Preamble

Influence of Varied Drought Types on Soil Conservation Service Within the Framework of Climate Change: Insights from the Jinghe River Basin, China

BAI Jizhou¹, LI Jing¹*, RAN Hui¹, ZHOU Zixiang², DANG Hui¹, ZHANG Cheng¹, YU Yuyang¹

¹College of Geomatics, Xi'an University of Science and Technology, Xi'an 710054, China
²[Affiliation not fully specified in original text]

Abstract

Severe soil erosion and drought are the two main factors affecting ecological security on China's Loess Plateau. Investigating the influence of drought on soil conservation service is crucial for regional environmental protection and sustainable development. However, research on their coupling relationship remains limited. This study focuses on the Jinghe River Basin as a case study, conducting quantitative evaluations of meteorological, hydrological, and agricultural droughts (represented by the Standardized Precipitation Index (SPI), Standardized Runoff Index (SRI), and Standardized Soil Moisture Index (SSMI), respectively) using the Variable Infiltration Capacity (VIC) model, and quantifying soil conservation service using the Revised Universal Soil Loss Equation (RUSLE) for historical (2000–2019) and future (2026–2060) periods under two Representative Concentration Pathways (RCPs) (RCP4.5 and RCP8.5). We further examined drought influences on soil conservation service at annual and seasonal scales. The NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP) dataset was used to predict and model hydrometeorological elements in the future period under the RCP4.5 and RCP8.5 scenarios.

The results showed that during the historical period, annual-scale meteorological drought exhibited the highest intensity, while seasonal-scale drought was generally weakest in autumn and most severe in summer. Drought intensity for all three types will increase over the next 40 years, with a greater increase under the RCP4.5 scenario than under RCP8.5. Furthermore, the intra-annual variation in drought intensity becomes smaller under both future scenarios relative to the historical period (2000–2019). Soil conservation service exhibits a distribution pattern characterized by high levels in the southwest and southeast and lower levels in the north, which has remained consistent across both historical and future periods. Over the past 20 years, intra-annual variation indicated peak soil conservation service in summer and lowest levels in winter; total soil conservation in the Jinghe River Basin displayed an upward trend, with the 2019 value being 1.14 times higher than in 2000. The most substantial impact on soil conservation service arises from annual-scale meteorological drought, which remains consistent across both historical and future periods. Additionally, at the seasonal scale, meteorological drought exerts the highest influence on soil conservation service in winter and autumn, particularly under the RCP4.5 and RCP8.5 scenarios. Compared to the historical period, soil conservation service in the Jinghe River Basin will be significantly more affected by drought in the future period in terms of both affected area and magnitude of impact.

This study provides beneficial attempts to evaluate and predict the dynamic characteristics of watershed drought and soil conservation service, as well as the response of soil conservation service to different drought types. Clarifying the interrelationship between the two is the foundation for achieving sustainable development in relatively arid and severely eroded areas such as the Jinghe River Basin.

Keywords: meteorological drought; hydrological drought; agricultural drought; soil conservation service; Variable Infiltration Capacity (VIC) model; Revised Universal Soil Loss Equation (RUSLE); Jinghe River Basin

1 Introduction

The frequent occurrence of extreme climate under global warming disrupts water resource distribution, posing formidable threats to the global ecological environment, particularly in arid and semi-arid areas (Leal et al., 2021). Drought is characterized by an imbalanced distribution of water resources resulting from insufficient precipitation in a region (Yang et al., 2005). Within the context of climate change, drought has inflicted incalculable and potentially irreversible harm on ecosystems (Fensham et al., 2009). The influence of drought on ecosystems is far-reaching and has gradually become a research focus (Gampe et al., 2021).

Drought alters the structural and functional attributes of dryland ecosystems, including microbial communities, plant productivity, and nutrient cycling processes (Huang et al., 2015). Berdugo et al. (2020) evaluated 20 ecosystem functions and attributes responsive to worsening drought, finding that drought led to an abrupt decline in various ecosystem attributes, including vegetation productivity, soil fertility, vegetation coverage, and vegetation richness. Drought is considered the most significant hazard resulting from climate change, necessitating urgent drought monitoring and assessment (Carle, 2015). The drought index represents the most commonly utilized method for drought monitoring (Cao et al., 2023) and can be categorized into meteorological, hydrological, agricultural, and socio-economic indices depending on drought type (Maity et al., 2016).

The meteorological drought index primarily relies on precipitation, evapotranspiration, and other data to characterize water scarcity resulting from insufficient rainfall (Zhou et al., 2013), such as the Standardized Precipitation Index (SPI). The hydrological drought index relies on calculations of runoff and groundwater from hydrological models such as the Soil and Water Assessment Tool (SWAT) to depict decreases in rivers or reservoirs resulting from insufficient surface water or groundwater (Zhang et al., 2019), such as the Standardized Runoff Index (SRI). The agricultural drought index is mainly related to factors such as soil moisture and vegetation coverage, describing situations where crops cannot grow normally due to insufficient soil moisture (Pan et al., 2023), such as the Standardized Soil Moisture Index (SSMI). The socio-economic drought index describes water scarcity resulting from human activities (Kimwatu et al., 2021), such as the Socio-economic Drought Index (SEDI).

Scholars have constructed hundreds of drought indices, but common indices only target specific drought types, and significant differences exist in statistical methods and applicable spatiotemporal scales (Wang et al., 2022). Consequently, it is imperative to explore methods for monitoring multiple drought types simultaneously in a region.

In addition to drought, severe soil erosion represents another significant threat in arid and semi-arid areas (Terwayet Bayouli et al., 2003). Climate change has increased drought frequency, thereby exacerbating soil erosion (Ciampalini et al., 2020). Soil conservation service refers to the ability of ecosystems to control soil erosion and retain sediment (Masroor et al., 2022). As an essential regulating service in ecosystems, it provides a healthy environment for soil formation and plant growth and represents an important guarantee for preventing land degradation and reducing flood risk (Liu et al., 2019). Currently, a unified quantitative indicator in the academic community quantifies soil conservation service as the difference between potential (maximum soil erosion without surface vegetation and soil conservation measures) and actual soil erosion (Zheng et al., 2021). The concrete calculation process is mostly based on soil erosion models such as the Revised Universal Soil Loss Equation (RUSLE), Integrated Valuation of Ecosystem Services and Trade-offs (InVEST), and SWAT models. Among these, the RUSLE model is the most widely used (Olorunfemi et al., 2020). It contains the basic elements involved in the soil erosion process and is suitable for steep, undulating terrain, making it advantageous for soil conservation assessment.

Drought has produced varying degrees of impact on different ecosystem services (Pravalie et al., 2014) and has resulted in declining quality of ecosystem services provided to humans (Berdugo et al., 2020). However, limited research exists on drought effects on soil conservation service. Han et al. (2019) applied Pearson's correlation coefficient to examine drought effects on freshwater ecosystem services at different time scales, finding that inter-annual and seasonal droughts decreased water yield and soil conservation in Guizhou Province, China. Bai et al. (2021) employed grey relation analysis to investigate relationships between soil conservation service and drought, vegetation, and other factors, showing that soil conservation service was closely related to drought, followed by vegetation. In summary, existing research mainly concentrates on qualitative analysis of the drought-soil conservation relationship, lacking precise quantitative expression, and prediction research under future scenarios requires exploration.

The Loess Plateau in China is highly susceptible to climate change impacts and possesses a fragile ecological environment, with the Jinghe River Basin being typical. Is there a link between climate change-related droughts and ecologically relevant soil conservation service in this area? Does this relationship exhibit variations across different temporal and spatial scales? Consequently, this study takes the Jinghe River Basin as a target to investigate drought characteristics at different time scales based on meteorological, hydrological, and agricultural drought indices and to predict effects of varied drought types on soil conservation service under two Representative Concentration Pathways (RCP4.5 and RCP8.5). The main research contents include: (1) constructing a Variable Infiltration Capacity (VIC) model and a generalized standardized index (SI) applicable to the Jinghe River Basin; (2) assessing temporal and spatial variability of drought at seasonal and annual scales; (3) estimating soil conservation service at different time scales using the RUSLE model; and (4) assessing drought impacts on soil conservation service. Elucidating these matters will aid managers in comprehending pivotal challenges that impede regional development and in devising focused remedies.

2.1 Study Area

The Jinghe River Basin (106°14′–108°42′E, 34°46′–37°19′N; elevation 219–2908 m; Fig. 1) covers an area of 45,421 km². Situated in a transition zone between semi-arid and subhumid climates, this area experiences high evaporation rates with an average annual temperature of approximately 8°C. Average annual precipitation ranges from 350 to 600 mm, exhibiting significant inter-annual variation and uneven spatial and temporal distributions, gradually decreasing from south to north with locally heavy rainfall concentrations. The basin experiences a high incidence of drought and frequent disaster events. Dominant soil types are loessal soil and heilu soil, with loess thickness varying from 50 to 250 m. These soils have loose structures with numerous voids, making them highly susceptible to erosion and easily dispersed by flowing water.

2.2 Data Sources

This study utilized various datasets including Digital Elevation Model (DEM), land use, soil data, Normalized Difference Vegetation Index (NDVI), daily runoff, meteorological observation data, NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP) dataset, and vegetation library file. Detailed information appears in Table 1.

All data were uniformly preprocessed through projection transformation, clipping, and resampling in ArcGIS 10.2. The arithmetic mean of daily runoff observation data for each month was calculated as the monthly average runoff (m³/s), serving as validation data for the VIC model.

Accurate climate projection models are crucial for assessing regional climate change influence and disaster warnings (Khan et al., 2021). To enhance simulation and prediction capabilities, researchers have undertaken extensive studies utilizing advanced technologies such as the Coupled Model Intercomparison Project 5 (CMIP5) (Zeng et al., 2020). The NEX-GDDP dataset comprises downscaled global climate scenarios derived from General Circulation Model (GCM) runs conducted under CMIP5 across two greenhouse gas emission scenarios (RCP4.5 and RCP8.5). This dataset includes historical climate simulations (1976–2005) and future climate projections (2026–2060). Historical data were used to validate GCM suitability, while future data predicted drought conditions and assessed soil conservation service. Multi-model ensembles improve simulation capability and are widely used for future climate simulation and prediction (Wu et al., 2018). This study adopted a multi-model ensemble of CNRM-CM5, INMCM4, MIROC-ESM, and MRI-CGCM3 models from the NEX-GDDP dataset. Correlation coefficients between multi-model mean and meteorological observation data for monthly precipitation, minimum temperature, and maximum temperature were 0.78, 0.99, and 0.98, respectively (Fig. S1), demonstrating that the selected ensemble accurately models and predicts hydrometeorological features of the study area.

Considering VIC model characteristics and Jinghe River Basin hydrological features, this study adopted a spatial resolution of 0.03°×0.03°, dividing the basin into 4,967 grid cells. Meteorological data for individual grid cells were obtained through inverse distance weight interpolation using both meteorological station data and NEX-GDDP dataset. Soil data were acquired from the Harmonized World Soil Database (HWSD) and uniformly classified using the FAO-90 system. The dominant soil type within each grid cell was determined based on percentage coverage, with its parameters serving as the grid cell's soil attributes. The vegetation library file, containing physical and chemical property parameters for each vegetation type, was acquired from the official VIC model website (https://vic.readthedocs.io/en/master/) using default settings.

2.3.1 Construction of the VIC Model

The VIC model, initially proposed by Wood et al. (1992), originally consisted of two soil layers (VIC-2L). Researchers subsequently incorporated an additional surface layer responsive to rainfall, forming the current VIC-3L model (Xie et al., 2003). The VIC model is a large-scale hydrological model that incorporates the soil water retention curve at grid scale to account for spatial heterogeneity in soil infiltration capacity (Liang et al., 2003). Furthermore, it accommodates both infiltration and saturation excess runoff, enhancing versatility (Xie et al., 2003). This model integrates easily with climate models and finds extensive application in assessing effects of human activities and climate change on the water cycle (Mahto and Mishra, 2020). Therefore, this study investigated drought conditions based on VIC model simulations. The model operated by gridding the study area and simulating internal runoff processes using meteorological (daily precipitation, temperature, wind speed), soil (type and physical-chemical properties), and vegetation type (derived from land use) data for each grid cell. The eight-direction (D8) algorithm defined flow direction during hydrological routing model operation.

The Nash-Sutcliffe efficiency (NSE), relative error (RE), and Kling-Gupta efficiency coefficient (KGE) are widely acknowledged indicators for assessing hydrological model simulation accuracy. Consequently, these three indicators evaluated VIC model performance. Due to challenges obtaining runoff data, the model was calibrated using measured data from a restricted period (preheating: 2000–2005; calibration: 2006–2010; validation: 2011–2015). Calibrated parameters appear in Table 2. Figure 2 depicts accuracy evaluation of runoff simulation data from Zhangjiashan Hydrological Station. Results indicated NSE exceeded 0.71, RE was below 0.01, and KGE surpassed 0.86 during calibration; while validation yielded NSE exceeding 0.83, RE below 0.05, and KGE surpassing 0.89. These results confirm the VIC model reasonably represents hydrological process characteristics in the Jinghe River Basin. Therefore, calibrated model outputs (runoff and soil moisture) are reliable for drought studies (Zhang et al., 2017). Similar to recent studies (e.g., Zhang et al., 2023), this study assumes a model calibrated with limited observed data remains applicable across broader temporal scopes.

2.3.2 Drought Index Calculation Based on the VIC Model

Choosing a suitable drought index is critical for accurate drought assessment and prediction (Khatiwada and Pandey, 2019). Index construction requires selecting appropriate statistical methods based on variable characteristics. SPI and SRI use Gamma distribution probability to calculate drought characteristics (Zhou et al., 2013), while SSMI and Standardized Precipitation Evapotranspiration Index (SPEI) employ log-logistic probability distribution (Shi et al., 2015). Spatiotemporal scales differ among indices: SPI, SRI, and SSMI allow drought calculation at various time scales, whereas Palmer Drought Severity Index (PDSI) and Crop Moisture Index (CMI) have relatively fixed time scales (Mu et al., 2013). To compensate for large uncertainty caused by Gamma distribution due to different optimal distributions of hydrological elements, Farahmand and AghaKouchak (2015) suggested a generalized SI based on non-parametric distribution (i.e., Gringorten plotting position). This method uses an empirical distribution function to standardize marginal probabilities of drought-related variables, including soil moisture, precipitation, and surface runoff. The SI does not require predetermined parameter distribution functions, nor does it necessitate parameter estimation or fit evaluation. Furthermore, when examining multiple drought types simultaneously, SI can mitigate conflicting statistical assumptions and ensure comparability across different drought types at both spatial and temporal scales.

Therefore, following Farahmand and AghaKouchak (2015), this study calculated three drought indices (SPI, SRI, and SSMI) using SI. SPI, SRI, and SSMI correspond to meteorological, hydrological, and agricultural droughts, respectively. Drought occurrence and severity were determined using classification criteria listed in Table 3 (Administration of Quality Supervision, Inspection and Quarantine of People's Republic of China and Standardization Administration of China, 2017).

This study selected drought indices at 3- and 12-month scales to characterize seasonal-scale and annual-scale droughts. Based on precipitation and VIC model simulations of runoff and soil moisture, SI values for SPI, SRI, and SSMI were calculated from 1981 to 2019. Additionally, meteorological data under two future scenarios (RCP4.5 and RCP8.5) were input into the VIC model to calculate drought indices for the future period (2026–2060). To evaluate SI applicability, drought records during 1981–2019 were compiled by referring to the Chinese Dictionary of Meteorological Hazards (Wen and Ding, 2008). Comparison revealed correlation coefficients between simulated and measured drought occurrence months of 0.80 at annual scale and 0.72 at seasonal scale, suggesting the SI effectively captures drought occurrence.

This study introduced drought intensity to explore drought characteristics. Specifically, drought intensity evaluates severity during a certain period and is reflected by the SI value (Eq. 1):

$$S = \sum_{i=1}^{m} |SI_i|$$

where $S$ is drought intensity; $m$ is drought occurrence frequency; and $SI_i$ is the absolute SI value corresponding to the three drought types at the $i$th drought occurrence. Larger $S$ values indicate more severe drought.

2.3.3 Quantification of Soil Conservation Service

Soil conservation service protects sensitive and fragile regional ecology and environment, representing an important regulating service. In terms of retaining soil and minimizing erosion, it focuses on ecosystem capacity to hold soil (Liu et al., 2020). A common quantification approach calculates the difference between potential and actual soil erosion (Bai et al., 2022). This study used the RUSLE model to estimate soil conservation amount (Maqsoom et al., 2020), as shown in Equation 2:

$$B_c = R \times K \times LS \times (1 - C \times P)$$

where $B_c$ is annual soil conservation modulus (t/(hm²·a)); $R$ is rainfall erosivity factor (MJ·mm/(hm²·h·a)); $K$ is soil erodibility factor (t·h/(MJ·mm)); $LS$ is topographic factor (slope gradient and length); $C$ is vegetation cover and crop management factor; and $P$ is soil conservation practice factor.

The rainfall erosivity factor ($R$) represents rainfall's erosion potential. The empirical formula proposed by Wischmeier et al. (1965) was adopted:

$$R = \sum_{i=1}^{12} 1.5 \log_{10}\left(\frac{PRE_i^2}{PRE}\right)$$

where $PRE_i$ denotes monthly precipitation in the $i$th month (mm); and $PRE$ represents annual precipitation (mm).

The soil erodibility factor ($K$) signifies soil erosion sensitivity to external forces such as rainfall impact. The Erosion Productivity Impact Calculator model estimates $K$ by considering soil mechanical composition and organic carbon content (Williams et al., 1983):

$$K = \left{0.2 + 0.3 \exp\left[-0.0256 \times SAN \times \left(1 - \frac{SIL}{100}\right)\right]\right} \times \left(\frac{SIL}{CLA + SIL}\right)^{0.3} \times \left[1 - \frac{0.25 \times SOC}{SOC + \exp(3.72 - 0.95 \times SOC)}\right] \times \left[1 - \frac{0.7 \times (1 - SAN/100)}{1 - SAN/100 + \exp(-5.51 + 22.9 \times (1 - SAN/100))}\right]$$

where $SAN$ is soil sand content (%); $SIL$ is soil silt content (%); $CLA$ is soil clay content (%); and $SOC$ is soil organic carbon content (%).

The topographic factor ($LS$) was calculated using the method from Wischmeier et al. (1978):

$$LS = \left(\frac{\lambda}{22.13}\right)^\alpha \times \left(10.8 \sin\theta + 0.03\right) \quad \text{for } \theta < 5°$$
$$LS = \left(\frac{\lambda}{22.13}\right)^\alpha \times \left(16.8 \sin\theta - 0.50\right) \quad \text{for } 5° \leq \theta < 10°$$
$$LS = \left(\frac{\lambda}{22.13}\right)^\alpha \times \left(21.9 \sin\theta - 0.96\right) \quad \text{for } \theta \geq 10°$$

where $L$ is slope length factor; $S$ is slope gradient factor; $\lambda$ is horizontal projection length of slope; $\alpha$ is slope length exponent; $\beta$ is ratio of rill to interill erosion; and $\theta$ is slope (°) extracted from DEM.

The vegetation cover and crop management factor ($C$) was calculated using the formula proposed by Cai et al. (2000):

$$C = \begin{cases}
1 & \text{if } f_c = 0.0\% \
0.6508 - 0.3436 \log_{10}(f_c) & \text{if } 0.1\% < f_c \leq 78.3\% \
0 & \text{if } f_c > 78.3\%
\end{cases}$$

where $f_c$ is vegetation coverage (%); $NDVI$ is Normalized Difference Vegetation Index; and $NDVI_{max}$ and $NDVI_{min}$ are maximum and minimum NDVI values in the study area.

The soil conservation practice factor ($P$) reflects soil and water conservation measures, indicating the ratio of soil loss after implementing special measures to that when planting along the slope. $P$ values were assigned based on previous research (Sun et al., 2013), considering land use and slope characteristics (Table 4).

Soil conservation amounts were calculated seasonally and annually from 2000 to 2019 using the RUSLE model. For seasonal-scale calculations, $K$, $LS$, $C$, and $P$ factors were consistent with annual-scale methods. The $R$ factor followed a similar approach to Equation 3 but involved summing precipitation for three months within each quarter. For future simulations, $K$ and $LS$ factors remained unchanged, while $C$ and $P$ were determined using 2019 as a baseline. The $R$ factor utilized precipitation data under different future scenarios (RCP4.5 and RCP8.5).

2.3.4 Effects of Drought on Soil Conservation Service

This study characterized drought influence on soil conservation service by measuring the degree of change in average soil conservation during severe drought years compared to the overall study period (Eq. 12). The five worst years for each drought type were identified for both historical (2000–2019) and future (2026–2060) periods under RCP4.5 and RCP8.5 scenarios using drought intensity results:

$$V = \frac{B_{c,dy} - B_c}{B_c} \times 100\%$$

where $V$ is the degree of change (%) in average annual soil conservation for severe drought years relative to the complete study period, representing annual-scale drought impact; $B_{c,dy}$ is average annual soil conservation in severe drought years (t/(hm²·a)); and $B_c$ is average annual soil conservation for the complete study period (t/(hm²·a)).

Furthermore, by referring to annual-scale drought impact evaluation, this study assessed seasonal drought impacts on soil conservation service based on the same principle, substituting annual values in Equation 12 with season-specific soil conservation data.

3.1.1 Annual-Scale Drought Characteristics

This study calculated meteorological, hydrological, and agricultural drought indices using VIC model output data. Spatial variations in drought intensity were observed during the historical period (2000–2019) and future period (2026–2060) under RCP4.5 and RCP8.5 scenarios (Fig. 3). During the historical period, meteorological drought intensity ranged from 2.47 to 3.92, with high-value areas in the north, southeast, and southwest. However, little difference in meteorological drought intensity across the region indicated limited variability in available water resources within the basin confines. Hydrological drought intensity fluctuated between 2.16–4.91. Agricultural drought intensity ranged from 1.81–4.82, with spatial distribution consistent with hydrological drought. High-value areas were concentrated in the southwestern Liupanshan Mountain region.

Compared to the historical period, the future period may witness increased intensity for all three drought types, exhibiting more pronounced impact under RCP4.5 (Fig. 3d–f) than RCP8.5 (Fig. 3g–i). Under RCP4.5 and RCP8.5, meteorological drought intensity ranges became narrower (4.31–4.79 and 3.66–3.94, respectively). Moreover, spatial differences between hydrological and agricultural drought intensities were not obvious in the future period, with high and low value distributions tending toward uniformity. However, agricultural drought intensity displayed negligible north-south variation under RCP8.5. The exacerbation of drought may further affect sustainable development in the Jinghe River Basin, warranting increased attention.

3.1.2 Seasonal-Scale Drought Characteristics

From 2000–2019, average seasonal-scale meteorological drought intensity was 1.02 in spring, 1.01 in summer, 0.52 in autumn, and 0.83 in winter. Meteorological drought intensity was highest in spring and summer, followed by winter and autumn (Fig. 4a–d). Average seasonal-scale hydrological drought intensity was 0.96 in spring, 1.00 in summer, 0.52 in autumn, and 0.77 in winter. Among hydrological droughts, summer drought was most intense, followed by spring and winter droughts, with autumn drought least intense (Fig. 4e–h). Average seasonal-scale agricultural drought intensity was 0.94 in spring, 1.13 in summer, 0.53 in autumn, and 0.70 in winter. Among agricultural droughts, summer drought was most intense, followed by spring, winter, and autumn droughts (Fig. 4i–l). Generally, seasonal-scale drought was weakest in autumn and most severe in summer, with little spatial distribution differences among the three drought types.

Under RCP4.5, spatial distribution of drought intensity remained relatively consistent across seasons for each drought type. Compared to the historical period, spatial differences in meteorological drought intensity in spring, summer, and autumn exhibited further reduction in the future period, with slight variation in the northern winter region (Fig. S2a–d). Spatial differences in hydrological drought intensity surpassed those in meteorological drought intensity (Fig. S2e–h). High-value hydrological drought intensity areas in summer concentrated in the central basin, while low-value winter areas distributed in the southern part. Spatial differences between seasonal-scale agricultural droughts were as inconspicuous as hydrological and meteorological droughts, with similar intensity ranges across seasons and uniform distribution within the basin.

Under RCP8.5, spatial differences in drought intensity across seasons were greater than under RCP4.5, showing scattered distributions (Fig. S2m–x). Drought intensity in all seasons was lower than under RCP4.5, consistent with annual-scale results. Similar to RCP4.5, RCP8.5 showed low meteorological drought intensity values in the northern winter region and low hydrological drought intensity values in the southern basin.

3.2.1 Annual Variations in Soil Conservation Service

Average annual total soil conservation was 1.24×10⁸ t during 2000–2019. From an inter-annual perspective, total soil conservation showed an upward trend, increasing by 8.16×10⁷ t from 2000 to 2019, with the 2019 value being 1.14 times that of 2000 (Fig. 5). The highest total soil conservation occurred in 2013 (2.96×10⁸ t), while the lowest was in 2016 (4.36×10⁷ t). During 2000–2019, the $R$ factor was highest in 2013 and lowest in 2016, corresponding to annual precipitation patterns.

Spatial patterns of average annual soil conservation modulus for 2000–2019 show high-capacity areas mainly in the southwestern, southern, and eastern basin regions (Fig. 6). These areas are dominated by woodlands and shrubs with high vegetation coverage, especially near high-altitude Liupanshan Mountain, where human disturbance intensity is relatively small. Conversely, the northern basin features alternating cultivated land and grassland patterns, resulting in low vegetation coverage, limited soil fixation capacity, and weak conservation. In the middle basin, cultivated land dominates with great human activity interference; most land was bare in winter and spring, exhibiting weak soil conservation capacity that presented a strip distribution near river channels.

Relative changes in average annual total soil conservation were calculated for future periods (2030s, 2040s, 2050s, and 2060s) under different climate scenarios compared to 2019. Under climate change, soil conservation under RCP4.5 and RCP8.5 will decrease. Under RCP4.5, average annual total soil conservation in the 2030s, 2040s, 2050s, and 2060s is expected to decrease by 54.3%, 35.0%, 41.0%, and 38.9%, respectively. Under RCP8.5, mean values will decrease by 42.1%, 34.8%, 47.7%, and 25.7%, respectively, with the largest reduction in the 2050s.

3.2.2 Seasonal Variations in Soil Conservation Service

Seasonal-scale soil conservation was estimated using the RUSLE model for 2000–2019 (Fig. 7). In spring, soil conservation exhibited an overall upward trend, with largest values in 2014 and 2015. In summer, a slight increasing tendency occurred, with significant increases in 2003, 2013, and 2018. Autumn recorded largest values in 2011 and 2014. Winter soil conservation showed its largest value in 2008. These increases can be attributed to precipitation increases.

From a spatial perspective, soil conservation modulus in spring ranged from 0.00–24.00 t/hm² during 2000–2019 (Fig. 8). Areas with modulus below 3.00 t/hm² encompassed 93.0% of the basin. High-value areas (≥10.00 t/hm²) distributed in southern and northern margins with high vegetation coverage, accounting for only 7.0% of the area. Summer modulus ranged 0.00–100.00 t/hm², with areas <50.00 t/hm² accounting for 92.3% (mainly cultivated land and grassland). Areas >50.00 t/hm² accounted for 7.7%, distributing in southwestern, southern, and eastern margins. Autumn modulus was 0.00–100.00 t/hm², with 99.2% of the basin within 0.00–50.00 t/hm². Winter modulus was 0.00–0.43 t/hm² (0.00–0.10 t/hm² in most areas). Overall, spatial variation was greatest in summer, followed by autumn, spring, and winter.

Comparing soil conservation service under different climate scenarios with 2019 values revealed seasonal-scale response characteristics (Table 5). Seasonal averages under both scenarios exhibited downward trends compared to 2019. Notably, under RCP4.5, spring soil conservation in the 2040s will increase by 21.6%, and winter soil conservation in the 2030s will increase by 200.1%, while decreases are most obvious in winter because less precipitation means slight changes cause exponential soil conservation responses.

3.3.1 Impact of Annual-Scale Drought on Soil Conservation Service

Based on drought identification and soil conservation service quantification, we estimated the degree of change in average soil conservation during the five most severe drought years compared to the historical period 2000–2019 (Fig. 9a–c). During severe meteorological drought years, average annual soil conservation was lower than historical averages in most areas, with reductions >25.0%. Only a few northern and southeastern areas showed increases of 0.0%–25.0%. Supplementary data analysis revealed precipitation levels in these specific regions were comparatively higher than other basin areas during corresponding years, making meteorological drought impact less pronounced. Hydrological drought impact on soil conservation service was smaller than meteorological drought. During severe hydrological drought years, more than half the basin experienced soil conservation declines. Compared to meteorological drought, areas with change between –50.0% and –25.0% contracted, while areas with change between –25.0% and 0.0% increased, providing evidence that hydrological drought influence concentrated at lower levels. During severe agricultural drought years, soil conservation reduction areas mainly distributed in the northern region, accounting for about half the basin area. On the eastern edge, soil conservation increased slightly.

Under different droughts, soil conservation declined in most areas. Analysis found that drought intensity corresponding to a few abnormal regions (where average soil conservation during severe drought years exceeded historical averages) was smaller than in other regions, confirming that more serious drought leads to lower soil conservation capacity.

Further investigation examined the degree of change in average annual soil conservation during severe drought years compared to the future period under RCP4.5 and RCP8.5 (Fig. 9d–i). Under RCP4.5, impacts of meteorological, hydrological, and agricultural droughts on soil conservation service will show insignificant differences. During severe meteorological drought years, soil conservation will decrease, with change < –30.0% in approximately two-thirds of the basin. During severe hydrological drought years, soil conservation will decrease by >20.0%, with more serious decreases in the west than east. During severe agricultural drought years, soil conservation will generally decrease by >20.0%.

Under RCP8.5, the degree of change in average annual soil conservation will be –10.0%–0.0% for meteorological drought. Soil conservation during severe hydrological drought years also shows decreases, with change mainly between –20.0% and –10.0%. During severe agricultural drought years, soil conservation will decrease by 30.0%–40% in the east, 10.0%–20.0% in the north, and 20.0%–30.0% in other areas. Overall, meteorological and agricultural drought impacts under RCP8.5 are greater than under RCP4.5, while hydrological drought impact is smaller than under RCP4.5.

3.3.2 Impact of Seasonal-Scale Drought on Soil Conservation Service

To further explore seasonal-scale drought influences, this study quantified the degree of change in average seasonal soil conservation during the five most severe drought years compared to 2000–2019 averages (Fig. 10).

Over the last two decades, spring and autumn were more severely affected by meteorological drought (Fig. 10a–d). During severe meteorological drought years, most areas showed decreasing soil conservation in spring, summer, and autumn, except the northern part which showed increases (Fig. 10a–c). In winter, southern region increases did not exceed 50.0% (Fig. 10d). Summer and autumn were more severely affected by hydrological drought (Fig. 10e–h). During severe hydrological drought years, soil conservation in most areas reduced by >50.0% in summer and autumn, with only small northern areas showing increases (Fig. 10f and g). Slight increases occurred in certain regions during spring and winter (Fig. 10e and h). Winter and summer were more severely affected by agricultural drought (Fig. 10i–l). In the Jinghe River Basin, winter soil conservation reduction exceeded 50.0% (Fig. 10l). Degree of change showed similar spatial distribution in spring and summer, with significant increases in central and eastern regions while other areas mainly decreased (Fig. 10i and j). In autumn, northern region increases occurred but reduction remained dominant in most areas (Fig. 10k).

To investigate future seasonal-scale drought impacts, this study computed the degree of change in average seasonal soil conservation during severe drought years compared to the future period under RCP4.5 and RCP8.5 (Fig. S3). More severe drought conditions are expected in the future compared to the historical period (Fig. 3). Under both scenarios, winter and autumn are most severely affected by meteorological drought, followed by spring and summer. In spring and autumn, soil conservation reduction under RCP8.5 will be greater than under RCP4.5, while in summer, reduction under RCP8.5 is smaller than under RCP4.5. In winter, reduction under both scenarios will exceed 80.0%.

4.1 Model Simulations of Drought and Soil Conservation Service

Based on multi-source data, the VIC model, SI, and RUSLE model simulated spatial distribution patterns and temporal changes of drought and soil conservation service in the Jinghe River Basin.

This study strictly calibrated and validated data and models before use, comparing results with previous studies. Ran et al. (2020) verified VIC model applicability in the Jinghe River Basin, prompting its selection for hydrological process simulation. Using runoff data from Zhangjiashan Hydrological Station near the watershed outlet, we calibrated VIC model parameters. Simulated and observed values met requirements of RE<0.05, NSE>0.83, and KGE>0.89 for the validation period, indicating the calibrated model reflects actual runoff characteristics. We calculated SI based on VIC model simulations and analyzed three drought types using SPI, SRI, and SSMI. Considering spatiotemporal scale variations among these indices, we introduced SI to ensure comparability. Droughts from 1981–2019 were simulated, with the initial 20-year period used for verification and the subsequent 20 years dedicated to analyzing drought-soil conservation associations. Consistency between SI-represented drought occurrence months and measured data showed correlation coefficients of 0.80 at annual scale and 0.72 at seasonal scale, indicating SI accurately reflects drought occurrence. Additionally, spatial drought intensity distribution is consistent with Zhang et al. (2016), with severe drought areas mainly in the western and northern basin. In summary, the drought indices established in this study accurately characterize basin drought status in both time series and spatial distribution.

The RUSLE model is commonly used to estimate soil conservation service. However, previous studies were limited by land use and precipitation data time accuracy, often calculating soil conservation at annual scale in discontinuous years (Liu et al., 2020). This study used land use datasets (2000, 2005, 2010, 2015, and 2018) and monthly precipitation data to calculate annual and seasonal soil conservation for 20 consecutive years (2000–2019) and forecast future changes, expanding RUSLE model application scope.

The distribution pattern of soil conservation service in the Jinghe River Basin exhibits higher levels in southwestern and southeastern regions and lower levels in the north. Total soil conservation is increasing overall, with the largest distribution in summer and smallest in winter. These findings align with Zheng et al. (2021) based on RUSLE and Yu et al. (2022) based on SWAT.

4.2 Mechanism of Drought Impacts on Soil Conservation Service

Since soil conservation service quantification is based on soil erosion, analysis proceeds from this perspective. Prolonged drought may lead to soil exposure, erosion, land degradation, and eventual desertification (Sidiropoulos et al., 2021). Firstly, intensified drought diminishes vegetation coverage and abundance, disrupts microbial community balance, impairs natural land function, and weakens plant soil consolidation capacity (Otkin et al., 2016). Secondly, drought increases soil hydrophobicity and reduces soil water infiltration capacity, leading to increased surface runoff and soil erosion (Gazol et al., 2018). Thus, drought suppresses soil conservation service indirectly by altering soil properties and vegetation conditions.

Furthermore, precipitation—closely related to drought—is the most direct external factor for soil erosion among meteorological factors. Bai et al. (2022) found correlation coefficients between precipitation and soil conservation service exceeding 0.80. In the Jinghe River Basin, drought generally corresponds to reduced precipitation. Combined with the RUSLE model, during drought periods the $R$ factor decreases, reducing potential erosion and the amount of soil needing maintenance, ultimately decreasing soil conservation service. This is consistent with our conclusion that meteorological drought has the greatest impact on soil conservation service. Drought influence involves many aspects including climate, hydrology, vegetation, soil, and project management. However, this study's principal aim is evaluating drought effects on soil conservation service under climate change conditions. Therefore, future scenario simulations solely utilized climate scenario data, neglecting underlying surface modifications. Consequently, this investigation focused on analyzing soil conservation service changes and drought implications under various climate scenarios. To comprehensively understand intricate impact mechanisms, establishing a more holistic land-atmosphere feedback model and conducting further field observations and experiments becomes imperative.

4.3 Limitations and Prospects

The VIC model parameters in this study were calibrated using only publicly measured runoff data from relatively short years (2006–2015), introducing certain limitations. Short calibration periods may fail to capture long-term hydrological trends, leading to overly optimistic or pessimistic future scenario assessments. Additionally, short periods may not cover rare extreme events. If the model does not calibrate enough extreme events, it may not accurately simulate their occurrence and impact under future scenarios. Therefore, in areas with abundant hydrological data, parameters should be calibrated based on longer time series, using time-varying parameter tuning methods (Li et al., 2019) or introducing data from other sources for calibration (Gou et al., 2021).

The RUSLE model is an empirical model lacking ability to depict physical processes, limiting further analysis of drought effect mechanisms. Therefore, introducing process-based SWAT and Water Erosion Prediction Project (WEPP) models to fully consider soil particle processes from stripping through transport to deposition would refine drought impact analysis. Moreover, future research should integrate precise soil erosion quantification through runoff plot experiments and isotopic tracing methods to calibrate soil conservation service model outcomes.

Additionally, simulation reliability is directly influenced by climate model result quality. Using GCM outputs as meteorological forcing data for hydrological models under different scenarios can lead to error accumulation. Due to variations in mechanisms, initial conditions, and parameterization schemes, different GCMs exhibit significant regional performance variations. Currently, Coupled Model Intercomparison Project Phase 6 (CMIP6) involves the largest number of models. However, Zhang and Chen (2021) found that prediction differences among CMIP6 models are even greater than in CMIP5 when comparing uncertainties in precipitation and temperature from 24 GCMs, proposing that more GCMs are needed to ensure robust climate projections. Conversely, CMIP5 has a more mature research foundation, especially for verifying GCMs suitable for reproducing regional climate. Therefore, this study selected climate models from CMIP5. With climate projection development, introducing more comprehensive climate models will enhance research result credibility based on climate model outputs, enabling systematic analysis and management of different modes and scenarios.

4.4 Recommendations

Based on research regarding drought impacts on soil conservation service at different time scales, past drought occurrence greatly reduced soil conservation, and future drought may further reduce it. To reduce ecological environment deterioration from declining soil conservation, this study offers targeted suggestions:

1. Establish scientific drought monitoring systems to strengthen early warning and forecasting. Timely drought information enables policymakers and stakeholders to make informed decisions minimizing negative impacts on soil conservation.

2. Develop and utilize modern water-saving technologies such as artificial rainfall, drip irrigation, and mulching. These enhance water efficiency and reduce soil conservation service vulnerability to drought, especially in arid and semi-arid areas like the Jinghe River Basin.

3. Adjust vegetation structure and enhance soil root holding capacity to mitigate soil conservation service decline during drought periods and improve regional climate. These measures improve water retention, reduce evaporation and erosion risks, and create microclimates promoting precipitation infiltration, leading to improved soil quality and ecological restoration.

4. Strengthen biological, engineering, and farming measures. Implementing horizontal terraces, appropriate afforestation, grass-shrub intercropping, and grass-tillage rotation practices according to local conditions can effectively enhance soil conservation service. These strategies promote land management practices that minimize erosion, increase water infiltration, and maintain soil fertility.

5 Conclusions

This study examined temporal and spatial features of past and future drought and soil conservation service in the Jinghe River Basin, predicting drought impacts on soil conservation service. Soil conservation service may be severely affected by future drought.

The northern and western regions exhibit high drought intensity throughout the year. Intensity of meteorological, hydrological, and agricultural droughts varies seasonally. Soil conservation service distribution shows high levels in the southwest and southeast and low levels in the north. Soil conservation is largest in summer and smallest in winter. Soil conservation declines in severe drought years, more pronounced during meteorological drought than hydrological or agricultural drought. Additionally, seasonal results show that more severe drought creates stronger restrictions on soil conservation service. This study enhances comprehension of drought-soil conservation service interplay, establishing a foundation for developing effective strategies to mitigate future drought impacts on soil conservation service in the Jinghe River Basin.

Conflict of Interest

The authors declare no known competing financial interests or personal relationships that could have influenced the work reported in this paper.

Acknowledgements

This work was supported by the National Natural Science Foundation of China (42071285, 42371297), the Key R&D Program Projects in Shaanxi Province of China (2022SF-382), and the Fundamental Research Funds for the Central Universities (GK202302002).

Author Contributions

Conceptualization: BAI Jizhou, LI Jing; Methodology: BAI Jizhou, RAN Hui; Formal analysis: BAI Jizhou, RAN Hui; Writing - original draft: RAN Hui; Writing - review and editing: BAI Jizhou, DANG Hui; Funding acquisition: BAI Jizhou, LI Jing; Resources: ZHANG Cheng, YU Yuyang; Supervision: LI Jing, ZHOU Zixiang. All authors approved the manuscript.

References

Administration of Quality Supervision, Inspection and Quarantine of People's Republic of China, Standardization Administration of China. 2017. Classification of Meteorological Drought (GB/T 20481-2017). [2023-01-15]. https://openstd.samr.gov.cn/bzgk/gb/newGbInfo?hcno=D2281945A96E8185F67EDC9E7A698049. (in Chinese)

Bai J Z, Zhou Z X, Zou Y F, et al. 2021. Watershed drought and ecosystem services: Spatiotemporal characteristics and gray relational analysis. ISPRS International Journal of Geo-Information, 10(2): 43, doi: 10.3390/ijgi10020043.

Bai J Z, Zhou Z X, Li J, et al. 2022. Predicting soil conservation service in the Jinghe River Basin under climate change. Journal of Hydrology, 615: 128646, doi: 10.1016/j.jhydrol.2022.128646.

Berdugo M, Delgado-Baquerizo M, Soliveres S, et al. 2020. Global ecosystem thresholds driven by aridity. Science, 367(6479): 787–792.

Cao Y Q, Zhao Z M, Zhang D, et al. 2023. Applicability analysis of two comprehensive drought meteorological indexes in growing period of Maize in Liaoning Province. Pearl River, 1–16. [2023-01-24]. http://kns.cnki.net/kcms/detail/44.1037.TV.20231227.1550.002.html. (in Chinese)

Carle J. 2015. Climate Change Seen as Top Global Threat. Pew Research Centre. [2023-10-12]. https://www.pewresearch.org/global/2015/07/14/climate-change-seen-as-top-global-threat/.

Ciampalini R, Constantine J A, Walker-Springett K J, et al. 2020. Modelling soil erosion responses to climate change in three catchments of Great Britain. Science of the Total Environment, 749: 141657, doi: 10.1016/j.scitotenv.2020.141657.

Farahmand A, AghaKouchak A. 2015. A generalized framework for deriving nonparametric standardized drought indicators. Advances in Water Resources, 76: 140–145.

Fensham R J, Fairfax R J, Ward D P. 2009. Drought-induced tree death in savanna. Global Change Biology, 15(2): 380–387.

Gampe D, Zscheischler J, Reichstein M, et al. 2021. Increasing impact of warm droughts on northern ecosystem productivity over recent decades. Nature Climate Change, 11(9): 772–779.

Gazol A, Camarero J J, Jiménez J J, et al. 2018. Beneath the canopy: Linking drought-induced forest die off and changes in soil properties. Forest Ecology and Management, 422: 294–302.

Gou J J, Miao C Y, Samaniego L, et al. 2021. CNRD v1.0: A high-quality natural runoff dataset for hydrological and climate studies in China. Bulletin of the American Meteorological Society, 102(5): 929–947.

Han H Q, Gao H J, Huang Y, et al. 2019. Effects of drought on freshwater ecosystem services in poverty-stricken mountain areas. Global Ecology and Conservation, 17: e00537, doi: 10.1016/j.gecco.2019.e00537.

Huang J P, Yu H P, Guan X D, et al. 2015. Accelerated dryland expansion under climate change. Nature Climate Change, 6(2): 166–171.

Khan F, Pilz J, Ali S. 2021. Evaluation of CMIP5 models and ensemble climate projections using a Bayesian approach: a case study of the Upper Indus Basin, Pakistan. Environmental and Ecological Statistics, 28(2): 383–404.

Khatiwada K R, Pandey V P. 2019. Characterization of hydro-meteorological drought in Nepal Himalaya: a case of Karnali River Basin. Weather and Climate Extremes, 26: 100239, doi: 10.1016/j.wace.2019.100239.

Kimwatu D M, Mundia C N, Makokha G O, et al. 2021. Developing a new socio-economic drought index for monitoring drought proliferation: a case study of Upper Ewaso Ngiro River Basin in Kenya. Environmental Monitoring and Assessment, 193(4): 213, doi: 10.1007/s10661-021-08989-0.

Leal Filho W, Azeiteiro U M, Balogun A L, et al. 2021. The influence of ecosystems services depletion to climate change adaptation efforts in Africa. Science of the Total Environment, 779: 146414, doi: 10.1016/j.scitotenv.2021.146414.

Li Y Y, Chang J X, Luo L F, et al. 2019. Spatiotemporal impacts of land use land cover changes on hydrology from the mechanism perspective using SWAT model with time-varying parameters. Hydrology Research, 50(1): 244–261.

Liang X, Xie Z H, Huang M Y. 2003. A new parameterization for surface and groundwater interactions and its impact on water budgets with the variable infiltration capacity (VIC) land surface model. Journal of Geophysical Research: Atmospheres, 108(D16): 8613, doi: 10.1029/2002JD003090.

Liu T, Zhou Z X, Zhu Q, et al. 2020. Spatiotemporal change of soil conservation service in Yanhe Watershed. Research of Soil and Water Conservation, 28(1): 93–100. (in Chinese)

Liu Y, Zhao W W, Jia L Z. 2019. Soil conservation service: concept, assessment, and outlook. Acta Ecologica Sinica, 39(2): 432–440. (in Chinese)

Mahto S S, Mishra V. 2020. Dominance of summer monsoon flash droughts in India. Environmental Research Letters, 15(10): 104061, doi: 10.1088/1748-9326/abaf1d.

Maity R, Suman M, Verma N K. 2016. Drought prediction using a wavelet based approach to model the temporal consequences of different types of droughts. Journal of Hydrology, 539: 417–428.

Maqsoom A, Aslam B, Hassan U, et al. 2020. Geospatial assessment of soil erosion intensity and sediment yield using the Revised Universal Soil Loss Equation (RUSLE) model. ISPRS International Journal of Geo-Information, 9(6): 356, doi: 10.3390/ijgi9060356.

Masroor M, Sajjad H, Rehman S, et al. 2022. Analysing the relationship between drought and soil erosion using vegetation health index and RUSLE models in Godavari middle sub-basin, India. Geoscience Frontiers, 13(2): 101312, doi: 10.1016/j.gsf.2021.101312.

Mu Q Z, Zhao M S, Kimball J S, et al. 2013. A remotely sensed global terrestrial drought severity index. Bulletin of the American Meteorological Society, 94(1): 83–98.

Otkin J A, Anderson M C, Hain C, et al. 2016. Assessing the evolution of soil moisture and vegetation conditions during the 2012 United States flash drought. Agricultural and Forest Meteorology, 218–219: 230–242.

Pan Y, Zhu Y H, Lü H S, et al. 2023. Accuracy of agricultural drought indices and analysis of agricultural drought characteristics in China between 2000 and 2019. Agricultural Water Management, 283: 108305, doi: 10.1016/J.AGWAT.2023.108305.

Pravalie R, Sîrodoev I, Peptenatu D. 2014. Changes in the forest ecosystems in areas impacted by aridization in south-western Romania. Journal of Environmental Health Science and Engineering, 12(1): 2, doi: 10.1186/2052-336X-12-2.

Ran H, Li J, Zhou Z X, et al. 2020. Predicting the spatiotemporal characteristics of flash droughts with downscaled CMIP5 models in the Jinghe River basin of China. Environmental Science and Pollution Research, 27(32): 40370–40382.

Shi B L, Zhu X Y, Hu Y C, et al. 2015. Spatio-temporal variations of drought in Henan Province over a 53-year period based on standardized precipitation evapotranspiration index. Geographical Research, 34(8): 1547–1558. (in Chinese)

Sidiropoulos P, Dalezios N R, Loukas A, et al. 2021. Quantitative classification of desertification severity for degraded aquifer based on remotely sensed drought assessment. Hydrology, 8(1): 47, doi: 10.3390/hydrology8010047.

Sun W Y, Shao Q Q, Liu J Y. 2013. Soil erosion and its response to the changes of precipitation and vegetation cover on the Loess Plateau. Journal of Geographical Sciences, 23(06): 1091–1106.

Terwayet Bayouli O, Zhang W C, Terwayet Bayouli H. 2023. Combining RUSLE model and the vegetation health index to unravel the relationship between soil erosion and droughts in southeastern Tunisia. Journal of Arid Land, 15(11): 1269–1289.

Wang D Y, Zhang W, Lu C J, et al. 2022. Construction and precision evaluation of comprehensive drought index based on meteorological and remote sensing vegetation information. Geomatics and Information Science of Wuhan University, doi: 10.13203/j.whugis20220237. (in Chinese)

Wen K G, Ding Y H. 2008. Chinese Dictionary of Meteorological Hazards. Comprehensive Volume. Beijing: Meteorological Press, 1–948. (in Chinese)

Wood E F, Lettenmaier D P, Zartarian V G. 1992. A land-surface hydrology parameterization with subgrid variability for general circulation models. Journal of Geophysical Research: Atmospheres, 97(D3): 2717–2728.

Wu Q, Jiang X W, Xie J, et al. 2018. Multimodel superensemble prediction of air temperature in southwestern China during 2020–2050 based on CMIP5 data. Journal of Arid Meteorology, 36(6): 971–978. (in Chinese)

Xie Z H, Su F G, Liang X, et al. 2003. Applications of a surface runoff model with horton and dunne runoff for VIC. Advances in Atmospheric Sciences, 20(2): 165–172.

Yang X L, Liu G S, Yang X G, et al. 2005. The modification of palmer drought severity model for Gansu Loess Plateau. Journal of Arid Meteorology, 23(2): 8–12. (in Chinese)

Yu Y Y, Li J, Zhou Z X, et al. 2022. Spatial pattern optimization of ecosystem services based on Bayesian networks: a case of the Jing River Basin. Arid Land Geography, 45(4): 1268–1280. (in Chinese)

Zeng P, Sun F Y, Liu Y Y, et al. 2020. Future river basin health assessment through reliability-resilience-vulnerability: thresholds of multiple dryness conditions. Science of the Total Environment, 140395, doi: 10.1016/j.scitotenv.2020.140395.

Zhang H B, Gu L, Xin C, et al. 2016. Investigation on the spatial-temporal variation of drought characteristics in Jinghe River Basin. Journal of North China University of Water Resources and Electric Power (Natural Science Edition), 37(3): 1–10. (in Chinese)

Zhang S N, Wu Y P, Sivakumar B, et al. 2019. Climate change-induced drought evolution over the past 50 years in the southern Chinese Loess Plateau. Environmental Modelling & Software, 122: 104519, doi: 10.1016/j.envsoft.2019.104519.

Zhang S B, Chen J. 2021. Uncertainty in projection of climate extremes: a comparison of CMIP5 and CMIP6. Journal of Meteorological Research, 35(4): 646–662.

Zhang Y Q, Zheng H X, Zhang X Z, et al. 2023. Future global streamflow declines are probably more severe than previously estimated. Nature Water, 1(3): 261–271.

Zheng T, Zhou Z X, Zou Y F, et al. 2021. Analysis of spatial and temporal characteristics and spatial flow process of soil conservation service in Jinghe Basin of China. Sustainability, 13(4): 1794, doi: 10.3390/SU13041794.

Zhou Y, Li N, Ji Z H, et al. 2013. Temporal and spatial patterns of droughts based on Standard Precipitation Index (SPI) in Inner Mongolia during 1981–2010. Journal of Natural Resources, 28(10): 1694–1706. (in Chinese)

Appendix

Fig. S1 Verification results of the NEX-GDDP dataset for monthly precipitation (a), monthly minimum temperature (b), and monthly maximum temperature (c) from 1976 to 2005. NEX-GDDP, NASA Earth Exchange Global Daily Downscaled Projections; $r$, Pearson's correlation coefficient.

Fig. S2 Spatial distribution of multi-year average meteorological drought intensity (a–d and m–p), hydrological drought intensity (e–h and q–t), and agricultural drought intensity (i–l and u–x) at seasonal scale under RCP4.5 and RCP8.5 scenarios in the future period (2026–2060). RCP, Representative Concentration Pathway. SPI3, SRI3, and SSMI3 correspond to seasonal-scale meteorological, hydrological, and agricultural droughts, respectively.

Fig. S3 Degree of change in average seasonal soil conservation in severe drought years relative to the future period (2026–2060) under RCP4.5 (a–d and m–p) and RCP8.5 (e–h and q–t) scenarios for meteorological, hydrological, and agricultural droughts. SPI3, SRI3, and SSMI3 correspond to seasonal-scale meteorological, hydrological, and agricultural droughts, respectively.