Surface Water

Abstract

This chapter assesses surface water changes due to climate change and human activities, by particularly examining runoff and streamflow. Changes in the hydrological cycle due to climate change and human intervention can lead to diverse environmental impacts and risks. Fresh water is the agent that delivers many of the impacts of climate change on society. As the major component of freshwater systems, surface water has been significantly altered across basins in terms of spatial and temporal characteristics. The comprehensive understanding of the current status of surface water in the LMRB, such as the distributions and patterns of runoff changes across the Lancang-Mekong River Basin was completed through the high-resolution river network extraction and sophisticated hydrological models. Significant but different trends were found in the seasonal and annual runoff from the LMRB due to different reasons. Over the period of 1971–2010, the annual streamflow shows a general downward trend due to the continued enhancement of human activities. Runoff in the dry season is found to increase faster than the mean annual runoff. As for the spatial distribution, significant trends in streamflow were observed mainly in the middle basin and east of the lower basin. Superimposed on the substantial seasonal cycles is the noticeable lake shrinkage in recent years, especially the Tonle Sap Lake. Evidently decreased inundation was found in most years in the recent two decades from 2000 to 2018. An evident decreasing trend in runoff caused by climate change in the high correlation zone of the Tonle Sap Lake, mainly due to the precipitation decreasing, indicates that climate change contributed to the decrease in water level in the Tonle Sap Lake in addition to human activities. In addition to the decreases in the runoff, streamflow and water level in the Tonle Sap Lake, a significant (p < 0.05) downward trend in the baseflow was also found from 1980 to 2007. Unlike the historical changes in runoff, previous studies projected with high confidence an increasing trend for streamflow in the LMRB, regardless of the climate forcings and models used. However, the flow regime is highly susceptible to a variety of drivers, e.g., dam construction, irrigation expansion, land-use change and climate change. Substantial changes are expected in both annual and seasonal flow, along with a generally increasing trend. Although hydropower development exhibits a limited influence on total annual flows, it has the largest seasonal impact on streamflow, with an increase in the dry season and a decrease in the wet season, by outweighing those of the other drivers.

3.1 Introduction

Despite rich water resources (~8,000 m³/cap/yr), the LMRB faces significant challenges due to the high variability in runoff, both in terms of timing and location (MRC, 2010). It is imperative to comprehend how runoff patterns respond to the impacts of climate change and human interventions. This understanding is crucial for ensuring the availability of water, food, and energy resources in the region, as well as for achieving long-term sustainability. Therefore, a thorough assessment of changes in the runoff regime of the LMRB is necessary to assist in understanding the influence of regional climate changes and human activities on water availability.

The LMR originates in the Tibetan Plateau, a region that is extremely sensitive to climate change (Chen et al., 2015; Kuang & Jiao, 2016). Climate change has already left a significant imprint on the hydrology of the LMRB in recent decades (Lyon et al., 2017; Phi Hoang et al., 2016). Over the past half century, the basin has experienced increased temperatures as well as increased rainfall during the flood period and reduced rainfall during the dry period. Additionally, the rapid economic growth, rising food demands, and increasing energy requirements in riparian countries have driven substantial changes in land use and land cover, especially due to extensive agricultural expansion and hydropower development across the basin (Johnston & Kummu, 2012). Climate change and human interventions have substantially reshaped the basin's runoff patterns, resulting in more frequent extreme events and extended dry periods (Thilakarathne & Sridhar, 2017). The lack of upstream inflow during the dry season exacerbates the risk of saltwater intrusion, impacting downstream delta ecosystems, domestic water supplies, and agricultural production (Smajgl et al., 2015). Simultaneously, intense and widespread precipitation events have led to severe flooding, causing damage to crops and infrastructure, and disrupting the functions of the downstream delta (Cosslett & Cosslett, 2014).

The changes in the runoff regime of the basin have led to the degradation of essential natural resources in the region, including fish, water, and land, upon which millions of people depend (Chea et al., 2016). In addition, the climate change impact on water has been projected to intensify in the near future, and the spatial and year-to-year distribution will be more uneven in the basin (Hoang et al., 2019). Superimposed by the effects of human activities, thereby challenging sustainable development in the region. Therefore, there is an urgent need to deepen our understanding of changing runoff patterns to facilitate collaborative efforts across borders and synthesize scientific advancements for the benefit of the region's sustainable future.

3.2 Runoff Changes in the Basin

3.2.1 River Networks Geometry in the Basin

The geometry of river networks fundamentally constrains the discharge process and thus has prominent impacts on water resource distribution. An expanded role for river networks is increasingly recognized due to more evidence that small streams process and store considerably more terrestrial materials than previously thought. However, the attempts to elucidate changes in terrestrial materials, including runoff, in a basin have been limited by modelling and observation at coarse resolutions. With Earth Observation (EO) data increasingly available, this section presents a novel imagery-based methodology to measure the geometry of river networks at finer resolutions and of more dimensions. Using the proposed methodology, the high-resolution river network geometric features including river networks, surface area, width, and depth have been delineated, which contribute to a more complete understanding of the distributions and patterns of the runoff changes across the LMRB.

3.2.1.1 River Networks

The extraction of river networks plays a pivotal role in addressing fundamental inquiries pertaining to the hydrological dynamics of a watershed's surface. This practice, deeply rooted in the field of hydrology, has traditionally been employed for the purpose of river flow modeling (David et al., 2011; Lin et al., 2018; Yamazaki et al., 2013). Nevertheless, the utilization of high-resolution drainage networks has seen a steady rise, finding applications not only in the realm of large-scale hydrological predictions but also in spatially comprehensive research endeavors. These include the evaluation of flood inundation, dam failure scenarios, and reservoir operations (Lehner & Grill, 2013; Shin et al., 2020; Yamazaki et al., 2019). Such applications hold particular significance in enhancing our comprehension of runoff patterns and water resource management within the watershed.

To extract high-resolution river networks for the LMRB that further facilitate surface water assessments, a new method is proposed, namely Remote Sensing Stream Burning (RSSB) (Wang et al., 2021). Enabled by RSSB, the basin-scale drainage networks are extracted at the highest 10-m resolution with the integration of Sentinel-2 imagery (Fig. 3.1). Compared to river networks, drainage networks provide additional flow information, such as flow direction and flow accumulation.

Fig. 3.1

Lancang-Mekong River networks extracted by RSSB (r represents the curvature radius of a meander). The river order is shown in Strahler stream order

Table 3.1 illustrates the distribution of river length and river network density for drainage networks extracted using the RSSB method. It is observed that both river length and network density exhibit a general adherence to the conventional power law pattern, as documented by Leopold and Maddock (1953). Of particular note is the high-resolution approach, which successfully delineates nine stream orders (ω = 9) in contrast to the coarse-resolution method. This observation underscores the efficacy of the newly proposed technique, as it not only enhances the accuracy of flowline representation but also substantially augments the level of detail within the network.

Table 3.1 Statistics of Lancang-Mekong River networks, including stream order ω (in Strahler order), number n_ω, mean length L_mean, total length L_total, and the river networks density D

3.2.1.2 Surface Water Area

Surface water area is one of the most perceivable indicators of water resources, offering a means to conduct quantitative assessments of human-induced modifications within a watershed, such as the linkage between river engineering and lake losses, and the coupling of water loss with long-term droughts (Pekel et al., 2016). Such applications aid in categorizing transitions in land surfaces, including conversions from land to water, water to land, the permanence of land, or the enduring presence of water, as described by Donchyts et al. (2016). These analyses provide essential support for research and evaluations related to flood inundation, land reclamation, and sea-level rise, particularly in regions of environmental and societal significance (Müller et al., 2016). Additionally, the detection of water plays a pivotal role as an initial step in numerous applications, including the mapping of land-use and land-cover (Arino et al., 2012; Chen et al., 2014), predicting waterborne epidemic disease (Smith et al., 2013), managing flood hazards, estimating water scarcity and assessing water quality (Dottori et al., 2016; Liu et al., 2016; Olmanson et al., 2016; Vanham et al., 2018).

The surface water area of the basin is highly fluctuating (Fig. 3.2). The total surface water area estimated by MuWI method based on both Landsat and Sentinel-2 data (Wang et al., 2018) is approximately between 20,000 and 30,000 km². Variations in surface water area are generally synchronized with the flood and drought cycle in the basin. For example, a devastating flood occurred in 2000 when the surface water area was high, while the 2015 drought, the most severe drought in the past three decades, coincided with a low surface water area. The frequency of the cycle appears to have decreased and stabilized in the past decade, which may imply that the regulation capacity of the increasing number of dams has come into effect.

Fig. 3.2

The total monthly surface water area in Lancang-Mekong River basin

Surface water areas are disproportionally distributed in the six countries within the transboundary basin (Fig. 3.3). Although more than one fifth (21.5%) of the basin lands are located in China, China shares an insignificant portion of the total surface water area (3–5%). In contrast, the downstream country, Cambodia, holds less basin land than China, but accounts for the most surface water area (55–60%) among the six countries.

Fig. 3.3

The yearly average surface water areas of six countries in the Lancang-Mekong River basin

3.2.1.3 River Width and Bathymetry

The river width and river bathymetry (depth) are the two fundamental geometric dimensions of the river networks. The river expands geometrically downstream due to erosion from the accumulated flow. This geometric expansion with the increasing flow often follows a power-law relationship, which is recognized as the well-known theory of hydraulic geometry (Leopold & Maddock, 1953). The river width and depth of the LMR (Fig. 3.4) follow the pattern of expansion in general where the magnitude of the major stem is considerably larger than tributaries.

Fig. 3.4

Width and depth of the Lancang-Mekong River

In particular, the LMRB is characterised by diverse fluvial geomorphology with valley-constrained regions upstream and bedrock-constrained areas downstream (Meshkova & Carling, 2012). The gradient of the upper Lancang River is approximately 2 m/km, more than ten times that of the lower Mekong River, indicating that more convergent topography exists upstream while divergent but well-defined banks are prevalent downstream (Pokhrel et al., 2018a, 2018b). Therefore, the upstream river channels are relatively narrow but deep.

3.2.2 Runoff Modelling in the Basin

3.2.2.1 Runoff Simulation with WAYS

Hydrological models are the most common tools for runoff simulation. They simplify the characterisation of real-world systems and describe the rainfall–runoff relations. Hydrological components and water storage in land surface, soil, and groundwater reservoirs are idealised in the model (Bierkens, 2015). In the basin considered here, the runoff is simulated by a sophisticated large-scale hydrological model, WAYS (Water And ecosYstem Simulator) that considers the spatial heterogeneity of the root zone during the hydrological simulation (Mao & Liu, 2019). The WAYS model is developed by the core members of the Strategic Priority Research Program of the Chinese Academy of Sciences “Climate Change and Water Resources in the Great River Regions in Southeast and South Asia” (project number XDA 20060400), and is tailored for the hydrological processes modelling in the basin.

WAYS is a process-based hydrological model, implemented in Python, which assumes water balance at the grid cell level and simulates the hydrological processes in a fully distributed way. The WAYS model works on a daily time step, and the model structure consists of five conceptual reservoirs: the snow reservoir Sw (mm) representing the surface snow storage, the interception reservoir Si (mm) expressing the water intercepted in the canopy, the root zone reservoir Sr (mm) describing the root zone water storage in the unsaturated soil, the fast response reservoir Sf (mm), and the slow response reservoir Ss (mm). Two lag functions are applied to describe the lag time from the storm to peak flow (TlagF) and the lag time of recharge from the root zone to the groundwater (TlagS). In addition to the water balance equation, each reservoir also has process functions to connect the fluxes entering or leaving the storage compartment (so-called constitutive functions). A schematic representation of how the hydrological processes are modeled in WAYS is shown in Fig. 3.5. Traditional hydrological models simulated soil hydrology with a layer-based scheme that cannot reflect the influence of the heterogeneity in the root zone, but the WAYS model assimilates the separately derived root zone storage capacity and thus is able to consider the impacts of the spatial heterogeneity of root zone in soil hydrology. More details about WAYS can be found in Mao and Liu (2019).

Fig. 3.5

Model structure of the WAYS model (Mao & Liu, 2019)

Using the newly developed WAYS model, some basic hydrological variables, such as precipitation, temperature, and specific humidity, were simulated from 1971 to 2010. The WAYS model depicts the dynamics of the hydrological variables every day at a spatial resolution of 0.5°, which allows for an in-depth understanding of the changes in the hydrological system, including the runoff changes. In order to more intuitively represent the dynamics of the simulated hydrological variables in the entire basin, the variables are averaged from a daily scale to a monthly scale, which is shown in Fig. 3.6. In addition, the spatial pattern of the hydrological variables can also be revealed based on the simulations (see Fig. 3.7).

Fig. 3.6

The observed (precipitation) and simulated (Runoff, Evaporation, and Root Zone Water Storage (RZWS)) time series of fundamental hydrological variables at a monthly scale

Fig. 3.7

The spatial pattern of the observed and simulated annual hydrological variables: a precipitation, b runoff, c root zone water storage, d evaporation

3.2.2.2 Runoff Simulation with a Multi-model Framework

In addition to the WAYS model, nine other state-of-the-art large-scale models (CLM4, DBH, H08, LPJmL, MATSIRO, MPI-HM, PRC-GLOBWB, VIC, and WaterGAP2) were applied to simulate the runoff for uncertainty assessment. Including the WAYS model, all the selected models participated in the second phase of the Inter-Sectoral Impact Model Inter-Comparison Project, which offers a framework for consistently investigating the impacts of climate change across affected sectors and spatial scales (ISIMIP2a) (Warszawski et al., 2014). All models were driven by the same climate forcing (Global Soil Wetness Project Phase 3 data) (GSWP3) using a spatial resolution of 0.5° from 1 January 1971 to 31 December 2010 on a continuous run on a daily scale. The GSWP3 dataset was generated based on the 20th Century Reanalysis Project, and has been widely used in several studies conducting hydrological simulations (Masaki et al., 2017; Tangdamrongsub et al., 2018; Veldkamp et al., 2017). The WaterGAP and WAYS models were calibrated prior to the hydrological simulation (Alcamo et al., 2003; Mao & Liu, 2019), while the other eight models were not calibrated specifically for the ISIMIP2a simulations, and their default model parameters were therefore used in the runoff simulations. All models were treated as independent, although many of them shared similar structures and parameterisations: for example, some were similar with respect to their fundamental approach to simulating evapotranspiration, representing water exchanges in soil across the basin, and modelling snow melting. The basic differences in the models with respect to simulating land-surface hydrological processes are presented in Table 3.2, and detailed descriptions of the models applied in this work are provided by references associated with each model cited in the table.

Table 3.2 Technical description of the ten evaluated global-scale hydrological models

To assess the accuracy of the hydrological models, a rigorous verification process was conducted. Monthly runoff data from the International Satellite Land Surface Climatology Project Initiative II University of New Hampshire/Global Runoff Data Centre (ISLSCP II UNH/GRDC) were employed for validation purposes. These data, available at a spatial resolution of 0.5° and spanning the period from 1986 to 1995, served as a benchmark for evaluating the performance of model simulations within the basin. The ISLSCP II UNH/GRDC dataset, often referred to as UNH-GRDC, is a composite of runoff data generated through a combination of water balance model estimates and the assimilation of observed discharge data from gauge stations. While it retains the spatial characteristics of the water balance, it is influenced and constrained by observed records from these monitoring stations (Fekete et al., 2011). Importantly, the UNH-GRDC dataset serves as a standardized reference dataset in the ISIMIP2a initiative for model validation purposes, as established by Warszawski et al. (2014).

Prior to examining changes in runoff patterns, an evaluation of the hydrological models used for runoff simulation was conducted against reference runoff data. This evaluation commenced with an analysis of the models’ performance through the simulation of monthly runoff time series. Subsequently, the models’ capabilities in replicating runoff at various return periods were assessed. Results indicated that all models were able to replicate the observed monthly runoff time series, and the seasonal runoff cycles were particularly well duplicated by the models. However, relatively large uncertainties were observed in high-value runoff simulations during summer seasons (with a wider spread among the models) in comparison with the low-value simulations (as depicted in Fig. 3.8). Although uncertainties existed in the model simulations, the multi-model ensemble mean agreed well with the reference runoff data. During the evaluation process, the performances of the models were further evaluated using a set of transferrable benchmarks. In order to overcome the problem that, generally, different metrics are only suitable for assessing individual characteristics of a simulated time series, and to enable consistent comparisons, six commonly used metrics were applied (the relative bias, normalised root mean square difference (RMSD), correlation coefficient, normalised standard deviation, centered RMSD, and the Nash–Sutcliffe coefficient of efficiency (NSE)), and some were standardised prior to conducting comparisons. These metrics were then used to assess the relative performance of each model in different aspects, and the results were presented in three types of diagrams (a target, a radar, and a Taylor diagram).

Fig. 3.8

Simulated basin-average monthly runoff time series by ISIMIP2a models (dashed lines), model ensemble mean (solid blue line), and UNH-GRDC runoff reference data (solid black line)

The ten selected models and the multi-model ensemble were evaluated to determine their ability to reproduce the observed monthly runoff time series. In addition, the model simulated monthly runoff time series and the corresponding ensemble mean were compared with reference data (UNH-GRDC runoff data) for the period from 1986 to 1995 (see Fig. 3.8). Results indicated that all models were able to replicate the observed monthly runoff time series, and the seasonal runoff cycles were particularly well duplicated by the models. However, relatively large uncertainties were observed in high-value runoff simulations during summer seasons (with a wider spread among the models) in comparison with the low-value simulations (as depicted in Fig. 3.8). Although uncertainties existed in the model simulations, the multi-model ensemble mean agreed well with the reference runoff data.

Detailed model evaluations revealed that the ensemble mean of the model was better than that of the single model in terms of monthly time series, seasonal cycles, and runoff at different return periods. Particularly, the model ensemble mean was also capable of modelling variability in the runoff time series. Accordingly, the model ensemble mean was used to analyse runoff regime changes in the basin, and then quantify the uncertainty associated with the model based on ten model simulations (Fig. 3.9).

Fig. 3.9

Diagrams showing statistics used in model evaluations: a target diagram for relative bias and normalised root mean square difference, b radar diagram showing the Nash–Sutcliffe coefficient of efficiency, c Taylor diagram showing the correlation coefficient, normalised standard deviation of errors, and centered RMSD

The comprehensive model evaluations unveiled that the model ensemble mean displayed superior performance compared to the individual models in replicating monthly time series, capturing seasonal cycles, and estimating runoff across various return periods. Notably, the model ensemble mean exhibited a remarkable capacity for modeling the variability within the runoff time series. Consequently, the analysis of runoff regime changes within the basin was carried out using the model ensemble mean. Subsequently, assessments were based on the results of the ten individual model simulations to quantify the uncertainties associated with the modeling process.

3.2.3 Historical Changes in Runoff

The changes in watershed runoff in the LMRB are firstly analysed by using hydrological simulations of the ten models. Based on five hydrological indicators, the characteristics of runoff changes within the basin from 1971 to 2010 were investigated. Mean Annual Runoff (MAR) was used to assess the overall runoff changes on a yearly scale and during the wet and dry seasons, respectively. The 95th percentile runoff (Q95) and the 5th percentile runoff (Q5) were applied to assess the high value and low value of runoff changes in the basin, respectively, and the annual 7-day maxima runoff (MAX-7) and annual 7-day minima runoff (MIN-7) were used to appraise the runoff regime changes relating to extreme events (Danneberg, 2012).

Based on the model ensemble mean, the average MAR in the LMRB was approximately 655 mm/yr for the period 1971–2010 and MAR increased by 8.0% (52.61 mm) during this period (Fig. 3.10). However, there was only a slight annual increase in MAR, at an average rate of 0.2% (1.32 mm/yr) and the trend detected was not significant. For the entire basin, different hydrological indicators showed different change ratios for the period 1971–2010. All hydrological indicators from all models demonstrated an increasing change trend for the basin, with the exception for MIN7 and Q95 indicators, which exhibited lower runoff values. However, some models demonstrated decreased trends with the median value of multiple models indicating an increasing trend. Models also showed relatively high agreements for change trend detections of MAR, MAX7, Q5, and runoff in the wet season. The highest model agreement was observed with respect to the MAR trend detection, where the smallest spread range was found among model estimates. In contrast, large uncertainties in model estimates were observed for change trend detections of low runoff values (MIN7 and Q95) and runoff in the dry season, particularly the trend in the dry season, which ranged from 7.6 to 34.9%. Overall, although uncertainties existed, the model ensemble mean based estimates indicated that runoff in the basin increased during the period 1971–2010 with respect to low values, high values, MAR, and runoff in both dry and wet seasons. The change in MAR (8.1%) exhibited an increasing magnitude, similar to the changes in MAX7 (8.5%), and Q5 (8.0%), indicating higher runoff. For the model average, low flow with respect to minimum runoff over seven consecutive days (MIN7) and runoff that exceeded 95% of the time series (Q95) exhibited the lowest increasing ratio with change values of 2.2 and 1.7%. Runoff during the dry season showed the greatest increase (17.7%) for the period 1971–2010, while runoff during the wet season increased slightly (approximately 6.2%).

Fig. 3.10

Changes in different hydrological indicators from the ten hydrological model simulations. The box-whiskers represent the 0th, 25th, 50th, 75th, and 100th percentiles of the distribution in changes for each hydrological indicator

Spatially, the trend in Mean Annual Runoff (MAR) exhibited a distinct gradient across the basin, with a pronounced increasing trend in both the upper and lower basin areas, which contrasted with the prevailing decreasing trend observed in the middle basin. Additionally, a small region within the lower basin displayed a decreasing trend (see Fig. 3.11). The trends observed in Maximum 7-Day Runoff (MAX7) and the 5th percentile runoff (Q5) displayed broadly similar patterns to those of MAR. However, when it comes to trends in low flow, specifically for Minimum 7-Day Runoff (MIN7) and the 95th percentile runoff (Q95), there were slight variations in spatial distribution compared to other hydrological indicators. Notably, more pronounced negative trends were evident in the middle and lower basin regions, albeit with relatively lower local variability.

Fig. 3.11

Spatial distribution of change trends in a MAR, b MIN7, c MAX7, d Q95, and e Q5, based on the model ensemble mean

Significant trends were observed mainly in regions that showed positive trends for annual runoff and high flow, particularly in the lower basin. In contrast, there was a significant negative trend for low flow and a less significant positive trend throughout the domain, which was particularly visible in the middle and lower basin. In addition to the differences in the significance tests, large differences were also observed in the model agreements for trend detection with respect to annual runoff, low flow, and high flow. For most of the region, the models consistently detected trends in annual runoff and high flow, aligning with the trend direction indicated by the model ensemble mean. This alignment encompassed both positive and negative trends, with all models demonstrating the same directional consistency. However, for low flow, there were more noticeable inconsistencies between the model estimates of trends. Throughout the entire domain, the models only agreed in a few pixels (mostly with respect to a negative trend), while the disagreement among models for low flow trends was widespread across the upper and lower basin.

In addition to our multi-model analysis of runoff changes, we provide a summary from the literature regarding streamflow patterns. It was observed that, during the time span from 1960 to 2010, there existed a general downward trend in annual streamflow within the basin. However, after 2010, no clear trend was detectable, although the confidence level associated with such a trend was low, as indicated by Ruiz-Barradas and Nigam (2018). Most of the studies conducted on historical streamflow in the basin reported a decreasing trend, while a minority of studies indicated the opposite—an increasing trend in streamflow. These discrepancies in findings can be attributed to variations in data sources and methodologies employed in each study, as summarized in Table 3.3.

Table 3.3 Changes in streamflow over the Lancang-Mekong River Basin and its upper (LRB) and lower (MRB) parts

Studies have indicated that the drivers behind streamflow alterations vary across different regions and time periods. Climate change emerged as a primary catalyst for changes in streamflow within the Lancang-Mekong River Basin (LMRB) before 2010, whereas human activities, particularly dam construction, became more influential after 2010. Climate change predominantly governed alterations in annual streamflow during the transitional period from 1992 to 2009, accounting for 82.3% of the changes, while human activities contributed to 61.9% of the streamflow changes in the post-impact period from 2010 to 2014, as outlined by Li et al. (2017). When considering annual streamflow and water-level variations, the hydrological response within the Lancang River Basin is observed to be more sensitive to climate factors than to human activities, especially when compared to the Mekong River Basin (Li & He, 2008). This discrepancy underscores the escalating impact of intensive human activities on hydrological processes, particularly within the Mekong River Basin in recent years (Shin et al., 2020).

3.2.4 Historical Impacts of Dams on Streamflow

Streamflow in the Mekong River has been altered by dams, both in the mainstream and tributaries (Han et al., 2019; Pokhrel et al., 2018a; Räsänen et al., 2017; Shin et al., 2020). Specifically, upstream flow regulation by dams has resulted in reduced peak flow and increased low flow, attenuating the flood pulse amplitude. Such changes in streamflow patterns at various mainstem and tributary locations within the Mekong River Basin have been investigated by numerous studies using either observed streamflow records or basin-wide hydrological modelling. For example, Li et al. (2017) examined the observed streamflow at five gauging stations for the pre-development (1960–1991), transition (1992–2009), and post-development (2010–2014) periods and found that the dam filling and operation reduced streamflow in the upper portion of the basin, but such an impact was relatively small at the Stung Treng station in the downstream. Importantly, they reported that dam operations, especially the cascade dams in the Lancang River in China, reduced wet season flow and increased dry season flow resulting in a unique seasonal variation compared to the pre-development period. Numerous other studies have conducted similar analysis suggesting that the impact of upstream dams have already been felt in terms of alterations in streamflow signatures even in the mainstream Mekong (e.g., Campbell, 2007; Cochrane et al., 2014; Han et al., 2019; Räsänen et al., 2017; Zhao et al., 2012). These studies have used different statistical techniques to detect the changes in streamflow in a particular year or during a given period and attribute the change to dam construction. For example, the changes in streamflow during 2010–2014 period have been linked primarily to the construction of large dams (i.e., the Jinghong, Xiaowan, Gongguoqiao, and Nuozhadu) in the Lancang River by assuming that filling of new reservoirs with high storage capacity directly affected downstream flows (Li et al., 2017). Such effects of Lancang cascade dams have been felt the most in the immediate downstream regions; the effects tend to decrease downstream because of larger flow accumulation from the tributaries and relatively small storage compared to the high flow volume in the far downstream.

The observation-based studies have provided crucial insights into the changes in streamflow and its seasonal signatures. However, it is challenging to attribute the recorded changes explicitly to climate variability or dam construction by using only observational data. Hydrological modelling can fill this gap by providing a framework where simulations can be conducted with and without considering dams—given the same climate conditions—such that the direct impacts of dams can be estimated by using the difference between two such simulations. However, very limited such studies have been conducted to date because of the challenges in simulating the complex and interconnected river-floodplain-reservoir processes over the entire basin. Among few such studies is that by Shin et al. (2020) that used a newly developed, high-resolution (~5 km grid) hydrodynamic model called the CaMa-Flood-Dam to explicitly simulate the effects of climate variability and dams over the entire Mekong basin. The model is based on the global hydrodynamics model CaMa-Flood (Yamazaki et al., 2014) and a new reservoir inundation and release scheme (Shin et al., 2019).

The study found that the impact of dams significantly increased after 2010 because the basin-wide reservoir storage capacity doubled from 2010 to recent years. In particular, river flows at various mainstem locations in the middle and lower reaches have been increasingly altered by dams in recent years (Fig. 3.12). This rapid increase in storage capacity came primarily from the completion of the Lancang cascade dams (Hecht et al., 2019; Pokhrel et al., 2018a). The study by Shin et al. (2020) also explicitly simulated water levels across the basin and inundation both the upstream and downstream of dams. Consistent with the changes in streamflow, the study reported a noticeable change in water levels downstream of dams, primarily after 2010 (Fig. 3.13). Their model explicitly simulated inundation dynamics in the natural rivers and floodplains as well as the upstream of dams. The model realistically captured the flood occurrence behind the major dams across the basin (Figs. 3.14 and 3.15) that depicted the influence of dam regulation at different levels on the flood besides climate change. Another study analysed the changes in streamflow due to climate change and dams by combining a hydrological model and observed discharges (Han et al., 2019); however, their model did not explicitly simulate reservoir operation. They quantified the impact of climate change and dams, reporting that during the 1987–2014 period the mean annual streamflow declined by ~ 6% compared to the 1980–1986 period. During the 1987–2007 period, only 43% of these changes were attributed to dams (~57% to climate change), but the contribution of dams rose drastically to 95% during the 2008–2014 period.

Fig. 3.12

River discharge simulated by CaMa-Flood-Dam model for three selected stations in the mainstem Mekong. Seasonal average for the periods of a 1979–2009 and b 2010–2016, and two dry years c before and d after basin-wide reservoir storage capacity doubled in 2010. NAT, DamIND-full, DamIND-low, and DamIND-opt denote simulations without dams (i.e., natural setting), considering dams with reservoirs at full level, considering dams with reservoirs at low level, and considering dams with reservoirs at the optimised regulation level, respectively. Figure modified after Shin et al. (2020)

Fig. 3.13

Same as in Fig. 3.12 but for water level

Fig. 3.14

Simulated flood occurrence at 3-arcsec (~90 m) resolution for the entire MRB (Shin et al., 2020). Labeled black boxes indicate regions for which a zoomed-in view is presented in Fig. 3.15. Red circles indicate the locations of dams simulated in the CaMa-Flood-Dam model

Fig. 3.15

Comparison of inundation dynamics simulated by CaMa-Flood-Dam model (left; 1979–2016 period) with the Global Surface Water (GWS) data (right; Pekel et al. (2016); 1984–2018 period). Results are shown as flood occurrence for the regions indicated in Fig. 3.14. Red circles indicate dam locations. Figure modified after Shin et al. (2020)

These findings suggest that the impacts of dams on streamflow were rather small until the late 2000s but have substantially increased in recent times since the completion of cascade dams in the Lancang river. Indeed, the total basin-wide active dam storage before 2010 amounts to only about 2% of the mean annual flow volume (Hecht et al., 2019), which increased rapidly after 2010 (Shin et al., 2020) and is expected to rise further to about 19% of annual mean flow volume by the mid-2020s (Hecht et al., 2019). This increase is expected to come not only from the continued dam construction in the Lancang river but also from the construction of several large dams in the lower basin including the recently completed Xayabouri Dam (Stone, 2011, 2016) and controversial Luang Prabang dam that is under construction (Fumagalli, 2020). Dam construction in the Laos and Cambodia portions of the Mekong Basin remains a highly contested issue and whether and how many of the proposed dams will be constructed in the coming decades remains highly uncertain. However, hydrological and hydrodynamic simulations clearly suggest that the fear of killing the Mekong by altering the magnitude, timing and duration of the Mekong flood pulse is a reality if many of the dams were to be built (Pokhrel et al., 2018b). If the mainstream flow were to be regulated by upstream dams, the hydrology of the Tonle Sap Lake—including the flow reversal in the Tonle Sap River—could be largely disrupted, also bringing major changes in flood dynamics in the Mekong Delta (Pokhrel et al., 2018b) and directly impacting fisheries across the Lower Mekong, especially in the Tonle Sap Lake region (Burbano et al., 2020). Some approaches have been suggested to minimise downstream impacts, especially on fisheries (Sabo et al., 2017), but the practical aspects of such engineering approaches remain unexplored (Williams, 2018).

3.2.5 Projected Changes in Streamflow

Unlike historical streamflow changes, previous studies have consistently projected an increasing trend in streamflow within the Lancang-Mekong River Basin (LMRB), regardless of the climate forcings and models employed. However, it is important to note that the flow regime in this basin is highly susceptible to various drivers, including dam construction, irrigation expansion, land-use changes, and climate change. Substantial changes are anticipated in both annual and seasonal flow patterns, with an overall increasing trend (Hecht et al., 2019; Hoang et al., 2019). Notably, hydropower development, while exerting a limited influence on total annual flow, has the most significant seasonal impact on streamflow, leading to an increase in the dry season and a decrease in the wet season, surpassing the effects of other drivers (Hoang et al., 2019). Furthermore, studies suggest that climate change may lead to a 15% increase in annual streamflow, while irrigation expansions could result in a slight decrease of 3% in annual streamflow over the period from 2036 to 2065 compared to the period from 1971 to 2000. These projections were based on statistically downscaled data from the Coupled Model Intercomparison Project Phase 5 (CMIP5) and utilized a distributed hydrological model, VMod, with a spatial resolution of 5 km × 5 km (Hoang et al., 2019). Taking future dam development into account, the change ratio in the dry season (70% increase) surpasses that in the wet season (15% decrease). In the 3S tributary, streamflow is projected to increase by 96% in the dry season and decrease by 25% in the wet season, highlighting higher streamflow sensitivity to climate change and human activities in the 3S system compared to the entire LMRB (Shrestha et al., 2016).

It is important to note that scenarios for streamflow changes exhibit spatial variability, especially within the Mekong River Basin (Liu et al., 2022). While an increasing streamflow trend is projected for the future of LMRB, uncertainties remain substantial. For instance, studies have reported varying projections, including an annual runoff increase ranging from 4 to 90% by the 2030s compared to the historical period (1951–2000), based on different global climate models (GCMs) (Eastham et al., 2008). Other studies, using CMIP5 datasets for the near future (2036–2065), have reported relatively modest changes in mean annual flow, ranging from 3 to 10% in the LMRB (Phi Hoang et al., 2016; Västilä et al., 2010).

Furthermore, projections indicate that the magnitude and frequency of extreme high-flow events are expected to increase, while low-flow events are anticipated to become less frequent, primarily due to the impacts of climate change (Phi Hoang et al., 2016). This shift could potentially heighten flood risks within the basin. However, it's worth noting that the massive construction of hydropower facilities, which has altered discharge patterns, is expected to exert a more substantial influence on hydrography in the next few decades compared to climate change (Lauri et al., 2012). Additionally, different patterns of hydrological changes may be observed in different subbasins of the basin, and the expected change ratios vary by location (Phi Hoang et al., 2016). Moreover, the number of wet days is projected to increase by the end of the twenty-first century (2080–2099), potentially increasing flood risk while benefiting water utilization during dry periods (Kiem et al., 2008).

3.2.6 Uncertainties in Streamflow Simulation

Due to the constraints of time and cost associated with large-scale and long-term field observations, hydrological models (HMs) and land surface models (LSMs) are valuable tools for simulating and managing water resources. Uncertainties in a hydrological simulation are inevitable due to the difference between the natural hydrological processes and model descriptions. Thus, uncertainties must be considered to reflect the reliability of models.

To assess uncertainties in model simulation in the Lancang-Mekong River Basin, observed discharge data from seven hydrological stations were used to evaluate ten HMs and LSMs from the Inter-Sectoral Impact Model Intercomparison Project (ISI-MIP2a). The simulated discharge data forced by Global Soil Wetness Project 3 (GSWP3) data in the ISI-MIP2a simulation round were selected. To capture the diverse aspects of hydrological regimes and their associated uncertainties, we considered simulated discharge series at various percentiles, including the 5th percentile (Q5), 25th percentile (Q25), 50th percentile (Q50), 75th percentile (Q75), and 95th percentile (Q95). These percentiles provide insights into extremely low discharge (Q5), the median discharge (Q50), high flow conditions (Q95), and additional discharge information in the form of Q25 and Q75, contributing to a comprehensive evaluation of the uncertainties inherent in different hydrological scenarios (Fig. 3.16; Table 3.4).

Fig. 3.16

Location of the seven hydrological stations

Table 3.4 General information of the ten evaluated Global Hydrological Models (GHMs) and Land Surface Models (LSMs)

For Q5 (Fig. 3.17), large deviations occurred between the simulated and observed discharge series. Discharge curves simulated by different models were more divergent than that of high flow (Fig. 3.18). The model ensemble discharge and the observed discharge displayed high consistency at most stations for all percentiles. Systematic errors occurred at Q5 for CLM4, H08 and LPJmL, where these models simulated a much smaller discharge than the observed and other models. As for high discharge percentiles, the simulated curves were more concentrated, which indicated more realistic simulations and smaller uncertainties.

Fig. 3.17

Comparison of the annual discharge series of observed and simulated discharges at the 5th percentile. a Chiang Saen, b Luang Prabang, c Nong Khai, d Mukdahan, e Pakse, f Stung Treng, and g Kratie

Fig. 3.18

Comparison of the annual discharge series of observed and simulated discharges at the 95th percentile. a Chiang Saen, b Luang Prabang, c Nong Khai, d Mukdahan, e Pakse, f Stung Treng, and g Kratie

Dispersion of the simulated discharge series reflects the uncertainties in discharge simulations among different models. The large deviations between the selected models indicated that uncertainties in discharge simulation for lower percentiles were much greater than that for higher percentiles.

The analysis of statistical metrics consistently revealed a pattern of decreasing model uncertainty as we moved from lower percentiles to higher percentiles (Fig. 3.19). Furthermore, all the models exhibited significant correlations with the observed discharge series, with most models achieving an R-squared (R2) value greater than 0.60 for all stations. Notably, several models surpassed an R2 value of 0.80 for stations located downstream of the river, including WaterGAP2, MPI-HM, H08, MATRISO, and WAYS (Table 3.5). These results signify that the simulated discharge series produced by all the models satisfactorily replicate the observed series. In contrast to the single model series, the model ensemble series consistently outperformed at all stations. Generally, R2 values tended to increase as we moved closer to the river's estuary but exhibited a decline for stations in proximity to the estuary, such as Stung Treng and Kratie. Figure 3.19b demonstrated that the majority of Nash–Sutcliffe Efficiency (NSE) values exceeded 0.40, indicating that the model simulations could be considered reliable. Similar to R2, the model ensemble displayed higher NSE values than the individual models at most stations. WaterGAP2 emerged as the top-performing model across all stations based on NSE and even outperformed the model ensemble at stations in Luang Prabang, Pakse, and Kratie. Additionally, Δμ represented negative deviations at Chiang Saen and Luang Prabang stations, while positive deviations were observed at Nang Khai and Kratie stations for most of the models. Δσ indicated deviations from the standard deviation between the simulated discharge series and the observed data. Notably, H08 and ORCHIDEE exhibited significantly different Δσ values compared to other models. H08 displayed larger Δσ values than the other models at all stations, while ORCHIDEE demonstrated the opposite performance. The DBH model exhibited a substantial positive deviation in Δμ but performed well in Δσ.

Fig. 3.19

Comparison of model performances with different metrics at different percentiles. a R2, b NSE, c Δμ, and d Δσ. Details of the definition of metrics could be found in Chen et al. (2021a, 2021b)

Table 3.5 Model performances at seven hydrological stations based on the Nash–Sutcliffe efficiency coefficient (NSE), correlation coefficient (R2), mean deviation in mean (Δμ), and mean deviation in standard deviation (Δσ). Details of the rank system could be found in Chen et al. (2021a, 2021b)

In terms of model performance rankings based on the scoring system, WaterGAP2 secured the top position, followed by WAYS, PCR-GLOWBW, MPI-HM, and MATRISO, which ranked 2nd, 3rd, 4th, and 5th, respectively. On the other hand, ORCHIDEE received the lowest ranking, primarily due to its poor performance in Δμ. The CLM4 model exhibited less favorable performance, particularly in terms of Nash–Sutcliffe Efficiency (NSE), with values of 0.18 at Chiang Saen and 0.24 at Kratie. Additionally, the CLM4 model displayed negative deviations for Δμ at Chiang Saen (−0.46) and Luang Prabang (−0.39). These results indicated that the simulated discharge series for the CLM4 model diverged significantly from the outcomes of other models. As we moved closer to the estuary, both NSE and R2 values for most models approached 1, indicating improved model performance. However, there was a decline in these values at the Kratie station. Furthermore, negative Δμ values were observed for most models at Chiang Saen, Luang Prabang, and Pakse, suggesting that these models consistently underestimated the magnitude of discharge series at these stations. Nevertheless, as the stations moved closer to the estuary, models with negative Δμ values decreased in number, and only two models (MATRISO and ORCHIDEE) displayed negative values for Δμ at the Kratie station. This shift underscores the enhancement in model performances as we moved closer to the estuary.

The models had poor performances for low discharge percentiles, although the simulated performances improved as discharge percentiles increased. The model performances in terms of discharge simulations generally improved with the distance to the estuary for all discharge percentiles. For the Lancang-Mekong River Basin, the discharge increases from upstream to downstream, which can partly explain the better performance downstream. However, the models had difficulties in simulating discharge for the river sections close to the estuary. The complex processes between the freshwater and saline water bodies may be the cause of this difficulty. The results suggest that current models have limits in extreme hydrological event simulations, which is vital for water resources management. It also indicates that current models are limited in extreme hydrology event prediction, which usually brings huge losses to the economy and society.

3.3 Baseflow Estimation and Change in the Basin

Streamflow in a river consists of two components, namely baseflow and stormflow. Baseflow refers to the component of streamflow originating from groundwater storage and other delayed sources (Hall, 1968). It represents the flow within a stream that would persist even in the absence of direct runoff resulting from rainfall. As a result, baseflow is an important source of water for a river, especially in dry seasons. Baseflow estimation has been achieved through isotopic and chemical tracer methods (Genereux, 1998; Muñoz-Villers et al., 2016). However, these tracer methods are often costly and labor-intensive when applied in field measurements (Lott & Stewart, 2016). To address these challenges, various mathematical methods have been developed for baseflow estimation that do not require the use of tracers. These methods include graphical approaches (Institute of Hydrology, 1980; Sloto, 1996) and digital filter methods (Anand Tularam & Ilahee, 2008; Chapman, 1991; Eckhardt, 2005; Furey & Gupta, 2001; Huyck et al., 2005; Lin et al., 2007; Maxwell, 1996). These techniques provide alternative means of estimating baseflow efficiently and cost-effectively (Fig. 3.20).

Fig. 3.20

Schematic diagram of baseflow. Total streamflow in a river consists of two components, namely baseflow and stormflow

In this chapter, the baseflow of two typical hydrologic stations in the Mekong River Basin, namely Yongjinghong and Kratie, were estimated and projected using mathematical methods. The Yongjinghong Station is located in the Upper Mekong River Basin, and the Kratie Station is located in the Lower Lancang-Mekong River Basin.

3.3.1 Comparison of Baseflow Estimation Methods

3.3.1.1 Baseflow Evaluation Criterion

Because of the lack of baseflow observation data, it is difficult to evaluate the accuracy of different baseflow separation methods through the observed baseflow. In this chapter, a robust mathematical evaluation method is employed to evaluate the accuracy of different separation methods. The main guideline is as follows. When the quick flow (interflow and overland flow) of a basin ceases on a certain day, the streamflow is completely replenished by the baseflow, and the streamflow of that day is equal to the baseflow. The daily streamflows of these days can then be used as the baseflow benchmark to assess different baseflow separation methods. According to Brutsaert (2008), days when streamflow is completely replenished by baseflow (hereafter baseflow days) can be selected through the following four steps:

• (1) Exclude days with streamflow $\frac{dy}{dt}\ge 0$, where $\frac{d{y}_{i}}{dt}=\frac{{y}_{i+1}-{y}_{i-1}}{2}$.

• (2) Exclude two days before and three days after the day with streamflow $\frac{dy}{dt}\ge 0$.

• (3) Exclude five days after high flow events that were identified by flood peaks greater than the 90th quantile of all daily streamflow observations (Cheng et al., 2016).

• (4) Exclude days followed by a day with smaller $\frac{dy}{dt}$, namely $\frac{{d}^{2}y}{d{x}^{2}}<0$.

These four steps have two purposes. The first three steps are to exclude the days when the streamflow may contain quick flow. The last step is to exclude the days when daily streamflow violates the pattern of baseflow recession during dry periods, namely followed by a larger $-\frac{dy}{dt}$ (Xie et al., 2020). Baseflow days selected by the four steps are shown in Fig. 3.21.

Fig. 3.21

Selecting baseflow days according to the four steps. The red points are excluded through the first three steps. The blue points are excluded through the last step. The daily streamflow observations are from the Yongjinghong station in 1980

The baseflow days (Black points in Fig. 3.21) were used as the baseflow benchmark to evaluate the accuracy of different baseflow separation methods, based on the evaluation metrics of Kling-Gupta Efficiency (KGE) (Knoben et al., 2019) and relative bias (BIAS):

$$KGE=1-\sqrt{(r-1{)}^{2}+\left(\frac{{\sigma }_{m}}{{\sigma }_{o}}-1\right)^{2}+\left(\frac{\underline{{Q}_{m}}}{\underline{{Q}_{0}}}-1 \right)^{2}}$$

(3.1)

$$BIAS=\frac{{\sum }_{1}^{n}({Q}_{oi}-{Q}_{mi})}{{\sum }_{1}^{n}{Q}_{oi}}$$

(3.2)

where r is Pearson’s correlation between the selected baseflow and the corresponding estimated baseflow. ${\sigma }_{o}$ is the standard deviation of the selected baseflow, and ${\sigma }_{m}$ is the standard deviation of the corresponding estimated baseflow. $\underline{{Q}_{0}}$ is the mean value of the selected baseflow, and $\underline{{Q}_{m}}$ is the mean value of the corresponding estimated baseflow. Q_oi and Q_mi are the selected baseflow and the corresponding estimated baseflow on the ith day, respectively.

3.3.1.2 Baseflow Separation Methods

In this chapter, 9 baseflow separation methods were evaluated, including 5 digital filter methods, namely the Chapman method, the LH method, the Eckhardt method, the EWMA method and the CM method, and 4 graphic methods, namely the UKIH method and three HYSEP methods. The digital filter methods are grounded on the assumption that baseflow constitutes the low-frequency component of streamflow, which exhibits a slow response to precipitation events, while quick flow represents the high-frequency component, reacting rapidly to precipitation. In contrast, the graphic methods identify specific low-flow points within a streamflow hydrograph, connect these points to form a continuous baseflow line, and subsequently constrain this baseflow line beneath the streamflow hydrograph to derive the baseflow hydrograph. For a more comprehensive understanding of these methods, their principles and specific details are presented in Table 3.6.

Table 3.6 The principles and details of 9 baseflow separation methods

3.3.1.3 Comparisons of Baseflow Separation Methods

Reservoir construction has significantly affected streamflow observations in the Lancang-Mekong River Basin since 2008. Thus, 2007 was selected as the last year of the baseflow separation in this study. Daily streamflow observations for the two hydrologic stations, namely Yongjinghong and Kratie, from 1980 to 2007 were obtained from the Mekong River Commission (https://portal.mrcmekong.org/home).

To evaluate the accuracy of the 9 baseflow separation methods, the baseflow points obtained through the four steps in Sect. 3.3.1.1 and the baseflow estimated using the 9 methods introduced in Sect. 3.3.1.2 were compared in the two hydrologic stations. Table 3.7 shows the evaluation result of the 9 baseflow separation methods for the two hydrologic stations. For Yongjinghong Station, the Eckhardt method has the largest value of KGE and the smallest value of BIAS among the 9 methods, with values of 0.86 and 5.98% respectively. For Kratie Station, the Eckhardt method also has the largest value of KGE and the smallest value of BIAS among the 9 methods, with values of 0.93 and 5.81% respectively. Generally, the Eckhardt method has the best performance to estimate baseflow for the two hydrologic stations. The good performance indicates that it is reliable to use the Eckhardt method in estimating baseflow for the two hydrologic stations in the LMRB.

Table 3.7 The evaluation result of the 9 baseflow separation methods

3.3.2 Baseflow Estimation in the Basin

Using the Eckhardt method, the baseflows of the two hydrologic stations from 1980 to 2007 were estimated. From 1980 to 2007, the annual average runoff of the two hydrologic stations, namely Yongjinghong and Kratie, was 388 mm and 649 mm, respectively. The annual average baseflow of the two hydrologic stations was 199 mm and 359 mm, respectively. The annual average BaseFlow Index (BFI), namely the ratio of baseflow to streamflow, of the two hydrologic stations was 0.51 and 0.55, respectively (Table 3.8). The annual BFI of the Yongjinghong Station from 1980 to 2007 showed a significant (p < 0.05) downward trend, with a value of −0.001 yr⁻¹, while the annual BFI of the Kratie Station showed a significant (p < 0.05) upward trend, with a value of 0.09 yr⁻¹ (Fig. 3.22). The annual baseflow of the Yongjinghong Station from 1980 to 2007 showed a nonsignificant (p = 0.22) downward trend, with a value of −0.77 mm/yr, while the annual baseflow of the Kratie Station showed a significant (p < 0.05) upward trend, with a value of 7.22 mm/yr.

Table 3.8 Annual average runoff, baseflow and BFI for the two hydrological stations from 1980 to 2007

Fig. 3.22

The annual baseflow and BFI (a, b), and average monthly runoff, baseflow and BFI (c, d) from 1980 to 2007 for the two hydrological stations Yongjinghong (a, c) and Kratie (b, d)

From 1980 to 2007, the maximum and minimum average monthly BFI of Yunjinghong Station were in December (0.75) and June (0.34), respectively (Fig. 3.22). The maximum and minimum average monthly baseflow of Yunjinghong Station were in September (34 mm) and April (6 mm), respectively. The maximum and minimum average monthly BFI of Kratie Station were in December (0.62) and June (0.46), respectively. The maximum and minimum average monthly baseflow of Kratie Station was in September (80 mm) and February (7 mm), respectively.

3.3.3 Influencing Factors of Baseflow [Xiaomang Liu]

For a basin, climatic factors have the most direct impact on the baseflow (Brutsaert, 2005). Climatic factors influence baseflow by altering rates of evapotranspiration, infiltration and recharge, and timing of snowmelt runoff (Tague & Grant, 2009; Winograd et al., 1998). Additionally, baseflow is also influenced by different basin characteristics, including climate conditions, soils, topography, and land cover.

Figure 3.23 shows scatterplots of monthly baseflow versus four climatic factors, namely precipitation (Pr), surface shortwave radiation (SSR), wind speed (u), and air temperature (Ta), for the two hydrologic stations, namely Yongjinghong and Kratie. For Yongjinghong Station, significant (p < 0.05) positive correlations were found between baseflow and the two climate factors, namely Pr and Ta, and the values of Pearson's correlation (r) were 0.56 and 0.50, respectively. Significant (p < 0.05) negative correlations were found between baseflow and the other two factors, namely SSR and u, and the values of r were −0.24 and −0.60, respectively. For Kratie Station, both Pr and Ta were significantly (p < 0.05) positively correlated with baseflow, and the values of r were 0.67 and 0.39, respectively. Both SSR and u were significantly (p < 0.05) negatively correlated with baseflow, and the values of r were −0.47 and −0.46, respectively. Thus, baseflow is significantly affected by the four climatic factors in the Lancang-Mekong River Basin.

Fig. 3.23

The relationships between the monthly baseflow and the four climatic factors for Yongjinghong Station (a–d) and Kratie (e–h). r value is the Pearson correlation coefficient. The shaded bands represent 95% confidence intervals for the regressions

3.3.4 Projected Change in Baseflow

Although the Eckhardt method can accurately estimate baseflow, the method needs daily streamflow as input, while accurate daily streamflow estimates are not available for the future. Therefore, models are needed to simulate the future baseflow. Past studies have shown that mechanism models, such as hydrological models and land surface models, have low accuracy in simulating baseflow (Bai et al., 2016). This is due to the groundwater simulation of the hydrologic model and the land surface model being relatively simple (Lo & Famiglietti, 2010). In this study, a machine learning approach, namely the Long Short-Term Memory network, was used to estimate the future baseflow based on data from the Coupled Model Intercomparison Project phases 6 (CMIP6).

3.3.4.1 The Long Short-Term Memory (LSTM) Network

The LSTM network is a state-of-art machine learning approach for time series forecasting (Hochreiter & Schmidhuber, 1997). It has the advantage of remembering information for a long period, namely long-time memory (Kratzert et al., 2018; Shen, 2018; Zhang et al., 2018). This advantage benefits monthly baseflow estimation and prediction by learning long-term dependencies between baseflow and previous basin conditions.

The Long Short-Term Memory (LSTM) network is structured as a collection of interconnected memory blocks, with each block consisting of several key components: a cell state, input gate, output gate, and forget gate, along with the hidden state. The cell state functions as the system's memory, retaining crucial information. The three gates, namely the input gate, output gate, and forget gate, enable the network to selectively store and retrieve important information from past time steps while discarding irrelevant data (Kratzert et al., 2018). The details of the LSTM network can refer to Hochreiter and Schmidhuber (1997) and Kratzert et al. (2019). In this study, the LSTM network was constructed using the deep learning toolbox available in MATLAB. Four related climatic factors, namely monthly precipitation (Pr), air temperature (Ta), surface shortwave radiation (SSR), and wind speed (u), were used to estimate and predict monthly baseflow based on Sect. 3.3.3.

3.3.4.2 Data and Method

Historical data of the four variables from 1980 to 2014 were obtained from Princeton Global Meteorological Forcing Dataset (Sheffield et al., 2006; Zhang et al., 2018), with a spatial resolution of 0.5°. Future data on these variables from 2015 to 2100 were obtained from 26 general circulation models (GCMs) in the CMIP6 (Table 3.9). Simulations from four shared socioeconomic pathways (SSPs), drawn from Tier 1 of ScenarioMIP: SSP1-2.6 (+2.6 W/m² imbalance; low forcing sustainability pathway), SSP2-4.5 (+4.5 W/m²; medium forcing middle-of-the-road pathway), SSP3-7.0 (+7.0 W/m²; medium- to high-end forcing pathway), and SSP5-8.5 (+8.5 W/m²; high-end forcing pathway), were used (O’Neill et al., 2016). The bilinear interpolation method was used to downscale all the variables to a common horizontal grid at 0.5° × 0.5° resolution. To proceed with further analysis with reduced biases, the perturbation method was used to perform bias correction against observed data.

Table 3.9 The information of the 26 GCMs in the CMIP6

The classic split sample test scheme (KlemeŠ, 1986) was used for calibration and validation of the LSTM. The available data in the basin was split into two sub-periods, namely sub-period I and sub-period II, which were used to calibrate and validate the LSTM, respectively. The LSTM was calibrated in sub-period I (1980–1999) and validated in sub-period II (2000–2007). The Nash–Sutcliffe efficiency (NSE) between simulated and observed baseflow was taken as the objective function to train the LSTM. The KGE and BIAS were used to evaluate the accuracy of model estimation.

3.3.4.3 Estimating and Predicting Baseflow with the LSTM Model

Generally, the LSTM model performed well in the calibration and validation periods for the two hydrologic stations, namely Yongjinghong and Kratie (Table 3.10). In the calibration period, the KGE values between the observed and simulated baseflow for the two hydrologic stations are 0.90 and 0.92, respectively. The BIAS values between observed and simulated baseflow for the two hydrologic stations are 1.3% and 1.6%, respectively. In the validation period, the median KGE values for the two stations are 0.87 and 0.75 respectively, and the BIAS values are 11.0% and 16.8% respectively. Thus, the trained LSTM model was used to estimate the future monthly baseflow.

Table 3.10 Performance of the LSTM model for the two hydrological stations

The trained LSTM model was used to estimate and predict the monthly baseflow from 1980 to 2100 for the two hydrologic stations. Figure 3.24 shows the time series of annual baseflow and BFI for the two hydrological stations from 1980 to 2100. Annual baseflows for the two hydrological stations in the four scenarios, namely SSP1-26, SSP2-45, SSP3-70, and SSP5-85, all have increasing trends, and the BFI in the four scenarios all have a slightly increasing trend. Table 3.11 shows the average annual baseflow and BFI for the two hydrological stations from 2015 to 2100. It could be found from Table 3.11 that the volume of baseflow from the Yongjinghong station upstream is much lower than that of Kratie station downstream, while the BFI is just slightly lower. And with the intensification of climate change and human activities, the baseflow at both the upstream and downstream increases and that at the upstream increases faster than that of the downstream, but the BFIs keep consistent, implying that the total streamflow doesnot have a similar increasing trend.

Fig. 3.24

Annual baseflow and BFI from 1980 to 2100 for the Yongjinghong station (a, b) and Kratie station (c, d) for the four scenarios. The shading denotes the 95% confidence intervals of the 26 models

Table 3.11 Annual average baseflow and BFI for the two hydrological stations from 2015 to 2100 in the four scenarios

3.4 Dynamics of Inundation Area and Water Turbidity in Tonle Sap Lake

3.4.1 Inundation Area Detection

Tonle Sap Lake (TSL) in Cambodia stands as the largest lake in Southeast Asia, playing a pivotal role as one of the world's most productive lake-wetland systems. This remarkable ecosystem supports approximately 1.7 million people who depend on it for their livelihoods. What sets TSL apart is its distinctive “flood pulse” phenomenon, marked by seasonal water level fluctuations between the wet and dry seasons, creating a periodically inundated floodplain. This dynamic floodplain offers unique habitats for seasonally migratory fish species and receives a vital influx of nutrients from the Mekong River. It serves as a critical source of freshwater resources and preserves essential habitats for numerous endangered species. Furthermore, the flood regime of TSL exerts a significant influence on land cover changes, such as delineating the extent of cropland in the floodplain and impacting alterations in forest cover. Consequently, Tonle Sap Lake holds the status of being the “heart of the lower Mekong” as regional socio-economic development and the sustainability of the ecosystem profoundly rely on the intricate dynamics of this “flood pulse.”

The boundary of Tonle Sap Lake was firstly defined before the inundation area extraction, which is the buffered extent that was larger than the maximum possible inundation area of the open water body of the lake (Lin & Qi, 2017). This definition is different from previous studies that also considered the entire floodplain as Tonle Sap Lake (Arias et al., 2012; Frappart et al., 2018; Sakamoto et al., 2007), and the currently used boundary excluded most of the area in the floodplain of Tonle Sap Lake. Such exclusion is because of the difficulty in estimating the surface area of the entire floodplain when using optical remote sensing data, where the water is hidden beneath the flooded forest. The inundation areas were extracted based on a normalised difference vegetation index (NDVI) (Mcfeeters, 1996; Verhoef, 1996). Note that the NDVI thresholds to separate water and land may differ among images. To overcome this challenge, a self-developed interactive graphical user interface (GUI) by Hou et al. (2018) was used to determine the optimal image-specific threshold. We further visualised the resulting land/water boundaries to assure the best consistency with the largest contrasts over NDVI images.

3.4.2 Modelling Inundation Areas and Their Change

Inundation in the Tonle Sap Lake region is governed by the (1) reversed flow in the Tonle Sap river, (2) inflow from the lake tributaries, and (3) direct rainfall on the lake system. The lake’s floodplains extend into 12,000–15,000 km² area during the wet season, storing 50–80 km³ of water, which shrinks to ~2,400 km² during the dry season with a water storage of 1.5–3.0 km³. The lake water levels during these wet-dry transitions vary between ~1.4 to ~9.0 m (Arias et al., 2012; Chen et al., 2021a, 2021b; Frappart et al., 2018; Kummu & Sarkkula, 2008; Kummu et al., 2014; Pokhrel et al., 2018a). Besides the permanently flooded lake portion, substantial areas in its periphery are flooded seasonally with varied flood occurrence during average, wet, and dry years (Fig. 25X; Dang et al., 2022). The dry–wet variation in flooded areas within a year serves as an important detention reservoir to provide increased dry season flow in the Mekong Delta region. The flooded areas vary vastly not only seasonally but also from year to year depending on regional climate variability and the water levels in the mainstream Mekong River that drive the flow reversal in the Tonle Sap River. On an average basis over long terms, ~54% of the inflow to the Tonle Sap Lake comes from the Mekong River either through flow reversal in the Tonle Sap River or by overland flooding, and the rest is contributed by inflow from the tributaries (~34%) and precipitation over the lake (~12%) (Kummu et al., 2014) (Fig. 3.25).

Fig. 3.25

Average flood occurrence (% time during a year) of the Tonle Sap Lake area in a dry year (a, 2015), long-term average (b, between 1979 and 2016) and a wet year (c, 2000)

Numerous studies have examined how the inundated areas in the Tonle Sap Lake floodplain have been changing in the past few decades by using hydrological-hydrodynamic modelling and remote sensing data. For example, Lin and Qi (2017) mapped the open water areas in the Tonle Sap Lake from 2001 to 2015 using remote sensing products and showed large inter-annual variability, also noting a consistent decline in open water areas during that period. They attributed such shrinking of the lake to the rapid increase in dam construction in the Mekong River Basin during the same period, but their study did not explicitly isolate the effects of dams versus climate change and variability. Another recent study (Frappart et al., 2018) used remotely sensed data to map inundation extents during the 1993–2017 period, finding that interannual anomalies of the lake surface water storage variations are more related to precipitation fluctuation outside of the Tonle Sap watershed with discharge from the Mekong River being the major influence. The study by Ji et al. (2018) used the Modified Normalised Difference Water Index (MNDWI) based on MODIS satellite data for 2000–2014 period. They suggested a decline in water surface area, especially after 2008, by 8.3% and 1.5% during the flood and dry seasons, respectively. This study also indicated a more dominant role of rainfall in the Mekong River Basin than that of the rainfall in the lake watershed on the variation of water areas in the lake, but also noted that the construction and operation of new dams in the Lancang river could not be directly linked to the decline in the lake area. Instead, they indicated that the increased runoff due to dam release during the dry season could have mitigated the decline in surface area during the dry season. These findings are in line with a potential increase in dry-season flow and water levels when the mainstream Mekong flow is regulated by upstream dams (Pokhrel et al., 2018b).

Chen et al. (2021a, 2021b) conducted a study revealing notable declines in water levels and inundation areas during the dry season and throughout the entire year since the late 1990s. These declines occurred alongside increased sub-decadal variability in the region. The study also identified decreasing probabilities of encountering high inundation areas and increasing probabilities of encountering low inundation areas for the period from 2000 to 2019 when compared to the return period of inundation areas for the years 1986 to 2000 and 1960 to 1986. Furthermore, the research unveiled a shift in the mean seasonal cycle of daily water levels, with a 10-day shift in the dry season and a 5-day shift in the wet season between the periods 2000–2019 and 1986–2000. The study also established significant correlations and changes in coherence between water levels and large-scale atmospheric circulations, including El Niño-Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), and Indian Ocean Dipole (IOD). These findings suggest that atmospheric circulations exerted influences on the flood pulse at various time scales. Additionally, changes in discharge at the Mekong mainstream were observed, indicating that anthropogenic factors may have played a role in impacting the high water levels in the lake. In summary, the study points to a diminishing flood pulse in the Tonle Sap Lake region since the late 1990s. These previous studies assume that water infrastructure development and climate change are the main factors affecting the inundation extent and duration in the Tonle Sap Lake region. However, a recent study by Ng and Park (2021) that used remote sensing products for 1980–2018 period highlighted the role of intensified local sand mining at Phnom Penh and Prek Kdam, which could have lowered the riverbed at the entrance from the Mekong mainstream to the lake and significantly impacted lake inundation dynamic. While these studies have provided crucial information on the changing inundation dynamics of the Tonle Sap Lake, the results suffer from uncertainties arising from missing data, cloud contamination, effect of vegetation, and inherent uncertainties in satellite products.

Hydrological modelling can fill the data gaps by providing spatially complete and temporally continuous simulations; however, realistic simulations require accurate input data and model parameters, which are not abundantly available for the Tonle Sap Lake region. Numerous modelling studies have been conducted for the Tonle Sap Lake. Kummu et al. (2014) presented a detailed modelling and water balance analysis of the Tonle Sap Lake system using an integrated framework that employed a digital bathymetry model, water level-area-volume relationship and the EIA 3D hydrodynamic model (Kummu et al., 2006). They provided a detailed water balance of the lake, including inundated areas, timing of flow reversal in the Tonle Sap River, and various other related hydrodynamic attributes of the lake. They also suggested that the lake water level is primarily governed by water levels in the Mekong River, and noted that a relatively small change in water level would inundate large areas of the floodplain.

3.4.3 Water Turbidity Estimation

To investigate the potential impacts of lake inundation changes on water turbidity, the concentrations of total suspended sediments (TSS) were quantitatively retrieved using remote sensing images. Various methods have been developed previously to estimate the TSS concentrations of inland and coastal waters using satellite observations, with the algorithms ranging from empirical (Doxaran et al., 2002; Feng et al., 2012; Hou et al., 2018; Nechad et al., 2010) to semianalytical approaches (Dekker et al., 2001). The underlined theory of these algorithms is the sensitivity of the TSS concentration to red and NIR reflectance (Feng et al., 2012; Tassan, 1994), where the signals increase with TSS increases, owing to the enhanced backscattering of suspended particles (Babin et al., 2003). In this study, a red band-based algorithm, previously used in both Tonle Sap Lake (Hoshikawa et al., 2019) and various other global waters (Miller and McKee 2004), was established in this work by using concurrent MODIS reflectance and in situ TSS concentrations, expressed as follows:

$$ {\text{TSS }}\left( {{\text{mg L}}^{{ - {1}}} } \right) \, = { 1}0.{\text{32 e}}^{{{21}.{72} \times \Omega }} \left( {\Omega = {\text{ R}}_{{{645}}} } \right) $$

(3.3)

where R₆₄₅ is the MODIS surface reflectance product in the 645 nm band, which has been proven to be effective in TSS estimation in lacustrine waters (Feng et al., 2018). Indeed, the feasibility of this algorithm can be indicated by the high correlation (R² = 0.84), small root mean square error (34.2%) and large TSS range (8.6–398.0 mg L⁻¹) (see Fig. 3.26).

Fig. 3.26

Calibration of a remote sensing model to retrieve TSS concentrations in Tonle Sap Lake using MODIS red band (Wang et al., 2020)

3.4.4 Long-Term Evolution of Inundation Area and Water Turbidity

Figure 3.27a displays the monthly mean inundation areas of Tonle Sap Lake for the period between 1988 and 2018, determined from both Landsat (red) and MODIS (black) data. Additionally, the monthly mean climatological inundation areas, which represent multiyear monthly means estimated using MODIS data, are plotted as green dashed lines. Points falling above the green line indicate that the current month's inundation value exceeded the monthly climatology, and vice versa. To analyze inundation trends over the past three decades, monthly anomalies were calculated as deviations from the monthly climatologies (in percent), as shown in Fig. 3.27b. Throughout the observed period, the inundation area of Tonle Sap Lake exhibited considerable variability, ranging from 3599.8 km² in October 2001 to 2304 km² in March 2013. These values experienced rapid fluctuations due to pronounced seasonal changes influenced primarily by shifts in regional precipitation and interactions between the river and the lake (Frappart et al., 2018).

Fig. 3.27

Monthly mean inundation area of Tonle Sap Lake between 1988 and 2018 obtained using the MODIS and Landsat observations (Wang et al., 2020)

However, superimposed on these substantial seasonal cycles is a noticeable trend of lake shrinkage in recent years. There is clear evidence of decreased inundation in most years over the past two decades (see Fig. 3.27a, b). Specifically, the decreasing patterns from 2000 to 2018 were consistent between MODIS and Landsat observations, despite differences in their data availabilities. This shrinking trend is further underscored by the consistent declines in the annual mean (8.22 km² per year, significant at P < 0.05), annual minimum (5.93 km² per year, significant at P < 0.05), and annual maximum (17.82 km² per year, significant at P < 0.05) inundation areas, as derived from MODIS-extracted data for the period between 2000 and 2018 (see Fig. 3.28). Furthermore, although statistically insignificant, a decreasing trend was also observed in the annual maximum and minimum ratio (P > 0.05), indicating a reduced strength of the flood pulse between the dry and wet seasons during the MODIS observational period (see Fig. 3.28d).

Fig. 3.28

MODIS observed annual a maximum, b minimum, c mean and d maximum/minimum inundation areas of Tonle Sap Lake (Wang et al., 2020)

Figures 3.29 and 3.30 show the annual mean TSS maps and zonal mean values of Tonle Sap Lake from 2000 to 2018. The annual mean TSS concentration of the entire lake showed a statistically significant increasing trend between 2000 and 2018 (see Fig. 3.30a, 7.92 mg L⁻¹ yr⁻¹, P < 0.05). In terms of seasonal patterns, significant TSS increasing trends were detected in quarters 1 and 4, whereas nonsignificant trends were identified in quarters 2 and 3 (Fig. 3.30b–e). Moreover, remarkable spatial heterogeneity was revealed in the TSS concentration maps. In particular, in most of the years, riverine estuaries in the southeastern, northwestern, and northern parts of the lake showed consistently higher values (sediment plume). The significant seasonal TSS dynamics can partially explain the spatial heterogeneities of the annual TSS maps. The zonal mean TSS concentration of the entire lake was generally <100 mg L⁻¹ (bluish to greenish) before 2004, and such values reached above 100 mg L⁻¹ in most of the later years. Spatially, TSS increase could be found in almost every location of the lake (see the last panel of Fig. 3.29).

Fig. 3.29

Annual mean TSS concentration maps of Tonle Sap Lake from 2000 to 2018. The last panel shows the change rate for the annual mean TSS concentration at each location in Tonle Sap Lake between 2000 and 2018 (Wang et al., 2020)

Fig. 3.30

Long-term mean values of TSS concentration in the Tonle Sap Lake. a annual mean, b–e quarterly mean from quarter 1–4 (Wang et al., 2020)

3.4.5 Drivers of Change in Inundation Area and Water Turbidity

Figure 3.31a reveals the identification of a high correlation zone (HCZ) marked with black dots, situated in the northern region of the Tonle Sap Lake drainage basin, depicted by yellowish to reddish coloring, encompassing more than 50% of the entire Mekong River Basin. Analyzing the change rate in annual mean precipitation between 2000 and 2016 (as shown in Fig. 3.31b), it becomes evident that approximately one-third of locations within the HCZ exhibit a statistically significant decreasing trend in precipitation. In contrast, most regions outside the HCZ display insignificant correlations. Notably, the year-to-year fluctuations in mean precipitation within the HCZ closely mirror those observed in the inundation area of Tonle Sap Lake, with a robust correlation (R2 = 0.67, significant at P < 0.05) (Fig. 3.31d). Similarly, a strong correlation exists between the annual mean runoff at Kratie station and the precipitation within the HCZ (R2 = 0.68, significant at P < 0.05) (Fig. 3.31e). Consequently, it can be deduced that the recent reduction in Tonle Sap Lake's inundation area is intimately linked with the decline in runoff in the Mekong River and the decrease in precipitation within the HCZ.

Fig. 3.31

Correlations (R2) between the annual mean precipitation at each location in the Mekong River Basin and the annual mean inundation area of Tonle Sap Lake between 2000 and 2016. b Annual mean precipitation change rate from 2000 to 2016 for each location in the Mekong River Basin; c same as b but with a period from 1988 to 2000. The black dots in (a–c) represent pixels with statistically significant (P < 0.05) precipitation trends. d Long-term annual mean precipitation throughout Tonle Sap Lake (P_lake), the Mekong River Basin (P_Mekong), and the HCZ (P_HCZ). e Long-term patterns and correlations between annual mean precipitation of the HCZ (P_HCZ) and runoff discharge of the Mekong River at Kratie (Wang et al., 2020)

The HCZ is situated in the lower basin of the Mekong River, predominantly outside the drainage basin of Tonle Sap Lake. Recent reductions in precipitation within the Mekong River Basin have previously been linked to El Niño/La Niña events, as well as the Indian and Western North Pacific Monsoons (Frappart et al., 2018). Previous research has suggested that the decreased runoff from the Mekong River was primarily a consequence of climate change rather than human interventions, such as upstream dam construction in China. In contrast, there were no significant trends in precipitation for most of the Mekong River Basin between 1988 and 2000 (Fig. 3.31c), which could potentially explain the stabilized inundation observed during this period (see Fig. 3.25). To assess the relative impacts of three potential factors on the interannual inundation changes of Tonle Sap Lake, a multiple general linear model was employed (Tao et al., 2015). These factors included the precipitation of the HCZ, the number of dams, and the evapotranspiration (ET) of the lake's drainage basin. The analysis revealed that the relative contributions were 76.1% for HCZ precipitation, 6.9% for the number of dams, and 2.0% for ET, respectively. These findings underscore the predominant role of HCZ precipitation changes in driving the interannual dynamics of the lake's inundation.

The water turbidity of the Tonle Sap Lake is likely to be controlled by two factors: (1) Sediment resuspension, which can be attributed to external forces such as wind activity within the lake, sediment discharge within the lake basin, as well as internal forces related to hydrodynamics (Hoshikawa et al., 2019); and (2) exchanges of sediments between the Tonle Sap Lake and the Mekong River. Satellite observations showed a pronounced increase in water turbidity, which was likely due to the lake shrinkage induced hydrodynamic changes. For example, higher chances of sediment resuspension from the bottom can be expected when water depth decreases, even if other external factors are stable. Indeed, it was further confirmed that the validity of this hypothesis by the statistically significant correlations between the annual TSS concentration and inundation area (R2 = 0.41 for quarter 1 and R2 = 0.49 for quarter 4, both with P < 0.05, see Fig. 3.32a, d). Such correlations agreed well with the results of a former study (Hoshikawa et al., 2019), where statistically significant relationships were detected between water depth and TSS in dry seasons. Therefore, the inundation shrinkage (i.e., water depth decline) has caused an increase in sediment resuspension through either wind or gravity flow and therefore lead to the recent increase in water turbidity in Tonle Sap Lake (Siev et al., 2018). In contrast, the TSS trend and TSS- inundation correlations in quarters 2 and 3 were insignificant, which were associated with the reversed flow of the Mekong River to Tonle Sap Lake that was intervented by human activities during these periods (Fig. 3.32b, c). Numerically, sediment flux from the Tonle Sap River to the lake varies between 5.1 and 6.4 Mt year⁻¹, whereas the magnitude of reversal sediment discharge from the lake to the river was about three times smaller (Kummu & Sarkkula, 2008; Sok et al., 2021). As such, the lake turbidity could be substantially modulated by the sediment-rich flows from the Mekong River, which smears the inundation shrinkage-induced impacts in wet seasons. Nevertheless, physical modelling and additional in situ hydrological measurements are required to determine the underlined mechanisms and to quantify the exact contributions of various drivers on the inundation and water turbidity.

Fig. 3.32

Relationships between the seasonal mean TSS and inundation areas of Tonle Sap Lake. a–d Quarters 1–4 (Wang et al., 2020)

3.4.6 Projected Change in Inundation Area

Numerous studies have used hydrological models to simulate and quantify the future changes in the Tonle Sap Lake’s inundation dynamics under various scenarios representing both climate variability and water infrastructure development plans. For example, Västilä et al. (2010) used multiple models including GCM, VIC, EIA to examine the effects of changes in sea level and Mekong mainstream discharge under climate change on the Lower Mekong flood pulse during 2010–2049 period. They found that water levels in the Lower Mekong, including the Tonle Sap Lake, would increase in the future, leading to higher annual flooded areas. In particular, annual maximum water depth and flooded areas increased during average and dry years and decreased during wet years. The study also reported that flood duration will be likely to increase slightly with greater flooding starting earlier and lasting longer with flood peaks arriving earlier in average hydrological years. Arias et al. (2012) evaluated the impact of water infrastructure development and climate change by using the MRC Decision Support Framework for multiple scenarios of progressive stages in comparison to simulations by Västilä et al. (2010). They reported that while hydropower development could reduce flood extent by up to 1,200 km², climate change is expected to increase flood extent by up to 1,000 km². They also noted that during average years in the future, water levels in the lake during October–November may increase due to climate change but reduce due to dam construction. The largest changes may occur during dry years, and the areas most impacted would be those at the fringe of the open water with flood duration of 9–10 months, and halfway between open water and the edge of the floodplain, flooded for ~4 months. A recent study also showed similar results based on hypothetical dam simulations, indicating that regulation of mainstream Mekong flow by dams may increase areas flooded for over 7 months and reduce those flooded for less than 5 months (Pokhrel et al., 2018b). Similar findings have been reported by Yu et al. (2019) using CAESAR-LISFLOOD system and by Try et al. (2020) using RRI model. Further, Arias et al. (2014) identified that areas that currently have long periods or are permanently inundated throughout the year are likely to expand while seasonally inundated areas will be decreased. They also found that the hydrological alteration of the hydropower system in the 3S basin could have similar effects as the Lancang dam cascade and the cumulative effect of development in both areas will cause significant disruption to the inundation pattern of the lake.

3.5 Past and Future Changes in Climate and Water Resources in the LMRB

3.5.1 Climate of the Lancang-Mekong River Basin

3.5.1.1 Past and Future Warming Trends

In the past decade, there have been notable and confidently increasing trends in the annual mean temperature across the LMRB (Fan et al., 2015). These warming trends in both the Lancang River Basin and Mekong River Basin have surpassed the global average temperature rise, which was reported as 0.17 °C per decade since 1981 by Hartfield et al. (2018).

Between the early 1980s and 2010, there were no statistically significant changes in annual maximum and minimum temperatures observed over the Lancang River Basin (Fan et al., 2015). However, it's noteworthy that both annual maximum and minimum temperatures exhibited the same warming trend direction as the mean annual temperature in the Mekong River Basin during the same period (Lutz et al., 2014). Among the seasons, the highest rate of warming trends was observed during winter (December–February) across both the Lancang River Basin (Fan et al., 2015) and the Mekong River Basin (Lutz et al., 2014) from 1981 to 2010. It's worth mentioning that the Lancang River Basin had already been experiencing warmer winters prior to 1981, particularly during the period from the 1960s to the early 2000s (You et al., 2010).

Projections for the twenty-first century indicate statistically significant warming trends in mean annual temperature over the Lancang-Mekong River Basin (Kingston et al., 2011; Lacombe et al., 2012). These trends are expected to be more pronounced in the northern and southern parts of the basin (Lauri et al., 2012). However, it's important to note that the extent of temperature change varies depending on the climate scenario used in the models. Over the Mekong River Basin, a warming trend of 0.01–0.03 °C per decade is projected (Zhou et al., 2013), while the Lancang River Basin is expected to experience slightly more evident and consistent warming (Kingston et al., 2011). Projections suggest that by 2050, the daily maximum temperature over the Mekong River Basin is likely to increase, with estimates ranging from 1.6 °C in the northern and southwestern regions to 4.1 °C in the southeastern areas, where the historical climate has been cooler than in the central part of the basin (Zhou et al., 2013). Consequently, an increase in the frequency of annual hot days (daily maximum temperature >33 °C) is anticipated, particularly in the southern part of the Mekong River Basin (Västilä et al., 2010). Regarding seasonal temperature changes, projections indicate a fairly homogeneous increase in temperatures across the Mekong River Basin, with a warmer climate expected during wet seasons (1.7–5.3 °C) compared to dry seasons (1.5–3.5 °C) for the near future (2020–2050) (Zhou et al., 2013). Meanwhile, daily mean temperatures across the Lancang River Basin are projected to be higher during dry seasons (7.5–10.5 °C) than during wet seasons (6.0–7.5 °C) under the 6 °C warming scenario. Furthermore, the warming trend is expected to extend to higher elevations, especially above 400 m, in the Mekong River Basin during this century (Zhou et al., 2013).

3.5.1.2 Uncertainty in Estimated Past and Projected Future Precipitation

Previous studies have reported moderately increasing trends in annual precipitation over the Lancang-Mekong River Basin (LMRB) in recent decades, although with varying levels of confidence (Lacombe et al., 2013). One recent study found a wet but statistically insignificant trend of 24.8 mm/decade in annual precipitation over the LMRB during the period 1983–2016 (Chen et al., 2019). In contrast, from 1981 to 2007, annual precipitation based on daily gridded (0.25° × 0.25°) APHRODITE (Asian Precipitation-Highly Resolved Observational Data Integration Towards Evaluation of Water Resources) data showed a significant increasing trend of 52.6 mm/decade over the Mekong River Basin (Lutz et al., 2014). Similarly, there was a significant increase (14.5 mm/decade) in annual precipitation over the Lancang River Basin during the period 1981–2010, based on in situ precipitation records at seven meteorological stations (Liu et al., 2022). These findings suggest that while there are differences in estimates based on different datasets, there has been an increasing trend in annual precipitation in the LMRB in recent years. There is a consensus, with high confidence, that significant increases in annual precipitation are expected across the LMRB over the next 30–50 years (Lacombe et al., 2012). Variability in annual precipitation is also projected to increase in this basin (Lauri et al., 2012). This high confidence in projected wetting trends is primarily attributed to future global warming, which is likely to enhance the transport of water vapor from the Indian Ocean and the western Pacific Ocean towards the LMRB, resulting in increased precipitation across the region (Zhang et al., 2017). Depending on the emissions scenario, these projected wetting trends in annual precipitation over the Lancang-Mekong River Basin range from 2.5–8.6% to 1.2–5.8% per year. For instance, annual precipitation is expected to increase by 35–365 mm (3–14%) over the Mekong River Basin by 2050 (Zhou et al., 2013) and by approximately 10% over the Lancang River Basin under the 2 °C warming scenario.

In terms of monthly precipitation, projections indicate increases over the Lancang River Basin for all months by 20–60% under warming scenarios of 2–6 °C, except for April, which shows a projected decrease of 16–40% (Kingston et al., 2011). With moderate confidence, it is expected that precipitation will increase during the wet season (May–October) over the Mekong River Basin by 2050 but decrease in the dry season (November–April) (Zhou et al., 2013). Additionally, there is a likelihood that precipitation will shift from higher to lower elevations, such that historical annual precipitation levels of 1,500 mm recorded at an elevation of approximately 280 m may be observed at elevations of around 80 m (Zhou et al., 2013) (Fig. 3.33).

Fig. 3.33

Changes in temperature and precipitation over the a LMRB, b upper part of the LMRB (LRB) and c lower part of the LMRB (MRB), based on the published literature

3.5.2 Water Resources in the LMRB: Historical Changes and Future Projections

3.5.2.1 Annual Mean Discharge

A general trend of decreasing annual streamflow was identified in the Lancang-Mekong River Basin (LMRB) over the period of 1960–2010, although this trend is associated with low confidence. However, no clear trend has been observed after 2010 (Ruiz-Barradas et al., 2018). It's worth noting that different studies have reported varying trends in historical streamflow in the LMRB, with some indicating a decrease and others suggesting an increase. These discrepancies can be attributed to differences in data sources and methodologies used in each study.

The changes in streamflow in the LMRB are the result of a combination of climate change and human activities, and the contributions of these factors vary across regions and time periods. Climate change was a dominant driver of streamflow alterations in the LMRB before 2010, accounting for 82.3% of the changes during the transition period of 1992–2009. In contrast, human activities, primarily dam construction, played a more significant role after 2010, contributing 61.9% of the changes in streamflow during the post-impact period of 2010–2014 (Li et al., 2017). Notably, the hydrological response of the Lancang River Basin appears to be more sensitive to climate factors than human activities when compared to the Mekong River Basin, underscoring the increasing impact of intensive human activities on hydrological processes in the Mekong River Basin, particularly in recent years (Shin et al. 2020).

Projections for future streamflow changes exhibit spatial variability, particularly within the Mekong River Basin (Liu et al., 2022). While an increasing trend in streamflow is anticipated for the Lancang-Mekong River Basin, significant uncertainties persist. Some studies project a 21% increase in annual runoff by the 2030s compared to the historical period (1951–2000) based on 11 global climate models (GCMs) (Eastham et al., 2008). In contrast, Västilä et al. (2010) reported only a 4% increase in annual flow by the 2040s in the Lancang-Mekong River Basin, using dynamically downscaled data from the ECHAM4 climate model. Other investigations based on CMIP5 datasets for the near future (2036–2065) have also indicated relatively small changes in mean annual flow, ranging from 3 to 10% in the Lancang-Mekong River Basin (Hoang et al., 2016; Västilä et al., 2010).

Furthermore, the magnitude and frequency of extremely high-flow events are projected to increase, while low-flow events are expected to occur less frequently, particularly as a result of climate change (Hoang et al., 2016). These more frequent extreme high-flow events could pose increased flood risks in the Lancang-Mekong River Basin. It's worth noting that the extensive construction of hydropower projects, which has led to changes in discharge, is anticipated to have a more significant impact on hydrography in the basin compared to climate change over the next 20–30 years (Hoang et al., 2019; Lauri et al., 2012) (Fig. 3.34).

Fig. 3.34

Changes in streamflow over a the LMRB, b LRB, and c MRB, based on published works

3.5.2.2 Groundwater

Global groundwater data from the International Groundwater Resources Assessment Center (IGRAC) indicate that approximately 0.55 km³ of groundwater was extracted from the Lancang-Mekong River Basin (mainly from the Mekong River Basin) in 2000 (Wada et al., 2010). However, it's important to note that this estimate is significantly lower than what has been reported by country-based statistics (Ha et al., 2015). This discrepancy could be attributed to the fact that the global database from IGRAC may not fully account for groundwater use by individual households across the basin (Pokhrel et al., 2018a, 2018b).

The groundwater system in the Lancang-Mekong River Basin (LMRB) is primarily influenced by changing hydrological conditions and intensive human activities, both of which impact the groundwater balance in terms of recharge and withdrawal (White, 2002). Over a 30-year monitoring period in the Mekong Delta, a significant decline in groundwater levels has been observed (IUCN, 2011). Particularly in Ca Mau Province, Vietnam, groundwater levels have dropped by as much as 10 m since 1995 (IUCN, 2011). In the river delta region of Vietnam, groundwater levels have consistently decreased at a rate of approximately 0.3 m per year, as documented by data from nested monitoring wells. This decline in groundwater levels has also led to land subsidence in the region, occurring at an average rate of about 1.6 cm per year (Erban et al., 2013).

The principal factors driving these declining trends in groundwater levels can be attributed to increased water demand and reduced water supply (IUCN, 2011). The growing population and expansion of agriculture have generated a higher demand for freshwater resources, intensifying the exploitation of groundwater. Additionally, the supply of clean water in this region has decreased (IUCN, 2011). Reduced groundwater recharge is primarily a result of changes in land use, including deforestation and increased cultivation of fields, which reduce the groundwater recharge ratio accordingly (White, 2002).

3.5.2.3 Potential Environmental and Social Impacts of Water Resource Changes

Anticipated changes in the water resources of the Lancang-Mekong River Basin (LMRB) are likely to have significant implications for sustainable water management. These projected changes in the basin's flow regime are expected to have negative consequences across several dimensions.

Firstly, substantial alterations to flow regimes can disrupt aquatic ecosystems by changing the distribution of vegetation, the natural habitats of native species, and fish migration patterns (Arias et al., 2012). Dams, in particular, are expected to profoundly impact fish abundance and catch in the lower reaches of the Mekong River Basin, which can have implications for dietary protein consumption (Burbano et al., 2020). Furthermore, reduced streamflow during the wet season may impede overland water flows that trigger the natural sedimentation process on floodplains, affecting flood-recession agriculture. Decreased sedimentation will also reduce the nutrients carried by sediment during flood events, further impacting crop yields (Hoang et al., 2019).

It is projected that water use in the LMRB will significantly increase due to rapid socioeconomic development and population growth, outpacing the increase in available water resources (Eastham et al., 2008). This could lead to growing challenges related to water security, with an increasing number of people facing water stress. Moreover, studies have shown that regions with significantly regulated flows due to dams tend to experience downstream shifts in water scarcity hotspots (Veldkamp et al., 2017).

The demand for groundwater in the LMRB is expected to surge under climate change conditions, as surface water becomes less accessible. This intensification of groundwater extraction could lead to large-scale land subsidence, potentially resulting in the release of arsenic from deep groundwater through vertical migration (Wagner et al., 2012). This poses risks to crop yields and human health in the future (Merola et al., 2014).

Despite these negative effects of an altered water system, there are some positive impacts to consider. For instance, an increase in streamflow during the dry season (Shin et al., 2020) could help alleviate water stress for agriculture (Son et al., 2012). Higher water levels during the dry season can prevent saltwater intrusion downstream, particularly in the vulnerable Mekong Delta (Smajgl et al., 2015). Additionally, the relatively lower water levels during the wet season induced by dams imply reduced flood risks along the river, especially in the main floodplains of the Mekong Delta (Pokhrel et al., 2018a, 2018b).