Spatiotemporal Analysis of Drought and Its Driving Factors in the Yellow River Basin Based on a Standardized Precipitation Evapotranspiration Index

Abstract

As a natural disaster, drought can endanger global ecology, socio-economic systems, and sustainable development. To address sudden droughts in the future, assess drought disasters, and propose mitigation measures, in-depth research on the spatiotemporal variations in and driving factors of meteorological drought is essential. To study drought in the Yellow River Basin, we calculated the multi-scale Standardized Precipitation Evapotranspiration Index (SPEI), derived from monthly meteorological data recorded at weather stations from 1968 to 2019. We examined the features of drought and its driving factors using the trend-free pre-whitening Mann–Kendall (TFPW-MK) test and Sen’s slope estimator, as well as a drought frequency analysis, center of gravity migration model, standard deviation ellipse model, and geographic detector. Our analysis shows that (1) from 1968 to 2019, the Yellow River Basin exhibited a shift from aridity to increased moisture on an annual basis, with the smallest SPEI of −1.47 in 2002 indicating a moderate drought; SPEI3 showed a growing tendency in all seasons, particularly in winter (0.00388/year), followed by spring (0.00214/year), summer (0.00232/year), and fall (0.00196/year). The SPEI3 exhibited higher fluctuations in frequency compared to the annual-scale SPEI12; (2) in terms of spatial variability, there was no significant change in drought conditions at any scale, with the probability of a drought event being greater in the eastern and northwestern portions of the watershed. The epicenter of the drought exhibited a tendency to migrate southwestward; (3) among the seven driving factors, land use and night lighting were the dominant factors affecting drought conditions, with driving force values of 0.75 and 0.63, respectively.

1. Introduction

Characterized by its wide-ranging impact, high frequency, and long duration, drought has become a global concern [1], profoundly affecting ecosystems, socioeconomic conditions, and sustainable development worldwide [2]. The escalation of global warming and the actions of humans is leading to an increasing frequency and severity of droughts, rendering it one of the most urgent environmental concerns of our time [3]. Among the different types of droughts—meteorological, hydrological, agricultural, and socioeconomic—meteorological drought serves as the foundation for the other three. The latter categories often result from the progression of meteorological droughts. Therefore, monitoring meteorological droughts plays a crucial role in providing early warnings for the other drought categories. To achieve a deeper understanding of how drought has evolved in the study region, a thorough investigation of the temporal and geographical aspects of drought as well as its underlying causes is required [4,5]. This is especially urgent and important in the context of potential future intensifications of droughts, making the monitoring of regional droughts and the formulation of mitigation and prevention strategies more critical than ever [6]. The second-largest river basin in China, the Yellow River Basin, provides water for around one-third of the state’s population [7]. Drought in the Yellow River Basin has led to ecological degradation, reduced agricultural yield, water shortages, and economic losses, posing a serious risk to sustainable development. As climate change intensifies, the severity of droughts increases, making it urgent to address the issue through improved water resource management, ecological restoration, and drought resilience measures. To urgently address the issues posed by drought, specific guidance and strong assistance may be provided for future drought risk assessments and catastrophe management by comprehensively investigating the spatiotemporal features and facilitating variables of droughts in the Yellow River Basin [8].

Drought indexes have become crucial tools in measuring, monitoring, and analyzing drought severity. Among them, the Standardized Precipitation Index (SPI) [9], the Palmer Drought Severity Index (PDSI) [10], and the Standardized Precipitation Evapotranspiration Index (SPEI) [11] are popular choices in research and practice due to their superior applicability and scientific rigor [11,12,13,14]. Compared to SPI and PDSI, the SPEI offers greater flexibility and features a multi-temporal scale. By incorporating potential evapotranspiration data into the SPI framework, SPEI brings together the strengths of both SPI and PDSI while also accounting for the impact of temperature in drought assessment [11]. There are two commonly used evapotranspiration models for calculating the SPEI: the Thornthwaite model and the Penman–Monteith model. The choice of evapotranspiration model can lead to varying results, which in turn affects the accuracy and effectiveness of the SPEI calculation and its application [15,16]. The Thornthwaite model is relatively simple, relying primarily on temperature data to estimate potential evapotranspiration. A more intricate and precise approach that accounts for numerous meteorological variables, like temperature, humidity, wind speed, and sun radiation, is the Penman–Monteith model. This model provides a comprehensive approach to calculating potential evapotranspiration, resulting in an SPEI that more accurately reflects drought severity. However, due to its high data requirements, the calculation process is more complex and necessitates extensive meteorological data [11,15,16]. Chen et al. [17] found that, in China’s arid regions, the Penman–Monteith-based SPEI model (SPEI_PM) runs better in terms of supervising droughts than the Thornthwaite-based SPEI (SPEI_TH). SPEI_PM more effectively captures the drought variability in the study area.

Many scholars have already conducted extensive drought analyses based on SPEI, thus providing insights into the drought patterns and characteristics of various regions. Wang et al. [18] utilized the SPEI in Inner Mongolia; this study examined how drought conditions impacted vegetation over various time spans from 1982 to 2015. They discovered that the likelihood of reduced vegetation production rose as the severity of the drought increased. In 2018, Tian et al. [19] employed the SPEI to examine the extent, duration, and spatiotemporal distribution of the drought in southeast Australia. Mondal et al. [20] assessed drought characteristics in Southeast Asia based on the SPEI and found that drought frequency and impact area increased with increasing radiation intensity under different climate scenarios, aligning with the historical drought records. This implies that, in global warming, the SPEI is suited for supervising and rating droughts. Zhou et al. [21] calculated the SPEI based on two potential evapotranspiration models, Penman–Monteith and Thornthwaite, and found that both SPEI_TH and SPEI_PM can assess the regional drought conditions in humid areas, while SPEI_PM functions better in semi-arid and arid regions.

To investigate the spatial and temporal dynamics of drought within the Yellow River Basin, Huang et al. [22] formulated one non-parametric multivariate drought index that combines the SPI and standardized streamflow index (SSFI). However, the effect of temperature on drought was overlooked, making the assessment less accurate in the context pf global warming. Although Zhu et al. [23] used the PDSI to study drought severity in the Yellow River Basin, the PDSI lacks multi-timescale qualities and is thus unable to adequately depict the multi-scale drought features in this area. Wang et al. [24] conducted an analysis of the spatiotemporal modes of drought in the Yellow River Basin using an SPEI derived from the Thornthwaite model, which does not account for all relevant factors. Furthermore, the majority of previous studies have concentrated on the spatiotemporal distribution of droughts, paying less attention to the underlying causes and long-term patterns in the basin.

This study examines the Yellow River Basin to gain a clearer understanding of drought’s spatiotemporal patterns and the key factors driving these changes. We employ the SPEI as a drought assessment tool, using SPEI12 to evaluate annual-scale spatiotemporal changes in drought and SPEI3 to assess seasonal-scale changes. We systematically analyze the drought conditions in the basin from 1968 to 2019 and use geographic detectors to evaluate the extent to which seven factors influence drought.

2. Materials and Methods

2.1. Study Area

The Yellow River Basin flows through nine provinces and regions: Qinghai, Sichuan, Gansu, Ningxia, Inner Mongolia, Shanxi, Shaanxi, Henan, and Shandong (32–42° N, 96–119° E). The river extends 5464 km, featuring a notable vertical elevation difference of up to 4480 m, and encompasses 795,000 square km in total, making it China’s second-largest river basin. The basin features a diverse range of climates, transitioning from typical plateau and mountain climates in the west to temperate continental and temperate monsoon climates in the east [25]. The area has significant seasonal fluctuations in temperature and 200–600 mm of precipitation on average each year. The mean yearly temperature extends from −4 °C to 14 °C, with the area enjoying many hours of sun [26]. In recent years, the basin has shown an increasing trend in temperature across all seasons, with spring having the most pronounced rise. Meanwhile, the overall evapotranspiration has shown a decreasing trend. Looking at the Yellow River Basin as a whole, there has not been a significant change in precipitation; however, there is a rising pattern of rainfall during the colder months and early spring. Vegetation is more densely distributed in the southeast and less so in the northwest, with the dominant land use types including forests, grasslands, and farmland. Compared to the land use types in the Yellow River Basin in 1968, there were minimal changes up to 2019. The land use type map for 2019 is illustrated in Figure 1.

2.2. Data

The China Meteorological Data Service Center (http://data.cma.cn/) provided the monthly meteorological data from 78 stations in the Yellow River Basin from 1968 to 2019. The data include maximum temperature (°C), minimum temperature (°C), average temperature (°C), precipitation (mm), wind speed (m/s), sunshine duration (h), latitude and longitude (°), and elevation (m). For individual stations with missing data, RClimDex was used to validate and process the input meteorological data. This was accomplished by substituting the missing values with historical averages and data from proximate meteorological stations, thus enhancing the data quality [27]. The DEM data used for the Yellow River Basin were sourced from the Geospatial Data Cloud (http://www.gscloud.cn/), while the land use data for the Yellow River Basin were obtained from the Resource and Environmental Data Cloud Platform (http://www.resdc.cn/). The distribution of meteorological stations is shown in Figure 2.

2.3. Methods

2.3.1. Standardized Precipitation Evapotranspiration Index

The SPEI is generated by taking the discrepancy between monthly precipitation and possible evapotranspiration and normalizing it to attain the final assessment index. The SPEI features multi-temporal scales. The following is the calculation procedure [28]:

(1) The formula for the monthly potential evapotranspiration ($PET$) is shown below:

$$PET={\displaystyle \frac{0.408\Delta \left(R_n-G\right)+\gamma \frac{900}{T+273}{U}_2\left(e_a-e_d\right)}{\Delta +\gamma \left(1+0.34U_2\right)}}$$

(1)

In the formula, $PET$ is the potential evapotranspiration, ∆ is the slope of the temperature versus saturated vapor pressure curve, $U_2$ represents wind speed, $e_a$ is the air saturation vapor pressure, $e_d$ is the actual air vapor pressure, $T$ is the average temperature, $\gamma$ is the psychrometric constant, $R_n$ is the net radiation, and $G$ is the soil heat flux density.

(2) To determine the monthly difference between evapotranspiration and precipitation we used

$$D_i=P_i-{\left(E{T}_0\right)}_i$$

(2)

$$\left\{\begin{array}{c}{D}_{i,j}^k={\displaystyle \sum _{m=13-k+j}^{12}{D}_{i-1,m}}+{\displaystyle {\sum }_{m=1}^j{D}_{i,m}}\begin{array}{cc}& \end{array}j<k\\ D_{i,j}^k={\displaystyle {\sum }_{m=j-k+1}^j{D}_{i,m}\begin{array}{cc}& \end{array}j\ge k}\end{array}\right.$$

(3)

In the equation, $D_i$ is the cumulative discrepancy between precipitation and evapotranspiration with the time scale used for assessment, and $D_{i,j}^k$ represents data beginning in the j-th month of the i-th year, and accumulated over $k$ months.

(3) Studies have shown that the log-logistic probability distribution function provides the best fit when fitting a $D_i$ data series.

$$f\left(x\right)={\displaystyle \frac{\beta }{\alpha }}{\left({\displaystyle \frac{x-\gamma }{\alpha }}\right)}^{\beta -1}{\left[1+{\left({\displaystyle \frac{x-\gamma }{\alpha }}\right)}^{\beta }\right]}^{-2}$$

(4)

In the equation, $\alpha \mathrm{and}\beta \gamma$ are capable of being attained through a linear moment (L-moment) approach, and $\Gamma$($\beta$) is the Gamma function.

$$\left\{\begin{array}{c}\beta ={\displaystyle \frac{2{\omega }_1-{\omega }_0}{6{\omega }_1-{\omega }_0-6{\omega }_2}}\\ \alpha ={\displaystyle \frac{\left({\omega }_0-2{\omega }_1\right)\beta }{\Gamma \left(1+\frac{1}{\beta }\right)\Gamma \left(1-\frac{1}{\beta }\right)}}\\ \gamma ={\omega }_0-\alpha \tau \left(1+{\displaystyle \frac{1}{\beta }}\right)\tau \left(1-{\displaystyle \frac{1}{\beta }}\right)\end{array}\right.$$

(5)

The cumulative probability density function of $D_i$ can be calculated.

$$F\left(x\right)={\left[1+{\left({\displaystyle \frac{x-\gamma }{\alpha }}\right)}^{\beta }\right]}^{-1}$$

(6)

(4) We standardized the result of the above expression to a standard normal distribution. The probability of $D_i$ being more than a certain value was $P=1-F\left(x\right)$, and the probability-weighted moment was $\omega =\sqrt{-2\ln \left(P\right)}$. In the formula, $c_0$ = 2.515517, $c_1$ = 0.802853, $c_2$ = 0.010328, $d_1$ = 1.432788, $d_2$ = 0.189269, $d_3$ = 0.001308.

When $P$ ≤ 0.5,

$$SPEI=\omega -{\displaystyle \frac{c_0+c_1\omega +c_2{\omega }^2}{1+d_1\omega +d_2{\omega }^2+d_3{\omega }^3}}$$

(7)

When $P$ > 0.5,

$$SPEI=-\left(\omega -{\displaystyle \frac{c_0+c_1\omega +c_2{\omega }^2}{1+d_1\omega +d_2{\omega }^2+d_3{\omega }^3}}\right)$$

(8)

Referencing the “Meteorological Drought Classification”, [29] the obtained SPEI drought index is categorized as shown in

Table 1. SPEI drought classification.

Grade	Type	SPEI
1	Extremely humid	SPEI ≥ 2.0
2	Severe humidity	1.5 ≤ SPEI < 2.0
3	Moderate humidity	1.0 ≤ SPEI < 1.5
4	Mild moisture	0.5 ≤ SPEI < 1.0
5	Approaching normal	−0.5 < SPEI < 0.5
6	Mild drought	−1.0 < SPEI ≤ −0.5
7	Moderate drought	−1.5 < SPEI ≤ −1.0
8	Severe drought	−2.0 < SPEI ≤ −1.5
9	Extreme drought	SPEI ≤ −2.0

below

2.3.2. Drought Frequency

We then determined how frequently drought episodes, P, occurred in the basin between 1968 and 2019.

$$P={\displaystyle \frac{n}{N}}\times 100\%$$

(9)

Here, $P$ represents how often drought occurs, while $n$ indicates the number of drought events within the sequence, and $N$ represents the total count of data points within the sequence.

2.3.3. Sen’s Slope

Sen et al. [30]. introduced the Sen’s slope estimator, a non-parametric approach for analyzing trends in time series data. This method is robust against missing data and outliers.

$$Slope=Median\left[{\displaystyle \frac{\left(x_j-x_i\right)}{j-i}}\right],\forall 1\le i\le j\le n$$

(10)

Here, $x_j$ is time series data, and so is $x_i$. $n$ stands for the length of the time series [31].

2.3.4. Trend-Free Pre-Whitening Mann–Kendall Test

The TFPW-MK (Trend-Free Pre-Whitening Mann–Kendall) test is an improvement to the classic Mann–Kendall (MK) trend test method, which is mainly used to deal with time series data with autocorrelation. Since positive autocorrelation in a time series increases the significance test value of the trend, which can lead to misclassification, TFPW-MK removes the effect of coupling between the trend and serial autocorrelation by de-trending and pre-whitening to improve the accuracy of the test results. The method first removes the trend component from the time series through linear regression, and then performs an MK test on the whitened residuals to ensure that the detected trend is independent of the intrinsic autocorrelation properties of the series. Specific calculation steps can be found in references [32,33].

2.3.5. Centroid Shift Model

The centroid shift model can capture the spatiotemporal clustering patterns and dynamic migration characteristics of droughts. By comparing the centroid positions of drought distribution over the course of the study, the patterns of spatial changes in drought distribution in the basin can be identified. The equation used to determine the drought centroid is given below:

$$X_t={\displaystyle \frac{{\displaystyle \sum _{i=1}^n\left(P_{it}{X}_{it}\right)}}{{\displaystyle \sum _{i=1}^n{P}_{it}}}}$$

(11)

$$Y_t={\displaystyle \frac{{\displaystyle \sum _{i=1}^n\left(P_{it}{Y}_{it}\right)}}{{\displaystyle \sum _{i=1}^n{P}_{it}}}}$$

(12)

In the equation, $X_t$ and $Y_t$ represent the longitudinal and latitudinal coordinates of the drought centroid, respectively; $P_{it}$ indicates the drought severity of the i-th pixel in year t; and $X_{it}$ represents the longitude coordinate of the drought centroid for the i-th value in year t, while $Y_{it}$ represents the latitude coordinate.

2.3.6. Ordinary Kriging Model

Ordinary kriging is a geostatistical interpolation method widely used across various fields. It models the spatial autocorrelation of known data points to predict values at unknown locations, serving as a classic general linear regression model. The specific calculation formulas are provided below:

$$\stackrel{\wedge }{z}\left(s_0\right)={\displaystyle \sum _{i=1}^n{\lambda }_{i}z\left(s_i\right)}$$

(13)

$\stackrel{\wedge }{z}\left(s_0\right)$ is the value at location $s_0$ to be interpolated. $z\left(s_i\right)$ refers to the sampled values, and ${\lambda }_i$, determined by the semivariogram modeling, refers to the weights to be assigned to each unsampled location.

2.3.7. Geodetector

The geodetector is aimed at detecting and applying the spatial heterogeneity, serving as a statistical approach to uncover driving factors behind spatial differentiation [34]. The Excel plugin downloaded from the GeoDetector website was used for data analysis and processing. The differentiation and factor detection components of this tool can identify spatial heterogeneity and assess the explanatory capacity of a specific factor for this change, determined by the q-value [35]. The expression is given below:

$$q=1-{\displaystyle \frac{{\displaystyle {\sum }_{h=1}^L{N}_h{\sigma }_h^2}}{N{\sigma }^2}}=1-{\displaystyle \frac{SSW}{SST}}$$

(14)

$$SSW={\displaystyle \sum _{h=1}^L{N}_h{\sigma }_h^2\begin{array}{cc}& \end{array}}SST=N{\sigma }^2$$

(15)

In the expression, $h=1$, $L$ refers to the stratification of variable $Y$ or the factor. $N_h$ and $N$ represent the number of units in layer $h$ and the entire region, separately. ${\sigma }_h^2$ and ${\sigma }^2$ are the variances in $Y$ within layer $h$ and the entire region, separately. $SSW$ and $SST$ indicate the sum of within-layer variances and the total variance of the total area, separately. If the $q$-value is high, between 0 and 1, there will be a stronger effect of the specific factor.

3. Results

This Section shows the results of a spatiotemporal analysis and evaluation of drought conditions on both annual and seasonal scales. Section 3.1.1 analyzes the trends in SPEI12 from 1968 to 2019, indicating an overall upward trend with notable fluctuations during specific periods. Section 3.1.2 examines the seasonal characteristics of SPEI3 in the Yellow River Basin, revealing that drought conditions in all seasons are on the rise, with significant seasonal variability. Winter was found to be the least prone to drought, while autumn was the most drought-prone period, underscoring the critical influence of seasonal climatic variations on drought conditions. Section 3.2.1 investigates the spatial and temporal patterns of drought in the Yellow River Basin through statistical tests, interpolation, and centroid migration models, identifying regional disparities, a southwestward shift in drought centroids, and stable clustering trends, which highlight risks to agriculture and water resource management. Section 3.2.2 focuses on seasonal drought patterns, uncovering distinct seasonal and spatial variations. Spring and autumn exhibited higher drought frequencies, notable centroid migration trajectories across seasons, and relatively stable drought conditions in summer and autumn, emphasizing the necessity of tailored drought management strategies. Finally, Section 3.3 identifies land use and nighttime lights as the primary factors influencing drought in the Yellow River Basin, with precipitation, temperature, and population also playing significant roles, whereas sunshine duration and per capita GDP had minimal impacts.

3.1. Temporal Characteristics of Drought

3.1.1. Annual Scale

SPEI12 had a general upward trend between 1968 and 2019, increasing at a rate of 0.00362 per year (Figure 3). This trend may be related to climate change, shifts in precipitation patterns, and variations in potential evapotranspiration.

As shown in Figure 3, the magnitude of changes in SPEI12 exhibits distinct variations across different periods. Notably, the fluctuation amplitudes during 1968–1973 and 2002–2008 are more pronounced. SPEI12 reached a peak value of 1.74 in 2012, indicating significantly wetter conditions, which resulted in multiple flood events during the flood season [36]. In contrast, the SPEI value in 2002 was −1.47, indicating a moderate drought condition [37]. The SPEI values in 1983, 1989, and 2012 all exceeded one, at 1.12, 1.58, and 1.74, respectively, indicating moderately wet conditions. These wetter years were favorable for agricultural production, water resource management, and ecosystem protection. On the other hand, the years 1969, 1971, 1972, 1997, 2002, and 2006 exhibited moderate drought conditions, necessitating appropriate water resource allocation and drought mitigation measures.

3.1.2. Seasonal Scale

A more accurate understanding of seasonal fluctuations in drought is possible when analyzing the drought features in the Yellow River Basin by splitting the temporal scale into spring, summer, fall, and winter. The following seasonal divisions were used in this research: March to May (spring), June to August (summer), September to November (autumn), and December to February of the following year (winter). We were able to establish a map of seasonal variation in SPEI3 for the Yellow River Basin to illuminate these changes (Figure 4).

As shown in Figure 4, the SPEI3 values exhibit an increasing trend across all seasons, increasing by 0.00214 per year for spring, 0.00232 per year for summer, and 0.00196 per year for autumn. Winter shows a more pronounced annual growth percentage of 0.00388. The increasing tendency at the seasonal scale indicates that long-term climate change has impacted drought situations in the basin. Seasonal fluctuations are also significant, with SPEI3 having a higher frequency of change than SPEI12. This variability is related to seasonal changes in precipitation and evapotranspiration. Precipitation and potential evapotranspiration are key factors influencing SPEI3, and their interannual and seasonal variations directly affect the drought index. Comparing the seasonal and annual scale drought characteristic maps reveals significant differences. The annual scale SPEI12 changes appear smoother, reflecting the overall trend of drought conditions throughout the year. In contrast, the seasonal scale SPEI3 shows a higher fluctuation frequency, indicating alternating periods of drought and wetness within the research area. Winter has the lowest chance of drought in the basin, in light of a deep examination of the seasonal SPEI3 data, which shows that winters with SPEI values less than −1 had the fewest number of incidents. This phenomenon is related to the lower evapotranspiration rates and relatively higher precipitation during winter. Autumn, on the other hand, has the largest frequency of SPEI values below −1, suggesting that it is the season with the biggest risk of drought. This is particularly concerning as autumn is typically a critical time for crops with high water demand. Drought occurrence during this period can severely impact agricultural production. Spring, as a transitional season with gradually increasing temperatures and precipitation, shows an upward trend in SPEI3, indicating a reduction in drought risk. Summer, traditionally a wet season, also exhibits an upward trend in SPEI3, with the relatively high precipitation reducing the likelihood of drought.

3.2. Spatial Characteristics of Drought

3.2.1. Annual Scale

In this work, the TFPW-MK test and Sen’s slope were applied to explore SPEI12. Then, our results were subjected to Kriging interpolation analysis in ArcGIS, producing the trend analysis map shown in Figure 5. It is clear from Figure 5 that the Yellow River Basin has notable regional variations in drought and wetness fluctuations. The eastern and core basin portions are indicative of most regions with an elevated SPEI12, suggesting a shift from formerly dry to wet conditions over the research period. This shift is likely associated with increased precipitation, decreased potential evapotranspiration, or a combination thereof. Conversely, areas with a decreased SPEI12 are concentrated in the northern and western basin, reflecting a transition from humid to arid conditions. This trend poses challenges to local water resource supply, agricultural production, and ecosystem balance.

The TFPW-MK test results for the SPEI12 data from 78 meteorological stations in the Yellow River Basin indicate that none of the stations exhibited statistically significant trends during the study period. However, 48 stations show a non-significant increasing trend, suggesting a tendency towards wetter conditions, while 30 stations display a non-significant decreasing trend, indicating a potential shift towards drier conditions. These findings suggest that while localized variations are evident, the overall trend across the basin remains stable.

To enhance the frequency and spatial distribution of drought occurrences at various stations within the basin, Table 1 categorizes the SPEI from 1968 to 2019 into nine distinct classes. For statistical and analytical reasons, the years classified as mild, modest, acute, and extreme drought are considered drought years in this study. We created a regional distribution map of the frequency of drought in the Yellow River Basin (Figure 6) via thoroughly classifying and analyzing the SPEI drought index.

As displayed in Figure 6, areas with an elevated drought frequency are chiefly located in the east and northwest of the basin, showing a frequency as high as 23%. This indicates the widespread and severe nature of drought in these regions during the study period. In contrast, areas with lower drought frequencies lie in the western and northeastern basin, showing the lowest frequency of only 2%.

The major plains and low hills of the eastern Yellow River Basin are primarily responsible for the increased frequency of droughts in the region. This topography supports agricultural productivity, but it also facilitates rapid water evaporation and runoff, making it difficult for moisture to remain in the soil for extended periods. Moreover, the irregular seasonal pattern of rainfall in the eastern area—especially the low precipitation in winter and spring, which are crucial for crop water needs—intensifies the mismatch between water supply and demand. In contrast, the western basin of the Yellow River experiences a lower frequency of drought. This region, characterized by plateaus and mountainous terrain, benefits from higher elevations that promote moisture condensation and precipitation. Additionally, snowmelt from the high mountains provides supplementary water resources, contributing to a relatively higher availability of water. The western region also has lower evaporation rates due to cooler temperatures and higher altitudes, which aid in the retention of water resources.

Additional variables may potentially impact the regional drought frequency distribution in the Yellow River Basin, according to further studies. Drought features, for example, may be impacted by basin-wide activities connected to the development and application of water resources, like agricultural irrigation and reservoir building and management. Drought frequency may also be impacted by human activities, including variations in green cover and land use. In terms of climate change worldwide, the drought probability in some areas may rise due to an increase in climate extremes. To more accurately understand and predict the risk of drought in the basin, future research needs to integrate these multiple factors and employ advanced drought monitoring and prediction models. In order to improve the basin’s ability for sustainable development and drought resilience, there should also be an emphasis on methods for managing water resources and modifying agricultural operations during dry spells.

The SPEI values for the years 1970, 1980, 1990, 2000, and 2010 were systematically calculated using the centroid migration model in order to provide a thorough examination of the spatial variation features of drought occurrences in the basin. By applying the centroid migration model, the centroid positions of drought events can be determined, and connecting these centroids chronologically produces a centroid migration trajectory map. As shown in Figure 6, this map visually presents the geographic distribution of drought cases and their long-term trends in the Yellow River Basin.

From 1970 to 2010, the centroid of drought events in the basin was primarily concentrated in the central region of the basin (Figure 7). However, over time, the centroid demonstrated a continuous migration pattern rather than remaining stationary. This migration was influenced by various factors, including climatic changes, human activities, and hydrological cycle variations.

In 1970, the centroid was located near the center of the basin, indicating that this region experienced the most severe drought impacts during this period. The central location reflects the concentration of extreme drought events caused by a combination of low precipitation levels and high evapotranspiration rates, which were particularly prominent in the core areas of the basin. Additionally, limited water resource management strategies at the time contributed to the heightened sensitivity of this region to drought conditions. By 1980, the centroid had shifted slightly to the northeast, reflecting minor changes in climate conditions and their effects on drought distribution. In 1990, the centroid migrated slightly to the southwest, indicating a gradual redistribution of drought occurrences, possibly driven by local shifts in precipitation patterns.

By 2000, the centroid had moved further to the southwest, a shift likely attributable to changes in precipitation regimes and the availability of water resources. In 2010, the centroid continued its southwestward migration over a more significant distance, suggesting an increased spread of drought risk toward the southwestern part of the basin.

The standard deviation ellipse diagrams for the years 1970, 1980, 1990, 2000, and 2010 are presented in this work in order to better investigate the distribution features, geographical clustering, and changes in spatial centroid and direction of drought emergence in this basin (Figure 8).

The parameters of the standard deviation ellipse are generally stable, showing only slight variations. The area of the ellipse does not exhibit significant fluctuations, while the length of the semi-major axis shows a trend of increase. Drought episodes mostly occurred in the middle of the Yellow River Basin between 1970 and 2010. Over time, both the quantity and direction of drought events have shown some variations, closely related to the region’s climatic characteristics, hydrological and geological conditions, and land use patterns. As a key agricultural area in the basin, the frequent emergence of drought cases within the central region presents a remarkable risk to local agricultural manufacturing and food safety. The migration trajectory and direction of the spatial centroid of droughts align with the outcomes predicted by the centroid migration model, demonstrating the model’s effectiveness and accuracy in obtaining the spatial trend of drought events.

3.2.2. Seasonal Scale

The trend analysis findings for each season were obtained by using the Sen’s slope estimator and TFPW-MK test to the SPEI3 information for spring, summer, fall, and winter (Figure 9).

The Yellow River Basin’s SPEI3 index clearly demonstrates seasonal change, as seen in Figure 9. Across all four seasons—spring, summer, autumn, and winter—none of the stations exhibit statistically significant trends. This indicates that while localized variations in wetting and drying tendencies are present, the overall changes remain non-significant throughout the study period. Furthermore, in each season, the spatial distribution of trends shows a mix of stations with increasing (wetting) and decreasing (drying) tendencies, reflecting the heterogeneity of hydroclimatic conditions within the basin.

Despite these overarching similarities, notable seasonal differences emerge in the proportion and spatial distribution of wetting and drying trends. In spring, a larger number of stations (45) display an increasing trend compared to those showing a decreasing trend (33), suggesting a slight dominance of wetting tendencies. Conversely, summer exhibits the opposite pattern, with 34 stations showing increasing trends and 44 displaying decreasing trends, indicating a more widespread tendency towards drying conditions. This drying trend becomes even more pronounced in autumn, where 47 stations show decreasing trends compared to 31 with increasing trends. Winter presents a relatively balanced distribution, with 38 stations exhibiting increasing trends and 40 showing decreasing trends, indicating minimal seasonal bias toward wetting or drying.

Considering the seasonal and spatial variability in drought trends observed in the Yellow River Basin, targeted measures should be implemented to mitigate potential impacts. In spring, regions with stronger wetting trends, particularly in the central and southern parts of the basin, should focus on enhancing water storage infrastructure, such as reservoirs and rainwater harvesting systems, to retain excess water for use during drier seasons. During summer and autumn, when drying trends are more pronounced in the northern and northeastern regions, efforts should prioritize the adoption of drought-resistant crop varieties, the optimization of irrigation systems, and the promotion of water-saving agricultural practices. In winter, areas experiencing localized wetting trends in the southeastern part of the basin can leverage this water availability to recharge groundwater resources and enhance soil moisture for the following growing season. These strategies, tailored to the spatial and seasonal characteristics of the region, can effectively address the diverse drought challenges across the basin.

There are notable seasonal and regional differences in drought incidence, as shown by the statistical study of drought frequency in the Yellow River Basin in four seasons (Figure 10). The distribution areas with high drought frequencies differ by season, but overall, spring and autumn exhibited relatively higher drought frequencies, with autumn showing the highest frequency by 24.5%. Droughts in spring were widespread across the basin, with a relatively narrow range of frequency variations, indicating a certain level of universality in the basin.

In contrast, drought frequency distributions in summer and winter are more localized. While winter droughts are widely distributed and demonstrate notable regional differences in drought risk, summer droughts are mostly in the southern and western parts of the basin. Overall, the pronounced drought frequencies in spring and autumn highlight the critical importance of these seasons for drought risk management in the Yellow River Basin, presenting significant challenges for regional water resource distribution and management.

Using the centroid migration model, we simulated the centroids of drought case at the seasonal scale, as indicated in Figure 11. The research provided fresh information on the seasonal movement of drought events in the basin by revealing that the centroids of drought events vary be-tween seasons within the same year. The migration trajectory of the centroid for spring drought events indicates that, over time, the centroid predominantly shifted in the northeast, southwest, southeast, and ultimately southwest directions. The most significant shift occurred between 2000 and 2010, which is related to increasing spring temperatures, rising evaporation rates, and growing agricultural water demand. These factors together may have contributed to a heightened risk of spring droughts. During summer, the centroid migration pattern is characterized by a distinct trajectory, proceeding the northeast, southwest, and west, and returning to the northeast. In the Yellow River Basin, summer normally becomes the wettest season; however, centroid migration suggests that there is still a chance of dryness in the center area during this time. During autumn, the centroid migration trajectory progressed southwest, northeast, northwest, and subsequently southwest, with the most pronounced shift occurring between 2000 and 2010. The intensification of autumn droughts is associated with reduced precipitation and relatively increased evaporation during this season. During winter, the centroid migration trajectory was southwest, southwest, southwest, and subsequently southeast, with the most significant displacement occurring between 1980 and 1990. Observing and analyzing the centroid migration trajectories for all four seasons reveals the seasonal variations in aridity risk in the Basin. These findings are crucial for developing seasonal water resource management strategies, optimizing agricultural irrigation plans, and implementing effective drought risk mitigation measures.

For the years 1970, 1980, 1990, 2000, and 2010, standard deviation ellipses were calculated for each season (Figure 12). In the standard deviation ellipses for the four seasons, the parameters for spring, summer, and autumn are generally stable, with only slight variations. It can be seen that, from 1970 to 2010, droughts predominantly affected the core area of the Basin. In winter, the area of the ellipse shows a trend of increasing fluctuations, while the length of the semi-major axis gradually increases. The standard deviation ellipses for 1970 and 1980 showed a significant reduction in coverage, indicating a high degree of spatial variability and suggesting that drought events in these years were more spatially concentrated. On the other hand, the summer and fall standard deviation ellipse distributions did not exhibit any notable changes, suggesting that the middle Yellow River Basin’s drought conditions remained rather stable over these seasons.

3.3. Analysis of Drought Drivers

In this study, seven influencing factors from meteorological, topographical, and socio-economic domains were selected. Remote sensing data on night-time lighting was used as a factor to consider the impact of human activities on drought [38]. Factor detection analysis was performed on the data from 2005, 2010, and 2015 to explore their impact on drought in the Yellow River Basin.

Table 2. Factor detection results for 2005, 2010, and 2015.

Code	Name	Definition	2005 q-Value	2010 q-Value	2015 q-Value	Average Value
1	Land use	Land use type	0.69	0.70	0.86	0.75
2	Night lights	Remote sensing data on night-time lighting	0.63	0.58	0.69	0.63
3	Per capita GDP	GDP per capita for the year	0.13	0.26	0.51	0.30
4	Population	Total population at the end of the year	0.51	0.52	0.51	0.51
5	Average precipitation	Average annual precipitation	0.55	0.54	0.59	0.56
6	daylight hours	Annual sunshine hours	0.02	0.70	0.41	0.38
7	average temperature	Average annual temperature	0.62	0.41	0.62	0.55

displays our results.

Based on the factor detection results for 2005, 2010, and 2015, the following observations can be made:

In 2005, land use types significantly affected drought in the Yellow River Basin, with a q-value of 0.69, whilst sunshine duration had a minimal influence, with a q value of only 0.02. In 2010, both land use type and sunshine duration had the largest influence, each showing a q-value of 0.70, whereas per capita GDP had the least impact, with a q value of 0.26. By 2015, land use type remained the most influential factor with a q value of 0.86, while sunshine duration exerted a minimum influence showing a q-value of 0.41.

Averages of the results from the three years reveal the following ranking of the seven driving factors based on their q values: land use (0.75) $>$ nighttime lights (0.63) $>$ average precipitation (0.56) $>$ average temperature (0.55) $>$ population (0.51) $>$ sunshine duration (0.38) $>$ and per capita GDP (0.30). This analysis indicates that land use exerts a minimal influence on drought severity in the basin, followed by nighttime lights, with both having q values greater than 0.60. This suggests that land use type and nighttime lights act as the primary factors driving drought in the basin. Average precipitation, average temperature, and population also have significant effects on drought variations, as their q values are all above 0.50. On the other hand, the q values for sunlight duration and per capita GDP are lower, suggesting that these variables exert a less remarkable influence upon drought than the other variables.

4. Discussion

The SPEI is used to better understand the spatiotemporal development of drought and its underlying causes in the Yellow River Basin. SPEI12 and SPEI3 were employed to analyze the spatiotemporal characteristics of drought at the seasonal and yearly scales, respectively, covering 1968 to 2019. The results show that, in general, there has been no significant change in the drought trend in the Yellow River Basin. At the seasonal scale, the SPEI values fluctuated more frequently, a finding that is the same as that of Jin et al. [39] They used SPEI and PDSI to comparatively analyze and study the drought conditions in the Yellow River Basin. The year 2012 saw the highest SPEI12 value, indicating a significant wet condition, resulting multiple flooding events [36]. On the other hand, 2002 had the lowest SPEI12 score, indicating a mild drought situation that was probably caused by global warming affecting the local climate [40].

From a seasonal perspective, there was also no significant change in drought conditions across the four seasons, but again, there was a great deal of variability between seasons. In winter, humidity increased from the SPEI perspective, as there were fewer years with SPEI values below −1 in this season, probably due to lower evapotranspiration rates [41]. In contrast, autumn recorded the most years with low SPEI values, with a drought occurrence frequency of 24.5%, making this season’s drought particularly impactful on agricultural production. Fu et al. [42] also found that autumn experienced the most severe droughts, with drought conditions more pronounced compared with other seasons.

Subsequent geographical study showed that there is a considerable amount of spatial variability in the danger of drought throughout the basin, with eastern areas experiencing more frequent droughts than the western ones. This finding is consistent with Xu et al.’s [43], primarily on account of the marked decline in precipitation in the east. Geng et al.’s results also indicated that the interannual and seasonal variations in drought in the basin are shifting towards wetter conditions, with the most pronounced autumn droughts in the middle of the basin [44]. Additional research supported the seasonal drought’s geographical distribution features in the Yellow River Basin. This has significant ramifications for managing drought management and water resources. Unlike the space–time analysis on the Yellow River Basin conducted by Zhou et al. [45], this study further analyzes the long-term variation in the space of drought centers and the change in drought-prone areas using the centroid migration model and the standard deviation ellipse model. Understanding seasonal drought patterns can help in developing more targeted mitigation measures.

When discussing the driving factors of drought, both natural and human factors play important roles [46,47]. This study analyzed seven potential influencing factors and found that land use types and nighttime human activities had the most significant impacts on drought variation [48]. A novel approach in this study was the use of nighttime light data as a proxy indicator for nighttime human activity, revealing the considerable influence of human activities on drought [49]. This emphasizes how important it is to properly account for how human activity and environmental variables interact when developing methods for managing and responding to droughts.

In the analysis, land use was considered a driving factor and compared with other factors. In the future, different land uses can be analyzed as individual factors to assess their impacts on drought conditions.

5. Conclusions

This study conducted a systematic analysis of the spatial and temporal evolution of drought in the Yellow River Basin from 1968 to 2019 using the SPEI and investigated the driving mechanisms behind drought changes in detail. The key conclusions are outlined as follows: From 1968 to 2019, at the annual scale, the results indicate that no station exhibits a statistically significant long-term trend, suggesting overall stability in drought conditions during the study period. However, the seasonal-scale analysis reveals pronounced variations, with distinct wetting and drying trends observed across different regions and seasons. For instance, spring exhibits a dominance of wetting trends, while summer and autumn show more widespread drying tendencies, particularly in the northern and northeastern regions. These findings underscore the complexity of drought dynamics in the basin, where localized seasonal changes may not be reflected in annual trends.

Drought events were more frequent in the eastern and northern parts of the basin, and the center, frequency distribution, and clustering of these events showed a spatial shift towards the southwest over time, highlighting the spatial complexity of drought changes.

The application of geographic detectors and other methods revealed the main factors affecting drought conditions in the basin and their degrees of influence. Among these, land use type and nighttime lighting were the dominant factors, emphasizing the direct and indirect impacts of human activities on drought conditions. In addition, factors such as average precipitation, average temperature, population, sunshine hours, and per capita GDP also influenced drought changes to varying degrees, yet with weaker impacts.