Remote-Sensed Determination of Spatiotemporal Properties of Drought and Assessment of Influencing Factors in Ordos, China

Abstract

Ordos drought impacts are complex; the Geodetector model is able to explore the interaction between impact factors. Based on the drought severity index (DSI), this study explored the spatio-temporal dynamics and changing trends of drought, and analyzed the driving factors of DSI spatial differentiation by using the Geodetector model. The results show that: the evapotranspiration (ET) and normalized difference vegetation index (NDVI) in Ordos showed a significant increasing trend (p < 0.05). The increasing rates were ET (4.291 mm yr⁻¹) and NDVI (0.004 yr⁻¹). In addition, the interannual variation of the DSI also showed a significant increase, with a trend change rate of 0.089. The spatial pattern of ET and the NDVI was low in the southwest and high in the northeast, and the spatial pattern of potential evapotranspiration (PET) was high in the southwest and low in the northeast, while the distribution of the DSI was dry in the west and wet in the east. The spatial differentiation of the DSI was mainly affected by five factors: air temperature, precipitation, land use type, soil type, and the digital elevation model (DEM), with q exceeding 0.15, which were the main driving factors of drought in the Loess Plateau. Under the interaction of multiple factors, the four combinations of temperature and the DEM, precipitation and the DEM, sunshine duration and the DEM, and relative humidity and the DEM jointly drive drought, in which precipitation (0.156) ∩ DEM (0.248) has the strongest influence on drought occurrence, and q reaches 0.389. This study directly informs specific drought management strategies or ecological conservation efforts in the region.

1. Introduction

Among natural disasters, drought represents one of the most frequently occurring and damaging, with detrimental effects on human wellbeing and agriculture, primarily due to increased temperatures and reduced rainfall [1,2,3]. Droughts also stunt vegetation growth, and can result in forest fires and reduced crop yields [4,5,6]. Hence, accurate and rapid determinations of drought conditions in an area of interest can guide management of agricultural production and drought mitigation.

Initial attempts at drought monitoring through the calculation of a drought index have predominantly relied on ground-based meteorological information or manual measurements. However, this approach suffers from the low numbers of stations, uncertainty through interpolation, and limited representativeness and spatiotemporal coverage [7]. In addition, raster meteorological data is used to calculate drought index [8]. In comparison, remote-sensed data through satellite technology offers rapidly accessible data characterized by a wide spatiotemporal coverage and continuous measurements at both spatial and temporal scales. Consequently, remote-sensed data offer clear advantages for regional-scale drought assessments [9,10,11]. Widely used contemporary remote-sensing drought monitoring indices include the conditional vegetation temperature index (VTCI) [12], the vegetation water supply index (VSWI) [13], the crop water scarcity index (CWSI) [14], the perpendicular drought index (PDI) [15], the modified perpendicular drought index (MPDI) [16], the temperature vegetation drought index (TVDI) [17], and the drought severity index (DSI) [18]. Among the above indices, the DSI integrates surface evapotranspiration information and vegetation response, and is highly accurate with a clear physical definition. Consequently, the DSI is recognized as offering important advantages over other drought indices [19] and has been widely applied and validated in large- and mesoscale drought monitoring studies [19,20,21]. The DSI is a type of agricultural drought [22]. Past drought assessment studies include Gang et al. (2019) [23], who assessed the drought conditions of global grassland ecosystems from 2000 to 2011, determining that most remained in a near-normal condition. Khan et al. (2021) [24] applied the DSI to determine that droughts show repeating patterns at a global scale, with droughts more extreme and frequent in Australia, Africa, and Asia than in South America and North America. Wang et al. (2020) [25] identified the relationship between evapotranspiration and soil moisture to mainly determine the relationship between DSI and soil moisture, thereby accurately describing typical drought processes in Inner Mongolia. An analysis of drought characteristics in Shanxi using the corrected DSI by Liu et al. (2020) [26] showed that the corrected DSI was highly correlated with drought in the study area. While there have been many past studies on remote-sensed drought indices, there has been very little focus on Ordos, China. This has contributed to a reduced accuracy of meteorological drought monitoring in Ordos, exacerbated by limited stations with an uneven spatial distribution. The application of the DSI remote-sensing drought index represents on approach to overcoming the above barriers to assessing drought in Ordos. In addition, the traditional correlation analysis method cannot accurately describe the complex response of drought to driving factors. The Geodetector model can detect the differentiation of space, but also can reveal the driving force behind it, is not limited by the time lag effect, and can be coupled with natural and other influencing factors for analysis [27,28,29]. At present, drought research in Ordos mainly focuses on short periods, while long-term drought monitoring research mainly focuses on the influence of meteorological factors such as precipitation and temperature, and rarely considers the influence of landform or human activities on the occurrence and distribution of drought.

The drought characteristics and drought vulnerability of Ordos City are mainly manifested in aspects such as sparse precipitation, high evaporation, drought climate, and water resource shortages, etc. These characteristics produce local water resource shortages, agricultural production limitations, ecological environment deterioration, and the limitation of social and economic activities under drought conditions. Therefore, it is necessary to take comprehensive measures to deal with the problem of drought, improve the capacity for disaster prevention and reduction, and ensure the sustainability of local economic and social development and the ecological environment. The aim of the present study was to assess the spatiotemporal properties of drought in Ordos and the influencing factors. The main objectives of the current study were to: (1) characterize the spatiotemporal evolution of drought in Ordos and seasonal changes; (2) quantitatively evaluate the influence of each factor on the DSI and their interactions, and; (3) identify the influences of major climatic factors on the DSI. The results of the present study can help further understand changes to the drought status in Ordos, with the broader contribution of acting as a scientific reference for regional disaster prevention and mitigation.

2. Materials and Methods

2.1. Study Area

Ordos City is in southwest Inner Mongolia Autonomous Region, hinterland of the Ordos Plateau, between 37°35′24″~40°51′40″ N and 106°42′40″~111°27′20″ E. The city falls in a typical temperate arid and semi-arid continental climate zone, with multi-year mean temperature and precipitation of 6.2 °C and 348.3 mm, respectively, and 70% of precipitation concentrated between July and September. As shown in Figure 1, the dominant land-use types in the study area are grassland.

2.2. Data Sources

For remote-sensing data, the moderate-resolution imaging spectroradiometer (MODIS) data products were MOD16A2 and MOD13A3, respectively, from the U.S. National Aeronautics and Space Administration (NASA) (https://ladsweb.modaps.eosdis.nasa.gov/, accessed on 1 February 2024). The ET, PET, and NDVI data were extracted, respectively. The spatial resolution of ET and PET was 0.5 km, and the temporal resolution was 8 days; the spatial resolution of the NDVI data was 1 km, and the temporal resolution was 16 days. The grid data of temperature, precipitation, sunshine duration, relative humidity, and wind speed were obtained from the National Earth System Science Data Center, National Basic Conditions Platform for Science and Technology (http://www.geodata.cn/). The spatial resolution of the data was 1 km and the temporal resolution was 1 month. Data for the digital elevation model, land use type, soil type, and surface roughness were obtained from the National Earth System Science Data Center (https://www.geodata.cn/). Among them, slope was extracted according to DEM. Finally, ARCGIS and MRT were mainly used to realize data concatenation, projection conversion, resampling and clipping, and maximum value synthesis. The spatial resolution was unified to 1 km, and the temporal resolution was 1 month.

2.3. Methods

2.3.1. DSI

The DSI integrates the NDVI:PET and ET:PET ratios to consider the state of vegetation and deficits in crop water. Previous studies have applied the DSI to monitor meteorological and agricultural drought. The DSI was calculated as follows [30]:

$$Z_{NDVI}={\displaystyle \frac{NDVI-\overline{NDVI}}{{\delta }_{NDVI}}}$$

(1)

$$Z_{ET/PET}={\displaystyle \frac{ET/PET-\overline{ET/PET}}{{\delta }_{ET/PET}}}$$

(2)

$$Z=Z_{NDVI}+Z_{ET/PET}$$

(3)

$$DSI={\displaystyle \frac{Z-\overline{Z}}{{\delta }_Z}}$$

(4)

where the NDVI and ET/PET are for a set period during the study period; $\overline{NDVI}$ and ${\delta }_{NDVI}$ represent the mean and standard deviation of the NDVI during the study period, respectively; those for ET/PET are $\overline{ET/PET}$ and ${\delta }_{ET/PET}$; those for deviation in Z are $\overline{Z}$ and ${\delta }_Z$. The DSI is proportional to the degree of wetness.

2.3.2. Linear Trend Estimation

Trend analysis through linear regression has been widely applied in drought assessment, and can be calculated as follows [31]:

$${\theta }_{Slope}={\displaystyle \frac{n\times {\displaystyle {\sum }_{i=1}^{n}i\times A_i-{\displaystyle {\sum }_{i=1}^{n}i\times {\displaystyle {\sum }_{i=1}^n{A}_i}}}}{n{\displaystyle {\sum }_{i=1}^n{i}^2-{\left({\displaystyle {\sum }_{i=1}^{n}i}\right)}^2}}}$$

(5)

where A_i represents the mean value of a particular factor in year i, including the ET, PET, NDVI, and DSI; n is the study length.

2.3.3. Geodetector Model

The Geodetector model is an emerging statistical method which identifies the interaction between multiple elements based on spatial differentiation, and reveals the influencing factors behind the above relationships. The core idea of this model is that if an independent variable has a significant impact on the dependent variable, then they should have similar distribution characteristics in space.

The main methods of this study included analyzing the spatial distribution of the DSI in the study area and the influencing factors of the DSI. In order to achieve this goal, the methods of differentiation and factor detector, and interaction detector were adopted in this paper. The differentiation and factor detector aims to investigate the spatial differentiation of dependent variable factor Y, and the explanatory ability of independent variable factor X to the spatial differentiation of dependent variable factor Y [32]:

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

(6)

where L represents stratification of Y or X; N and σ² represent the total samples in the study area and the discrete variance of the entire area, respectively; and N_h and ${\sigma }_h^2$ represent the sample number and the discrete variance of area h, respectively, with h = 1, 2, 3, …, n. Interaction detectors are predominantly applied to detect interactions between influencing factors and whether these interactions act synergistically, antagonistically, or independently on the dependent variable.

The interaction detector was used to identify the interaction between different influence factors, in short, to explore whether the combined action of two factors increased or decreased the explanatory power of the dependent variable factor Y. First of all, the exploration also needed to calculate the explanatory power q value of the two independent variable factors X for Y, then calculate the q value after the interaction of the two independent variable factors, and finally compare the two q values. The results are divided into the following 5 categories (

Table 1. Classification of interaction detector interaction types.

Basis of Judgment	Interaction
q(X1∩X2) < min[q(X1), q(X2)]	Nonlinearity attenuation
min[q(X1), q(X2)] < q(X1∩X2) < max[q(X1), q(X2)]	Single-factor nonlinearity decreases
(X1∩X2) > max[q(X1), q(X2)]	Two-factor enhancement
q(X1∩X2) = q(X1) + q(X2)	Independent
q(X1∩X2) > q(X1) + q(X2)	Nonlinear enhancement

)—q(X₁∩X₂): q(X₁) interacts with q(X₂); min[q(X₁), q(X₂)], the minimum value of q(X₁) and q(X₂); max[q(X₁), q(X₂)], the maximum value between q(X₁) and q(X₂); q(X₁) + q(X₂): sum of q(X₁) and q(X₂).

3. Results

3.1. Characteristics of Spatial and Temporal Variability of Drought

As shown in Figure 2, there were no significant inter-year fluctuations in the ET, PET, NDVI, and DSI in Inner Mongolia during the study period, with significant increasing trends in the ET, PET, and NDVI in Ordos from 2001 to 2020 (p < 0.05) of 4.291 mm yr⁻¹, 0.299 mm yr⁻¹, and 0.004 yr⁻¹, respectively. There was an overall significant increasing trend in the inter-year variation in the DSI; a good fitting linear regression with a slope of 0.089 and R² of 0.639. This result could be attributed to the positive correlation between the DSI and degree of wetness, and suggests that Ordos experienced reduced drought conditions during the study period.

Clear spatial heterogeneity was evident in the distributions of the multi-year mean ET, PET, NDVI, and DSI in Ordos from 2001 to 2020 (Figure 3), with the ET and NDVI increasing from southwest to northeast, PET showing the opposite pattern, and DSI increasing from west to east. The multi-year average ET, PET, NDVI, and DSI varied from 26.515 to 445.730 mm, 430.650 to 1.690.810 mm, 0.028 to 0.816, and −0.230 to 0.209, respectively. The DSI representing drought conditions (low DSI) were mainly concentrated in central Ordos, whereas ET and PET showed the opposite spatial distribution. Areas of higher vegetation cover were mainly in the Yellow Belt and in eastern Ordos.

As shown in Figure 4, ET varied spatially from −7.194 to 18.515 mm yr⁻¹ (Figure 4a), with an increasing trend in ET evident in 72.35% of the area covered by vegetation, and this trend was significant in 67% of the area (p < 0.05) (Figure 5a); a decreasing trend in ET was evident in 27.65% of the area. Spatial variability of PET ranged from −48.358 to 28.323 mm yr⁻¹ (Figure 4b), with an increasing trend in PET in 64.36% of the area covered by vegetation, with this trend significant in 19.34% of this area (p < 0.05) (Figure 5b); and there was a decreasing trend in PET in 35.64% of the area, concentrated in the southwest near the Maowusu Sandland. The spatial variability of the NDVI in Inner Mongolia ranged from −0.037 to 0.038 mm yr⁻¹ (Figure 4c), with an increasing trend in the NDVI concentrated in east Ordos and a decreasing trend concentrated in central and western Inner Mongolia, with this trend significant in 28.94% of this area (p < 0.05) (Figure 5c). The rate of change in DSI ranged from −0.151 to 0.171 yr⁻¹ (Figure 4d), with an increasing trend in DSI in 86.98% of the area covered by vegetation, with this trend significant in 74.65% of this area (p < 0.05) (Figure 5d); and a decreasing trend in DSI was evident in 13.02% of the area, concentrated in central Ordos.

As shown in Figure 6, the spatial change in the DSI indicated varying degrees of drought in Ordos over the last two decades, as well as a decreasing trend in drought. The year in which drought encompasses the largest area of 93.23% was 2001; there was a decreasing trend in normal drought years, with maximal normal drought conditions encompassing 34.95% of the total area in 2011; there was an increasing trend in wet area, with the maximum wet area of 85.57% in 2013. The results showed that droughts dominated Ordos until 2009, after which wet conditions have dominated; maximum areas of drought and normal conditions occurred in 2017, after which there were slight decreasing trends in drought and normal conditions.

3.2. Analysis of Drought Drivers

3.2.1. Detection Factor Impact Analysis

The factor detector can identify the effect of each factor on the DSI, expressed as a power of determinant (PD) value which is proportional to the relationship between the factor and the DSI, and a p-value identifying the significance of the relationship (Figure 7). The results identified temperature and precipitation to be the climate factors with the strongest relationships with the DSI, with a PD of 0.141 and 0.156, respectively, and mean explanatory power exceeding 10%. Sunshine hours and relative humidity had lower relationships with the DSI, with explanatory power of 2.8% and 5.7%, respectively. The DEM and soil type were the surface factors with the strongest relationships with the DSI, with PD values of 0.248 and 0.189, respectively. The temperature, precipitation, soil type, land use type, and DEM showed the strongest relationships with the DSI, indicating the importance of these factors in regulating drought.

3.2.2. Interaction of the Drivers

While the explanatory power of temperature on the DSI was 14.1% (Figure 8), the interactions between temperature and other factors showed nonlinear increasing explanatory power on the DSI. The explanatory factor for the DSI of temperature (0.141) ∩ relative humidity (0.057) was 0.263; temperature (0.141) ∩ DEM (0.248) was 0.375; and temperature (0.141) ∩ soil type (0.189) was 0.317. All above interactions had a DSI explanatory power exceeding 20%. The DSI explanatory power of precipitation of 15.6% was enhanced through interactions with the wind speed, sunshine hours, relative humidity, DEM, land use type and soil type; precipitation (0.156) ∩ DEM (0.248) was 0.389; and precipitation (0.156) ∩ soil type (0.189) was 0.354. The DSI explanatory power of the above two interactions exceeded 35%. While the DSI explanatory power of surface roughness was 2.1%, this was increased through interactions with soil type and wind speed to above 10%. These results stress the important effects of interactions between factors in regulating the DSI.

3.2.3. Impact of Climatic Factors on DSI

As shown in Figure 9, the DSI mainly showed a negative correlation with precipitation in 69.64% of the study area, with the remaining area showing a positive correlation, and these correlations significant in 43.58% of the area (p < 0.05). While the DSI mainly displayed a positive correlation with temperature, a large area in the center of the study area showed a negative correlation. Temperature increases in moisture-deficient areas cannot contribute to evaporation resulting in an increase in PET, and a greater disparity between PET and ET and a reduced DSI. While the DSI showed positive and negative correlations with temperature in 72.25% and 27.75% of the total area, respectively, these correlations were significant in only 0.2% of the area (p < 0.05).

4. Discussion

4.1. Characteristics of Spatial and Temporal Variations of Drought in Ordos

Global warming has led to frequent extreme weather events [33,34,35], and Ordos City is no exception. Climate warming has enhanced evapotranspiration, further aggravating the trend of aridification, and significantly changing the regional ecological environment. Using the DSI to study the characteristics of drought change and its influencing factors is helpful to promote the sustainable development of Ordos agriculture and the ecological environment. The assessment of drought in Ordos from 2001 to 2020 in the present study highlighted the performance of drought mitigation over the last two decades. The results of the present study showed decreasing and increasing trends in the DSI in the west and east, respectively, consistent with the findings of Liang et al. (2019) [36]. The results suggest the highest probability of drought in central-western Ordos, with this probability decreasing outward. In contrast, the highest probabilities of mild drought were in the southwest and east, which contradicts the findings of Peng et al. (2023) [37] who identified the highest probability of drought in the west based on the standardized precipitation index (SPI). This conflicting result can possibly be attributed to differences in the drought evaluation indices used, the basis of the drought class classification, and the study period [38,39,40]. A previous study assessed spatiotemporal properties of drought in Inner Mongolia based on the standardized precipitation-evapotranspiration index (SPEI) according to data obtained from meteorological stations. This approach may not be able to accurately describe drought at regional scales or in areas with very few meteorological stations. Therefore, the present study applied the DSI integrating the ET, PET, and NDVI to evaluate drought. This integration of ET, PET, and the NDVI into the DSI can likely explain the contradictions [41,42,43,44]. The results of the present study indicated mild drought in Yijinhuoluoqi Banner, Dalateqi Banner, Dongsheng City, Hangjin Banner, and Ertok Banner; moderate drought was identified in areas with dominant land use types of grassland, desert grassland, sand, and desert, including eastern Hangjin Banner, northeastern Ertok Banner, and Yijinhuoluoqi Banner. Severe drought was identified in Yijinhuoluoqi Banner and the surrounding areas, which could be attributed to a low rainfall and high temperature. This high evapotranspiration influences the vertical flux of energy and soil moisture, thereby changing soil filtration and triggering drought [45]. Zhungeer Banner and northern Wushenqi Banner were unique, with typical grassland in the former and sandy land in the latter preventing conditions of low rainfall and high temperature causing drought.

4.2. Analysis of the Drivers of the Ordos Drought

The present study considered vegetation within the assessment of drought in Ordos since drought has considerable and complex impacts on vegetation [46,47,48]. This is because drought reduces soil moisture, thereby influencing vegetation growth and development [49]. The knock-on effects are reductions in the productivity and carbon fixation capacity of vegetation [50] as well as changes to vegetation structure and function. This can lead to a disruption of vegetation stability and diversity [51] and to degradation of the ecosystem through interference with energy and material cycles [52]. The responses of vegetation to drought differ among the different vegetation types, including regulation of water absorption, movement, and transpiration by changing root depth, leaf size, stomatal conductance, and osmotic pressure [53]. There is also a negative relationship between altitude and drought given the increase in precipitation at higher altitudes [54]. However, this positive relationship between altitude and precipitation only occurs up to a certain altitude, above which the relationship becomes negative [55]. Consequently, the influences of various factors on drought are complex, and drought can be regulated by climate factors, vegetation, topography, and geographic location [56,57,58]. Especially in the quantitative study of multifactor interactions, the existing methods often lack in-depth quantitative analysis of data, and it is difficult to reveal the interaction between multiple variables, multicollinearity and heteroscedasticity, and ignore the complex interaction between drought and the environment. In order to reveal the driving mechanism of drought change more accurately, the Geodetector model can understand the complex relationship between drought and environment more comprehensively and deeply. Some previous studies examining correlations between soil moisture and climate factors in Ordos found that increasing precipitation and temperature promotes ET, resulting in an increase in the DSI. Their findings suggest increases in precipitation and temperature act to mitigate drought, with the former having the larger impact in Ordos, consistent with the conclusions of Zhong et al. [59]. Given the patterns of change in climate in Ordos from warm-drying to warm-humidification, the dependency of soil moisture on precipitation has decreased. The influence of air temperature on soil condition in the human zone exceeded that of precipitation, and directly regulated water vapor evaporation from the soil and evapotranspiration.

4.3. Uncertainties and Limitations

The results of the present study can act as a reference for effective drought mitigation measures in the future, and can guide management of ecological risk to Ordos posed by climate change. The DSI currently uses 0.5 as the weight coefficient of ET/PET and the NDVI based on experience, without considering the differences in their respective contributions to drought in time and space. There is a time lag in the response of the NDVI to drought, and soil moisture reflects the comprehensive effect of surface precipitation and evapotranspiration, which is one of the key indicators of drought monitoring. Subsequent studies should link the NDVI and ET/PET with soil moisture data to improve the synchronicity of the three responses to drought, and further study the difference in the three contributions to drought in time and space. Based on this, the weight coefficients of the three should be adjusted, and the DSI should be optimized to improve the accuracy of drought monitoring. Given the complexity of drought and the large number of influencing factors, the application of simple meteorological indicators or remote sensing monitoring methods cannot comprehensively reflect the dynamic characteristics of drought. In contrast, the DSI can be considered an effective indicator of drought due to its consideration of remote-sensed indicators and meteorological drought indices. Future research should further improve the utilization rate of remote-sensing data. These studies can help alleviate future agricultural droughts.

5. Conclusions

Drought is a meteorological phenomenon characterized by significant impacts on human activities. In this work, Ordos showed different degrees of drought over the last two decades, with a decreasing trend in drought area, the largest area of drought in 2001 (93.23%), and a decreasing trend in normal drought conditions, with the largest area of normal drought area in 2011 (34.95%). Temperature, precipitation, soil type, land use type, and elevation were the main factors driving aridification in the study area. Interactions between elevation and temperature, precipitation, sunshine hours, and relative humidity were the dominant interactions driving aridification. To sum up, the process of the study area is a multifactor, multilevel complex system. A comprehensive consideration of the drivers and their interaction mechanisms is needed to fully understand and effectively address the problem of drought. The influence of the DEM on temperature, precipitation and other factors is helpful to further study the formation mechanism of drought in Ordos. Through the study of these interactions, a more accurate meteorological drought prediction model can be established to provide a scientific basis for coping with drought disasters. There are desertification problems in some areas of Ordos. The interaction between the DEM and climate factors has an important influence on the development of desertification. The government will be able to formulate corresponding desertification prevention and control policies according to the degree of desertification and climate characteristics in different elevations. For example, in the low DEM and dry areas, the construction of windproof and sand-fixing projects could be strengthened to control the expansion of desertification; in areas with high DEM and relatively high rainfall, afforestation and other measures could be adopted to improve the ecological environment and reduce soil erosion.