Evaluation of spatiotemporal variation and driving forces of baseflow coupled with CMIP6 and LUCC in the source region of the Yangtze River, China

Global climate change is intensifying regional and global water cycles, and may significantly impact fragile socioeconomic and ecological environmental systems (Zhang et al. 2017). The water cycle is impacted by various factors, such as the climate, land use, and human activities, and its evolution exhibits complex coupling characteristics with high uncertainty and high randomness (Jian et al. 2023; Zeng et al. 2024). By combining different shared socioeconomic pathways (SSPs) with representative concentration pathways and incorporating the impact of socioeconomic development, Coupled Model Intercomparison Project Phase 6 (CMIP6) was released in 2021/2022, providing important data support for future climate-change projections and evolution mechanisms (O'Neill et al. 2016; Güven et al. 2024). The spatial resolution and physical parameters of CMIP6 are better optimized than those of CMIP5 (Eyring et al. 2016; Meinshausen et al. 2017). Anthropogenic activities, particularly land use, have an appreciable influence on the hydrological cycle. Therefore, research has begun to focus on the hydrological processes of basins under the combined impact of land use and land cover change (LUCC), and climate change (Da Silva Cruz et al. 2022). Owing to the interaction between climate change and human activities, global land use has undergone substantial changes (Gao et al. 2024). Land use change simulation models are widely used to project future regional land use. Many studies have used the Conversion of Land Use and its Effects at Small regional extent (CLUE-S), Cellular Automata (CA), and Future Land Use Simulation (FLUS) models to research land-use pattern evolution, hydrology, and water resources (Qi et al. 2024; Qiao et al. 2024). The CMIP6 and FLUS models provide highly credible simulation ranges to investigate climate change and LUCC (Guo, H. et al. 2022; Kang et al. 2023; Wang et al. 2024), providing a basis for analyzing the spatiotemporal evolution and driving mechanisms of hydrological components under changing environmental conditions.

Baseflow, which comprises previous rainfall and is recharged via slow interflow, is a considerable component in the hydrological cycle of a basin, providing a sustainable water source for rivers and thereby maintaining life in and around them (Guo, X. et al. 2022). Baseflow is also the most stable source of river runoff, particularly during periods of drought or low precipitation, and is essential for maintaining the ecological environment of rivers (Ficklin et al. 2016; Shao et al. 2022). Therefore, it is critical to identify the spatiotemporal evolution characteristics of baseflow to understand the subsurface water cycle and the pattern of evolution in changing environments. Following increased recognition of the significance of baseflow in the water cycle of rivers, researchers have begun to examine the dynamic characteristics of baseflow and the factors that influence it (Hu et al. 2017; Waterman et al. 2022; Li et al. 2023). Sun et al. (2021) analyzed seasonal and spatial variations in baseflow and identified the factors contributing to these changes. Ahiablame et al. (2017) investigated the spatiotemporal change of annual baseflow in the Missouri River Basin, considering the impact of climate change and agricultural land use. Mao et al. (2024) studied spatiotemporal changes of baseflow at the basin scale using the Hydrograph Separation Program (HYSEP), United Kingdom Institute of Hydrology (UKIH), and digital filter methods. Waterman et al. (2022) analyzed the spatiotemporal changes in runoff and baseflow in an interregional precipitation gradient basin. Wu et al. (2019) used the elastic coefficient method to identify the factors driving changes in baseflow on the Loess Plateau, finding that the fluctuations in the baseflow index (BFI) were mainly dominated by climatic factors. Adnan et al. (2022) employed hydrological simulations to reveal the spatiotemporal characteristics of the baseflow of the Gilgit River and quantified the average relative contributions of snow meltwater, glacier meltwater, and rainfall runoff to the baseflow. Yang et al. (2020) identified the response of baseflow in different geomorphology types in the midstream of the Yellow River to ecological construction and climate change. There were positive correlations between baseflow and precipitation, snowfall, sediment concentration, surface slope, elevation, grassland proportion, and wasteland but negative correlations between baseflow and temperature, potential evapotranspiration, silt and clay content, agriculture proportion, and shrubland (Rumsey et al. 2015; Huang et al. 2021).

The source region of the Yangzte River (SRYR) is located in Zaduo County, Qinghai Province, which is characterized by high altitudes and low temperatures. It is often referred to as the ‘Changjiang Water Tower’ owing to its importance as a water resource and a water ecological security barrier (Xu et al. 2019). Recently, the SRYR has experienced considerable impacts from global warming and anthropogenic activities (Cui et al. 2023). These changes have resulted in altered hydrological processes, which could have detrimental effects on the ecosystem (Rounce et al. 2023). The baseflow, which is the primary source of runoff in the SRYR, plays a critical role in sustaining this region's ecological environment (Wu et al. 2024). Wu et al. (2024) used the Budyko and correlation analysis method to investigate the dynamic characteristics and influencing factors of baseflow in the SRYR. Previous research has primarily focused on temporal analysis of baseflow characteristics (Shao et al. 2022; Li & Fan 2023), and few studies have attempted to explore the spatial variability of baseflow and its future change. The distributed hydrological model offers new opportunities for research on the spatialization and quantification of baseflow. Some scholars have employed the Soil and Water Assessment Tool (SWAT) model to examine hydrological processes in the SRYR. In a prior study conducted by Luo (2021), a SWAT model was implemented in the SRYR to quantitatively identify the response characteristics of hydrological processes to ecological engineering. Yuan, Z. et al. (2019) utilized the SWAT model to estimate the spatiotemporal distribution characteristics of water resources in the SRYR, while Ahmed et al. (2022) applied the model to simulate the runoff process in the SRYR and quantitatively analyze the impact of climate change and LUCC on streamflow. These findings indicate that the SWAT model is an effective tool for hydrological simulation in the SRYR and can be widely used.

The construction of a SWAT model requires the preparation of terrain, land use pattern, soil property, and hydrometeorological data. Meteorological data were acquired from the Qinghai Hydrology and Water Resources Forecasting Centre (QHHWC), and hydrological data were collected from the Zhimenda gauging station. The aforementioned data are presented in Table 1. It is noted that the spatial distribution characteristics of land use in the SRYR in 1980, 1990, 2000, 2010, and 2020 can be found in the Supplementary Material.

In our study, nine CMIP6 model datasets were selected as SWAT model inputs (Table 2). Four SSPs, namely ssp126, ssp245, ssp370, and ssp585, were selected to represent different socioeconomic development path models (Jian et al. 2023). Bilinear interpolation and quantile mapping (QM) methods were used to downscale and correct the model data. Finally, the multi-model ensemble (MME) method was used to further improve the accuracy of model prediction.

LUCC is affected by anthropogenic activity and natural events. A series of factors related to the natural ecology and social economy, including digital elevation model (DEM), slope, aspect, precipitation, temperature, water distribution, gross domestic product (GDP), population density, and distance to roads, was selected as driving factor data for the FLUS model (Table 3). The spatial distribution of these main driving forces of LUCC can be found in the Supplementary Material.

The SWAT model has been widely applied to carry out hydrological simulations in cold regions based on physical mechanisms, as it can simulate the snowmelt runoff process well (Shukla et al. 2021; Zhao et al. 2022; Biswas & Biswas 2024). It performs distributed hydrological process simulation by dividing the basin into sub-basins and hydrological response units (HRUs) (Zhu et al. 2022). In our study, the SRYR was divided into 31 subcatchments and 305 HRUs. The land use patterns were categorized into six major types, and the soil characteristics were divided into 13 categories. We completed the performance evaluation of the SWAT model by using the coefficient of determination (R²) and the Nash–Sutcliffe efficiency (NSE) (Chen et al. 2023).

The Theil–Sen median is a non-parametric trend calculation method that provides a robust estimation of the data trend. The Mann–Kendall (MK) method is a non-parametric method suitable for analyzing trends in meteorological and hydrological series with a non-normal distribution, and is not affected by a few interference values (Farris et al. 2021). The Theil–Sen median has high calculation capacity and is not sensitive to measurement errors and outliers. It is applicable for the trend analysis of long-term hydrometeorological data series. In our study, Theil–Sen was employed to calculate the time-series trend value, and MK to detect its significance (Da Silva et al. 2015; Xue et al. 2022).

For ⁠, the time series data has a notable trend if ⁠.

The FLUS model was proposed by Liu et al. (2017) to simulate future land use pattern changes based on a comprehensive consideration of natural and human factors. The FLUS model couples back-propagation artificial neural networks and roulette selection to enhance the performance of various LUCC modeling techniques. In our study, the FLUS model was applied to predict the land use spatial distribution of SRYR in 2040. First, based on historical data, the Markov model predicts the numbers of various future land use patterns in the SRYR. Second, taking the influence of various driving forces, including precipitation, temperature, terrain, distance to roads, and GDP, into account, the FLUS model is applied to estimate the general transformation probability of a cell. Then, the transformation type of the cell is determined, and the future land use spatial pattern of SRYR is obtained.

To estimate the land use spatial pattern of the SRYR in 2040 based on the land use pattern in 2020, the Markov method was applied to forecast the requirement of a variety of land use patterns in 2020. The land use spatial distribution pattern of the SRYR in 2020 was obtained through parameter setting and execution of CA to estimate the type of cellular transformation. The estimated results were compared with the actual pattern in 2020. Under the condition that the kappa coefficient was greater than 0.7, based on the baseline data in 2020, the FLUS model was used to estimate the land use spatial distribution in 2040.

The trends of annual precipitation, temperature, snowmelt, potential evaporation, baseflow, and BFI, and whether these trends were significant (α = 0.05) are presented in Table 4. The results indicated a significant rising trend in precipitation and temperature, whereas other factors showed an insignificant increasing trend.

The intra-annual distribution data showed that precipitation, temperature, and baseflow followed a unimodal distribution pattern, with maximum values occurring in July. In contrast, BFI and snowmelt displayed a bimodal distribution pattern. The BFI exhibited a downward trend from January to April, followed by a subsequent increase, reaching a first peak in May, a decrease again until June, and a steady increase culminating in a second peak in November. In contrast, snowmelt presented a rising trend from January to April, with a first maximum occurring in May, followed by decreasing and increasing trends with a second maximum in October.

In addition, the baseflow and BFI exhibit a robust spatial correlation with LUCC. As demonstrated in Figure 15(b), the correlation coefficients between baseflow and grassland and water are 0.83 and 0.95, respectively, while those of BFI are 0.72 and 0.69, indicating that LUCC significantly influences baseflow change. Our findings revealed that higher baseflow and BFI values were observed in the upstream region of the Dangqu River. The Chadan Wetland, the largest peat wetland in the SRYR, is situated in the upstream portion of the Dangqu River. Wetlands serve as natural storage reservoirs for groundwater (Erwin 2009), which could potentially contribute to the observed higher baseflow and BFI values in the region. The land use type also exhibits a significant influence on the spatial characteristics of baseflow and BFI. The highest BFI occurred in the Tongtian River because of its good vegetation coverage, which reached a grassland coverage of 89%. The lowest BFI values throughout the SRYR occurred for the Chumaer River. This area is mainly covered by bare land, with a coverage of 48%. Moreover, the compacted or exposed soils in the upper Chumaer River can limit the infiltration capacity of water into the ground (Owuor et al. 2016). Overall, the spatiotemporal distributions of baseflow and BFI in the SRYR were impacted by a variety of factors, such as precipitation, temperature, human activities, vegetation coverage, and terrain.

Undoubtedly, the prediction of meteorological factors is largely uncertain due to limitations in climate models (Thomas et al. 2023). To address this, bias correction and the MME method are commonly used to reduce uncertainty in CMIP6 model data. In our study, the QM and MME techniques effectively decreased the uncertainty of the CMIP6 model's downscaling data at each meteorological station. However, the downscaling and bias correction methods employed in this research were statistical in nature and did not consider the impact of spatial factors on climate models. Future studies should be conducted to further explore alternative spatial dynamic downscaling methods to overcome this limitation. Moreover, the FLUS model utilized only nine driving forces to estimate land use patterns in 2040, neglecting snow cover and frozen soil, which introduced uncertainty into the simulation results. Although the simulation accuracy in this study was high, it did not account for the influences of glaciers and permafrost, which also contributed to the uncertainty. Although the SWAT model demonstrated strong simulation accuracy, it similarly failed to consider the impact of glaciers and frozen soil, leading to uncertain results. To enhance model adaptability and improve land use predictions, future research should incorporate additional driving factors and apply multiple hydrological models for comparative analysis.

Our study proposes a coupled framework, which combines the SWAT model with CMIP6, FLUS, and two-parameter digital filtering to evaluate the spatiotemporal evolution and response mechanism of baseflow in the SRYR. The principal conclusions are as follows.

• (1) The changes in land use patterns were minimal before 2000, and the main change, particularly the transition from bare land to grassland in the mid-lower reaches of the Tongtian River and the transition from grassland to bare land in the upper reaches of the Tuotuo River, occurred from 2000 to 2010.

• (2) Temporally, baseflow exhibits a significant upward trend under ssp126, ssp245, and ssp370 from 2015 to 2100, and in contrast, an insignificant upward trend occurred under ssp585. Baseflow exhibits a unimodal intra-annual distribution pattern, and BFI presents a bimodal distribution pattern.

• (3) Spatially, baseflow increases gradually from northwest to southeast. Under all scenarios, the maximum baseflow values principally occurred in the downstream region of the Tongtian River. Baseflow values under ssp126 and ssp245 are greater than those under ssp370 and ssp585. Under all scenarios, the Tongtian River exhibited the highest BFI, followed by the Dangqu, Tuotuo, and Chumaer rivers. In comparison, the BFI values of the Dangqu River and the Tuotuo River under ssp370 and ssp585 are lower than those under ssp126 and ssp245.

• (4) The spatiotemporal distributions of baseflow and BFI are impacted simultaneously by a variety of factors, such as precipitation, temperature, human activities, vegetation coverage, and terrain.

In summary, our findings provide important insight for the analysis of spatiotemporal evolution and the driving mechanisms of water resources in the SRYR, and can also serve as a guide for similar studies in other watersheds.

Conceptualization, methodology, and software: H.R., L.S., G.W., and W.T.; formal analysis, validation, and investigation: H.R. and L.S.; resources and data curation: H.R., X.Y., F.N., M.W., and M.Z.; writing – original draft preparation: H.R.; writing – review and editing: W.T., F.N., M.Z., and N.J.; visualization: H.R. and N.J.; supervision: L.S.; project administration and funding acquisition: G.W. and H.R. All authors have read and agreed to the published version of the manuscript.

This work was supported by the National Key Research and Development Programs of China (2022YFC3201700), the Fundamental Research Funds for Central Public Welfare Research Institutes (CKSF20241012/SZ), the Major Science and Technology Projects of the Ministry of Water Resources in 2022 (SKS-202239), and the National Natural Science Foundation of China (52009006).

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.