Future Climate Change Increases Streamflow and Risks of Hydrological Hazards in the Pearl River Basin

Abstract

Understanding and predicting climate change impacts on the terrestrial water cycle is essential for water resources management and hazard prevention. This study aims to project future runoff of a densely-populated river basin, the Pearl River Basin (PRB), under different Shared Socioeconomic Pahway (SSP) scenarios, by combining the Soil and Water Assessment Tool (SWAT) model and the CMIP6 climate projections. Results show that climate change will significantly increase the runoff of the PRB, with changing rates of 0.21, 0.20, 0.11, and 0.17 mm/month/year for low- to high-emission scenarios SSP126, SSP245, SSP370, and SSP585, respectively. Future runoff exhibits strong seasonal and spatial variability due to complex changes in precipitation and potential evapotranspiration across the basin. The PRB may experience higher flood risks during the wet season under all SSP scenarios, driven by a ~15% increase in runoff during the wettest month during 2061–2100 relative to that of 2021–2060. Conversely, drought risks may escalate in the East River Sub-basin of the PRB during the dry season under the high-emission scenarios (SSP370 and SSP585), with a ~20% reduction in runoff during the driest month during 2061–2100 relative to that of 2021–2060. The highest-emission scenario (SSP585) may lead to the most drastic hydrological changes, including increased risks of flooding and drought across different parts of the PRB. Our findings suggest intensified water cycling and increased hydrological risks in the PRB under a changing climate, highlighting the necessity of future water resource management to consider potential climate change impacts to mitigate the risks of floods and droughts effectively.

1. Introduction

Climate change is increasing the frequency and intensity of natural disasters, causing severe damage to our society’s well-being [1]. For example, in 2019 and 2020, Australia experienced the most significant drought and forest fires on record, resulting in considerable wildlife losses, destruction of natural habitats, and extensive property damage [2]. In June 2024, 5 billion people worldwide experienced extreme heat. Record-breaking temperatures soared to around 50 °C in regions including the Middle East, India, and parts of South America, setting new records in these areas [3]. Moreover, frequent flood disasters have occurred in recent years, not only in areas with abundant rainfall, such as southern China, Indonesia, and Sudan [4,5,6], but also in regions known for scarce precipitation, including northern China, Libya, and the Middle East [7,8,9]. Unusually extreme rainfall has led to catastrophic flooding in these areas, resulting in loss of life and significant economic damage. Such increasing frequency of extreme weather events serves as a warning of the potentially more severe negative impacts of climate change [10].

Reliably predicting future streamflow and runoff is crucial for efficient water resources management and mitigation of hydroclimatic hazards (e.g., droughts and floods) [11]. The terrestrial hydrological cycle is sensitive to driving forces, such as climate conditions and anthropogenic activities [12,13]. For example, global runoff has shown significant changes in the 20th century and is projected to further increase under future climate change [14,15,16]. Consequently, predicting future hydrology under a changing climate is necessary for informing the management of highly variable water resources.

Climate change affects the terrestrial water cycle by altering both water falling on land surfaces and water fluxes leaving land surfaces in the form of water vapor. Increasing air temperature as a result of the elevated greenhouse gas (GHG) concentrations could lead to substantial variations in precipitation and evapotranspiration (ET) [17,18]. Although the overall amount of global rainfall has not changed significantly over the past few decades, variations in horizontal energy transport are altering historical spatial patterns of precipitation, leading to greater heterogeneity over time and space [19,20,21]. Consequently, areas receiving more precipitation tend to have higher runoff, and vice versa [22,23,24,25]. However, rising temperatures also increase the evaporative water demand of the air and lead to higher ET, which tends to reduce runoff [26,27,28]. How those counteracting processes interact with each other in shaping the spatial and temporal patterns of the terrestrial water cycle has not been sufficiently investigated.

Understanding the impacts of climate change on hydrology is critical for mitigating hydrological hazards. On the one hand, the increased atmospheric temperature could lead to more intense extreme precipitation as a result of the increasing water vapor in the atmosphere [29,30]. Previous studies have suggested that more precipitation will happen in rainfall-abundant areas in the future [19,31,32], contributing to the increased intensity of floods [33,34,35]. On the other hand, the warming climate also increases the drought risks by enhancing evaporative water loss, especially during dry seasons [36,37]. As a result, quantifying potential changes in flood and drought risks under future climate change scenarios is necessary for mitigating hydrological hazards.

Climate projections from General Circulation Models (GCMs) provide useful information for predicting future hydrological processes. Driven by different GHG emission scenarios, GCMs have been widely used to predict future climate conditions. Many studies have shown increasing trends in runoff at the basin level by driving hydrologic models with GCMs climate projections [38,39,40]. Stronger seasonal variability has also been identified under future climate conditions, as the wet season will have more riverine discharge as a result of increasing precipitation, while spring and winter will be drier because of the reduced rainfall [41,42,43]. Consequently, extreme weather events, such as heavy rainfalls and droughts, are expected to occur at unprecedented rates in the future. Examining extreme hydrological events in response to climate change is crucial as it can provide insights into the mitigation of future flood and drought risks, particularly in disaster-prone watersheds [44].

As the home of about 100 million people, the Pearl River Basin (PRB) has experienced increasingly frequent extreme precipitation and flood hazards in recent years. For example, the North River Sub-basin of the PRB experienced two large flood events in June 2022 and April 2024, with several stations along the Pearl River recorded the largest streamflow and highest water levels since they were built [45]. Such unusually frequent and severe floods caused significant damage to local properties, challenging the management of water for the well-being of society. Since climate change will persist in the future, it is crucial to quantify the temporal trends, magnitudes, and variability of future runoff in this region under different climate change scenarios to inform the management of water resources and mitigation of hydrologic hazards.

A few previous studies have attempted to investigate the potential impacts of climate change on the streamflow of the Pearl River based on CMIP5 climate projections [46,47], and indicated that different parts of the PRB may experience different changes in streamflow. It is necessary to update our understanding of the impacts of climate change on hydrology using the latest climate projections and new climate change scenarios. The primary objective of this study is to investigate how future climate change will affect hydrological processes in the PRB during 2021–2100. To that aim, the Soil and Water Assessment Tool (SWAT) is employed to simulate future hydrological processes using GCM climate projections as input. We aim to answer the following questions: (1) How will precipitation and evapotranspiration change in the PRB in the remainder of the 21st century? (2) How will future hydrological processes vary under different climate change scenarios in the PRB?

2. Methods

2.1. Study Area

The Pearl River is the third-longest river by drainage area and the second-largest river in terms of streamflow in China (Figure 1). The mainstream of the Pearl River is 2214 km long, with a drainage area of about 442,100 km². The PRB includes four Sub-basins, including the North River Sub-basin (NRB), the East River Sub-basin (ERB), the West River Sub-basin (WRB), and the Pearl River Delta Sub-basin (PRD), with areas of 38,363 km², 25,325 km², 351,535 km², and 26,041 km², respectively [45].

The basin terrain predominantly consists of hills and mountains in the upstream regions and plains in the coastal regions. The PRB is characterized by higher elevations in the west and lower elevations in the east. The northwestern region of the PRB is the Yunnan–Guizhou Plateau, while the Nanling Mountain covers the majority of the northern parts of the basin. The eastern and southeastern areas comprise a mix of hills and plains.

With a tropical and subtropical climate, the PRB has hot and rainy summers and mild winters. The average annual temperatures in the PRB range from 14 to 22 °C, with precipitation ranging from 1200 to 2200 mm/year. The spatial and temporal distributions of precipitation vary widely, contributing to the highly uneven seasonal distribution of streamflow, with the flood season from April to September accounting for approximately 80% of the annual streamflow and June, July, and August alone exceeding 50%. The streamflow also demonstrates significant interannual variability and uneven spatial–temporal distribution. This variability leads to frequent natural disasters in the basin, including floods, droughts, and salinity intrusion [45,48].

2.2. Climate Projections

The Coupled Model Intercomparison Project Phase 6 (CMIP6) is a global initiative that brings together climate modeling centers from around the world to predict future climate change. It builds on previous phases (CMIP1-CMIP5) and aims to enhance climate models, simulate a wide range of climate scenarios, and provide valuable data for climate research and policymaking [49]. CMIP6 involves various GCMs developed by different institutions, each conducting simulations based on different scenarios of greenhouse gas emissions, land use changes, and other factors [50]. To understand the impact of future climate change on the hydrology of the PRB, we employ GCM projections from 3 models, which are Australian Community Climate and Earth-System Simulator (ACCESS), Euro-Mediterranean Centre on Climate Change (CMCC), and Institute of Numerical Mathematics, Russian Academy of Sciences (INM) [51,52,53]. Previous investigations suggest reasonable performance of these models in simulating climate conditions in China [54,55,56]. Specifically, the ACCESS model tends to have lower precipitation projections than many other GCMs, but the bias of this model tends to be much lower than other models; the INM model has much lower bias in the minimum temperature projection than other models.

To drive the hydrologic model (detailed in Section 2.3), we obtain precipitation, daily maximum temperature, daily minimum temperature, solar radiation, vapor pressure, and wind speed from each GCM climate projection, and calculate the potential evapotranspiration (PET) with the Penman-Monteith method [57,58]. We utilize 4 shared socioeconomic pathways (SSPs) for each variable, which are SSP126, SSP245, SSP370, and SSP585, representing scenarios of a range of GHG emissions from low to high levels from 2015 to 2100. Specifically, SSP126 represents a more sustainable path; SSP245 is to develop in a way not shifting markedly from historical patterns; SSP370 refers to a low international priority for strong environmental management; SSP585 represents the fossil-fueled development. The detailed explanation of SSPs in CMIP6 can be found in O’Neill et al. [59]. Additionally, we collect historical simulation data from each GCM model covering the period from 1980 to 2014 for bias correction. To summarize, six climate variables with 4 SSPs from 3 models (a total of 12 sets of projections) are collected for future climate projection in the PRB. For the subsequent analysis, we take the average of the projections in each SSP from the three models to reduce the uncertainty associated with any single model.

Post-processing is necessary to reduce bias in the raw GCM projections. The climate data from the ERA5-Land dataset produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) is collected as the benchmark data [60]. We modified the spatial resolution of the GCM projections to be consistent with the ERA5-Land dataset (0.1°) using the nearest neighbor interpolation method. Then, we employed the ‘linear scaling’ method to correct bias in GCM projections, where the model outputs are added a delta difference or multiplied with a scaling factor to match the reference data. This method is efficient and robust for bias correction of large datasets and can preserve spatiotemporal patterns of the original data [61].

2.3. SWAT Model Setup, Calibration, Validation, and Future Projection

The Soil and Water Assessment Tool (SWAT) is employed to simulate future hydrological processes in the PRB. This model was designed to assess extensive and intricate watersheds over extended periods, considering varying land use, soils, and management practices [62].

This investigation is conducted using different datasets (

Table 1. Datasets for setting up SWAT simulations.

Datasets	Application	Resolution	Source
Terrain (Digital Elevation Model)	Watershed Delineation	250 m	Shuttle Radar Topography Mission
River Network	Stream Definition	-	Global Runoff Data Centre
Land Use/Land Cover	HRU Definition	30 m	China Land Cover Dataset in 2021
Soil	HRU Definition	1 km	Harmonized World Soil Database Version 2.0
Meteorology	Forcing	0.1°	ERA5-Land (for calibration and validation) & CMIP6 (for future projection)
Streamflow Observations	Calibration and Validation	-	Pearl River Water Resources Committee

). Digital elevation model (DEM) data from the Shuttle Radar Topography Mission (SRTM) is employed for delineating watersheds in the PRB [63]. River network data from the Global Runoff Data Centre (GRDC) Major River Basins is used to assist in identifying streams [64]. Land use/land cover (LULC) data and soil database are employed to define the hydrologic response units (HRUs) [65,66]. The ERA5-Land reanalysis climate data is collected for training SWAT and simulating historical streamflow, whereas the CMIP6 climate projections are utilized as input for projecting future streamflow. Historical streamflow records from 1975 to 2023 are collected for the calibration and validation of the model [45]. In this study, we use the three most downstream gauge stations of the WRB, NRB, and ERB for model calibration and streamflow simulation (Figure 1).

Using the input data described above and the SWAT model, we delineate 660 sub-basins and 8838 HRUs for the study area. We use “Sub-basin” to refer to the four different sub-regions of the PRB and “sub-basin” to refer to the watersheds defined by SWAT. We run SWAT at a monthly scale to obtain simulated streamflow for the study area. We split the observed streamflow data into calibration and validation periods, which are 1975–2004 and 2005–2023, respectively. SWAT-CUP Premium and its SPE (Swat Parameter Estimator) algorithm are employed for the SWAT calibration [67]. Based on previous studies [68,69,70,71], 13 parameters are selected to calibrate each of the three upstream Sub-basins (NRB, ERB, and WRB). The objective function of calibration is Nash-Sutcliffe efficiency (NSE). The calibrated model parameters were calibrated for each Sub-basin (

Table 2. SWAT calibration parameters in each Sub-basin.

Parameter	Description	Lower Bound	Upper Bound	Calibrated Parameters
				NRB	ERB	WRB
CN2.mgt *	SCS runoff curve number	−0.2	0.2	0.067	0.081	−0.024
ESCO.hru	Soil evaporation compensation factor	0.01	1	0.132	0.507	0.113
OV_N.hru *	Manning’s “n” value for overland flow	−0.5	6	3.240	3.594	5.948
ALPHA_BNK.rte	Baseflow alpha factor for bank storage	0	1	0.659	0.796	0.845
CH_N2.rte	Manning’s “n” value for the main channel	0.01	0.3	0.258	0.168	0.084
CH_K2.rte	Effective hydraulic conductivity in main channel alluvium (mm/hr)	0	500	95	163	214
GW_REVAP.gw	Groundwater “revap” coefficient	0.02	0.2	0.20	0.16	0.20
GW_DELAY.gw	Groundwater delay (days)	0	500	170	230	38
REVAPMN.gw	Threshold depth of water in the shallow aquifer for “revap” to occur (mm)	0	1000	100	213	50
GWQMN.gw	Threshold depth of water in the shallow aquifer required for return flow to occur (mm)	500	5000	5000	2500	5000
SOL_K.sol *	Saturated hydraulic conductivity (mm/hr)	−0.8	0.6	−0.129	−0.093	−0.793
SOL_AWC.sol *	Available water capacity of the soil layer (mm/mm)	−0.8	0.6	0.293	0.530	0.078
SOL_BD.sol *	Moist bulk density (g/cm3)	−0.8	0.6	0.109	0.535	−0.282

Note: * These parameters were adjusted as a percentage to preserve spatial diversity.

).

After obtaining the calibrated parameters, we validate the model performance using streamflow observations during 2005–2023. After calibrating and validating the SWAT model, we use the 12 sets of post-processed CMIP6 projections as weather inputs to run the SWAT model separately, resulting in 12 sets of future streamflow and runoff projections. When analyzing runoff in this study, we summed the streamflow from outlets and then divided it by the drainage areas to obtain the runoff. It is noted that for our analysis, we have also averaged the results across the three different models within each SSP scenario.

2.4. Evaluation Metrics

In analyzing the model’s accuracy, it is imperative to conduct a thorough evaluation using multiple evaluation metrics [72]. In this study, we choose 5 metrics to evaluate model performance, which are percent bias (PBIAS), RMSE (relative squared error)-observations standard deviation ratio (RSR), Pearson’s correlation coefficient (r), Nash Sutcliffe efficiency coefficient (NSE) [73], and Kling-Gupta Efficiency (KGE) [74]:

$$\mathbf{P}\mathbf{B}\mathbf{I}\mathbf{A}\mathbf{S}\left(\mathbf{\%}\right)=100\times {\displaystyle \frac{\frac{1}{\mathbf{T}}\sum _{\mathit{t}=1}^{\mathbf{T}}{\mathbf{Q}}_{\mathbf{o}}\left(\mathit{t}\right)-{\mathbf{Q}}_{\mathbf{s}}\left(\mathit{t}\right)}{\frac{1}{\mathbf{T}}\sum _{\mathbf{t}=1}^{\mathbf{T}}{\mathbf{Q}}_{\mathbf{o}}\left(\mathit{t}\right)}}$$

(1)

$$\mathbf{R}\mathbf{S}\mathbf{R}=\sqrt{{\displaystyle \frac{\sum _{\mathbf{t}=1}^{\mathbf{T}}{\left({\mathbf{Q}}_{\mathbf{o}}\right(\mathit{t})-{\mathbf{Q}}_{\mathbf{s}}\left(\mathit{t}\right))}^2}{\sum _{\mathbf{t}=1}^{\mathbf{T}}{\left({\mathbf{Q}}_{\mathbf{o}}\right(\mathit{t})-\overline{{\mathbf{Q}}_{\mathbf{o}}})}^2}}}$$

(2)

$$\mathbf{r}={\displaystyle \frac{\sum _{\mathbf{t}=1}^{\mathbf{T}}({\mathbf{Q}}_{\mathbf{o}}\left(\mathit{t}\right)-\overline{{\mathbf{Q}}_{\mathbf{o}}}){(\mathbf{Q}}_{\mathbf{s}}\left(\mathbf{t}\right)-\overline{{\mathbf{Q}}_{\mathbf{s}}})}{\sqrt{\sum _{\mathit{t}=1}^{\mathbf{T}}{({\mathbf{Q}}_{\mathbf{o}}\left(\mathit{t}\right)-\overline{{\mathbf{Q}}_{\mathbf{o}}})}^2\sum _{\mathit{t}=1}^{\mathbf{T}}{({\mathbf{Q}}_{\mathbf{s}}\left(\mathit{t}\right)-\overline{{\mathbf{Q}}_{\mathbf{s}}})}^2}}}$$

(3)

$$\mathbf{N}\mathbf{S}\mathbf{E}=1-{\displaystyle \frac{\sum _{\mathbf{t}=1}^{\mathbf{T}}{\left({\mathbf{Q}}_{\mathbf{o}}\left(\mathit{t}\right)-{\mathbf{Q}}_{\mathbf{s}}\left(\mathit{t}\right)\right)}^2}{\sum _{\mathbf{t}=1}^{\mathbf{T}}{\left({\mathbf{Q}}_{\mathbf{o}}\left(\mathit{t}\right)-\overline{{\mathbf{Q}}_{\mathbf{o}}}\right)}^2}}$$

(4)

$$\mathit{K}\mathit{G}\mathit{E}=1-\sqrt{{(\mathit{r}-1)}^2+{({\displaystyle \frac{\sum _{\mathit{t}=1}^{\mathit{T}}{\left({\mathit{Q}}_{\mathit{s}}\right(\mathit{t})-\overline{{\mathit{Q}}_{\mathit{s}}})}^2}{\sum _{\mathit{t}=1}^{\mathit{T}}{\left({\mathit{Q}}_{\mathit{o}}\right(\mathit{t})-\overline{{\mathit{Q}}_{\mathit{o}}})}^2}}-1)}^2{+({\displaystyle \frac{\overline{{\mathit{Q}}_{\mathit{s}}}}{\overline{{\mathit{Q}}_{\mathit{o}}}}}-1)}^2}$$

(5)

where ${\mathbf{Q}}_{\mathbf{s}}\left(\mathit{t}\right)$ and ${\mathbf{Q}}_{\mathbf{o}}\left(\mathit{t}\right)$ denote the stream simulation and observation (unit: mm/month) at time t; $\overline{{\mathit{Q}}_{\mathit{S}}}$ and $\overline{{\mathit{Q}}_{\mathit{o}}}$ mean the temporal average of simulations and observations. All these metrics are unitless. If simulations match observations perfectly, the PBIAS and RSR should be 0, while r, NSE, and KGE should be 1. When evaluating a time-series simulation, the criteria for a streamflow prediction model to be considered ‘satisfactory’ for each parameter are as follows: −25 < PBIAS (%) < 25, RSR < 0.7, NSE > 0.5, KGE > 0.5 [75,76]. When r > 0.7, it means that the simulations are strongly correlated with observations [77].

2.5. Mann–Kendall and Sen’s Slope Estimator Test

In this study, we employed the Mann–Kendall (MK) test and Sen’s Slope Estimator (SSE) to detect temporal trends in hydro-climate projections data. The MK test is a non-parametric statistical method used to detect monotonic trends without assuming any specific data distribution, which has been widely used in hydrometeorological studies [78,79,80,81].

3. Results

3.1. SWAT Model Performance

Based on the selected evaluation metrics, the SWAT model performs satisfactorily within all Sub-basins during both the calibration and validation periods, except for the PBIAS value (−30.83%) for the West river sub-basin during the calibration period. Underestimation of a few high streamflow events because of the static land use data in SWAT simulations might be the reason for the bias. However, previous studies suggested that this could happen in model calibration [82]. More importantly, the PBIAS for the validation period of the West river sub-basin drops to −6.92%, indicating satisfactory performance of SWAT in simulating streamflow in recent years. In general, the evaluation metrics indicate that the model is capable of reconstructing the magnitudes of streamflow well, giving us confidence about the performance of the SWAT model in simulating future streamflow (

Table 3. Performance of streamflow projections by SWAT during the calibration (1975–2004) and the validation period (2005–2023) in each Sub-basin.

	PBIAS (%)	RSR	r	NSE	KGE
Calibration (1965–1999)
NRB	−2.69	0.65	0.77	0.58	0.70
ERB	−0.60	0.51	0.88	0.74	0.87
WRB	−30.83	0.56	0.92	0.69	0.68
Validation (2000–2023)
NRB	23.04	0.61	0.88	0.63	0.52
ERB	18.07	0.52	0.89	0.73	0.71
WRB	−6.92	0.51	0.88	0.74	0.86

and Figure 2). Uncertainty analysis further indicates the capability of SWAT in simulating the variability of streamflow in the PRB (Figure S2).

3.2. Future Climate Change in the PRB

According to the CMIP6 climate projections, the PRB would experience dramatic changes in precipitation and PET during 2021–2100 in the context of global warming (Figure 3 and Figure S1). All scenarios demonstrate gradually increasing trends. The highest emission scenario (SSP585) shows the highest increasing rates among all scenarios, with about 20 mm/month and 15 mm/month for precipitation and PET, respectively, by the end of the 21st century, compared to the 2020s. The other three SSPs exhibit increases in precipitation and PET of less than 15 mm/month and 10 mm/month, respectively. Except for the SSP370 scenario, the other SSPs show a greater increase in precipitation than those in PET, while such increases in precipitation are comparable to those of PET in SSP370.

Precipitation and PET will be much higher in the late 21st century than in the early decades (Figure 4). Most of the sub-basins illustrate a significantly increasing trend in precipitation in the future (Figure S3a) and all sub-basins among all SSPs demonstrate significantly increasing trends in PET (Figure S3b). The average changing rates of SSP126, SSP245, SSP370, and SSP585 are 0.22, 0.26, 0.14, and 0.30 mm/month/year for precipitation and 0.08, 0.14, 0.16, 0.26 mm/month/year for PET, respectively. The CMCC model demonstrates substantial fluctuations, while the ACCESS model shows relatively lower interannual variability (Figure S4). There is a noticeable difference between the spatial pattern of precipitation and PET changes (Figure 4). The highest increase rate of precipitation is observed in the northeast of WRB and the southern part of NRB, while PET exhibits the highest increase rate in the southern parts of WRB.

3.3. Future Runoff in the PRB

In response to climate change, future runoff in the PRB is projected to increase steadily throughout the 21st century, particularly in the latter half of the century (Figure 5). Our projections indicate that the mean runoff in the PRB will experience an upward trend, with an anticipated increase exceeding 20% by the end of this century. The maximum monthly runoff is also expected to intensify, with a projected increase of approximately 50 mm/month in the late 21st century relative to historical levels. The temporal trend of runoff in each SSP is highly consistent with that of precipitation, while its relative magnitude among the four SSPs illustrates a different pattern from that of precipitation (Figures S4 and S5). Before the 2070s, at the decadal level, runoff does not show evident differences among the four SSPs. During the period from the 2070s to the 2090s, higher runoff can be found in the SSP126 compared to other SSPs, while SSP585 will lead to the highest runoff at the end of the 21st century.

Most of the sub-basins in SSP126 and SSP245 show significant increasing trends in runoff, while less sub-basins in SSP370 (416 out of 660) and SSP585 (473 out of 660) show increases in runoff (Figure 6). The changing rates of SSP126, SSP245, SSP370, and SSP585 are 0.21, 0.20, 0.11, and 0.17 mm/month/year, respectively. Although the increasing rate in precipitation of the SSP585 scenario is the highest among all scenarios, the high increasing rate in PET of this scenario will cancel out the increases in precipitation, leading to less significant increases in runoff (Figure 3). The spatial distributions in these scenarios demonstrate inconsistencies with those of precipitation and PET. In most sub-basins of the WRB, significantly increasing trends are found in all scenarios, with the highest changing rate in the northwest of the WRB found in the SSP585 scenario, and relatively higher changing rates in the southwest are found in the other three scenarios. In the NRB and ERB, SSP126 and SSP245 result in significant increasing trends, while SSP370 and SSP585 do not lead to significant trends of changes in most areas of these two sub-basins. Additionally, the increases in runoff in the NRB are greater than that of the ERB.

Trends of the projected runoff also show seasonal variability in all scenarios (Figure 7). Future runoff in the PRB tends to experience a larger increase in the late wet season (i.e., July to September), and such a trend is also found in precipitation in these months (Figure S6a). Under the SSP585 scenario, the hydro-climate response to climate change is more pronounced, exhibiting the highest rate of increases in precipitation, exceeding 0.7 mm/year, and a corresponding increase in runoff of more than 0.3 mm/year during these months. In addition, a lower increasing rate and no significant trend can be seen in the late dry season and the early wet season (e.g., January to May), while a more significant increasing trend of PET is found in this period (Figure S6b). During this period, the SSP370 and SSP585 scenarios both experience the greatest number of months with no significant increasing trends in future runoff. Notably, these two scenarios exhibit the most pronounced changes in evapotranspiration, with PET in the SSP585 scenario exceeding an increase of 0.4 mm/month/year, and the SSP370 scenario showing an approximate increase of 0.25 mm/month/year. Responses of other water fluxes (e.g., percolation, groundwater recharge, and total water yield) to climate change can be found in Figure S7.

3.4. Maximum and Minimum Runoff Under Future Climate Conditions

We counted the maximum and minimum monthly runoff for each year of the SWAT projections for the period 2021–2100 and also analyzed the spatial and temporal trends. Maximum runoff tends to increase across most parts of the PRB, but the increasing rates demonstrate large spatial variability (Figure 8). Under the four scenarios (SSP 126, SSP245, SSP370, and SSP585), there are significant increases in maximum runoff across 72.9%, 87.0%, 46.7%, and 70.0% of the 660 sub-basins, respectively. The highest emission scenarios (SSP585) will increase the maximum runoff by up to 0.47 mm/month/year, followed by SSP245 and SSP126 at 0.43 and 0.38 mm/month/year, respectively. However, for the SSP370 scenario, the increasing rate will be less significant than in other scenarios, with a rate of 0.21 mm/month/year. Most sub-basins of the WRB will experience significant increases in maximum runoff across all SSPs, with the highest increase found in the northeastern parts of the sub-basin. The NRB and ERB exhibit increasing trends in SSP126 and SSP245 but show no significant increases under the other two scenarios (SSP370 and SSP585).

Compared to the first half of the century (2021–2060), high-flow events in the late 21st century (2061–2100) are projected to increase by about 15% (Figure 9). During the first half of the century, the four SSPs will lead to similar magnitudes in high-flow events. However, in the late 21st century, the highest runoff magnitudes are projected under SSP585, which are approximately 20 mm/month higher than the second-highest projections under SSP126 and SSP245 and about 50 mm/month higher than the lowest projections under SSP370.

In contrast to the future trends for total and maximum runoff, the projections of annual minimum monthly runoff identify fewer regions exhibiting a statistically significant increasing trend, and decreasing trends are even found in some sub-basins in scenarios of SSP370 and SSP585 (Figure 10). In the low-emission scenarios SSP126 and SSP245, most sub-basins (441 and 287, respectively) demonstrate significant increasing trends in minimum runoff. However, in the two high-emission scenarios (SSP370 and SSP585), 104 and 130 sub-basins demonstrate decreasing trends, outnumbering the sub-basins (64 and 45, respectively) exhibiting increasing trends. Spatial distributions of sub-basins with significant reductions in minimum runoff are also consistent in these two scenarios, covering a vast area across nearly the entire ERB and the southeastern corner of the WRB. In addition, SSP585 exhibits a greater rate of decline in runoff in areas compared to that of SSP370.

Upon comparing the lowest range of runoff in the ERB between the late and mid-21st century (Figure 11), it is evident that under low-emission scenarios (e.g., SSP126 and SSP245), there are significant increases of approximately 10% in runoff. Conversely, under high-emission scenarios (e.g., SSP370 and SSP585), there are more pronounced decreases of roughly 20%. By the late 21st century, it is projected that SSP370 and SSP585 will yield approximately 5 mm less of the minimum runoff in the dry months compared to SSP126 and SSP245.

4. Discussion

4.1. Climate Change Impacts on the Terrestrial Water Cycle

Our investigation suggests that future runoff will increase as a result of warming temperatures and more precipitation in the PRB. Such findings are consistent with other studies conducted at a regional scale, which also projected increased runoff due to more precipitation in future climates [83,84,85]. In addition to precipitation, we also highlight the impacts of warming temperatures on runoff. Due to the increased water loss through ET induced by higher temperatures, higher precipitation of the high-emission scenario (SSP585) compared to other scenarios (SSP126 and SSP245) does not lead to more runoff. This finding is consistent with Guan et al. [40], which found that higher ET under warmer temperatures should be considered when evaluating climate change impacts on the terrestrial water cycle.

Our results are generally consistent with previous studies in the Pearl River Basin that reported intensified hydrological variability under climate change. The climate condition of our study area is dominated by two interacting climate systems: the Southwest Monsoon system and the Southeast Monsoon system. Upper streams of West River Sub-basin are mainly affected by the Southwest Monsoon system, which transport moisture from the Indian Ocean to the basin, whereas the other two sub-basins receive the precipitation transported by the Southeast Monsoon system from the Pacific Ocean. The GCM climate conditions indicate that more precipitation will be induced by the enhanced moisture transport of the Southwest Monsoon in western parts of the PRB, while reduced rainfall will occur because of the weaker Southeast Monsoon system in the East River Sub-basin, particularly under the SSP370 and SSP585 scenarios.

In particular, similar to Yan et al. [46], our results indicate wetter wet seasons and drier dry seasons, implying increased flood risk and aggravated drought stress. In addition, consistent with Liu et al. [47], we find significant warming and notable changes in runoff seasonality, with enhanced evapotranspiration under high-emission scenarios intensifying dry-season water stress, despite relatively weak long-term trends in precipitation. Under low-emission scenarios, however, precipitation increases may partially offset water loss through evapotranspiration, resulting in relatively stable minimum runoff.

Our findings also agree with previous studies concluding that changes in precipitation and ET have more direct impacts on the water cycle than other processes [86,87,88]. Warming temperatures enhance evapotranspiration and increase the humidity of the air, and accelerate snowmelt processes [89], speeding up the hydrological cycle and spatiotemporal changes in runoff [12,18]. These climatic variations then alter the water exchange between the atmosphere and the land surface system, which influences not only surface streamflow but also impacts soil moisture dynamics and groundwater patterns [90,91]. The model simulations conducted in this study corroborate those findings and suggest an intensified water cycle will occur in the PRB under future climate change scenarios.

4.2. Implications for Water Resource Management in the PRB

The results of our investigation provide valuable information for the management of water resources in the PRB. More frequent floods and droughts under a changing climate are expected to compound hydrological hazards in the densely populated Great Bay Area, which has been struggling to manage water resources for competing users. The altered spatial and s patterns of runoff projections shown in our simulations highlight the necessity of considering future climate change in water allocation. Under the two high-emission scenarios (SSP370 and SSP585), the western PRB tends to have higher runoff, while the northern and eastern regions would have decreased runoff. Another future change is the seasonal patterns in runoff. Under the two high-emission scenarios, runoff tends to increase more in summer than in spring, modifying runoff distributions in different months of the year. Future water resources management needs to prepare for those enlarging spatial and temporal disparities in water resources.

Hydraulic engineering is an effective way for humans to manage water resources [92]. In the PRB, a large number of dams and reservoirs were built to regulate flow and prevent flood hazards [93]. However, our results indicate that future increases in runoff could be unprecedented. Whether the facilities designed based on historical hydrological conditions could remain functional needs to be evaluated in future water resource management [94,95]. In addition, future design of hydraulic facilities needs to consider potential changes in climate conditions to increase their resilience [96,97]. Nature-based solutions, such as preserving and restoring mangroves and wetlands, could help attenuate flood risks. What is more, non-engineering solutions, such as water forecasting, will enable the public to be better prepared for hazards like flooding and heatwaves.

Although the PRB has a subtropical climate with high precipitation, it still struggles to provide sufficient water to support the large population, particularly in the downstream delta area, where the total population exceeds 80 million. Unfortunately, our projection indicated that drought risks will increase in ERB. Transferring water from areas with rich water resources to regions lacking water might be a solution to meet the increasing demand of the large population under the changing climate [98,99]. The recently implemented Pearl River Delta Water Resources Allocation Project is an example showing that hydraulic projects can be used to address challenges of water supply to support the increasing population [100]. Results of this study show that the eastern and northern parts of the PRB, which have large areas of agricultural lands, may also need such projects to mitigate potential droughts in those regions in the future. In addition, since the population of the Guangdong–Hong Kong–Macau Bay area is expected to further increase in the future, water resource planning for drinking water supply, the design of urban drainage systems, and the management of water pollution may also need to take climate change impacts into consideration.

4.3. Increasing Hydrological Risks Under a Changing Climate

Flooding has been a primary hydrological hazard in the PRB. Our investigation indicates that precipitation would lead to increased runoff across most parts of the PRB, leading to elevated flooding risks. This finding aligns with investigations from other studies that suggest rainfall-induced flooding is expected to become more severe in various regions in a changing climate [34,101]. Climate change enhances the atmosphere’s capacity to hold more moisture and alters the spatial and temporal distributions of water, contributing to an increasing trend in extreme precipitation events and thus a higher risk of the flood [102,103,104].

Unlike many studies establishing the relationship between extreme rainfall and flooding from a statistical perspective, this study conducts a spatiotemporal quantification of future runoff changes and identifies regions that are likely to face increased flood risks under various climate change scenarios. Our findings quantify the impact of climate change on flood risk and indicate that the maximum monthly runoff in certain sub-basins of the western PRB could increase by more than 50%. This information can provide valuable implications for the formulation of more precise flood management policies.

Although projections indicate that the PRB will generally experience increased runoff in the future, our research reveals that under high-emission scenarios (SSP370 and SSP585), there will be reductions in runoff during the dry seasons. This finding serves as a warning that even with abundant precipitation, attention must still be paid to the potential hydrological drought associated with the increased evapotranspiration under global warming. Previous studies also indicated that higher emission scenarios with greater warming potentials may lead to more intensive and frequent drought [105,106]. Previous research has primarily focused on regions with insufficient rainfall. However, our study projects that in the eastern part of the PRB, an area with annual precipitations close to 2000 mm, the minimum runoff could decrease by 20% by the end of the 21st century under the SSP370 and SSP585 scenarios. This projection indicates that this region needs to prepare not only for the gradually increasing risk of floods but also for the potential drought risks under high-emission scenarios.

4.4. Caveats and Future Work

Admittedly, our investigations are subject to uncertainties that need to be constrained to improve our understanding of hydrological responses to the changing climate. First, Hydrological processes are highly sensitive to the climatic driving forces, and the results and conclusions of this study are mainly based on the selected GCMs. Considering the large uncertainties in climate projections, it is necessary to include more GCMs, and use more ensemble members to provide a more comprehensive representation of climate uncertainties in the future.

Second, uncertainties of SWAT parameters should also be further evaluated. In this study, we conduct extensive model calibration by trying hundreds of parameter value combinations and choose the one fitting observations best to project future changes in hydrological processes (Figure S2). However, it is acknowledged that using a single set of parameter values could not represent uncertainties in the parameters well. This could be resolved by employing more parameter value combinations to conduct ensemble hydrological projections.

Third, in this study, we focus on the direct impacts of meteorological changes on key water fluxes (e.g., precipitation, ET, and runoff), while neglecting the indirect impacts of climate change on the water cycle [107]. For example, the plant physiology, distribution of plant species, and plant community structures are all sensitive to climate change [108,109]. How those changes will affect the hydrological cycle and need to be investigated to better understand the responses of terrestrial hydrology to climate change.

Fourth, human activities and their impact on the water cycle are not explicitly considered in this investigation. Dams and reservoir operations substantially alter the natural streamflow regimes [110,111]. Therefore, investigating the impact of anthropogenic activities on hydrology is crucial for reliably reproducing temporal patterns of streamflow, particularly for the simulation at fine temporal scales (e.g., hourly and daily). Although the simulated monthly streamflow in this study still produces reasonable seasonal changes, as evidenced by the satisfactory model performance in the calibration and validation periods (Table 3), we believe a better representation of reservoir operations could further increase the accuracy of streamflow simulations, especially at shorter time scales.

In addition, we use static land use data to evaluate future hydrological changes. However, urbanization, deforestation, and expansion of agricultural lands are likely to occur and will further alter runoff generation and ET [112,113]. Other human activities, such as water withdrawal for irrigation and drinking water supply, may also change substantially in the future, affecting many hydrological processes [114,115,116]. Therefore, evaluating the impacts of such human activities, along with the impacts of climate change, will improve our understanding and prediction of how the future water cycle will vary in response to multiple disturbances.

5. Conclusions

In this study, we employ the SWAT model to project hydrologic responses to future climate changes and the resultant hydrological risks in the PRB under different emissions scenarios. Our investigations indicate that runoff in the PRB will increase due to the concurrent increase in precipitation and evapotranspiration, causing more net water input into the basin. Higher GHG emissions tend to result in more precipitation and higher PET, resulting in varying increases in runoff among different climate change scenarios, i.e., increasing rates of 0.21, 0.20, 0.11, and 0.17 mm/month/year for SSP126, SSP245, SSP370, and SSP585 scenarios, respectively.

Our study also indicates the increasing risk of hydrological hazards in the PRB. In the future, the PRB is projected to face more severe extreme precipitation events, which could lead to more intense extreme streamflow and increasing flood risks. By the end of the 21st century, there will be an increase of approximately 15% in the runoff during the wettest month compared to the early to the first half of the 21st century. Under the high-emission scenarios, specifically SSP370 and SSP585, the runoff of eastern PRB during the driest month will decrease by approximately 20% by the end of the 21st century than the early to mid-century levels.

Spatial and temporal patterns in future runoff projected by this study can provide implications for the future management of valuable water resources to support China’s most important economic zone in achieving sustainable development goals. Hydrological risks in the context of climate change highlighted in this study suggest the necessity of preparing for unprecedented hazards in the future.