Future rainfall projections for the Lower Mekong Basin using CMIP6 dynamical downscaling Open Access

The Lower Mekong River Basin (LMB), spanning Laos, Thailand, Cambodia, and Vietnam, represents a critical ecological and socioeconomic region that supports approximately 70 million people dependent on its water resources. The LMB is recognized for its diverse climatic conditions, ranging from high-altitude continental and temperate regimes in the upper basin to tropical monsoonal climates in the lower basin. Rainfall patterns in the basin are primarily governed by the global monsoon system, with the Southwest Monsoon driving the distinct wet (May–October) and dry (November–April) seasons in the lower basin (Kiem et al. 2008; Vu et al. 2015). With global warming intensifying these challenges, the basin faces more frequent and severe floods and droughts, impacting natural resources, infrastructure, and settlements (Misra 2014). As a consequence, reliable projections of future changes in rainfall patterns over the LMB are crucial for effective water resource management, agricultural planning, and disaster risk reduction strategies. However, projecting future rainfall in this region remains challenging due to the complex topography, diverse climate regimes, and the influence of large-scale climate phenomena such as the El Niño–Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD) (Kingston et al. 2011; Vu et al. 2018; Jiang et al. 2022).

To date, existing research on rainfall projections for the LMB predominantly utilized downscaling the outputs of global climate models (GCMs) participating in the Coupled Model Intercomparison Project (CMIP). Both statistical and dynamical downscaling methods have been employed in these efforts to provide more localized and high-resolution projections. Early efforts, such as those by Lacombe et al. (2012), statistically downscaled CMIP Phase 3 (CMIP3) GCM outputs to explore potential shifts in seasonal rainfall patterns, projecting wetter wet seasons and drier dry seasons in the LMB. While these studies laid the groundwork, they faced limitations in capturing the intricate interactions between regional climate and larger phenomena like ENSO and IOD, as highlighted by Vu et al. (2018). Subsequent studies, such as the dynamical downscaling of CMIP Phase 5 (CMIP5) outputs by Fu et al. (2023) using a regional climate model (RCM), allowed for a more detailed understanding of rainfall patterns and trends projecting an overall increase in annual rainfall across the basin while highlighting substantial spatial variability.

It is important to highlight that the Coordinated Regional Climate Downscaling Experiment – Southeast Asia (CORDEX-SEA) project (Tangang et al. 2020) has so far played an important role in producing high-resolution projected climate information for Southeast Asia. In its earlier phases, CORDEX-SEA downscaled CMIP5 outputs to a resolution of 25 km for the whole of Southeast Asia (Phase 1) (Cruz et al. 2017; Tan et al. 2019; Supari et al. 2020; Tangang et al. 2020) and further to a resolution of 5 km (Phase 2) for specific regions, notably the Mekong River Delta (Hoang-Cong et al. 2022; Chung et al. 2023). These efforts generally projected an increase in mean annual rainfall over the basin in the future, with variations depending on the GCM and RCM used.

However, most existing studies on rainfall projections for the LMB have relied on earlier CMIP phases (CMIP3 and CMIP5). To date, to our knowledge, no study has applied a dynamical approach using the latest CMIP6 projections and the associated Shared Socioeconomic Pathways (SSPs) for the region, despite CMIP6 being the state-of-the-art framework for investigating future climate scenarios (Arias et al. 2021). Additionally, these previous research efforts have predominantly focused on mean rainfall trends, likely overlooking the critical spatial and temporal variability of extreme rainfall events. Furthermore, substantial differences remain between raw RCM outputs and observed rainfall (e.g., Hoang-Cong et al. 2022), highlighting the need for bias correction (BC) methods to improve the reliability of rainfall representation in the region (e.g., Trinh-Tuan et al. 2019).

Currently, Phase 3 of CORDEX-SEA activities is underway, focusing on downscaling CMIP Phase 6 (CMIP6) GCM outputs to a 25 km resolution over Southeast Asia for various SSP scenarios (Thanh 2023; Ngo-Duc et al. 2024). This study, utilizing high-resolution outputs of the latest aforementioned downscaling experiments combined with statistical BC, aims to provide insights into future rainfall patterns, including both mean and extreme rainfall, over the LMB. The results are expected to contribute to enhanced water resource management, climate adaptation, and disaster risk reduction strategies in this critical region. The rest of this paper is composed of three sections. The section ‘Data and Methodology’ describes the study area, datasets, and analysis methods. In the section ‘Results and Discussion’, we present the results, examining the model performance, as well as future spatial and temporal changes in mean and extreme rainfall. Finally, conclusions are presented in the last section.

In this study, we employed the non-hydrostatic version of the RCM RegCM4 (RegCM4-NH) (Coppola et al. 2021; Pichelli et al. 2021) to downscale outputs from the CNRM-CM6-1 model (referred to as CNRM) by the French National Centre for Meteorological Research (Voldoire et al. 2019) and the NorESM2-MM model (referred to as NorM) by the Norwegian Climate Center (Seland et al. 2020). Both CNRM and NorM are top-performing models for simulating temperature and rainfall across Vietnam's climate zones (Nguyen-Duy et al. 2023). Their downscaled experiments using the RegCM4-NH model, referred to as R.CNRM and R.NorM, were conducted at a horizontal resolution of 25 km × 25 km covering the Southeast Asian domain (89°E–147°E, 15°S–30°N) (Figure 1).

The RegCM4-NH model configuration included the Kain-Fritsch convective parameterization scheme, the UW planetary boundary layer scheme, the SUBEX large-scale precipitation scheme, the Zeng ocean–atmosphere heat transport scheme, and the CLM4.5 land surface scheme. This configuration was selected for Phase 3 of the CORDEX-SEA project based on sensitivity evaluations of various physical schemes within the region (Ngo-Duc et al. 2024).

The downscaling experiments were evaluated for the baseline periods from 1985 to 2014 and the future period from 2015 to 2100 under two SSP scenarios: SSP2-4.5 and SSP5-8.5. Specifically, the SSP2-4.5 scenario represents a future with balanced development across the globe and various sectors. It aims for a radiative forcing target of 4.5 W/m² by the end of the century, which will be achieved through emission reduction policies, renewable energy promotion, and land-use changes. In contrast, the SSP5-8.5 scenario depicts a future with high global development inequality and a radiative forcing target of 8.5 W/m² by the end of the century, reflecting limited or no significant actions to mitigate climate change (Arias et al. 2021).

The reference rainfall dataset used for evaluation and BC in this study is the Asian Precipitation – Highly-Resolved Observational Data Integration Towards Evaluation (APHRODITE) dataset (Yatagai et al. 2012), version 1101 and 1101EX_R1. APHRODITE is based on ground-based rain gauge observations and employs advanced algorithms to interpolate data onto a 0.25° × 0.25° grid. This dataset has been extensively used and accurately captured spatiotemporal rainfall patterns in the monsoon Asian region (Ono & Kazama 2011; Ang et al. 2022).

The performance evaluation of the downscaling experiments performance in simulating rainfall during the baseline period involves comparing simulated rainfall indices: total daily rainfall (R, mm/day), Simple Precipitation Intensity Index (SDII, mm/day), and maximum 1-day rainfall within a year (RX1day, mm/day) against reference values from APHRODITE. The evaluation uses several statistical metrics, including mean bias, root mean square error (RMSE), and spatial correlation coefficients.

To address systematic biases in the model outputs, a non-parametric quantile mapping (QM) BC technique (Gudmundsson et al. 2012; Trinh-Tuan et al. 2019) is employed. This technique involves aligning the cumulative distribution functions (CDFs) of the model data with those of the observations to derive a transfer function (TF). The TFs are initially derived for each grid point for each month using data from 1985 to 1999 as a training dataset. These TFs are then applied to the 15-year (2000–2014) period for evaluation purposes, allowing for a comprehensive assessment of the performance. This approach is favored due to its independence from predetermined statistical distributions.

To further optimize the TF construction process and align with best practices recommended by Reiter et al. (2016), TFs are subsequently derived for the entire baseline period (1985–2014) and applied to the future projection period (2015–2100). This approach ensures that the BC is tailored to the full range of observed climate conditions, thereby enhancing the robustness and reliability of future rainfall projections.

The peak rainfall months, June, July, and August (JJA), are shown in Figure 2(b), reflecting the intensity of the monsoon season. This seasonal pattern is essential for the basin's water resources, contributing significantly to the annual totals. The Southwest Monsoon brings moisture-laden air that encounters the Truong Son Mountain range, located along the western boundary of the LMB (see Figure 1), causing more rainfall on the windward side of the range and creating rain-shadow (foehn) wind effects on the leeward side (Nguyen-Le et al. 2014; Kusaka et al. 2024). These orographic influences are especially pronounced in areas such as Laos, where Vientiane and Sekong see high rainfall, while regions further from the mountain range, such as Thailand and Cambodia, receive significantly less precipitation.

Further examination of the mean rainfall distribution by latitude (Figure 2(d)) reveals a distinct latitudinal gradient, with rainfall amounts increasing toward lower latitudes within the basin. The highest rainfall amounts are concentrated around the latitudes 17–20°N and 13–16°N in the western Truong Son Mountain range. This gradient suggests the interplay of various factors, including moisture sources, topography, and atmospheric circulation patterns interacting with the monsoon system.

The observed spatial and temporal variability aligns with previously published research on the Mekong's hydroclimatic dynamics. High rainfall in Laos, particularly in Vientiane and Sekong, is attributed to the Southwest Monsoon and orographic enhancement (Lacombe et al. 2010; Chen et al. 2012), while regions farther from the mountain range, such as Thailand and Cambodia, are less affected by these processes, receiving less rainfall (Chou et al. 2009; Ngai et al. 2017). The dominance of the JJA season is driven by the seasonal migration of the InterTropical Convergence Zone (ITCZ) and associated monsoon circulation (Schneider et al. 2014; Keshtgar et al. 2020; Vásquez et al. 2022). This seasonal cycle, along with the latitudinal rainfall gradient, with lower latitudes receiving more rainfall, expresses the critical role of atmospheric and geographic factors in shaping the rainfall variability across the basin.

During the monsoon season (JJA), model performance is less accurate compared with annual simulations, with spatial correlation coefficients of 0.77 and 0.69 for R.CNRM and R.NorM, respectively. This lower accuracy can be attributed to the complex rainfall patterns and convective processes that are more challenging to simulate at a regional scale during this period. This observation aligns with the findings of Tan et al. (2019) and Ngo-Duc et al. (2017), who reported similar difficulties in simulating the monsoon rainfall over the Mekong region.

In contrast, the BC products exhibit substantial advantages. Especially during JJA, the spatial correlation coefficients increase from 0.77 and 0.69 to 0.94 and 0.92 for R.CNRM-QM and R.NorM-QM, respectively. The spatial patterns of rainfall from these BC products closely resemble those in the APHRODITE dataset, indicating the effectiveness of the BC method in reducing systematic biases in the original model simulations.

The simulation results after BC using QM demonstrate a significant improvement, with the seasonal cycles by R.CNRM-QM and R.NorM-QM closely aligning with APHRODITE, exhibiting correlation coefficients of 0.92 and 0.95, respectively. However, similar to the spatial distribution results (Figures 3 and 4), the BC outputs underestimate rainfall for all months. Both models also underestimate the seasonal variability of rainfall, with normalized standard deviation errors of 0.8 and 0.9 for R.CNRM-QM and R.NorM-QM, respectively. This underestimation of BC rainfall is consistent with the findings of previous studies (e.g., Lafon et al. 2013; Cannon et al. 2015; Kim et al. 2020; Sheau Tieh et al. 2022; Mbienda et al. 2023). Furthermore, it is worth mentioning that while the QM BC improves the statistical alignment of model outputs with observed data, it can introduce additional uncertainties by altering the physical consistency of climate processes (Cannon et al. 2015). Additionally, the TFs derived from the historical period, which are assumed to remain valid in the future, may amplify uncertainties in the projection results, particularly for extreme rainfall cases (Zhang et al. 2022; Vidrio-Sahagún et al. 2025).

After BC, the increasing trend in rainfall persists, albeit with a reduced intensity compared with the original model projections. Notably, in the final two decades of the 21st century, the BC products exhibit much smaller increasing rates compared with those in their respective original model projections. The R.CNRM-QM (R.NorM-QM) product shows an increase in annual rainfall of approximately 8.2 (16.9%) and 3.2% (3.8%) by the end of the 21st century compared under SSP5-8.5 and SSP2-4.5, respectively.

Further, indices like Rx1day and SDII reveal more pronounced increasing trends in the future over the LMB compared with mean annual rainfall, particularly during the latter half of the 21st century. While the changes under SSP2-4.5 display a relatively modest increase in both indices, their upward trends become more pronounced under SSP5-8.5. It is noteworthy that the BC products maintain the direction of the overall trends but with reduced magnitude. For instance, R.CNRM (R.NorM) exhibits an increase of 18.9 (26.1%) and 24.3% (22.5%) for RX1day and SDII, respectively, by the end of the 21st century under SSP5-8.5. For the same scenario and period, the BC products, R.CNRM-QM (R.NorM-QM) displays a less-pronounced increase of 10.1 (19.1%) and 18% (18.2%) for RX1day and SDII, respectively. The projected increases in both Rx1day and SDII, particularly under the SSP5-8.5 scenario, imply an intensification of rainfall extremes. This intensification can be attributed to the fact that the atmosphere can hold more moisture under global warming, about 7% more moisture per degree of warming according to the Clausius–Clapeyron equation (Berg et al. 2013; Lehmann et al. 2015), favoring a stronger rate of intensification for extreme events (Fischer & Knutti 2015; Robinson et al. 2021). The increase in rainfall extremes thereby heightens the risk of severe rainfall-related hazards, such as flooding and landslides, especially toward the end of the 21st century.

Our study's results are in line with prior research on rainfall patterns in the LMB that utilized CMIP3 and CMIP5 outputs. Specifically, Vu et al. (2015) used the weather research and forecasting (WRF) model to downscale CMIP3 GCMs, reporting a future intensification of the seasonal rainfall cycle, with wetter wet seasons and drier dry seasons. MONRE (2020), using CMIP5 downscaled products, identified similar future trends of increased rainfall during the wet season and decreased rainfall during the dry season in the LMB. They projected an annual rainfall increase of up to 20% by the century's end under the RCP8.5 scenario. Hoang-Cong et al. (2022), employing the RegCM4.7 model driven by the CMIP5 EC-EARTH model, observed an overall annual rainfall increase, notably during the wet season, and a decrease during the dry season, especially under higher emission scenarios. It is noteworthy that while MONRE (2020) applied BC to the model outputs, neither Vu et al. (2015) nor Hoang-Cong et al. (2022) implemented any BC. The agreement between our study's results and the prior research underscores the robustness of the projected changes in rainfall patterns in the LMB.

The projected changes in rainfall patterns over the LMB can be linked to shifts in large-scale atmospheric circulation and monsoon dynamics across Southeast Asia. During the JJASON months, the LMB receives a significant portion of its annual rainfall from the Southwest Monsoon. As global temperatures rise, increased land–sea thermal contrast and enhanced atmospheric moisture content can lead to a strengthening of the monsoon circulation (Kitoh et al. 2013). This intensification results in more robust moisture transport from the Indian Ocean and an increase in convective activity over the LMB, contributing to higher rainfall during the wet season. Conversely, during the DJFMAM months, the Northeast Monsoon system, originating from the Siberian high-pressure system, plays a crucial role in determining rainfall patterns over the region. Additionally, prior studies suggested a strengthening of the Siberian high and an associated intensification of the Northeast Monsoon flow under global warming (Pohl et al. 2011). This intensification can lead to drier conditions over the LMB during the dry season, as the northeasterly winds bring relatively drier air masses from the Asian continent. Furthermore, the projected changes in rainfall patterns over the LMB can also be influenced by shifts in the position and strength of the ITCZ and the Western North Pacific Subtropical High (WNPSH). The northward migration of the ITCZ, which is expected under a warming climate (Huang et al. 2013), can contribute to increased rainfall over the downstream regions of the LMB during the wet season. Meanwhile, the WNPSH, a semi-permanent high-pressure system that influences the East Asian monsoon, is projected to intensify and expand westward under future climate scenarios (He et al. 2015). This expansion can modulate the moisture transport and convergence patterns over the LMB, potentially contributing to the enhanced wet season rainfall in downstream regions.

Projected changes in rainfall patterns may vary due to differences in models, emission scenarios, and downscaling techniques (e.g., Zhao et al., 2023). However, our results indicate a consistent trend in the LMB: the region is expected to undergo a more intense seasonal rainfall cycle, with wetter wet seasons and drier dry seasons. This intensification is particularly pronounced in downstream regions like the Mekong Delta, especially under higher emission scenarios. Consequently, more flooding risks are anticipated during the wetter months of the year, JJASON, while the drier period of DJFMAM may experience heightened drought risks in the future.

This study projects future rainfall over the Lower Mekong Basin using CMIP6-downscaled outputs from the RegCM4-NH model, with boundaries defined by the CNRM-CM6-1 and NorESM2-MM GCMs. A non-parametric QM BC technique was employed to address systematic biases in the RCM outputs. TFs were developed using data from 1985 to 1999 as a training dataset and then applied to the 2000–2014 period for evaluation purposes. While the original model outputs capture the general spatial and seasonal rainfall patterns, they overestimate rainfall amounts. After applying the BC, model performance improved significantly, with spatial correlation coefficients rising from 0.86 and 0.81 to 0.92 and 0.94, closely matching the APHRODITE reference dataset.

The study's future projections for the LMB reveal a nuanced and progressive transformation of regional rainfall patterns, with distinct characteristics across different time horizons. In the mid-future period (2031–2060), climate change impacts emerge subtly, with less-pronounced scenario divergence between SSP2-4.5 and SSP5-8.5. This transitional phase shows early signs of seasonal rainfall shifts, including marginal increases in summer–autumn (JJASON) precipitation and slight decreases in winter–spring (DJFMAM) rainfall. By 2070–2099, under the high-emission SSP5-8.5 scenario, the models project a dramatic intensification of the seasonal rainfall cycle. Summer–autumn months are expected to experience rainfall increases up to 70 mm/month, while winter–spring periods will see decreases of 5–30 mm/month, potentially resulting in annual rainfall increases up to 20%. The seasonal rainfall cycle is expected to intensify significantly, particularly in midstream to downstream areas. Extreme precipitation indices such as Rx1day and SDII predict a 7% increase in moisture capacity per degree of warming. The downstream regions, especially Cambodia and Vietnam's Mekong Delta, are projected to experience the most pronounced changes in precipitation patterns. This shift, driven by interactions between atmospheric circulation patterns such as the Southwest Monsoon, Northeast Monsoon, and ITCZ, increases the risks of flooding during wet months and drought during dry periods.

These projections underscore a fundamental restructuring of the LMB's hydrological regime, highlighting the urgent need for adaptive strategies to mitigate potential ecological and socioeconomic impacts. The research reveals that climate change will not merely alter rainfall patterns but potentially redesign the entire water resource landscape of Southeast Asia's critical water system, with far-reaching consequences for agriculture, water management, and regional resilience.

While this study leverages CMIP6 models and high-resolution downscaling using RCMs, additional research is needed to address or to better understand uncertainties in rainfall projections over the region. Ultimately, the findings of this study emphasize the urgency of mitigating greenhouse gas emissions and proactively adapting to the anticipated changes in rainfall regimes over the LMB in the future.

This research was funded by the Ministry of Science and Technology of Vietnam under grant number NĐT/KR/21/18 and supported by the Vietnam National Foundation for Science and Technology Development (NAFOSTED) under Grant 105.06-2021.14. Additionally, we also acknowledge the support from the Asia-Pacific Network for Global Change Research (APN) support for the CARE for SEA megacities project (CRRP2023-08MY-Cruz). The authors would like to thank the editor and anonymous reviewers for their valuable and constructive comments to improve our manuscript.

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

The authors declare there is no conflict.