The meteorological data are based on gridded data interpolated from daily observations of more than 2,400 meteorological stations of the National Meteorological Information Centre (http://www.nmic.cn/en/) (Xu et al., 2009). The data are interpolated and superimposed by the anomaly approach from the climatological field and the anomaly field, respectively, with a spatial resolution of 250 m for the daily data.

This study is based on the Google Earth Engine cloud computing platform, utilizing the surface reflectance data from Landsat 5, Landsat 7, and Landsat 8 provided by the United States Geological Survey (USGS), with a spatial resolution of 30 m. The imagery data selected spans from 2000 to 2020. The accuracy of this dataset has been confirmed by Wang (2023), who validated its applicability in the Inner Mongolia region through field surveys.

The digital elevation data (DEM) is derived from the geospatial data cloud, using GDEMV 30M data, which is ASTER‘s satellite optical data production (https://www.gscloud.cn/search). The data set has high spatial resolution and can accurately show the terrain features.

The pixel-based dynamic extreme value method was employed to identify extreme temperature and precipitation events for this study (Lu et al., 2018). A 21-day period from 2000 to 2020, centered on the same date, was chosen, making a total of 420 days, and it was sorted. Based on the ranking results, 90% of temperature and 10% of precipitation, respectively, were set as thresholds for deciding extreme events. For example, if the precipitation on the center date of a year is below the threshold or the temperature is above the threshold, an extreme event is identified to have occurred on that date. In the notion, the frequency of extreme climate events is equal to the cumulative number of extreme event days in a year.

Theil-Sen median trend analysis is employed to examine trend variations in long-term time series, as its accuracy remains robust against outliers. The formula is as in Equation 1 (Gong et al., 2024): β = M e d i a n x j − x i j − i , j > i (1)

In the formula, β represents the slope. When it is greater than zero, it indicates that the series shows an upward trend; when it is less than zero, it indicates that the series shows a downward trend; Median refers to the median value.

The Coefficient of Variation (C_V) is a statistical metric that quantifies the relative dispersion within a dataset. It is calculated as the ratio of the standard deviation to the mean and is commonly expressed as a percentage. The formula is as in Equation 2 (Alharbi et al., 2019): C v = ∑ i = 1 n x i − x ¯ 2 x ¯ 2 n − 1 (2) where n is the number of years of study, x_i is the NDVI value of the ith year, and x ¯ is the average NDVI. The larger the C_v, the lower the NDVI stability.

The Hurst exponent serves as a critical tool for analyzing time series data, enabling the measurement of long-term memory and self-similarity within such datasets (Equation 3) (Gu et al., 2023): R S i = R i S i (3) where RS_i is the Hurst index, R_i is the NDVI range of vegetation, and S_i is the standard deviation of NDVI of vegetation. H value between 0.5 and 1 indicating memory enhancement, that is, a time series suggesting long-term memory, which is continuous, an H value between 0 and 0.5 indicating a weakened memory and anti-persistence, and an H value equal to 0.5 indicating that the time series can be described by random walk.

The Pearson correlation coefficient is employed to quantify the linear relationship between two variables (Equation 4) (Li Z. K. et al., 2024): R x y = ∑ i = 1 n x i − x ¯ y i − y ¯ ∑ i = 1 n x i − x ¯ 2 y i − y ¯ 2 (4) where R_xy represents the correlation coefficient between NDVI and drought frequency, x and y represent NDVI and drought frequency, respectively.

The geographical detector assesses the extent to which the independent variable explains the dependent variable by analyzing the geographical and spatial variability of both variables (Yin and Lin, 2024). In this study, the frequency of extreme temperature and precipitation events serves as independent variables, analyzed using the factor detection module of the geographic detector, while NDVI is the dependent variable (Equation 5) (Zhu and Ling, 2024): q = 1 − ∑ h = 1 L N h σ h 2 N σ 2 (5) where q is the explanatory power of the independent variable factor to the dependent variable; σ h 2 and σ 2 are the total variance and the influence factor variance; N and N_h are the total number of samples and the number of driving factor samples, respectively.

The long-term average frequency of extreme precipitation events is 230 days, exhibiting a slight overall growth trend at an average rate of 0.13 days/year (Figure 2). The highest frequency occurred in 2003 (247 days), whereas the lowest was recorded in 2007 (222 days).

The extreme temperature events exhibits significant variability from 2000 to 2020, contrasting with the trend observed in extreme precipitation events. The multi-year average stands at 14 days, indicating an overall downward trend with a decline rate of 0.32 days/year. The minimum extreme temperature events occurred in 2018, followed by 2003, the same year as the peak extreme precipitation events. Compared to the average frequency, the maximum frequency in 2010 reached 25 days, reflecting an increase of 11 days.

The extreme climate events in 2000, 2010, and 2020 exhibit a similar spatial distribution pattern. High-value areas are predominantly located in the northeastern region of the NYSM, and the extent of areas exceeding the multi-year average of extreme events from 2000 to 2020 demonstrates a gradual upward trend (Figure 3). Specifically, the areas surpassing the multi-year average accounted for 52.3%, 56%, and 62.4% of the NYSM in 2000, 2010, and 2020, respectively. Conversely, the central and western regions of the NYSM are identified as low-frequency zones for extreme precipitation events.

During the period from 2000 to 2020, areas with frequent extreme temperature events were distributed in the central and western regions, of which 38% of the area was higher than the multi-year average. The northeastern portion of the research region is characterized by a low-frequency area of extreme temperature events. Over time, the extent of this low-value area initially decreases before subsequently increasing. A spatial comparative analysis of extreme temperature and precipitation events reveals that the low-value area of extreme precipitation events corresponds to the high-value area of extreme temperature events. This phenomenon may be attributed to a corresponding reduction in precipitation within high-temperature regions.

Between 2000 and 2020, the vegetation quality of NYSM exhibited a progressive greening trend (Figure 4). The annual average NDVI was 0.35, with an average growth rate of 0.003. Over this period, the NDVI exceeded the multi-year average for 10 years, reaching a peak of 0.44 in 2018, followed by 0.43 in 2012. The lowest NDVI value, recorded at 0.29 in 2009, was also observed in 2001 and 2007 at 0.30.

An analysis of the spatial variation trend in vegetation NDVI revealed a general upward trend, with areas exhibiting growth accounting for 85% of the NYSM (Figure 5). The central and northeastern regions of NYSM, however, showed a decline in NDVI, constituting 14.8% of the total area. Overall, between 2000 and 2020, a greening trend was observed in the vegetation surrounding the NYSM.

According to the analysis of vegetation NDVI growth stability, areas with a coefficient of variation below 0.2 constitute 51.9% of the NYSM, predominantly located in its western and northeastern regions. This indicates that vegetation development in these areas is highly stable, resilient to extreme climate events, and exhibits significant tolerance to adverse climatic conditions. In contrast, the central region of the NYSM comprises the majority of areas with a relatively high coefficient of variation, suggesting that vegetation in this region is more vulnerable to extreme events.

Areas with a vegetation NDVI standard deviation below 0.04 are predominantly located in the western part of the study area, accounting for 18.3% of the NYSM. This indicates minimal vegetation variability in this region, likely due to the prevalence of unused land and sparse vegetation distribution, which results in a low standard deviation. In contrast, the northeastern region, comprising 13% of the NYSM, contains the majority of areas with a standard deviation exceeding 0.1, reflecting significant vegetation heterogeneity.

Across most of the NYSM, vegetation displayed a continuous state, with areas having a Hurst value greater than 0.5 comprising 12.17% of the region (Figure 6), while those with a Hurst index below 0.5 accounted for 87.83%. Combined with an analysis of vegetation NDVI trend changes (Table 1), the results revealed that 10.21% of the region showed consistent improvement, primarily in the northeast, the western Gobi region, and the southern vegetation restoration area. Conversely, 10.86% of the vegetation exhibited anti-sustainable degradation, predominantly in the Hunshandake Sandy Land and sporadically in the northeast of the NYSM. Random changes were observed in 0.53% of the area, while continuous degradation accounted for 1.9%. Notably, 76.5% of the region demonstrated anti-continuous improvement, nearly encompassing the entire NYSM.

The frequency of extreme precipitation events exhibits a positive correlation with NDVI (Figure 7). Within the NYSM, which is predominantly concentrated in the central and northeastern regions, areas showing a positive correlation constitute 93.7%. Notably, the northeastern region includes areas with correlation values exceeding 0.5. Conversely, the negative correlation between extreme precipitation event frequency and NDVI is primarily observed in the western region, representing 6.3% of the NYSM. Correlation analysis reveals that 25.5% of the NYSM passed the significance test, while 74.6% did not. These findings indicate that, although not highly significant, extreme precipitation events exert a measurable influence on vegetation.

A negative association was observed between the NDVI and extreme temperature events. This negative correlation explained a substantial portion of the NDVI variation, covering nearly the entire NYSM with an area ratio of 98.8%. Compared to precipitation, the significance test results more clearly demonstrate a significant relationship between the NDVI and extreme temperature events. Specifically, 47.8% of the NYSM failed the significance test, while the remaining areas, constituting more than half, passed.

The analysis of correlations between vegetation NDVI and extreme climate events across different land use types indicates that the six land use types predominantly exhibit significant positive correlations with extreme precipitation events, with no significant negative correlations observed. Conversely, extreme temperature events demonstrate negative correlations with most land use types, with significant negative correlations ranging from 21.9% to 60.1% across these types. Notably, no significant positive correlations are observed between extreme temperature events and land use types (Figure 8).

The geographic detector’s factor detector was employed to analyze the effects of extreme temperature and precipitation events on vegetation NDVI in the NYSM (Table 2). Between 2000 and 2020, the frequency of extreme precipitation events exhibited higher explanatory power for NDVI (0.45–0.52) compared to that of extreme temperature events (0.44–0.49). Both types of extreme events demonstrated significant explanatory power for NDVI, as evidenced by their passing the significance level test (p < 0.05) (Xu et al., 2019). Moreover, the relatively high explanatory power of extreme precipitation events underscores their substantial influence on NDVI.

This study examines the response of NDVI to extreme climate events across various land use types (Table 3). The explanatory power of extreme precipitation and temperature events on the NDVI of the six land use types is statistically significant (p < 0.05). Notably, extreme climate events exhibit the strongest explanatory power for the NDVI of grassland and unused land, with q-values exceeding 0.5. Specifically, the explanatory power of extreme precipitation events on the NDVI of cultivated land, forest land, and grassland surpasses that of extreme temperature events. Conversely, the explanatory power of extreme precipitation events on the NDVI of water bodies, construction land, and unused land is lower than that of extreme temperature events.