Assessing the impact of upstream reservoirs on streamflow in the Mekong River Basin Open Access

The Mekong River Basin (MRB) is a vital transboundary area spanning six countries, with the Upper Mekong Basin (UMB) in China and Myanmar, and the Lower Mekong Basin (LMB) in Laos, Thailand, Cambodia, and Vietnam. Covering a basin area of 832,000 km², the Mekong River is about 4,750 km long, with approximately half of it in China, locally known as the Lancang River (Xue et al. 2010). Thousands of years ago, the MRB had a large down-drift area and experienced rapid progradation around Cape Ca Mau, located approximately 200 km downstream from the river mouth. This process leads to the formation of the Mekong River Delta (MRD), which is the third largest delta in the world (Xue et al. 2010; Liu et al. 2017). The Mekong River, particularly the MRD, offers essential ecosystem goods and services, including water for consumption, agriculture, and a highly productive freshwater fishery. As a biodiversity hotspot, the MRB and MRD host numerous unique species. Though the riparian countries are diverse, the Mekong River unites them, benefiting millions. However, challenges including hydropower development, climate change, and land use changes threaten ecosystems, biodiversity, and dependent communities.

The outstanding problem, especially noticeable in the case of hydropower development, is that of the Mekong River. There are 455 dams (existing, under construction, and planned) on the Mekong River, of which 326 dams are operational (Shin et al. 2020). The cumulative effects of constructed dams on the mainstream and its tributaries are transforming the fundamental characteristics of the river regime with pervasive repercussions not only for natural systems but also for social systems and economies (Li et al. 2017; Thilakarathne & Sridhar 2017; Phung et al. 2021). According to Yun et al. (2020), the reservoirs across the Lancang River (the upper Mekong River located in China) reduced the annual average streamflow by 5% at the Chiang Sean Station (northern Thailand) from 2008 to 2016. Lu & Chua (2021) found that observed monthly water discharge in the dam period decreased by 35% during the wet season compared to pre-dam records. Dam construction has caused greater losses of biodiversity and fisheries than climate change in the LMB. The reduction of 276,847 and 178,169 tons of fish, 3.7 and 2.3% of rice, and 21.0 and 10.0% of maize will contribute to a decrease of 3.7 and 0.3% of the GDP of Cambodia and Vietnam, respectively (Yoshida et al. 2020).

Despite the profound impact on the downstream regions, the upstream dam construction continues (Pokhrel et al. 2018). The basin's hydropower reservoir storage may rise from about 2% of its mean annual flow in 2008 to 20% in 2025 (Hecht et al. 2019). Several dams located in different countries, each with varying physical features, governmental policies, and priorities for short- and long-term water resource management, may lead to conflicts in managing and sharing water resources across transboundary regions (Gleick 1993; Adger et al. 2005; Lebel et al. 2006; Eckstein 2009). Countries compete to protect their interests, and the advantage is higher for upstream countries, while downstream countries usually remain passive in using these water resources (Vu & Ranzi 2017). Along with differences in perspectives, priorities, and arrangements in water resources, limited or no data availability has become an additional concern in transboundary watersheds when there is usually no formal data-sharing agreement among parties (Sneddon & Fox 2006; Wilk et al. 2006; Voss et al. 2013).

The understanding of upstream reservoir activities in the MRB is currently limited. Previous major studies were based on assessments of observation data including ground and satellite data, and they can only exhibit a shift in flow patterns before and after the construction of dams, as long as the observation data exist (Räsänen et al. 2017; Timpe & Kaplan 2017). However, these studies have not examined the dynamics to isolate the changes caused by natural climate variability and human activities. To address these issues effectively, models can provide an enhanced understanding of the isolation of natural and human-induced changes (Pokhrel et al. 2017).

Modeling the hydrological processes and reservoir activities of the MRB poses substantial challenges due to its complex topography, the high computational demand involving long computation times and large output storages, data limitations including operation rules, and water downstream demand. Previous studies have worked on simulating streamflow considering dams' effects based on generic operation schemes within continental and global scale hydrological models with input data provided by the Research Program on Water, Land, and Ecosystem (WLE; https://wle-mekong.cgiar.org/). Although there have been significant strides in describing reservoir activities, issues remain regarding the coupling between hydrological and reservoir operation models and detailed assessment of the impacts of upstream reservoirs on the hydrologic balance over the MRB (Lauri et al. 2012; Ngo et al. 2018; Shin et al. 2020; Dang et al. 2022; Yuan et al. 2022). First, the hydrological and reservoir models are not well connected, in which the reservoir model is not integrated into the same system as the hydrological component. Such poor connections between models may result in longer time consumption including data transfer, model simulations, and the connectivity of hydrologic conditions (Kavvas et al. 2013; Connecteur et al. 2018). In addition, the models in different systems are incompatible with real-time and seasonal forecasting studies due to traditional manual connection processes. Regarding the assessment of the impacts of upstream reservoirs, detailed evaluations are still limited. Some studies applied reservoir and hydrological models to investigate the influence of reservoirs only in the upper Mekong Region, specifically the Sesan and Srepok Rivers in the LMB, or focused on selecting dams for such assessments.

A recent development that shows considerable promise is coupling reservoir operation and river floodplain utilizing CaMa-Flood-Dam (Shin et al. 2020). Dang et al. (2022) applied CaMa-Flood-Dam and the global hydrological model HiGW-MAT to simulate river–floodplain–reservoir systems over the MRB. These studies considered a large number of dams across the entire MRB. One issue with this coupled modeling is that dam operation was incorporated into the hydrodynamic model. This method still requires expensive computational resources due to long computation times and large output storage, making it incompatible with real-time forecasting and water resource assessment approaches. An alternative to decreasing expensive computations while aligning with water resource management is using a watershed model coupled with a reservoir model.

In this study, there are four main datasets described in Figure 1 including the global rainfall dataset, land surface dataset, dam information, and streamflow dataset (for calibration and validation).

The global rainfall dataset used in this study is APHRODITE (Asian Precipitation – Highly-Resolved Observational Data Integration Towards Evaluation). Rainfall data extracted from this dataset serve as input for the hydrology model. APHRODITE is a gridded global daily observed precipitation dataset with a resolution of 0.25° (∼25 × 25 km) (Yatagai et al. 2012). It provides long-term data covering the period from 1951 to 2015. The APHRODITE data, combined with configured parameters, were input into the WEHY model for further assessment of reservoir impacts on downstream areas. Along with the rainfall data, the second dataset includes land surface data, comprising elevation data extracted from digital elevation model (DEM), soil data obtained from SoilGrids parameters with a 1 km resolution (Hengl et al. 2014; Trinh et al. 2018), and land use/cover information retrieved from the Global Land Cover Characterization (GLCC) dataset (Loveland et al. 2000). The elevation parameters were delineated from the ASTER Global DEM with a spatial resolution of 30 m (Tachikawa et al. 2011) (Figure 2). Dam and reservoir information including location, purpose, commissioned year, height, and storage capacity are obtained from Dang & Pokhrel (2024). This information is collected from the databases of the Research Program on the WLE (https://wle-mekong.cgiar.org/) and the Stimson Center (https://www.stimson.org/2020/mekong-infrastructure-tracker-tool/). Eventually, the streamflow data for calibration and validation are collected from the Mekong River Commission (MRC) observed database.

Implementing models and simulating hydro-reservoir activities over the MRB are crucial but challenging for hydrologists due to the complexity of the Mekong's topography and governmental policies. The proposed approach involves coupling a physical-based hydrology model with dam operation, as shown in Figure 1. The output of this methodology is an assessment of each country's dam storage impacts on the downstream area of the MRB.

The following describes the implementation of data and models over the study area.

The WEHY model is selected to simulate hydrologic conditions over the MRB. This model has been developed since 2004 by a research group at UC Davis and is widely applied to different regions, particularly in transboundary river basins (Chen et al. 2004a, b; Kavvas et al. 2004,2006; Chen et al. 2011; Kure et al. 2013; Trinh et al. 2016a, b, 2020, 2022a, b, 2023). The WEHY is a distributed hydrologic-watershed model that can simulate parallel surface, subsurface, and groundwater discharge in the river network.

Furthermore, the WEHY enables applications to assess hydrologic conditions over transboundary regions based on a global dataset; thus, it can be applied to the Mekong River watershed including the upstream region of China, Thailand, Laos, Cambodia, and the downstream portion of Vietnam. The required input of WEHY includes atmospheric, topography, soil, and land use/cover data. To configure the WEHY model over the Mekong, the global data sources for topography, soil, and land use/cover were selected and processed over the target watershed.

The elevation, soil, and land use/cover parameters were extracted through the processing of Geographic Information System (GIS) data before being integrated with atmospheric data into model computational units (MCUs) (Kavvas et al. 2013). After configuration, the model is calibrated and validated by comparing it with the corresponding observation data. Once the hydrologic model is validated for the MRB, it can be coupled with a reservoir model to assess the impacts of upstream reservoirs on the downstream hydrologic balance.

The first step in implementing the WEHY model involves delineating the river network and the MCU or hillslopes over the MRB. The delineation was processed by using the GIS technique based on the obtained ASTER global DEM data, resulting in 306 MCUs and 153 stream reaches.

Based on the delineated MCUs and reaches, the second step involved processing inputs obtained from the GLCC and 1 km soil grid datasets to provide land use/cover and soil parameters. WEHY land surface parameter maps for roughness height, vegetation root depth, and leaf area index from January to December are presented in Figures SM1 and SM2 (in Supplementary Materials). The soil parameters were extracted from soil grids datasets, and the resulting product is exhibited in Figure SM3 (in Supplementary Materials), including six parameters including soil hydro-conductivity, mean volumetric water content at saturation of the soil, mean volumetric water content of the soil, mean bubbling pressure, pore size, and soil depth. These land use/cover and soil parameters are crucial for surface and subsurface flow simulations.

Based on the delineated land use/cover and soil parameter maps, all parameters are distributed heterogeneously, particularly soil depth, soil pore size, leaf area index, surface roughness, and root depth. This is due to different regional climate conditions and a wide range of topographies. Thus, to simulate hydrologic conditions, it is necessary to apply a distributed hydrologic model such as the WEHY model.

After the successful configuration of topography (river network and MCUs), soil, and land use parameters, the next step of the implementation process is the calibration and validation of the WEHY model.

In this study, the WEHY model applied configured topography (river network and MCUs), soil, and land use parameters covering the target watershed. However, some parameters needed to be adjusted throughout calibration such as river width, soil depth, and soil moisture at MCUs over the MRB. These values were adjusted based on observed river discharge data, with atmospheric inputs provided by APHRODITE. The adjusting process was based on changes in parameters and comparisons between observation and simulation data. The performance of the WEHY model was evaluated using two performance indicators suggested by Moriasi et al. (2015), including the coefficient of determination (R²) and the Nash–Sutcliffe efficiency (NSE). Their definitions are shown in Table A1 of Supplementary Materials.

While information on the specifications of many reservoirs in the Mekong region has been collected and published in previous studies, their actual operational rules are not widely accessible due to various reasons. Thus, to estimate the regulated outflow for each reservoir separately, we applied a well-known and popular method in reservoir operation, which is maximizing outflow from a reservoir through hydropower turbines (Lauri et al. 2012; Shin et al. 2020) while maintaining dam safety through storage control. The resulting operation will potentially overestimate the current real-life usage of these reservoirs since normal operational rules are generally more complex to meet multiple purposes, including maintaining base flow, supporting water demand, and preventing extreme floods, along with filling the reservoir each year while operating to meet committed electricity demand as well as regional laws and treaties. However, by applying this optimization scheme, we aim to uncover the maximum possible impact that large hydropower could impose on the MRB in each future scenario.

Last, to reduce the computational cost of optimization during the hydrodynamic simulation, we have utilized the natural condition (NAT, no dam) simulated discharge as input to optimize the relevant reservoir operation. We first performed NAT simulations with all scenarios and forcing inputs. Then, a complete time-series discharge at each dam site was extracted for each set-up to obtain Q_in. The potential outflow (Q_out) is optimized for each dam and used as input to guide the model in simulations with dams (DAM).

The reservoir model was developed as open-source software using a combination of Fortran and Python and was presented by Dang et al. (2022) under CaMa-Flood-Dam. In this study, the dam operation component was incorporated into the flow simulation by the WEHY model. First, the WEHY model simulates hill flow to obtain lateral and upstream flow conditions for the river network simulations conducted in WEHY's routing model. Then the obtained lateral and upstream flow conditions are provided as input for the dam operation component. For first-order reservoirs, the inflow is natural flow, while, for second or third-order reservoirs, the inflow is calculated based on the sequence of upstream reservoir activities.

Among the dams commissioned by 2022, 15 dams have been selected based on the following criteria: (1) primary purpose as hydropower dams; (2) located at the main stream of Mekong; and (3) storage capacity impact to the downstream area. The locations of these dams are shown in Figure 2, and more detailed dam specifications can be found in Table 1.

The calibration and validation of the WEHY model were performed by comparing simulated and observed data at selected discharge stations across the MRB. Five discharge stations were chosen, as shown in Table 2, based on data availability, time step, and their locations in different countries along the main stream. The observed data were collected from the MRC database, with records available from 1961 to 1979 at both daily and monthly intervals. Table 3 presents the statistical criteria, including the coefficient of determination (R²) and the NSE coefficient. According to the classification by Moriasi et al. (2015), the simulation performance for daily and monthly streamflow falls within the ‘very good’ range based on an R² coefficient of ≥0.85. However, when assessed using the NSE coefficient, the simulation performance is classified as ‘very good’ for monthly streamflow calibration (NSE ≥ 0.82), ‘good’ for monthly streamflow validation (0.75 ≤ NSE ≤ 0.865), and ‘satisfactory’ for daily stream validation (0.61 ≤ NSE ≤ 0.78).

These figures demonstrate that the model-simulated outflow matches well with the observations, while the elevation level hydrographs align reasonably with the observations. This calibration and validation confirm that the distributed hydrologic model WEHY can produce unimpaired historical flow data over the MRB successfully. The WEHY simulated flow data were then input into the reservoir model to reconstruct impaired flow and assess the upstream reservoirs' effects on the hydrologic balance.

After the successful implementation of upstream dam reservoirs into the WEHY model, it can simulate impaired hydrological conditions over the MRB and assess the impacts of the upstream reservoirs on streamflow in the downstream areas. Fifteen dams have been selected for this assessment, including six dams from China, five dams from Laos, two dams from Thailand, and two dams from Vietnam as shown in Figure 2. The assessment is conducted based on comparisons among different scenarios. There are five scenarios in this assessment including (1) Scenario 1 (S1) is unimpaired flow obtained from WEHY simulation without dam; (2) Scenario 2 (S2) is impaired flow obtained from WEHY coupled with six dams from China; (3) Scenario 3 (S3) is impaired flow obtained from WEHY coupled with five dams from Laos; (4) Scenario 4 (S4) is impaired flow obtained from WEHY coupled with two dams from Thailand; (5) Scenario 5 (S5) is impaired flow obtained from WEHY coupled with two dams from Vietnam. The Kratie Station was selected to evaluate upstream effects on the downstream due to its account for all impacts from upstream dams' operations.

After the successful implementation, calibration, validation, and coupling of WEHY and reservoir models, the next step is to assess the impact of upstream reservoirs on streamflow in the lower MRB through the five scenarios in the different seasons (dry and wet seasons).

It can be inferred from Figure 8 that China has the largest storage, accounting for 62.53% of the total storage in this study, Laos's storage accounts for 27.31%, while Thailand and Vietnam's storages are 6.96 and 3.21%, respectively. In both the dry and wet seasons, China's reservoir has the largest impact on the downstream, with of 42.13 and 50.46%, respectively. Laos's dam storage follows with the second largest impact, contributing 26.15 and 27.9% during the dry and wet seasons, respectively. Vietnam and Thailand have a smaller contribution, ranging from 14.92 to 16.80% during the dry season and from 9.8 to 11.84% during the wet season. It is evident that the percentage impact is directly proportional to the dam's storage; larger storage results in a larger impact, as depicted in Figures 8 and 9.

This study proposes an approach that can analyze the effects of each country's dam storage on downstream areas separately, utilizing a watershed model coupled with a reservoir model. The proposed approach can account for the complex interactions between hydrologic conditions and reservoir operations under different scenarios, thus enabling the evaluation of upstream dams' impacts on the downstream area of the MRB. However, this study only considered major dams (with a capacity larger than 3 billion m³) on the mainstream and did not account for all dam operations. To enhance the reliability of the approach, future studies should consider a larger number of dams and project these dam impacts downstream under climate change conditions.

This study aims to reconstruct historical streamflow and evaluate the impacts of major upstream reservoirs on streamflow in the downstream areas of the MRB. The reconstruction and evaluation were conducted by coupling the WEHY model with the reservoir model from the CaMa-Flood-Dam model. Initially, the WEHY model was implemented based on delineated hillslopes and river networks, with input from global datasets, including precipitation, soil, land use/cover, leaf area index, and DEM.

The WEHY model is calibrated and validated against the observed flow data from the MRC database. The calibration and validation results demonstrated strong performance across the MRB, as indicated by statistical criteria (R² and NSE) showing a high correlation. Thus, the WEHY model is deemed reliable for further applications. The simulated natural flow from WEHY was subsequently input into the reservoir model to simulate impaired flow conditions across the MRB. After successful configuration and validation, the coupled model was able to simulate various upstream dam activity scenarios and their impact on streamflow in downstream regions. The assessment of upstream reservoir impacts on streamflow in downstream regions aligns with previous analyses, indicating that reservoir activities tend to increase flow conditions in the dry season while decreasing them in the wet season. China's reservoirs, having the largest storage capacity, exert the most significant impact on downstream areas, particularly evident during the wet seasons of 2014 and 2015. Laos follows as the second largest contributor. In contrast, Thailand and Vietnam, with smaller storage capacities, have negligible impacts on downstream regions. A key conclusion is that a dam's impact is directly proportional to its storage capacity, with larger storage leading to greater downstream effects.

This study also underscores the importance of coupling a physically based hydrology model with the reservoir model to facilitate the comprehensive reconstruction and evaluation of dam impacts. Moreover, the proposed technique can simultaneously simulate rainfall, flow conditions, and dam operations, offering reliable forecasts and projections of flow conditions across the MRB. These insights could prove invaluable to management authorities during flood and drought events.

However, it is important to note that this study only considered major dams (with capacities larger than 3 billion m³) on the mainstream, without accounting for all dam operations. Therefore, further studies are needed to improve the reliability of the proposed technique by incorporating all dam operations.

This research was funded by the Vietnam National Foundation for Science and Technology Development (MOST), the Ministry of Science and Technology, under grant number NDT/KR/21/18. The authors also would like to thank the anonymous reviewers for their valuable and constructive comments to improve our manuscript.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.