Future rice farming threatened by drought in the Lower Mekong Basin

The Lower Mekong River basin (LMB) has experienced droughts in recent decades, causing detrimental economic losses and food security conundrums. This study quantified the impact of climate change on drought, and rainfed rice production in the LMB. The Soil and Water Assessment Tool (SWAT) and AquaCrop models were used to evaluate long-term drought indices and rainfed rice yields under historical and future climate conditions (1954–2099) with four climate models and two emission scenarios (RCP 4.5 and RCP8.5) from the Coupled Model Intercomparison Project Phase 5 (CMIP5). We found that rice yield might increase (24–43%) due to the elevated levels of atmospheric CO2 concentration (+ 34.3 to + 121.9%) and increases in precipitation. Contrastingly, considerable decreases in rice yield up to 1.5 ton/ha in the Vietnam Central High Plain (VCHP) region could be expected resulting from reduced precipitation by about 34% during drought years. To avert any major food crisis, an expansion of irrigation areas could be required to compensate for the expected reduction in rice yields. We conclude that a framework combining hydrology and crop models to assess climate change impacts on food production is key to develop adaptation strategies in the future.

in plants as a result from increased CO 2 levels 30 . Furthermore, these results pointed out that there were decreases in rice yields in the middle of twenty-first century without the irrigation and CO 2 fertilization effect in the Mekong Delta. However, considering CO 2 fertilization effect of RCP 4.5 and 8.5 emissions, increases in rice yield up to 23% were plausible, and that could compensate for yield declines by increasing temperatures 28 . Jiang et al. 31 assessed the impacts of climate change on rice production in the Mekong Delta using DSSAT and global climate models. The results implied that reductions of rainfed rice occurred due to increasing temperatures and seasonal variations of precipitation, but the elevated CO 2 resulted in remarkable increases in rice productions. Chun et al. 32 demonstrated that improved irrigation strategy and CO 2 fertilization effect could cause significant increases in rice yields (up to 8.2-42.7%) in the 2080s under RCP 8.5 in Cambodia and Thailand. Poulton et al. 33 found that CO 2 level and rainfall variability have significant influences on rainfed rice production in Cambodia after 2030. Also, they concluded that expanding irrigation areas would be required to compensate for the impacts of climate change.
Thus far, several studies concluded that a combination of CO 2 fertilization and irrigation strategies may lead to increases in rice production or at least compensate for the yield losses due to the impacts of climate change in the LMB. However, other studies showed different results in the regional scale analysis. Boonwichai et al. 34 simulated rainfed rice using the DSSAT crop simulation model in the Songkhram River Basin of Thailand, and rice yields reduced by 14% and 10% under the RCP4.5 and RCP8.5, respectively. The reductions in rice yields were mainly due to the increases in temperature and crop water use. Only a few studies demonstrated the impacts of climate change and resulting drought on rice yields, including for the LMB region where a significant cultivable land area was used for rice cultivation. Trisurat et al. 35 used the InVEST model (Integrated Valuation of Ecosystem Services and Tradeoffs) and estimated rice production in the LMB. They found that rice production would decrease up to 4.2% due to the extreme drought conditions, and these reductions were mostly confined to Thailand and Lao PDR. However, the lack of an integrated assessment using a physically-based hydrologic model, climate models, and crop simulation is evident for evaluating future drought and rice yield in rainfed areas 36 . This approach is essential to quantify the impacts of droughts and rice yield at a scale appropriate for decision making by explicitly accounting for increased biomass production from the CO 2 fertilization effect. It is also imperative to evaluate what-if scenarios for expanding irrigated areas to mitigate any crop loss due to drought and to avoid potential reductions in rice yield.
The objectives of this study are to evaluate drought and rice yield in the LMB at the provincial scale over the historical period  and to assess the impacts of drought on rainfed rice yield in the next 80 years (2020-2099) with two future climate projection periods (f1: 2020-2059, f2: 2060-2099). This study provides crop yield under future drought scenarios using soil moisture-based drought index, crop simulation model, and high-resolution meteorological dataset, and the results of this transnational study can serve as a viable guideline for planning water resources and land management for alleviating food security concerns in the face of natural and human-induced changes in LMB. Figure 1 outlined the overall approaches and methods that were used in this study. First, input data required for the Soil and Water Assessment Tool (SWAT) 37 and AquaCrop 38 models for the LMB were obtained. Second, the SWAT model was calibrated by comparing monthly streamflows from seven stations, and the AquaCrop model was calibrated using the observed rice yields for each province. Third, SWAT-driven soil moisture was used to compute a drought index for the historical and future periods. Forth, AquaCrop was used to estimate rice yields for the historical and future periods. Following this, the climate change impacts on droughts, rice productions, and required irrigation areas were evaluated based on the drought index and rice yield simulations.

Study area.
This study focused on the entire LMB. It covers an area of 642,000 km 2 , roughly consisting of 76% of the Mekong River Basin (Fig. 2a). The river originates in China and flows about 2300 kms to reach near Chiang Saen and this segment is called the upper Mekong River. Below this point, the river becomes the Lower

AquaCrop model and inputs.
AquaCrop is a crop growth model developed by the Land and Water Division of the Food and Agricultural Organization (FAO), and it is a process-based model that considers crop growth, atmospheric, soil, crop, and management conditions 38,51 . AquaCrop requires a relatively small number of inputs and parameters, and it simulates yield response to water of crops and is suited to explain conditions where water is a crucial factor in crop production 38 . AquaCrop has been widely used for investigating climate change impacts on crop production and food security in many regions in the world, particularly in the LMB 3,52 . Biomass production and yield formation in AquaCrop were based on Eqs. (1) and (2) 38 .
where B is the cumulative biomass production (ton/ha), WP * is the biomass water productivity normalized for climate, Tr i is the daily crop transpiration (mm/day), ETo i is the daily reference evapotranspiration (mm/day), Y  www.nature.com/scientificreports/ is the final crop yield (ton/ha), and HI is the harvest index. Raes et al. 38 pointed out that an increased CO 2 level, which induces CO 2 fertilization, derives slight decreases in the crop transpiration (Tr) and significant increases in the biomass water productivity (WP * ) for the AquaCrop simulation. Besides, a temperature is likely to increase the evaporating power of the atmosphere (ETo). AquaCrop requires soil properties, daily precipitation, minimum and maximum temperature, reference evapotranspiration (ETo), planting date, and the growing period, which are relatively a few parameters and inputs to simulate the crop yield in response to various water availability conditions 38,40 . For the historical and future simulations, soil properties were extracted from the FAO soil dataset 44 . Forcing datasets include daily precipitation, minimum and maximum temperature data and the same inputs used for SWAT simulations were again utilized after aggregating them for each province. Besides, other variables namely reference evapotranspiration (ETo) and CO 2 concentration for each RCP were derived from SWAT as PET and Nazarenko et al. 53 , respectively.
Model-driven drought index. SWAT simulated soil moisture was used to compute the multivariate standardized drought index (MSDI) 54 . MSDI is based on the joint distribution and probability of two hydro-meteorological variables such as precipitation and soil moisture. For this study, monthly soil moisture estimations of entire soil columns (about 2 m) from the SWAT model and precipitation data were used to evaluate historic and future drought conditions in the 74 provinces in the LMB.
The joint distribution of two variables (X and Y) is defined as Eq. (3): where p is joint probability of two variables, and MSDI is described as Eq. (4): where ∅ is the standard normal distribution. For the MSDI estimation, the Gringorten plotting position formulas is used to reduce the uncertainties in fitting parameters 55 , and the joint distribution was computed by Eq. (5); where n is the number of observations, and m k is the number of occurrences when the pair ( x i , y i ) is x i ≤ x k and y i ≤ y k . Supplementary Table S2 Table S3), and the SUFI-2 algorithm in the SWAT calibration and uncertainty assessment tool (SWAT-CUP) 56 was used. The model evaluation was performed based on the Nash-Sutcliffe coefficient (NS) 57 (Eq. 6).
where − Q o is the mean of observed discharge, Q s is simulated discharge, Q i o is observed discharge at time i. In addition, the model-driven MSDI was validated by the comparisons with a reference drought index (Palmer Drought Severity Index; PDSI) at a moderate drought category (PDSI < − 2, MSDI < − 1). Supplementary Figure S1 presents the selected locations for drought assessments, and they were highlighted as red points for PDSI and black areas for MSDI.
The model accuracy of the AquaCrop model was assessed by comparing observed and simulated rice yields for each province, and the performance was verified by calculating the root mean square error (RMSE) (Eq. 6), which has been widely used for crop simulation studies 3,40,58 . For the baseline simulations, planting date, harvest index (HI), and fertilizer stress were used as the calibration parameters, and they have been applied for the AquaCrop simulations (Supplementary Tables S4-S7 in the Supplementary Information) 3,54,59 . Based on the assessment of root mean square error (RMSE), the simulation was found to be satisfactory for the LMB region, and the detailed discussion is provided in the later sections.
where, simulated i and observed i are the simulated and observed rice yield, and N is the number of observations. The performance of the simulation improves as RMSE approaches zero.
Satellite-based soil moisture and rice yield. Soil moisture is an essential variable for monitoring agricultural production and drought, and it is appropriate to evaluate the impact of water deficit 54,60 . In particular, the regions of rainfed agriculture are largely affected by soil moisture deficits during the crop growing season 61 . In this study, the simulated rice yields and droughts were compared with a satellite-based root zone soil moisture observation from Soil Moisture Active Passive (SMAP, Level 4) 62 to understand the relationship between rice www.nature.com/scientificreports/ yield, precipitation, and drought conditions. SMAP provides three hourly soil moisture data with a 9 km spatial resolution from April 2015, and 3065 grids correspond to the rainfed areas in the LMB (Fig. 2c).

Impacts of drought on rice yield.
In this study, the crop yield analysis for the rainfed rice was performed at the provincial scale using the observed agricultural production data in the LMB (Fig. 2b). Observed crop yield data was obtained from the official statistical websites or reports of the countries in the LMB 24,40 . The study area contains 74 provinces in the LMB, and there are 20, 17, 22, and 15 provinces in Cambodia, Loa PDR, Thailand, and Vietnam, respectively. The average area of all the 74 provinces is 8079 km 2 , the largest province is Savannakhet in Lao PDR (21,407 km 2 ), and the smallest is Phnom Penh in Cambodia (373 km 2 ). However, the Vietnam Central High Plain (VCHP) region does not include the provinces entirely because the current boundaries are based on the Mekong River basin, not on a provincial border (Fig. 2).
To figure out the impacts of drought on rainfed rice, we divided drought and non-drought years and compared the rice yields. The drought and non-drought years were separated whether the MSDI values were below average or not. To mitigate the impacts of drought, rainfed areas would need irrigation as they serve as a shock absorbed during the times of failed monsoon. While the expanded irrigation shields the farmers from facing the losses in crop yield, it can potentially alleviate food security and livelihood concerns. The required irrigation area (RIA) was calculated to quantify how much additional irrigation areas would be needed to compensate for the yield losses during the drought years, and RIA was calculated by Eqs. (8) and (9).
where, YL is the yield losses (ton), Y n is the rice yield during the non-drought years (ton/ha), Y d is the rice yield during the drought years (ton/ha), A rice is the rice cultivated area for each province (ha), and RIA is the required irrigation area (ha).

Climate projections. The Coupled Model Intercomparison Project Phase 5 (CMIP5) climate model out-
puts are widely used to assess historical and future climate impacts with higher reliability in the projected precipitation and temperature. In this study, four CMIP5 climate models, GFDL-ESM2M, IPSL-CM5A-LR, MIROC-ESM-CHEM, and NorESM1-M were implemented with two representative concentration pathways (RCP4.5 and RCP8.5), and they captured a wide range of precipitation (+ 18.3 to + 47.0%) and temperature changes (+ 1.1 to + 4.6 °C) in the LMB. We directly used statistically downscaled and bias-corrected data at 0.25-degree resolution that were available by the intersectoral impact model intercomparison project (ISI-MIP) 63 .
As temperature and precipitation characterizations are needed for drought and crop growth modeling, especially under rainfed conditions, the evaluation of precipitation and temperature for their trends and future projections are deemed necessary. Figure 3 presents changes in mean values of precipitation and temperature for the four GCMs, RCPs, and future periods. Among the climate models and periods, GFDL-ESM2M and RCP4.5-f1 period was the driest model that showed the lowest precipitation increase (18.3%), while NorESM1-M and RCP8.5-f2 period was the wettest model, which presented the highest precipitation increase (47.0%). In addition, NorESM1-M and RCP4.5-f1 period was a cold model that showed the lowest temperature increase (1.1 °C), while MIROC-ESM-CHEM and RCP8.5 f2 was a hot model, which presented the highest temperature increase (4.6 °C). Supplementary Figure S2 in the Supplementary Information shows the spatial maps of historical annual precipitation and future changes for each climate model and period, and Supplementary Fig. S2a,b are the results of RCP4.5 and RCP8.5, respectively. For the historical period, annual precipitation ranged from 933 to 2070 mm. For the future periods, the means of four climate models in the entire LMB indicated that there were  Although the overall precipitation increases were expected, a significant reduction in precipitation during the crop growing season as projected could cause a yield reduction. Tables 1 and 2 show the total amount of precipitation during crop growing season for the historical and future periods and precipitation differences between drought and non-drought years for future periods. Table 1 present the mean values of four GCMs, and www.nature.com/scientificreports/ shows the precipitation differences for each climate model and the future period. In Table 1, the numbers below the precipitation amount indicate the number of droughts and non-drought years, and they are approximately 30 years that are half of the historical period. However, there were decreases in a number of drought years and increases in non-drought years for the future periods, which were particularly noticeable in the results of a climate model with high precipitation increases (e.g., NorESM1-RCP4.5 model and f2 period). In the following sections, this assessment serves as the basis for evaluating the effect of temperature and precipitation changes on drought and crop yield in the rainfed agriculture areas of the basin.

Results and discussion
SWAT and drought evaluations. The SWAT model was calibrated using SWAT-CUP 56 65 and multivariate standardized drought index (MSDI) at a moderate drought category (PDSI < − 2, MSDI < − 1) (Supplementary Table S2). Supplementary Figure S1 in the Supplementary Information shows the locations of PDSI and SWAT sub-watersheds, and the numbers in the yellow boxes correspond to specified points on the map. The model-driven MSDI generated 4.0, 10.1, and 15.9% of extreme, severe, and moderate droughts, and they captured between 56 and 73% of the drought occurrences in this period. Factors responsible for the discrepancies in these comparisons were manifold. First, the PDSI data was at 2.5-degree resolution, while the estimated MSDI was at 0.25-degree, primarily driven by the resolution of meteorological data. Second, the two different sources of meteorological data were applied for this assessment. For example, PDSI was calculated by the precipitation data from the National Centers for Environmental Prediction (NCEP) Climate Prediction Center (CPC) 66 , while the model-based MDSI used the other precipitation data 47 , which had 15.3% more precipitation. Finally, two drought indices represent different drought types. PDSI represents the meteorological drought and is based on an empirical method of a balance between moisture supply and demand 65 , while the MSDI is based on a probabilistic approach that is joint distribution of precipitation and soil moisture 54 .

AquaCrop model evaluation. Supplementary Tables S4-S7 in the Supplementary information present
the planting date, harvest index (HI), fertilizer stress (FS), calibration period, observed and simulated average yields, and RMSE for the AquaCrop simulations. The planting date of rainfed rice varies from April 20 to July 1, and it depends on the start of the wet season for each location 40 . HI is a function of water availability and crop variety 59 , and a higher value generates a higher rice yield. FS demonstrates deficit nutrient conditions in soil layers, and a higher value adversely affects the yield. For all provinces, HI varies from 23 to 37%, which was within the range of other applications in the LMB 40,59 . The mean values of RMSE for each country were 0.30, 0.30, 0.20, and 0.40 ton/ha for Cambodia, Lao PDR, Thailand, and Vietnam, respectively. The RMSE values were 15.6%, 9.7%, 8.4%, and 9.3% of the mean values of the simulated rice yields during the calibration period. For all provinces, RMSE varied from 0.10 to 0.61 ton/ha, which were 3.9-22.2% of the simulated rice yields, which were within the RMSE range of the previous studies in the LMB 40,67 . Therefore, the results implied that model performances of SWAT and AquaCrop were satisfactory for the simulation of rainfed rice in the LMB. Since those calibration parameters optimize the suitable planting date, water availability, and nutrient deficiency of the land, appropriate parameter estimations for each region are crucial for reliable simulations of the rice yields 40 . Furthermore, a well-calibrated hydrological model can be relied upon for accurate soil moisture and drought index computation, and the integrated hydrology model with crop simulation and climate models enables the assessment of future climate change impacts on drought and rice yields in the LMB.
Relationship between soil moisture and rice yield. Figure 4a,b present the average SMAP soil moisture, average standardized soil moisture index (SSI), rice yield, and precipitation amount during the crop growing season. SMAP soil moisture and model-driven SSI showed similar patterns for drier years such as 2015 and 2019, while a slight overestimation of SWAT-SSI for an average rainfall year is seen. General ascent in wetness was seen in SMAP soil moisture, SSI, precipitation, and that corresponded well with increases in rice yield from 2015 to 2017, and decreases from 2017 to 2019. Figure 4c,d show spatial maps of the average soil moisture from SMAP and SSI for each year. Despite localized anomalies, the larger Mekong region including the delta did not suffer significant soil moisture deficits due to low climate variability and better surface water management projects, which led to stable rice yields ( Supplementary Fig. S4d in the Supplementary Information). However, some regions showed fluctuation in rice yield and soil moisture patterns primarily driven by year-to-year monsoonal variability 68 . Those results imply that rainfed rice productions were primarily affected by precipitation and soil moisture conditions.

Climate change impacts on droughts and rice yields.
To further explore the future scenarios of production of rice, this study simulates long-term rice yield for the rainfed areas in the LMB, and Fig. 5 presents the time series of annual average simulated rice yield for the basin countries during the historical and future periods. During the historical period, trends in rice yields had a gradual increase. While comparing the earliest decade (1956)(1957)(1958)(1959)(1960)(1961)(1962)(1963)(1964)(1965) 40,67 , by up to 25% due to increased CO 2 concentration levels. Figure 6 presents non-linear scatter plots of crop yield and CO 2 concentration, which show a strong positive relationship in all countries and for both RCP4.5 and RCP8.5. Numerous studies have pointed out a positive fertilization effect of the elevated CO 2 concentration on crop production [69][70][71] . This study also confirms that CO 2 concentration is a primary driver of biomass productivity and rice yield under climate change conditions. However, this study does not evaluate some detrimental impacts of elevated CO 2 concentration on crop quality such as protein, micronutrients, and vitamin contents of rice 72,73 . Baker and Allen 74 and Allen et al. 75 www.nature.com/scientificreports/ also found the non-linear relationship between CO 2 concentration and rice yield, and explained that elevated CO 2 concentrations improved crop yields due to increased photosynthesis, and even further propounded that doubling of the CO 2 concentration will increase crop yields up to 50%. Allen et al. 76 tested a non-linear model to verify the impacts of the elevated CO 2 concentration on crop yields and found proportional yield decreases under the high CO 2 level that derived the non-linear relationship. From the investigation of long-term projections of rice yield, it is clear that simulated rice yield was generally found to be increasing in all countries; however, there were some sharp declines in rice yields for the future periods. The numbers of drought events in the future periods were presented in Table 1, and the sharp declines in rice yield occurred during the drought years. Supplementary Figure S6 presents the spatial patterns of yield differences between the drought and non-drought years for future periods and climate models. As shown in the tables and figures, there was an overall reduction in rice yields during the drought years, and the decreases were highlighted as yellow to red areas. Specifically, in the VCHP (Vietnam Central High Plain) and some provinces Lao PDR regions, there were more than 0.6 ton/ha reductions in rice yields, which were derived from the sharp declines of precipitation during the crop growing seasons (Tables 1, 2). On the contrary, Cambodia and Mekong Delta showed less than 6.1% precipitation decreases during the drought year, and rice yields were relatively unchanged (− 0. 15  Required irrigation areas and discussion. This study also explored an unconventional approach to mitigate the impacts of drought on food production by inverting yield loss to expanded areas for irrigation during drought years. Table 3 shows required irrigation areas (RIAs) computed both as area and percentage for paddy growing areas to compensate for the potential decreases in crop yields during the drought years, and Fig. 7 presents the spatial maps RIA. As defined in the "Methods" section, RIA is an additional irrigation area to compensate for the loss in rice yields during the drought years, and RIA% is the percentage of RIA needed to annex the existing paddy areas for each province (Eqs. 7,8). and the highest mean values of RIA% was evident in the VCHP (up to 18.8%) and Lao PDR (up to 4.0%) regions, while the Mekong delta region showed the lowest (0.1-2.1%).
The RIA% values were closely associated with the decreases in precipitation during the crop growing season, and the highest precipitation declines occurred in the VCHP region (− 15.3 to − 23.6%; Table 1). In addition, Lao PDR showed large decreases in precipitation (− 8.6 to − 13.6%). However, the Mekong delta showed lower RIA% as it might witness only moderate decreases (− 0.3%) or increases (0.2-2.0%) in precipitation.
From various climate projections of precipitation and temperature, which were used for soil moisture simulations by SWAT, the GFDL-ESM2M and MIROC-ESM-CHEM models presented drier and drought conditions   Fig. 7), which were deeply impacted by the reduced precipitation during the drought years (− 12.0 to − 15.4%; Table 3). Also, MIROC-ESM-CHEM-RCP4.5 resulted in high RIA% (12.5-12.7%), and they were also primarily due to decreases in precipitation during the drought years (− 29.8 to − 25.9%). On the contrary, MIROC-ESM-CHEM and NorESM1-M showed a smaller increase in RIA% (0.0-2.1%) that were caused by the modest precipitation declines (− 0.2 to − 3.0%) or increases (0.4-16.5%) in the Mekong Delta. Thus, the results imply that spatial heterogeneities and variabilities of precipitation between drought and non-drought years would lead to different responses in projected rice yields and areas in the LMB in the future, and higher uncertainties were probable in estimating the required irrigation areas as the climate model outputs were directly dependent on emission scenarios and climate models.
In the LMB, many dams have already been built and operated for various purposes (e.g., hydropower, irrigation, multi-purpose), and many dams are being planned in the future. In this study, the VCHP region was chosen as requiring the most RIA, and 28 dams were already commissioned for irrigation purposes. However, at least seven were also planned to be built for hydropower generation in the future 77 . Since the VCHP region needs the most RIA, an additional analysis was carried out to identify whether the additional dam for irrigation would fulfill the need. Besides, Lao PDR needs the second-largest RIA (up to 4.0% on average). However, since there were no currently operated or planned irrigation dams in the provinces, sound considerations and plans can potentially augment the need to meet crop production demand during the drought years.
Overall, the results of all climate models projected an increase in crop yields due to higher than historic precipitation and elevated CO 2 concentration in the atmosphere. However, the climate models used in our study suggested probable decreases in precipitation during drought years that could result in drastic decreases in crop yields, especially in the VCHP and Lao PDR regions. For all climate models and future periods, an average of 55 and 39 provinces for RCP4.5 and RCP8.5 were affected by the decrease in precipitation during the crop growing season of the drought years. This also suggested that the future precipitation patterns estimated by the climate models could necessitate expansion in irrigated areas due to climate extremes, despite a long-term gradual increase in precipitation increases were possible. Understandably, these impacts can be mitigated by adjusting the planting date and applying adaptation methods 3 , while the uncertainties persisted in climate model predictions, future droughts, and rice production.
This study did not evaluate diverse adaptation options under various stresses including salinity intrusion, and dam constructions that were widely considered in other studies, and further analyses are required to assess their role in water and food security in the basin. Mainuddin et al. 40 evaluated climate change impacts on rainfed rice yield in the LMB based on a crop growth simulation model (AquaCrop) and IPCC SRES A2 and B2 scenarios and reported the possibility for increased rice yield in Laos and Thailand due to the elevated CO 2 concentration. They also assessed various adaptation methods such as supplementary irrigation, adjusting planting date, and reduction in fertility stress, which compensated negative impacts on rice yield by climate change. While it was not assessed in any earlier studies, our study evaluated the need for expanded irrigation areas for the entire LMB in the face of climate change-induced hydrologic extremes. Other studies also assessed adaptation strategies to mitigate the adverse impacts of climate change on rice production in Asia 39,78 , and found yield losses were primarily caused by heat stress and drought recommended that adjusting sowing dates would effectively compensate for the declines of rice yield. Also, further studies are required to estimate the impacts of additional irrigation on the water as well as energy budgets and managements [79][80][81] . Finally, sea-level rise and salinity intrusion in the Mekong Delta were not considered while significant sea-level rise (up to 30 cm in 2050) and salinity intrusion are expected, and extensive areas will be negatively affected 82,83 . Future dam constructions are also expected to www.nature.com/scientificreports/ cause reductions in downstream reaches flows in the basin 84 , and this could negatively impact rice production with a reduction in flows to the delta region 85 .

Data availability
Simulation data developed in this study will be made available on request to the corresponding author.