. L., Ruhi, A., Holtgrieve, G. W., Elliott, V., Arias, M. E., Ngor, P. B., et al. (2017). Designing river flows to improve food security futures in the Lower Mekong Basin. Science , 358 (6368), eaao105 3. htt ps://doi.org/10.1126/science.aao1053 Sakamoto, T., Van Nguyen, N., Kotera, A., Ohno, H., Ishitsuka, N., & Yokozawa, M. (2007). Detecting temporal changes in the extent of annual flooding within the Cambodia and the Vietnamese Mekong Delta from MODIS time - s eries imagery. Remote Sensing of Environment , 109 (3), 295 313. https://doi.org/10.1016/j.rse.2007.01.011 Scanlon, B. R., Zhang, Z., Save, H., Wiese, D. N., Landerer, F. W., Long, D., et al. (2016). Global evaluation of new GRACE mascon products for h ydrolo gic applications. Water Resources Research , 52 (12), 9412 9429. https://doi.org/10.1002/2016WR019494 Schmidt, C. (2015). Alarm over a sinking delta. Science , 348 (6237), 845 LP 846. Retrieved from http://science.sciencemag.org/content/348/6237/845.ab stract Schmitt, R. J. P., Bizzi, S., Castelletti, A., & Kondolf, G. M . (2018). Improved trade - offs of hydropower and sand connectivity by strategic dam planning in the Mekong. Nature Sustainability , 1 (2), 96 104. https://doi.org/10.1038/s41893 - 018 - 0022 - 3 S hin, S ., & Paik, K. (2017). An improved method for single flow direct ion calculation in grid digital elevation models. Hydrological Processes , 31 (8), 1650 1661. https://doi.org/10.1002/hyp.11135 Shuai, P., Cardenas, M. B., Knappett, P. S. K., Bennett, P. C ., & N eilson, B. T. (2017). Denitrification in the banks of fluctuati ng rivers: The effects of river stage amplitude, sediment hydraulic conductivity and dispersivity, and ambient groundwater flow. Water Resources Research , 53 (9), 7951 7967. https://doi.or g/10.1 002/2017WR020610 Smajgl, A., Toan, T. Q., Nhan, D. K., Ward, J. , Trung, N. H., Tri, L. Q., et al. (2015). Responding to rising sea levels in the Mekong Delta. Nature Climate Change , 5 (2), 167 174. https://doi.org/10.1038/nclimate2469 166 Smajgl, Alex, Wa rd, J. R., Foran, T., Dore, J., & Larson, S. (2015). Visions, beliefs , and - sector and transboundary dynamics in the wider Mekong region. Eology and Society , 20 (2). https://doi.org/10.5751/ES - 07421 - 200215 Solander, K. C., Re ager, J. T., Thomas, B. F., David, C. H., & Famiglietti, J. S. (2016) . Simulating Human Water Regulation: The Development of an Optimal Complexity, Climate - Adaptive Reservoir Management Model for an LSM. Journal of Hydrometeorology , 17 (3), 725 744. https:/ /doi.o rg/10.1175/JHM - D - 15 - 0056.1 Sridhar, V., Kang, H., & Ali, S. A. (2019). Human - Induced Alterations to Land Use and Climate and Their Responses for Hydrology and Water Management in the Mekong River Basin. Water , 11 (6), 1307. https://doi.org/10.3390/w11 061307 Stone, R. (2016). Dam - building threatens Mekong fisheries. Sci ence , 354 (6316), 1084 1085. https://doi.org/10.1126/science.354.6316.1084 Syed, T. H., Famiglietti, J. S., & Chambers, D. P. (2009). GRACE - Based Estimates of Terrestrial Freshwater Discha rge fr om Basin to Continental Scales. Journal of Hydrometeorology , 10 (1), 22 40. https://doi.org/10.1175/2008JHM993.1 Syvitski, J. P. M., Vörösmarty, C. J., Kettner, A. J., & Green, P. (2005). Impact of humans on the flux of terrestrial sediment to the glo bal co astal ocean. Science , 308 (5720), 376 380. https://doi.org/10.11 26/science.1109454 Syvitski, J. P. M., Kettner, A. J., Overeem, I., Hutton, E. W. H., Hannon, M. T., Brakenridge, G. R., et al. (2009). Sinking deltas due to human activities. Nature Geos cience , 2 (10), 681 686. https://doi.org/10.1038/ngeo629 Taeb, A., Rea ger, J. T., Turmon, M., & Chandrasekaran, V. (2017). A Statistical Graphical Model of the California Reservoir System. Water Resources Research . https://doi.org/10.1002/2017WR020412 Takat a, K., Emori, S., & Watanabe, T. (2003). Development of the minimal a dvanced treatments of surface interaction and runoff. Global and Planetary Change , 38 (1 2), 209 222. https://doi.org/10.1016/S0921 - 8181(03)00030 - 4 Tanner, W. F. (1971). The River Profile. The J ournal of Geology , 79 (4), 482 492. https://doi.org/10.1086/6276 53 Tapley, B. D. (2004). GRACE Measurements of Mass Variability in the Earth System. Science , 305 (5683), 503 505. https://doi.org/10.1126/science.1099192 Timpe, K., & Kaplan, D. (2017). T he cha nging hydrology of a dammed Amazon. Science Advances , 3 (11), e1 700611. https://doi.org/10.1126/sciadv.1700611 Trung, L. D., Duc, N. A., Nguyen, L. T., Thai, T. H., Khan, A., Rautenstrauch, K., & Schmidt, C. (2018). Assessing cumulative impacts of the propo sed Lower Mekong Basin hydropower cascade on the Mekong River f loodplains and Delta Overview of 167 integrated modeling methods and results. Journal of Hydrology . https://doi.org/10.1016/j.jhydrol.2018.01.029 Spatial Dynamics of Gas Ebullition in a Temperate Water - Storage Reservoir. Water Resources Research , 53 (10), 8266 8276. https://doi.org/10.1002/2017WR020694 Urbaza ev, M., Cremer, F., Migliavacca, M., Reichstein, M., Schmullius, C., & Thiel, C. (2018). Potential of Multi - Temp oral ALOS - 2 PALSAR - 2 ScanSAR Data for Vegetation Height Estimation in Tropical Forests of Mexico. Remote Sensing , 10 (8), 1277. https://doi.org/1 0.3390/rs10081277 Veldkamp, T. I. E., Wada, Y., Aerts, J. C. J. H., Döll, P., Gosling, S. N., Liu, J., et al. (2 017). Water scarcity hotspots travel downstream due to human interventions in the 20th and 21st century. Nature Communications , 8 . https://doi.o rg/10.1038/ncomms15697 Voisin, N., Hejazi, M. I., Leung, L. R., Liu, L., Huang, M., Li, H. Y., & Tesfa, T. (2017). Effects of spatially distributed sectoral water management on the redistribution of water resources in an integrated water model. Water Resou rces Research , 53 (5), 4 253 4270. https://doi.org/10.1002/2016WR019767 Voisin, Nathalie., Li, H., Ward, D., Huang, M., Wigmosta, M., & Leung, L. R. (2013). On an improved sub - regional water resources management representation for integration into earth syst em models. Hydrology an d Earth System Sciences , 17 (9), 3605 3622. https://doi.org/10.5194/hess - 17 - 3605 - 2013 Vörösmarty, C. J., Sharma, K. P., Fekete, B. M., Copeland, A. H., Holden, J., Marble, J., & Lough, J. A. (1997). The storage and aging of continenta l runoff in large reser voir systems of the world . (F. A. O. of the UN, Ed.) (Vol. 26). Vörösmarty, C. J., Meybeck, M., Fekete, B., Sharma, K., Green, P., & Syvitski, J. P. M. (2003). Anthropogenic sediment retention: major global impact from registered riv er impoundments. The Su pply of Flux of Sediment along Hydrological Pathways: Anthropogenic Influences at the Global Scale , 39 (1), 169 190. https://doi.org/10.1016/S0921 - 8181(03)00023 - 7 Vörösmarty, C. J., McIntyre, P. B., Gessner, M. O., Dudgeon, D., Prusev ich, A., Green, P., et al. (2010). Global threats to human water security and river biodiversity. Nature , 467 (7315), 555 561. https://doi.org/10.1038/nature09440 Wada, Y., Wisser, D., & Bierkens, M. F. P. (2014). Global modeling of withdrawal, allocation a nd consumptive use of surface water and groundwater resources. Earth System Dynamics , 5 (1), 15 40. https://doi.org/10.5194/esd - 5 - 15 - 2014 Wada, Y., de Graaf, E. M. I., & van Beek, P. H. and climate impacts on glo bal water resources. Journal of Advances in Modeling Earth Systems , 8 (2), 735 763. https://doi.org/10.1002/2015MS000618 168 Wada, Y., Reager, J. T., Chao, B. F., Wang, J., Lo, M. H., Song, C., et al. (2016). Recent Changes in Land Water Storage and its Contrib ution to Sea Level Variations. Surveys in Geophysics , 38 (1), 1 22. https://doi.org/10.1007/s10712 - 016 - 9399 - 6 Wada, Y., Bierkens, M. F. P., de Roo, A., Dirmeyer, P. A., Famiglietti, J. S ., Hanasaki, N., et al. (2017). Human water interface in hydrological m odelling: current status and future directions. Hydrology and Earth System Sciences , 21 (8), 4169 4193. https://doi.org/10.5194/hess - 21 - 4169 - 2017 Walko, R. L., Band, L. E., Jill Baron, , F Kittel, T. G., Lammers, R., Lee, T. J., et al. (2000). Coupled Atm osphere Biophysics Hydrology Models for Environmental Modeling. Journal of Applied Meteorology , 39 (6), 931 944. https://doi.org/10.1175/1520 - 0450(2000)039<0931:CABHMF>2.0.CO;2 Wang, S., Qian, X., Han, B. P., Luo, L. C., & Hamilton, D. P. (2012). Effects of local climate and hydrological conditions on the thermal regime of a reservoir at Tropic of Cancer, in southern China. Water Research , 46 (8), 2591 2604. https://doi.org/10.1016/j.watre s.2012.02.014 Wang, W., Lu, H., Yang, D., Sothea, K., Jiao, Y., Gao, B. , et al. (2016). Modelling Hydrologic Processes in the Mekong River Basin Using a Distributed Model Driven by Satellite Precipitation and Rain Gauge Observations. PLOS ONE , 11 (3), e0152 229. https://doi.org/10.1371/journal.pone.0152229 Wang, W., Lu, H., Rub y Leung, L., Li, H. - Y., Zhao, J., Tian, F., et al. (2017). Dam Construction in Lancang - Mekong River Basin Could Mitigate Future Flood Risk From Warming - Induced Intensified Rainfall. Geo physical Research Letters , 44 (20), 10,378 10,386. https://doi.org/10.10 02/2017GL075037 Weron, R. (2014). Electricity price forecasting: A review of the state - of - the - art with a look into the future. International Journal of Forecasting , 30 (4), 1030 1081. ht tps://doi.org/10.1016/j.ijforecast.2014.08.008 Wild, T. B., & Loucks, D . P. (2014). Managing flow, sediment, and hydropower regimes in the Sre Pok, Se San, and Se Kong Rivers of the Mekong basin. Water Resources Research , 50 (6), 5141 5157. https://doi.org/ 10.1002/2014WR015457 Winemiller, K. O., McIntyre, P. B., Castello, L., Fluet - Chouinard, E., Giarrizzo, T., Nam, S., et al. (2016). Balancing hydropower and biodiversity in the Amazon, Congo, and Mekong. Science , 351 (6269), 128 129. https://doi.org/10.1126/ science.aac7082 Wisser, D., Fekete, B. M., Vörösmarty, C. J., & Schuman n, A. H. (2010). Reconstructing 20th century global hydrography: a contribution to the Global Terrestrial Network - Hydrology (GTN - H). Hydrology and Earth System Sciences , 14 (1), 1 24. h ttps://doi.org/10.5194/hess - 14 - 1 - 2010 169 Woldemichael, A. T., Hossain, F., Pielke, R., & Beltrán - Przekurat, A. (2012). Understanding the impact of dam - triggered land use/land cover change on the modification of extreme precipitation. Water Resources Research , 48 (9). https://doi.org/10.1029/2011WR011684 Wood, E. F., Roundy, J. K. , Troy, T. J., van Beek, L. P. H., Bierkens, M. F. P., Blyth, E., et al. (2011). Hyperresolution global land surface modeling: Meeting a grand challenge for al water. Water Resources Research , 47 (5). https://doi.org/10.1029/2010 WR010090 Xue, Z., Liu, J. P., & Ge, Q. (2011). Changes in hydrology and sediment delivery of the Mekong River in the last 50 years: Connection to damming, monsoon, and ENSO. Earth Surfa ce Processes and Landforms , 36 (3), 296 308. https://doi.org/10.1002/esp .2036 Yamagata, Y., Hanasaki, N., Ito, A., Kinoshita, T., Murakami, D., & Zhou, Q. (2018). Estimating water food ecosystem trade - offs for the global negative emission scenario (IPCC - RCP 2.6). Sustainability Science , 13 (2), 301 313. https://doi.org/10.1007/s 11625 - 017 - 0522 - 5 Yamazaki, D., Oki, T., & Kanae, S. (2009). Deriving a global river network map and its sub - grid topographic characteristics from a fine - resolution flow direction map. H ydrology and Earth System Sciences , 13 (11), 2241 2251. https://doi.org/ 10.5194/hess - 13 - 2241 - 2009 Yamazaki, D., Kanae, S., Kim, H., & Oki, T. (2011). A physically based description of floodplain inundation dynamics in a global river routing model. Water Res ources Research , 47 (4), 1 21. https://doi.org/10.1029/2010WR009726 Yama zaki, D., Baugh, C. A., Bates, P. D., Kanae, S., Alsdorf, D. E., & Oki, T. (2012). Adjustment of a spaceborne DEM for use in floodplain hydrodynamic modeling. Journal of Hydrology , 436 437 , 81 91. https://doi.org/10.1016/j.jhydrol.2012.02.045 Yamazaki, D., de Almeida, G. A. M., & Bates, P. D. (2013). Improving computational efficiency in global river models by implementing the local inertial flow equation and a vector - based river network map. Water Resources Research , 49 (11), 7221 7235. https://doi.org/10.1 002/wrcr.20552 Yamazaki, D., Sato, T., Kanae, S., Hirabayashi, Y., & Bates, P. D. (2014). Regional flood dynamics in a bifurcating mega delta simulated in a global river model. Geophysi cal Research Letters , 41 (9), 3127 3135. https://doi.org/10.1002/2014GL0 59744 Yamazaki, D., Trigg, M. A., & Ikeshima, D. (2015). Development of a global ~90m water body map using multi - temporal Landsat images. Remote Sensing of Environment , 171 , 337 351. ht tps://doi.org/10.1016/j.rse.2015.10.014 Yamazaki, D., Ikeshima, D., Taw (2017). A high accuracy map of global terrain elevations. Geophysical Research Letters . https://doi.org/10.1002/2017GL07287 4 170 Yamazaki, D., Ikeshima, D., Sosa, J., Bates, P. D., Allen, G., & Pave lsky, T. (2019). MERIT Water Resources Research , 2019WR024873. https://doi.org/10.1029/2019WR024873 Yang, G., Guo, S., Liu, P., Li, L., & Liu, Z. (2017). Multiobjective ca scade reservoir operation rules and uncertainty analysis based on PA - DDS algorithm. Journal of Water Resources Planning and Management , 143 (7), 1 13 . https://doi.org/10.1061/(ASCE)WR.19 43 - 5452.0000773 Yang, S. L., Milliman, J. D., Li, P., & Xu, K. (2011). 50,000 dams later: Erosion of the Yangtze River and its delta. Global and Planetary Change , 75 (1 2), 14 20. https://doi.org/10.1016/j.gloplacha.2010 .09.006 Yeh, P. J. - F., Swenson, S. C. , Famiglietti, J. S., & Rodell, M. (2006). Remote sensing of groundwate r storage changes in Illinois using the Gravity Recovery and Climate Experiment (GRACE). Water Resources Research , 42 (12). https://doi.org/10.1029/2 006WR005374 Zajac, Z., Revilla - Romero , B., Salamon, P., Burek, P., Hirpa, F., & Beck, H. (2017). The impact of lake and reservoir parameterization on global streamflow simulation. Journal of Hydrology , 548 , 552 568. https://doi.org/10.1016/j.jhydrol.2017.0 3.022 Zarfl, C., Lumsdon, A. E., Berlekamp, J., Tydecks, L., & Tockner, K. (2015). A global boom in hydropower dam construction. Aquatic Sciences , 77 (1), 161 170. https://doi.org/10.1007/s00027 - 014 - 0377 - 0 Zhao, F., Veldkamp, T. I. E., Frieler, K., Schewe, J., Os tberg, S., Willner, S., et al. (2017). The critical role of the routing scheme in simulating peak river discharge in global hydrological models. Environmental Research Letters , 12 (7), 075003. https://doi.org/10.1088/1748 - 9326/aa7250 Zhao, G., Gao, H. , Naz, B. S., Kao, S. C., & Voisin, N. (2016). Integrating a reservoir regulation scheme into a spatially distributed hydrological model. Advances in Water Resources , 98 , 16 31. https://doi.org/10.1016/j.advwatres.2016.10.014 Zhao, T., Richards, K. S., Xu, H., & Meng, H. (2012). Interactions between dam - regulated river flow and riparian groundwater: A case study from the Yellow River, China. Hydrological Processes , 26 (10), 1552 1560. https://doi.org/10.1002/hyp.8260 Zhou, T., Nijssen, B., Gao, H., & Lettenm aier, D. P. (2016). The Contribution of Reservoirs to Global Land Surface Water Storage Variations. Journal of Hydrometeorology , 17 (1), 309 325. https://doi.org/10.1175/JHM - D - 15 - 0002.1 Zhou, Y., Hejazi, M., Smith, S., Edmonds, J., Li, H., Clarke, L., et al . (201 5). A comprehensive view of global potential for hydro - generated electricity. Energy & Environmental Science , 8 (9), 2622 2633. https://doi.org/10.1039/C5EE00888C