Quantifying the Effects of Climate Variability, Land-Use Changes, and Human Activities on Drought Based on the SWAT–PDSI Model

Abstract

Much attention has recently been devoted to the qualitative relationship between climate factors and drought; however, the influences of climate variability, land-use/cover changes (LUCC), and other human activities on drought have rarely been quantitatively assessed. Based on the Soil and Water Assessment Tool (SWAT) model and the Palmer drought severity index (PDSI), this study presents a framework to quantify drought changes in an attribution study, and quantifies the effects of climate factors, LUCC, and other human activities on drought in a typical basin (Yihe River) in eastern China from 1980 to 2019. (1) The SWAT–PDSI results revealed a slight decreasing trend from 1980 to 2019, indicating that the degree of drought increased—especially in the middle of the basin. (2) The precipitation in the basin exhibited a downward trend (−2.7 mm/10 a), while the temperature exhibited a significant increasing trend (0.13 °C/10 a, p < 0.05). Over the past 40 years, LUCC in the Yihe River Basin was mainly characterized by a reduction in the dryland area (149 km²) and an increase in the built-up area (135 km²), which changed by −1.77% and 18.96%, respectively. (3) Climate fluctuation was the main driving factor of drought change, with a contribution rate ranging from 68 to 84%, and the contribution to drought gradually increased. Among the various factors, the contribution of temperature exceeded that of precipitation from 2010 to 2019, suggesting that temperature has become the most important climate factor affecting drought. The contribution rates of LUCC to drought changes over the periods 1990–1999, 2000–2009, and 2010–2019 were 7.8%, 18%, and 12.6%, respectively. This indicates that the relative contributions of other human activities to drought changes gradually decreased. This study refines the drought attribution framework, which could provide scientific support for the quantitative attribution of drought and the formulation of disaster prevention and reduction strategies.

1. Introduction

Against the background of climate warming, the frequency and intensity of drought events have gradually increased, threatening the sustainable development of the economy and society [1]. Over the past 50 years, drought has resulted in a 9–10% increase in agricultural losses [2]. From 2005 to 2015, drought caused 29 billion in losses in the agricultural markets of developing countries [3]. In addition, due to global changes, the evolution mechanisms of the water cycle are becoming increasingly complex, leading to changes in regional soil moisture content and river runoff and, thus, altering regional dry and wet conditions [4]. Therefore, the study of the spatiotemporal variation in drought and its driving mechanisms can not only improve the understanding of the evolution mechanisms of the water cycle, but also improve the level of disaster prevention and mitigation in response to climate change.

Because of the complexity of the timing, range, and intensity evolution of drought events, many scholars have proposed various drought indices to indicate the degree of drought changes to improve drought monitoring and evaluation [5,6,7]. McKee et al. proposed the standardized precipitation index (SPI), which can not only reflect changes in rainfall within a short period, but also reflect the evolution of long-term water resources [8]; however, this method used precipitation as the only parameter in drought assessment, ignoring the impact of other factors on drought [9,10]. Vicente-Serrano proposed the standardized evaporation index (SPEI) based on the SPI, considering the impact of the potential evapotranspiration (PET) on drought [11]. Shukla proposed the standardized runoff index (SRI), which uses runoff to replace precipitation in the SPI calculation method [12]. The required input data are limited and easy to obtain. Palmer proposed the Palmer drought severity index (PDSI), which is based on a two-layer soil model to calculate the water balance, considering factors such as precipitation, temperature, soil water content, and evapotranspiration [13]. In addition, estimation of evapotranspiration via the PDSI involves two factors, namely, the soil’s effective water-holding capacity, and runoff, which can more comprehensively describe drought [14]. Therefore, the PDSI is regarded as a milestone in the history of drought monitoring [15]. However, the effects of soil spatial heterogeneity, land-use types, and underlying surface conditions on drought processes are typically not considered in the estimation of evapotranspiration via the PDSI [16,17]. Distributed hydrological models can overcome this shortcoming. Hydrological models have been built according to watershed water cycle theory—especially distributed hydrological models. The water circulation system is typically adopted as the research object, and heterogeneities in the soil space, land-use type, underlying surface, etc., are considered, while a strong correlation between hydrological elements such as runoff, surface evaporation, soil moisture content, and climate can be revealed [18,19].

Against the background of global climate change and the increasing intensity of human activities, many scholars have considered various methods to study the attribution of drought, and have achieved substantive results. Gan used the continuous integral method and piecewise integral method to attribute drought events in Southwest China from 2009 to 2010 [20]. The experiment revealed that insufficient precipitation and high temperatures are the main driving forces of the formation and maintenance of drought. Xie proposed that mutation was an important manifestation of water climate change [21]. Through detection of mutation in precipitation in Southwest China, the causes of drought were analyzed. The detection results revealed that a reduction in precipitation was the main cause of drought. Kew studied the attribution of drought in East Africa by using a combination method of multiple statistical models and various datasets [22]. The research demonstrated that precipitation contributed more to drought than temperature. Li used quantile mapping to eliminate the systematic deviation of the simulated temperature and precipitation, and attributed drought combined with the SPEI; it was found that drought is closely related to increases in temperature [23]. Philip used the CLM and FLDAS to attribute drought events in the Netherlands [24]. He found that strong trends in temperature and global radiation have resulted in increased potential evapotranspiration, resulting in agricultural drought. Drought is a natural disaster with slow development and a long duration. However, against the background of climate change, the characteristics of drought are gradually changing [24]. Scholars have begun to devote attention to the impact of human activities on drought. Chiang studied global drought by using the SPI and SPEI generated on nine Coupled Model Intercomparison Project (CMIP) model sets, and confirmed that human activities affect drought characteristics [25]. Wang found that human activities significantly increased the intensity and speed of drought based on data of Phase 6 of the CMIP (CIMP6), resulting in more challenging drought prediction and early warning generation [26]. Deng studied the attribution of drought in China by constructing a comprehensive drought index (WSDI) and using Pearson’s correlation coefficient [27]. Wanders used the global hydrological and water resources model PCR-GLOBWB to simulate daily discharge globally [28]. The results showed a significant impact of climate change and human water use in large parts of Asia, the Middle East, and the Mediterranean, where the relative contribution of humans to the changes in drought severity could be close to 100%. The research determined that drought in China was mainly caused by a reduction in precipitation, and that human activities exacerbated drought. The above research on the attribution of drought—whether based on distributed hydrological models or mathematical statistical theory—exhibits two characteristics: (1) the selection of drought drivers is very simplistic, and the comprehensive impacts of temperature, precipitation, human activities, and other factors on drought cannot be considered at the same time; and (2) despite time-series analysis between drought occurrence and temperature, precipitation, human activities, and other correlation factors, the degree of influence remains unknown.

Therefore, this study chose the Yihe River Basin as an example, and the following research was performed: (1) the temporal and spatial variation characteristics of watershed meteorological drought were revealed based on the SWAT–PDSI model, which could reflect the influence of the underlying surface on drought processes; (2) the temporal and spatial variation characteristics of temperature, precipitation, and LUCC were analyzed via statistical methods; and (3) through controlled simulation experiments, the impacts of temperature, precipitation, LUCC, and other human activities on drought changes were quantitatively evaluated. The results of this research could be of great significance for the formulation of water resource management strategies in river basins.

2. Data and Methods

2.1. Study Area

The Yihe River is a major river of the Yishusi River system in the Huai River Valley. This river system is bounded by the Haihe Plain to the north and the North China Plain to the south (Figure 1). The basin exhibits an area of 17,000 km² and a total length of more than 500 km. The Yihe River is the mother river of Linyi, and originates from the Yiyuan County River in Shandong Province, passes through Yishui County, Yinan County, Linyi County, Mengyin County, Pingyi County, and other counties and cities, enters the Xinyi River at the village of Wu Lou in Pizhou County, Jiangsu Province, and reaches Yanwei Port in the Yellow Sea [29]. The basin belongs to a warm, temperate, semi-humid monsoon continental climate, which is alternately affected by continental air masses and marine air masses, with four distinct seasons, hot summers, cold winters, and other characteristic. The average annual precipitation of the basin is about 815 mm, and the precipitation is extremely uneven throughout the year, concentrated in summer and autumn [30]. The hilly area of the Yihe River Basin accounts for approximately 70% of the total basin area, and the plain area accounts for approximately 30% of the total basin area. The terrain is high in the northwest and low in the southeast. Mountains and valleys intersect, and the terrain is complex. The Yihe River is a mountainous river. The river is cut off during the dry period and flooded during the flood season, and the ecology is fragile. In recent years, with global warming, the complexity and non-stationarity of drought have been further aggravated, and extreme drought events in the basin have noticeably increased, seriously threatening the basin’s ecology and economy [31,32,33].

2.2. Data Sources

For the SWAT simulations, DEM and LUCC data were obtained from the Data Center of Resources and Environmental Science of the Chinese Academy of Sciences (https://www.resdc.cn/ (accessed on 10 July 2021)). Runoff data were retrieved from the China River Sediment Bulletin, which covers the period from January 2009 to December 2019 (http://www.mwr.gov.cn/sj/#tjgb (accessed on 15 July 2021)). Soil data were obtained from the Harmonized World Soil Database (HWSD) (https://www.fao.org/soils-portal/soil-survey/soil-maps-and-databases/harmonized-world-soil-database-v12/en/ (accessed on 25 July 2021)). Meteorological data were derived from the Global Forecast System Reanalysis (CFSR) dataset of the National Center for Environmental Prediction, which covers the period from January 1980 to December 2019. We selected precipitation, maximum/minimum temperature, relative humidity, and wind speed. The horizontal resolution of the CFSR is 0.5° × 0.5° (approximately 38 × 38 km) (https://climatedataguide.ucar.edu/climate-data/climate-forecast-system-reanalysis-cfsr (accessed on 1 July 2021)). Previous studies have demonstrated that the SWAT model based on CFSR data can accurately describe basin hydrological conditions [34]. CFSR data can reflect not only the daily precipitation, but also the topographic diversity [35,36]. Meanwhile, we attained data quality control and assessment of homogeneity using the RClimDex software package (Version 1.3, available at the ETCCDI website, http://etccdi.pacificclimate.org/software.shtml), which has been developed and maintained by Xuebin Zhang and Yang Feng at Environment Canada.

2.3. Methodology

2.3.1. SWAT–PDSI Model

First, based on the SWAT model and the PDSI principle, a localized SWAT–PDSI model was constructed. The SWAT model is a distributed hydrological model for large- and medium-sized basins that can simulate long timescales. This model is often used to simulate the hydrological response process, sediment yield, and transport of chemical pesticide pollutants in river basins [37]. The SWAT model divides the study area into sub-basins of different areas based on topographic data provided by a digital elevation model (DEM) and a minimum catchment area threshold. The sub-basins are further divided into hydrologic response units according to land-use and soil data. Therefore, this model can calculate hydrological characteristics at three spatial scales: watersheds, sub-watersheds, and hydrologic response units (HRUs) [38,39]. The hydrological process in the SWAT model mainly comprises two modules: the sub-basin module (runoff generation and slope convergence), and the convergence calculation module (channel convergence). In the flow concentration calculation module, the expression of the water balance is as follows [39]:

$$\mathrm{WYLD}=\mathrm{SURQ}\_\mathrm{CNT}+\mathrm{LATQ}+\mathrm{GWQ} - \mathrm{TLOSS}$$

(1)

where WYLD is the amount of water (mm) flowing into the main channel during the time step, SURQ_CNT is the contribution of surface runoff to the main channel runoff (mm) during the time step, LATQ is the contribution of lateral flow to the stream runoff (mm) during the time step, GWQ is the contribution of underground runoff to the total runoff of the main river course (mm) during the time step, and TLOSS is the amount of water loss (mm) transported by tributaries within the HRU through the riverbed during the time step.

$${ \mathrm{SURQ}\_\mathrm{GEN}=(\mathrm{SURQ}\_\mathrm{GEN}+\mathrm{SURQ}\_\mathrm{STOR}}_{\mathrm{i}-1})$$

(2)

where SURQ_GEN is the surface runoff (mm) generated in the HRU during the time step, and ${\mathrm{SURQ}\_\mathrm{STOR}}_{\mathrm{i}-1}$ is the surface runoff stored in the HRU or maintained from the previous day.

$$\mathrm{SURQ}\_\mathrm{GEN}=\frac{{{( \mathrm{R}}_{\mathrm{day}}{ - \mathrm{I}}_{\mathrm{a}} )}^2}{{\mathrm{R}}_{{\mathrm{day} - \mathrm{I}}_{\mathrm{a}}+\mathrm{S}}}$$

(3)

where R is the daily rainfall infiltration depth (mm), and Ia includes surface water storage (mm), interception (mm), and infiltration (mm). These values are determined by the SCS (Soil Conservation Service) runoff curve number method. S is the retention parameter (mm), while Ia represents the initial abstractions including surface storage, interception, and infiltration prior to runoff, which is commonly approximated as 0.2 S (mm).

$$\mathrm{LATQ}=0.024 \times \left( \frac{{2 \times \mathrm{SW}}_{1\mathrm{y}, \mathrm{excess}}{ \times \mathrm{K}}_{\mathrm{sat}} \times \mathrm{slp}}{\varnothing {\mathrm{d} \times \mathrm{L}}_{\mathrm{hill}}} \right)$$

(4)

where ${\mathrm{SW}}_{1\mathrm{y}, \mathrm{excess}}$ is the amount of water stored in the saturated zone of the mountain slope per unit of area (mm), Slp is the average gradient in the HRU (m/m), $\varnothing \mathrm{d}$ is the drainage porosity of the soil layer (mm/mm), ${\mathrm{L}}_{\mathrm{hill}}$ is the slope length (m), and K_sat is the saturated hydraulic conductivity (mm/h).

$$\mathrm{GWQ}=\frac{{8000 \times \mathrm{K}}_{\mathrm{sat}}}{{\mathrm{L}}_{\mathrm{gw}}^2}{ \times \mathrm{h}}_{\mathrm{wtbl} }$$

(5)

where ${\mathrm{L}}_{\mathrm{gw}}$ is the distance (m) from the ridge or HRU watershed of the groundwater system to the main channel, and ${\mathrm{h}}_{\mathrm{wtbl}}$ is the height of the water table.

The model was calibrated with the sequential uncertainty fitting optimization algorithm version 2 (SUFI-2) in SWAT–CUP, developed by the Swiss Federal Institute of Aquatic Sciences and Technology. The linear regression correlation coefficient (R²), Nash–Sutcliffe efficiency coefficient (ENS), and PBIAS were used to evaluate the relationship between the measured and simulated runoff values, and the model’s applicability in the study area could be determined. The specific equations are as follows:

$$\mathrm{ENS}=1 - \frac{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}} {\left({ \mathrm{Q}}_{\mathrm{i}}^{\mathrm{obs}}-{\mathrm{Q}}_{\mathrm{i}}^{\mathrm{sim} }\right)}^2 }{{{\displaystyle \sum }}_{\mathrm{i} =1}^{\mathrm{n}}{\left( {\mathrm{Q}}_{\mathrm{i}}^{\mathrm{obs}}-{ \mathrm{Q}}_{\mathrm{avg}}^{\mathrm{obs}} \right)}^{2 }}$$

(6)

$$\mathrm{PBIAS}=\frac{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}{(\mathrm{Q}}_{\mathrm{i}}^{\mathrm{obs}}-{\mathrm{Q}}_{\mathrm{i}}^{\mathrm{sim}})}{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n} }{\mathrm{Q}}_{\mathrm{i}}^{\mathrm{obs}}}$$

(7)

$${ \mathrm{R}}^2=\frac{{\left( {{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n} }\left({ \mathrm{Q}}_{\mathrm{i}}^{\mathrm{obs} }-{ \mathrm{Q}}_{\mathrm{avg}}^{\mathrm{obs}}\right)\left({ \mathrm{Q}}_{\mathrm{i}}^{\mathrm{sim} }-{ \mathrm{Q}}_{\mathrm{avg}}^{\mathrm{sim}} \right) \right)}^2}{{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}{{(\mathrm{Q}}_{\mathrm{i}}^{\mathrm{obs}}-{ \mathrm{Q}}_{\mathrm{avg}}^{\mathrm{obs}})}^2{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}{{( \mathrm{Q}}_{\mathrm{i}}^{\mathrm{sim}}-{ \mathrm{Q}}_{\mathrm{avg}}^{\mathrm{sim} })}^2}$$

(8)

where ${\mathrm{Q}}_{\mathrm{i}}^{\mathrm{obs}}$ is the observed runoff, ${\mathrm{Q}}_{\mathrm{i}}^{\mathrm{sim}}$ is the simulated runoff, ${\mathrm{Q}}_{\mathrm{avg}}^{\mathrm{obs}}$ is the average observed runoff value, and ${\mathrm{Q}}_{\mathrm{avg}}^{\mathrm{sim}}$ is the average simulated runoff value.

The simulation values were in the range of 0~1, and the closer the ENS value is to 1, the better the simulation effect. For 0.36 < ENS < 0.75, the simulation effect is essentially satisfactory [40]. PBIAS reflects the relative error between the simulated and observed values. The optimal PBIAS value is 0, and PBIAS > 0 indicates underestimated simulation results, while PBIAS < 0 indicates overestimated simulation values. ${\mathrm{R}}^2$ indicates the degree of correlation between the observed and simulated values. The closer R² is to 1, the more consistent the trend between the simulated and observed values [41].

Based on empirical parameters, the PDSI makes assumptions regarding the field water-holding capacity and water transfer between different layers, but does not comprehensively consider the actual precipitation runoff process from the perspective of the water cycle, while the SWAT model can overcome this drawback [42,43].

In this paper, the SWAT–PDSI drought model was constructed with the SWAT distributed hydrological model. Compared to the traditional two-layer soil model, the SWAT–PDSI model can be used to determine the surface runoff, potential evapotranspiration, actual evapotranspiration, and soil water content in each sub-watershed, which ensures a higher accuracy of the PDSI results [44]. The calculation equations are as follows:

$$\mathrm{d}=\mathrm{P} -\widehat{ \mathrm{P}}$$

(9)

$${\mathrm{k}}^{\prime }=1.5\lg \left[\frac{\overline{\mathrm{PE} }+\overline{\mathrm{R}}+\overline{\mathrm{RO}}}{\left(\overline{ \mathrm{P} }+\overline{\mathrm{L} }\right)\overline{ \mathrm{D}}}+\frac{2.8}{\overline{ \mathrm{D}}}\right]+0.5$$

(10)

$$\mathrm{k}=\frac{17.67}{{{\displaystyle \sum }}_1^{12} \overline{\mathrm{D}}{\mathrm{k}}^{\prime }}$$

(11)

$$\mathrm{Z}=\mathrm{Kd}$$

(12)

$${ \mathrm{X}}_{\mathrm{i}}=0{.897\mathrm{X}}_{\mathrm{i}-1}+\frac{{\mathrm{Z}}_{\mathrm{i}}}{3}$$

(13)

Because the PDSI attains varying levels of applicability in different climatic regions, the calculation equation developed by DC Palmer must be adapted outside the central United States [45]. The parameters of the PDSI were modified based on data for China measured by Liu Weiwei, and SWA–-PDSI values could be calculated in the study area [46]. The specific equation is as follows:

$${\mathrm{K}}^{\prime }=1.28151\lg \left[\frac{\overline{\mathrm{PE}}+\overline{ \mathrm{R}}+\overline{\mathrm{RO}}}{\left(\overline{ \mathrm{P} }+\overline{ \mathrm{L} }\right)\overline{ \mathrm{D}}}\right]+3.3027$$

(14)

$$\mathrm{k}=\frac{581.391}{{{\displaystyle \sum }}_1^{12} \overline{\mathrm{D}}{\mathrm{k}}^{\prime }} {\mathrm{k}}^{\prime }$$

(15)

$${\mathrm{X}}_{\mathrm{i}}=0{.9331\mathrm{X}}_{\mathrm{i}-1}+\frac{{\mathrm{Z}}_{\mathrm{i}}}{125.99}$$

(16)

where $\overline{\mathrm{PE}}$,$\overline{ \mathrm{R}}$,$\overline{ \mathrm{RO}}$,$\overline{ \mathrm{P}}$, $\overline{ \mathrm{L}}$, and $\widehat{ \mathrm{P}}$ are the multiyear monthly average potential evapotranspiration, actual water replenishment, runoff, precipitation, actual water loss, and suitable precipitation, respectively. K is an adjustable weight factor, K’ is the estimated value of K, D is the average absolute value of d, and Z is the water anomaly index.

Then, statistical methods such as the Mann–Kendall test method can be used to analyze the temporal and spatial variation trends of the SWAT–PDSI index.

2.3.2. Climate Factors and Land-Use Changes

The Mann–Kendall test method was used to analyze the temporal and spatial changes in temperature and precipitation, and modern statistics were used to analyze the characteristics of land-use changes in the basin in 1990, 1995, 2005, and 2015.

The Mann–Kendall test method assesses the trend of samples by constructing a standard normal distribution statistic (Z). For an independent and identically distributed time series X1, X2, X3, ···, Xn, n is the length of the time series, and the statistical variable S can be defined as follows [47,48]:

$$\mathrm{S}={{\displaystyle \sum }}_{\mathrm{k}=1}^{\mathrm{n} - 1}{{\displaystyle \sum }}_{\mathrm{j}=\mathrm{k}+1}^{ \mathrm{n}}{\mathrm{Sgn}(\mathrm{x}}_{\mathrm{j}}-{ \mathrm{x}}_{\mathrm{k}})$$

(17)

$$\mathrm{Sgn}\left({\mathrm{x}}_{\mathrm{j}}{-\mathrm{x}}_{\mathrm{k}}\right)=\left\{\begin{array}{c}{1, \mathrm{x}}_{\mathrm{j}}-{ \mathrm{x}}_{\mathrm{k}} > 0\\ { 0, \mathrm{x} }_{\mathrm{j}}-{ \mathrm{x}}_{\mathrm{k}}=0 \\ {-1, \mathrm{x}}_{\mathrm{j}}-{ \mathrm{x}}_{\mathrm{k}} < 0\end{array} \right.$$

(18)

The variance in the S statistic is determined as follows:

$$\mathrm{Var}\left(\mathrm{S}\right)=\frac{\mathrm{n}\left(\mathrm{n}-1\right) \times \left(2\mathrm{n}+5\right)}{18}$$

(19)

For n > 10, the standard normal statistic Z can be obtained as follows:

$$\mathrm{Z}=\left\{\begin{array}{c}\frac{\mathrm{S}-1}{\sqrt{\mathrm{Var}\left(\mathrm{S}\right)}}, \mathrm{S} > 0\\ 0, \mathrm{S}=0\\ \frac{\mathrm{S}+1}{\sqrt{\mathrm{Var}\left(\mathrm{S}\right)}}, \mathrm{S} < 0\end{array} \right.$$

(20)

where Z > 0 indicates that the sequence is an ascending sequence, while Z < 0 indicates that the sequence is a descending sequence. When |Z| equals 1.28, 1.96, and 2.32, the trend confidence reaches 90%, 95%, and 99%, respectively. In this paper, the significance level p was set to 0.05 for trend analysis.

2.3.3. Quantitative Attribution of Drought Changes

In this study, the average SWAT–PDSI value from 1980 to 1989 was used as a benchmark to study changes in drought and its attribution in the different years. The SWAT–PDSI model can reflect the effects of climatic factors such as precipitation, evapotranspiration, and temperature, along with human factors such as LUCC, on drought. Among these factors, when LUCC is used as a variable, the other elements remain unchanged:

$${ \mathrm{P}}_{\mathrm{LUCC}}{=\mathrm{P}}_{\mathrm{LUCCz} }- \mathrm{p}$$

(21)

where ${\mathrm{P}}_{\mathrm{LUCC}}$ is the change in the SWAT–PDSI value due to LUCC during a certain period, ${\mathrm{P}}_{\mathrm{LUCCz}}$ is the simulated SWAT–PDSI value when LUCC during period z is used as a variable, and pis the simulated SWAT–PDSI value during the benchmark period.

When climate factors are used as variables, the other factors remain unchanged:

$${ \mathrm{P}}_{\mathrm{cl}}{=\mathrm{P}}_{\mathrm{Cz} }- \mathrm{p}$$

(22)

where ${\mathrm{P}}_{\mathrm{cl}}$ is the total change in the SWAT–PDSI value attributed to climate change during a certain period, and ${\mathrm{P}}_{\mathrm{cz}}$ is the simulated SWAT–PDSI value when climate elements are used as variables during period z.

Simultaneously, we distinguished the impacts of temperature and precipitation on the SWAT–PDSI model. When precipitation is used as a variable, the other factors remain unchanged:

$${ \mathrm{P}}_{\mathrm{p}}{=\mathrm{P}}_{\mathrm{pz}}-\mathrm{p}$$

(23)

where ${\mathrm{P}}_{\mathrm{p}}$ is the SWAT–PDSI change caused by precipitation during a given period, and ${\mathrm{P}}_{\mathrm{pz}}$ is the simulated SWAT–PDSI value when the precipitation during period z is used as the only variable.

When the air temperature is used as a variable, the other factors remain unchanged:

$${\mathrm{P}}_{\mathrm{t}}{=\mathrm{P}}_{\mathrm{tz}}-\mathrm{p}$$

(24)

where ${\mathrm{P}}_{\mathrm{t}}$ is the SWAT–PDSI change due to the temperature during a given period, and ${\mathrm{P}}_{\mathrm{tz}}$ is the simulated SWAT–PDSI value when the temperature during the z period is used as a variable.

Based on the above analysis, the SWAT–PDSI change can be determined as follows:

$${ \mathrm{P}}_{\mathrm{c}}{=\mathrm{P}}_{\mathrm{t}}{+\mathrm{P}}_{\mathrm{p}}$$

(25)

$${ \mathrm{P}}_{\mathrm{Z}}{=\mathrm{P}}_{\mathrm{c}}{+\mathrm{P}}_{\mathrm{h}}$$

(26)

$${ \mathrm{P}}_{\mathrm{h}}{=\mathrm{P}}_{\mathrm{LUCC}}{+\mathrm{P}}_{\mathrm{other}}$$

(27)

where ${\mathrm{P}}_{\mathrm{Z}}$ is the total SWAT–PDSI change during a certain period z relative to the benchmark period, ${\mathrm{P}}_{\mathrm{h}}$ is the SWAT–PDSI change caused by human activities, and ${\mathrm{P}}_{\mathrm{other}}$ is the SWAT–PDSI change due to human activities except for LUCC.

When simulating drought changes in the study area, LUCC data for 1990 were used from 1980 to 1989, LUCC data for 1995 were used from 1990 to 1999, LUCC data for 2005 were used from 2000 to 2009, and LUCC data for 2015 were used from 2010 to 2019.

3. Results and Analysis

3.1. Temporal and Spatial Variations in Drought Based on the SWAT–PDSI Model

In terms of correlation, the R² ratio between the simulated and measured runoff values in the Yihe River Basin was higher than 0.7 during the calibration period (2009–2013) and the validation period (2014–2019), indicating that the change trend between the simulated and measured runoff values was consistent and the correlation was satisfactory. In terms of central variance, the ENS values were 0.72 and 0.70 during the calibration and validation periods, respectively, indicating that the dispersion between the simulated and measured runoff values was low. In terms of standard deviation, PBIAS was less than 20% during both the calibration and validation periods, indicating that the deviation between the simulated and measured runoff values was low. The above results indicate that the SWAT model could attain high simulation accuracy, and could be used to simulate the hydrological change processes in the Yihe River Basin (Figure 2).

From 1980 to 2019, the SWAT–PDSI value in the Yihe River Basin exhibited a decreasing trend (−0.022/10 a), indicating that the degree of drought increased with increasing temperature. Moreover, compared to the reference period (1980–1989), the drought change rate in the basin from 1990 to 1999 was high, at 50.99%. The drought change rates during the periods 2000–2009 and 2010–2019 were similar, at 37.98% and 38.42%, respectively (Figure 3). Spatially, the northeastern and southeastern parts of the Yihe River Basin tended to become wetter, and the maximum change rate of the SWAT–PDSI reached 0.204/10 a, while the central part tended to become drier, and the minimum change rate of the SWAT–PDSI reached −0.53/10 a (Figure 4).

3.2. Climate Fluctuation and LUCC

From 1980 to 2019, the average temperature in the Yihe River Basin experienced a significant increasing trend (0.13 °C/10 a, p < 0.05) (Figure 5a). Spatially (Figure 5c), most of the watershed reflected a warming area, and the temperature increase rate gradually increased from north to south. Among the various regions, the maximum temperature increase occurred in the southwestern part of the basin, reaching 0.26 °C/10 a.

From 1980 to 2019, the precipitation in the Yihe River Basin exhibited a downward trend (−2.7 mm/10 a, p < 0.05), the annual average precipitation ranged from 536.1 to 974.2.1 mm, and the multiyear average precipitation reached 755.1 mm. From the perspective of the spatial change rate of the precipitation in the sub-watershed, the precipitation in the northeastern and southeastern parts of the watershed exhibited a high increase rate, and the maximum value could reach 3.52 mm/10 a, while the precipitation rate at the center indicated a downward trend. Moreover, the maximum decline rate reached −4.85 mm/10 a (Figure 5d). Overall, over the past 40 years, the precipitation in the study area has sharply fluctuated, with a nonsignificant downward trend, and the average annual precipitation change rate reached −2.7 mm/10 a (Figure 5b).

Our research demonstrated that before 1990, the intensity of human activities such as LUCC was relatively low, while after 1990 the intensity of human activities gradually increased, and the impact on the regional climate further increased [49]. Therefore, choosing the land-use types in 1990 and 2015 as examples, the LUCC change characteristics of the Yihe River Basin were analyzed. In terms of the area proportion of the various land-use types, the land-use types in the study area in 1990 and 2015 mainly included dryland, followed by grassland and forestland. The area of built-up land reached 7.35% and 8.75%, respectively, while the area proportion of the other land-use types was small.

In terms of land-use change, from 1990 to 2015, the total land-use change in the Yihe River Basin involved an area of 428 km² (Figure 6 and Figure 7). First, the amount of dryland transfer out of the basin was the largest, with a total area of 149 km², accounting for 34.81% of the total land change area. Second, grassland and forestland were transferred out of the basin at 54 and 22 km², respectively, accounting for 12.61% and 5.14% of the total land change area, respectively. The amount of built-up land transfer was the largest, at 135 km², accounting for 31.54% of the total land change area, followed by water bodies at 45 km², accounting for 10.51% of the total land change area. From 1990 to 2019, the majority of the land-use changes in the Yihe River Basin entailed the conversion of dryland into built-up land and water bodies.

Based on the change rate of the land use, from 1990 to 2015, the increase rate of built-up land was the highest, at 18.96%, followed by that of unused land and water bodies, at 15% and 13.04%, respectively. The wetland reduction rate was the highest, at −15.38%; the next-highest values were attained for grassland and dryland areas, at −3.09% and −1.77%, respectively (Figure 8).

From the transfer of land-use changes in the Yihe River, the transfer quantities of water bodies, forest, and dryland were the most obvious during 1990–1995 and 1995–2005. Among them, the transfer of water to forests was the highest, accounting for 15.33% (636 km²) and 15.51% (700 km²) of the dominant change directions in the two periods, respectively. From 2005 to 2015, the transfer quantities of wetland, unused land, and built-up area were the most obvious, among which the transfer quantity of wetland to unused land was the most significant, accounting for 63.46% of the total transfer area (Figure 9).

3.3. Quantifying the Effects of the Temperature, Precipitation, LUCC, and Other Human Activities on Drought

From 1990 to 1999, the contribution rate of climate change to SWAT–PDSI variation was 68.3%, indicating that climate fluctuation was the main cause of drought changes during this period (Figure 10). Among the various factors, the contributions of changes in precipitation and temperature to drought variation were 38.53% and 29.77%, respectively, indicating that the contribution of precipitation to drought change was greater than that of temperature. LUCC slightly impacted drought, with a contribution rate of only 7.8%, while other human activities exerted a greater impact, with a contribution rate of 23.9%.

From 2000 to 2009, the contribution rate of climate change to SWAT–PDSI variation increased further, reaching 71.52%. This mainly occurred because the contribution rate of temperature to drought changes increased by 33.13%, while the contribution rate of precipitation essentially remained constant (38.39%). During this period, the impact of LUCC began to increase, and its contribution rate reached 18%, while the contribution rate of other human activities to drought change decreased to 10.48%.

From 2010 to 2019, the impact of climate fluctuation on the SWAT–PDSI value continued to increase, and the contribution rate reached 83.28%. Among the different factors, the contributions of precipitation and temperature fluctuations to drought changes reached 40.14% and 43.14%, respectively, suggesting that the relative contribution of temperature to drought change exceeded that of precipitation. At the same time, the degree of influence of LUCC decreased during this period, and the contribution rate reached only 12.6%, while the degree of influence of other human activities on drought variation decreased, and the contribution rate reached 4.12%.

The above results indicate that from 1990 to 2019, climate fluctuation was the main factor influencing changes in SWAT–PDSI values in the Yihe River Basin, with contribution rates higher than 68%, and an increasing trend. Among the various factors, the contribution rates of temperature and precipitation to drought changes gradually increased, but the relative contribution rate of temperature changes increased more rapidly. In other words, before 2010, precipitation fluctuation was the main climatic factor influencing drought changes in the Yihe River Basin, while after 2010 temperature gradually became the main climatic factor determining drought changes.

4. Discussion

Adopting the Yihe River Basin as an example, based on a distributed hydrological model (SWAT), the SWAT–PDSI drought model was constructed to further refine the drought attribution framework. Through a controlled coupling experiment, the degrees of influence of temperature, precipitation, LUCC, and other human activities on drought variation in the basin were quantitatively evaluated.

This study demonstrated that the degree of drought in the Yihe River Basin increased with increasing temperatures from 1980 to 2019, which is essentially consistent with the results obtained by Rajah [50]. Based on a dataset of 12,513 subgrade stations worldwide, changes in the global daily precipitation distribution were studied, and Wu used remote sensing and ground observation data, along with productivity models, to analyze the degree of drought in Northern China [51].

However, the impact of global warming on drought could be offset to a certain extent by the decline in evaporation attributed to wind speed and radiation reduction [14,52,53]. However, this study found that climate fluctuation remained the main driving factor of drought change in the Yihe River Basin, and its contribution rate varied between 68% and 84%. Most studies have demonstrated that precipitation more notably impacts drought than does temperature, but this study found that from 2010 to 2019, the influence of temperature on drought changes was greater than that of precipitation. The reasons for this may include the following aspects: (1) With continuous expansion of the urban scale, the urban heat island effect is gradually strengthened [30]. Impermeable surfaces widely distributed in cities can affect the circulation of energy, water, and chemical substances, resulting in a decrease in surface evapotranspiration and difficulties in surface heat emission. In addition, artificial heat emissions, reduction in the vegetation cover, and other factors could lead to an increase in the surface temperature [54,55,56]. This study determined that the urban construction land area in the Yihe River Basin had increased by 18.96% in 2015 compared to the 1990 level. (2) Recent studies have demonstrated that heatwaves and drought could produce a synergistic effect [57], while heat waves in North China have increased in recent years [58], which has further exacerbated the development of drought. (3) Although the contribution of temperature to drought exceeded that of precipitation after 2010, this does not suggest that the effect of precipitation on drought is decreasing. In contrast, in recent years, the intensity of the Asian polar vortex has significantly decreased in summer, resulting in a reduction in precipitation in North China, and this phenomenon has resulted in a continuous increase in the contribution of precipitation to drought variation [59].

A change in LUCC causes obvious variations in the properties of the underlying surface, including the surface albedo, roughness, vegetation leaf area index, and surface vegetation coverage [60]. A change in reflectivity results in a variation in ground solar radiation absorption. Since the ground is the main heating source of the atmosphere, a change in ground conditions could inevitably lead to a disruption in the atmospheric heat distribution balance and pressure distribution, affecting drought [60]. In contrast, a change in the vegetation cover of the land surface impacts evaporation and even cloud-forming rain conditions, which is also a reason for drought [61]. This study determined that the contribution of LUCC to drought change first increased and then decreased from 1990 to 2019. This may have occurred because the land-use change process in the study area was intense from 1990 to 2009, while after 2010 the land-use change process tended to remain constant. Choosing Linyi—a typical representative city in the study area—as an example, Wang found that the sum of the changes in the various land-use types in the city of Linyi from 1990 to 2000 reached 113.11% [62]. From 2000 to 2010, the total change proportion of the various land-use types in the city of Linyi was 85.98% [62]. Wang also found that the proportion of the various land-use types transformed into construction land in Linyi significantly decreased from 2014 to 2019, and the area of forest and grassland transformed into other land-use types reached 29.25 and 5.52 km², respectively—far smaller than the levels before 2014 [63]. This study revealed that the change rate of construction land from 1995 to 2005 reached 14.91%, while from 2005 to 2015 the change rate of construction land decreased to 4.69%. In addition, in 2011, the General Office of the State Council issued a notice on the implementation of relevant policies in the central region, referring to the old revolutionary base areas in Yimeng, Shandong (GBH [2011] No. 100), which clearly noted that Linyi should be strongly supported with regard to ecological construction and water system connectivity. In 2017, the city of Linyi passed the national pilot acceptance level of water ecological civilization city construction. Therefore, the government has implemented a series of measures to mitigate the impact of drought on ecological construction to a certain extent.

Previous studies have revealed that the impact of human activities on drought in China has further increased [64]. This study demonstrated that from 1990 to 2019, the contribution rate of other human activities (e.g., water conservancy facilities and irrigation) to drought changes in the Yihe River Basin decreased from 23.9% to 4.12%. However, this does not indicate that the impact of human activities other than LUCC on drought variation is decreasing, but that the relative contribution of human activities is lower than that of climate fluctuation. In contrast, in recent years, construction and the optimal layout of ecological engineering, optimal allocation of water resources, and improvement in management efficiency have further reduced the adverse impact of other human activities on drought to a certain extent.

This paper constructed a SWAT–PDSI model that could reflect the impact of the underlying surface on drought, refined the drought attribution framework, and quantitatively evaluated the contributions of climate, LUCC, and other human activities to drought. However, there remain deficiencies in this study. For example, only land-use change information of representative years was used, which could not fully reflect the land-use change process in the basin; contributing factors other than climate and LUCC to the impact of other human activities could yield uncertainty in the research results. These problems will be our main research focus in the future.

5. Conclusions

From 1990 to 2019, the SWAT–PDSI value in the Yihe River Basin decreased at a rate of −0.22/10 a, indicating drought intensification across the basin. The average annual precipitation decreased at a rate of −2.7 mm/10 a, and the decline rate in the central and western parts of the basin was the highest, reaching 4.85 mm/10 a. The temperature rose at a rate of 0.13/10 a, and the increasing trend of the temperature at the center of the basin was the highest, reaching 0.26 °C/10 a.

From 1980 to 2019, the increase rate of built-up land in the study area was the highest, at 18.96%, followed by that of unused land and water bodies, at 15% and 13.04%, respectively. The reduction rate of wetlands was the highest, at −15.38%, followed by that of grasslands and drylands, at −3.09% and −1.77%, respectively.

The impact of land-use changes on drought has increased overall, while the relative contribution of other human activities to watershed drought has gradually decreased. Climate fluctuation is the main factor of drought change in the Yihe River Basin, and its contribution to drought increases year by year. From 1990 to 2009, precipitation was the main factor influencing climate fluctuation. From 2010 to 2019, temperature became the main factor of drought change. LUCC contributed 7.8%, 18%, and 12.6% to drought during the periods 1990–1999, 2000–2009, and 2010–2019, respectively, while the relative contribution of other human activities to drought variation gradually decreased.

The SWAT hydrological model achieved a satisfactory simulation ability and a suitable simulation accuracy regarding the hydrological processes in the study area. The SWAT–PDSI model could be used to study the spatial heterogeneity in soil types, and could be widely employed to reflect the impact of drought on the process of land-use diversification.