China is located in the eastern part of the Eurasian continent, spanning 3°51′N–53°34′N in latitude and 73°29′E–135°04′E in longitude, with a total land area of about 9.6 million square kilometers (Figure 1A). From east to west, elevations rise from the plain in the east to the plateau in the west, showing a three-step distribution (Wang et al., 2017). China contains a wide variety of terrain types, including mountains, plains, basins and plateaus (Figure 1B); there is also a variety of climate types from tropical to frigid zones. Due to China’s clear geographical and regional differences and the influence of climate types, there are significant monthly, annual and interannual changes in temperature and precipitation. In general, there is more precipitation in the east and less in the west; temperatures are high in the south and low in the north, respectively. China’s rich climate diversity also leads to diversified vegetation communities and land use types. The main land use types include cropland vegetation, forest vegetation, grassland vegetation, desert and water bodies (Figure 1A). Cropland vegetation is primarily found in the northeast and central regions, while forest vegetation is predominantly located in the northeast and southeast regions. Grasslands are mainly distributed across the Inner Mongolia Plateau and Qinghai-Tibet Plateau, while deserts are primarily situated in the northwest. The vegetation types in this study were mainly grassland, forest and cropland. The land in northeastern China is mainly dominated by black soil, while the mountainous and plain areas in tropical and subtropical China are mainly characterized by yellow soil. In areas with high temperatures and abundant rainfall in China, the land is primarily red soil.

To explore the response of different vegetation system types to PET and P, this study referred to the research method of Ding et al. (2021) and roughly divided the study area into eight climate zones (Figure 1C): (I) Cold temperate coniferous forest, temperate monsoon climate; (II) Temperate coniferous forest and deciduous broad-leaved mixed forest, temperate monsoon climate; (III) Warm temperate deciduous broad-leaved forest, temperate monsoon climate; (IV) Subtropical evergreen broad-leaved forest, subtropical monsoon climate; (V) Tropical monsoon forest and tropical rain forest, tropical monsoon climate; (VI) Temperate grassland, temperate monsoon climate and temperate continental climate; (VII) Temperate desert, temperate continental climate; and (VIII) Vegetation in the cold areas of the Qinghai-Tibet Plateau, alpine plateau climate. Climate zone I is located in the northern part of the Greater Khingan Mountains, climate zone II includes the Northeast Plain and Changbai Mountain, climate zone III includes the North China Plain and the Loess Plateau, and the geographical types of climate zone IV are the lower reaches of the Yangtze River Plain, the Yunnan-Guizhou Plateau, the Sichuan Basin and the southeast hilly area. Climate zone VI includes the Inner Mongolia Plateau, climate zone VII includes the Junggar Basin and Tarim Basin, and climate zone VIII is almost entirely located on the Qinghai-Tibet Plateau (Figures 1B, C). Average annual potential evapotranspiration and annual average cumulative precipitation for each climate zones, as shown in Table 1 (Hao et al., 2019).

This study used NDVI as an indicator to characterize the dynamic changes of vegetation system types. The selected NDVI data set is the third-generation NDVI (NDVI 3 g) data developed from the Advanced Ultra High Resolution Radiation Sensor (AVHRR) by the Global Inventory Modeling and Mapping Research (GIMMS) group (http://ecocast.arc.nasa.gov/). Beck et al. (2011) and Tian et al. (2015) found that this was the best comprehensive NDVI product for monitoring long-term land vegetation dynamics. The spatial resolution of the data set is about 8km, and the temporal resolution is 1 month. The time series selected in this study was the latest data version from 1982 to 2015. Cihlar et al. (2004) and Zeng et al. (2013) corrected NDVI for orbital drift, viewing geometry, snow cover, volcanic aerosols, ozone cover, atmospheric water vapor, cloud cover, and other errors not related to vegetation change. Pinzon and Tucker (2014) calibrated parameters for AVHRR. We adopted the Maximum Value Composite (MVC) technique (Holben, 1986) to aggregate the time steps of the original NDVI sequence into a monthly scale to further reduce the impacts of clouds and airborne particles.

The monthly-scale meteorological data (1982–2015) used in this study came from 699 meteorological sites (Figure 1C) (http://data.cma.cn/site/index.html) and were downloaded from the China Meteorological Data Sharing Website. These include maximum temperature, minimum temperature, precipitation, wind speed, relative humidity and solar radiation. We eliminated outliers and impute missing records to ensure the quality of the data. A Digital Elevation Model (DEM) dataset with a spatial resolution of 1 km was downloaded from the NASA Space Shuttle Radar Topographic Mission (SRTM) website (http://www.glcf.umd.edu/). We supported spatial climate interpolation by providing terrain information. The thin-plate smoothing spline method (Hancock and Hutchinson, 2006) was used to evaluate the influence of elevation and atmospheric level on spatial climate interpolation, so all meteorological data were processed into a monthly scale grid cells layer with a spatial resolution of 8 km.

The vegetation ecosystem type data with a spatial resolution of 1 km were developed by the China Resource and Environment Data Cloud Platform (http://www.resdc.cn/Default.aspx) (Wei et al., 2018) (Figure 1A). Grid cells contain vegetation ecosystem type attributes in the area. We used most of the functions in the Resample Tool of ArcGIS 10.2 (ESRI, Redlands, California, USA) to resample it to match the spatial resolution of 8 km (Xu et al., 2018). This study considered three vegetation types: cropland, forest, and grassland.

The NDVI tendency rate represents the vegetation trends and change rate in each climate zone. A positive value reflects an increase in NDVI and a negative value reflects the opposite. In this study, the variation of average NDVI in eight climate regions showed a small fluctuation, and all showed a decreasing trend (Figure 2). The overall average change trend of each climate zone showed that climate zone IV had the lowest absolute value, with a median value of −0.003. Both climate zones I and II were in Northeast China and had mainly forest vegetation. From 1982 to 2015, rapid expansion of arable land and urbanization led to a reduction in forest area, resulting in a significant deterioration in the region’s vegetation status. This was evidenced by a notable decrease in NDVI and median NDVI tendency rates. The median and tendency rates of NDVI was 0.059 and −0.116, respectively. Although the main land cover type of climate zone VII was desert, NDVI continuously decreased with a median change tendency rate of −0.012. Climatic zone VIII was located in the Qinghai-Tibet Plateau and was mainly composed of grassland vegetation in the cold plateau. There was little NDVI change and the median change tendency rate was only −0.007. Vegetation in other climatic regions (III, V, VI) also decreased to varying degrees, with median NDVI tendency rates of −0.034, −0.026 and −0.024, respectively (Figure 2B). There were significant regional differences in NDVI change trends in climate zones I, II, IV and V, and the regional differences in NDVI change trends in climate zone VII were the lowest among all climate zones.

The grassland vegetation in the study area was mainly distributed in climate zones VI and VIII and the Dzungarian Basin located in climate zone VII. Grassland vegetation coverage in other climate zones was relatively low (Figures 1A, B). Almost all grassland vegetation NDVI and PET values in the study area were positive correlation (Figure 3A). Correlations between NDVI and PET in the northern Inner Mongolia Plateau, the eastern Qinghai-Tibet Plateau and the Dzungarian Basin in climate zone VII were very high, and the maximum partial correlation coefficients ranged from 0.597-0.968. At the same time, the partial correlation coefficients between NDVI and PET of sparse grassland vegetation in climate zone I are greater than 0.734. The correlations between NDVI and PET of grassland vegetation in the southern Inner Mongolia Plateau in climate zone VI were lower than in the northern area. However, partial correlation coefficient in the western the Qinghai-Tibet Plateau were lower than eastern. Moreover, the range of correlation coefficients was 0.394–0.597. The Tarim Basin in climate zone VII was mostly desert. Correlations between NDVI and PET in the western and southern margins were low, with coefficients from 0.394–0.597. We found that the grassland and climate factors (P and PET) had positive relationship. Similarly, NDVI and p values in almost all climate zones were positively correlated (Figure 3B). Correlations between NDVI and P in the northern part of the Inner Mongolia Plateau and the eastern part of the Qinghai-Tibet Plateau were the highest, ranging between 0.670 and 0.897 in most regions. However, correlations in the western part of the Qinghai-Tibet Plateau were significantly lower than in the east, with negative correlations in some areas and coefficients ranging from–0.240–0.162. At the same time, correlations in the southern Inner Mongolia Plateau were lower than that in the northern Inner Mongolia Plateau and the eastern Qinghai-Tibet Plateau, where the maximum coefficient was 0.670. Correlations between NDVI and P in the Loess Plateau were significantly less than those for PET (Figures 2A, B). NDVI and P correlations for sparsely distributed grassland vegetation in climate zone I were less than those for PET, all less than 0.670. Correlations in the central part of the Dzungarian Basin were greater than those for PET, ranging from 0.349–0.523. Correlations in other regions of the Dzungarian Basin were less than 0.349. Correlations in the Tarim Basin were less than 0.349, and the NDVI and P of grassland vegetation in some areas were negatively correlated.

The temporal and spatial variability of NDVI for grassland vegetation in the study area to PET and P were relatively large (Figures 3C, D). The time scale in the Inner Mongolia Plateau and the eastern Qinghai-Tibet Plateau affected by PET was significantly longer than the time scale of the response to P. The response time scales were 5–6 months and 1–2 months, respectively. The response time scale of NDVI in the southwestern corner of the Qinghai-Tibet Plateau to PET was shorter than that of the eastern region, around 3–4 months; however, the time scale of the response of grassland vegetation NDVI to P in the southwestern corner of the Qinghai-Tibet Plateau was longer than in the eastern region, around 7–8 months. The time scale of the response of grassland vegetation NDVI to PET in the Loess Plateau was also longer than the time scale of its response to P. Response time scales were 3–4 months and 1–2 months, respectively. The time scales for NDVI response in the Dzungarian Basin to PET and P were almost the same, around 1–2 months. The response time scale of NDVI of grassland vegetation to PET in most areas of the Tarim Basin was longer than the time scale of its response to P. The time scale of the NDVI response to PET for sparse grassland vegetation in climate zone I was longer than the time scale of its response to P. The NDVI of grassland vegetation on the Yunnan-Guizhou Plateau in climate zone IV responded to PET with a longer time scale, all of which were longer than 7 months, and more than 11 months in parts of the western region. However, the time scale of the NDVI response in the Yunnan-Guizhou Plateau to P was significantly shorter than the time scale of the response to PET, and the response time scale of the eastern region was around 1–2 months.

The effects of PET and P on NDVI for grassland vegetation in different climate zones in the study area was very different (Figure 4). The median partial correlation coefficients of NDVI and PET in climate zones VI and VIII where grassland vegetation was concentrated were 0.734 and 0.727, respectively. However, correlations between NDVI and P in climate zones VI and VIII were less than those for PET, and median partial correlation coefficients were 0.690 and 0.671, respectively. The distribution of grassland vegetation in climate zones I, II and III was very sparse. However, the correlations between NDVI and climate factors in different climate zones were relatively high, and the median partial correlation coefficients with PET were 0.860, 0.845 and 0.793 respectively. However, the partial correlation coefficients between NDVI and P for grassland vegetation in zones I, II and III were smaller, and median values were 0.634, 0.598 and 0.600, respectively. The spatial variability of NDVI in climate zones IV and VII to PET and P was significantly greater than in other climate zones (Figures 3A, B). The median partial correlation coefficients of NDVI and PET for grassland vegetation in climate zones IV and VII were 0.635 and 0.601, respectively. The grassland NDVI had higher correlation with P than PET, with median values of 0.463 and 0.323, respectively. Grassland vegetation in climate zone V had the lowest correlations between NDVI and PET, with a median value of 0.462, whereas correlations between grassland vegetation NDVI and P in this climate zone were the highest, with a median value of 0.690. The time scale of the response of grassland vegetation NDVI to PET and P in each climate zone was also very characteristic (Figures 4C, D). The response time scale of NDVI for grassland vegetation in climate zone VI to PET was slightly shorter than in climate zone VIII, and the median response time scales were 4 months and 5 months, respectively. Climate zones VI and VIII had the same time scale in response to P, with a median value of 2 months, and the time scale and spatial variability of the two climate zones were very small. The median time scales of grassland vegetation NDVI response to PET in climate zones I, II, and III were 3, 3, and 4 months, respectively, which were longer than the time scales of these climate zones to P response, and the median values were 1, 1 and 2 months. The time scales of the response of grassland vegetation NDVI to PET and P in climate zones IV and VII were clearly different (Figures 3A, B). The median time scales of grassland vegetation NDVI response to PET in climate zones IV and VII were 5 and 4 months, respectively, which were longer than the time scales of response to P, and median values were both 2 months. The response time scales of NDVI of grassland vegetation in climate zone V to PET and P were the longest, with median values 10 and 6 months, respectively.

The forest vegetation in the study area was mainly distributed in the Greater Xing’an Mountains, the Changbai Mountains, the Yun-Gui Plateau, the Changjiang Downstream Plain, and climate zone V. The forest vegetation in other areas was less distributed (Figures 1A, B). NDVI for forest vegetation in the entire study area was positively correlated with PET and P (Figures 5A, B). NDVI and PET of forest vegetation in the northern Greater Xing’an Mountains and Changbai Mountains were significantly positively correlated, with coefficients of 0.768–0.961. However, the correlations between NDVI and P in the northern Greater Xing’an Mountains and Changbai Mountains were significantly lower than correlations with PET, with coefficients from 0.517–0.638 and 0.389–0.638, respectively. The coefficients between NDVI and PET in the Loess Plateau were 0.768–0.961, significantly greater than those between NDVI and P (less than 0.517). Correlations between NDVI and PET in the entire Yun-Gui Plateau were significantly less than those with P. The coefficients between NDVI and PET in most regions were only 0.097-0.362. The response of NDVI in the Sichuan Basin to PET was stronger than its response to P. Coefficients between NDVI and P of forest vegetation in the Yun-Gui Plateau and western Sichuan Basin were only 0.097–0.251, which were significantly less than those with PET (0.362–0.639). However, the partial correlation coefficients between NDVI and PET in most areas of the Changjiang Downstream Plain were greater than 0.639, and greater than those for NDVI and P in this area (lower than 0.638). The response of forest vegetation NDVI to PET in southeast climate zone IV was significantly stronger than its response to P. Coefficients for NDVI and PET in this area were 0.511–0.639, whereas those for NDVI and P in this area were less than 0.389. Correlations between NDVI and PET and P in the Brahmaputra River Valley of climate zone V were almost the same, ranging from 0.362–0.639 and 0.389–0.638, respectively.

The time scale of the response of forest vegetation NDVI to PET and P in the study area also had large spatial variability (Figures 5C, D). The response time scale of NDVI to PET in the northern Greater Xing’an Mountains and Changbai Mountains was relatively short, around 3–4 months; the time scale of NDVI response to P in this area was only 1–2 months. The response time scale of the Loess Plateau to P was 1–2 months, which was less than the time scale of NDVI response to PET in this area (3–4 months). The time scale of the response of NDVI to PET in the Yun-Gui Plateau varied. The time scales NDVI response to PET in the central, western and eastern regions of the Yun-Gui Plateau were 11–12, 9–10 and 7–8 months, respectively. At the same time, the time scales of NDVI response to P in the central, western and eastern regions of the Yun-Gui Plateau were 5–6, 7–8, and 3–4 months, respectively. The time scale of NDVI response in the Sichuan Basin to P was around 1–2 months, while the time scale of the response to PET in this area ranged from 1 to 2 to 5–6 months. The time scale of NDVI response to PET in the Changjiang Downstream Plain was only 1–2 months, significantly shorter than the time scale of NDVI response to P in this area (5–8 months). The time scale associated with NDVI and P in the southeast of climate zone IV was greater than 7 months, significantly longer than the time scale associated with PET (3–4 months). The response time scale of NDVI to PET in the Brahmaputra River Valley in climate zone V was 9–12 months, while the time scale of NDVI to P in this area was only 5–6 months.

Correlations between NDVI of forest vegetation and climate factors were analyzed in different climatic regions (Figures 6A, B). NDVI values in climatic areas with dense forest vegetation distribution were very sensitive to PET, and the median partial correlation coefficients were 0.863 and 0.839, respectively; The partial correlation coefficients were significantly higher than those observed between NDVI and PET in climatic areas with the same dense forest vegetation (median values of 0.591 and 0.463, respectively). However, partial correlation coefficients between NDVI and P in climate zones I and II were 0.612 and 0.554, significantly less than coefficients with PET. Correlations between NDVI and P in climate zones IV and V were even lower, with coefficients of only 0.368 and 0.468. PET had strong impacts on forest vegetation NDVI in climate zones III and VI where forest was less distributed. The median partial correlation coefficients for NDVI and PET in the two regions were 0.817 and 0.813, respectively. Compared with the impact of PET on forest vegetation in climate zones III and VI, the influence of P is slightly lower. Coefficients for NDVI and P in climate zones III and VI were 0.569 and 0.676, respectively. Although the forest vegetation in climate zones VII and VIII was very scarce, correlations between NDVI and PET were high and median values were 0.751 and 0.786, respectively. The median value of the partial correlation coefficient between NDVI and P of forest vegetation in climate zone VIII is 0.714, significantly higher than that of NDVI and P (median value 0.400). The time scales of forest vegetation NDVI response to PET and P in the study area were mostly short-term and medium-term (Figures 6C, D). The median time scale of NDVI response to PET in climate zones I and II was 3 months, which was longer than the time scales of response to P (median values were 1 and 2 months, respectively). The time scale of NDVI response to PET in climate zone IV was very different from the time scale of NDVI response to PET in climate zone V, with median values of 3 and 10 months, respectively. However, the time scales of NDVI response to P in climate zones IV and V were the same, with a median value of 6 months. The median time scales of NDVI response to PET in climate zones III and VI were both 3 months, which were slightly longer than the time scales of NDVI response to P (median value was 2 months). The time scales of NDVI response to P in climate zones VII and VIII were relatively short (the median value was 2 months), whereas the time scales for PET response were mid-term with median time scales of 4 and 5 months, respectively.

The cropland vegetation in the study area was mainly distributed in the Northeast Plain, North China Plain, Loess Plateau, Sichuan Basin, Changjiang Downstream Plain and Yun-Gui Plateau (Figures 1A, B). Of the cropland in the study area, only a small portion of the NDVI results were negatively correlated with PET and P (Figures 7A, B). The areas with negative correlations between NDVI and PET were mainly distributed in the Yun-Gui Plateau and parts of the southern North China Plain. Negative correlations between NDVI and P were less common and included the Sichuan Basin and a small part of the southern North China Plain area. The response of NDVI in the Northeast Plain to PET was similar to its response to P, with partial correlation coefficients of 0.662–0.766 and 0.628–0.885, respectively. There were clear spatial differences in correlations between NDVI and PET and P of cropland vegetation in North China Plain. The spatial distribution of correlations between NDVI and PET and P in the northern and southern parts was small. The partial correlation coefficients between NDVI and P in the Loess Plateau were mostly 0.454–0.628, whereas those between NDVI and PET were 0.549–0.766. The response of NDVI in the Sichuan Basin to P was slightly weaker than its response to PET. Correlations between NDVI and PET in the central area of the Changjiang Downstream Plain were higher than those with P, with coefficients from 0.766–0.965 and 0.454–0.628, respectively.

In general, the time scale of the response of cropland vegetation NDVI to P in the northern part of the study area was significantly shorter than that in the south, whereas the time scale and spatial distribution of the response of NDVI to PET were different (Figures 7C, D). The response time scale of NDVI to P in the Northeast Plain and Loess Plateau was very short (1–2 months), but the time scale of response of NDVI to PET in these two regions was longer than 3 months. The time scale of the NDVI response to PET in the northern part of the North China Plain was 5–6 months, and the time scale of the response of NDVI to PET in the southern part of the region was shorter (1–2 months). The time scale of the NDVI response to P in the northern part of the North China Plain was shorter than in the southern part. The response time scale of the Sichuan Basin to PET and P was 1–2 months, and only some of the NDVI response time scale to PET was around 5–6 months. The response time scale of NDVI in the Changjiang Downstream Plain to P was greater than 5 months, longer than the time scale of response of NDVI to PET in this area (less than 4 months). The response time scale of NDVI on the Yun-Gui Plateau to PET was longer than the response to P.

The correlations between NDVI of cropland vegetation and PET were higher than P in different climatic regions (Figures 8A, B). Although cropland vegetation in climate zones I and VIII was sparsely distributed, the NDVI values in these two climate zones had strong correlations with PET, with median values of 0.852 and 0.750, respectively. In contrast, the median partial correlation coefficients for NDVI and P in climate zones I and VIII were 0.679 and 0.722, respectively. The partial correlation coefficients of NDVI and P in climate zones II and VI were similar, whereas median partial correlation coefficients of the two climate zones and PET are 0.805 and 0.742 respectively. The same was true for correlations between NDVI and P in climate zones II and VI, with median values of 0.665 and 0.723, respectively. The median value of the partial correlation coefficient between NDVI and PET in the climate zone with the most densely distributed cropland vegetation was 0.689, which was significantly higher than the coefficient for NDVI and P (the median value is 0.469) in this area. The response of NDVI to PET in climate zone IV was stronger than its response to P. The median values of partial correlation coefficients for NDVI, PET and P in this area were 0.651 and 0.367, respectively. Of all the climate zones, correlations between NDVI, PET and P in climate zone V were the lowest, with median values of 0.467 and 0.308, respectively. P in climate zone VII had little effect on NDVI in this area, with a median correlation of only 0.330. However, the partial correlation coefficients between NDVI and PET were relatively large (median value is 0.779). The median time scale of the response of NDVI to PET and P in different climate zones was less than 6 months (Figures 8C, D). The median time scales of the response of NDVI to PET in climate zones I and VIII were 3 and 5 months, respectively, and the spatial heterogeneity was the lowest. Climate zone IV had the shortest response time scale to PET (median 2 months), whereas climate zone V had the longest (median 6 months). The median time scales of response to PET in climate zones II, III, and VI with densely distributed cropland vegetation were 4, 3, and 5 months, respectively. The median time scale of response to PET in climate zone VII was 3 months. Except for the two climatic regions (IV and V) with the largest spatial heterogeneity in NDVI to P response time scale, the median time scale of the response to P in all other climatic regions was 2 months. The median time scales for climate zones IV and V were 5 and 6 months, respectively.

The grassland vegetation in the study area was mainly distributed in the temperate grassland area (VI), the cold vegetation area of the Qinghai-Tibet Plateau (VIII) and the Dzungarian Basin located in the temperate desert area (VII) (Figures 1A, B). There were positive correlations between NDVI and PET for grassland vegetation in almost all climate zones (Figure 3A). Higher cover level of grassland in Climate zone VI had higher partial correlation coefficient with PET than P (Table 1). The reduction of aerodynamic terms was significant (Han et al., 2012) and strongly impacted the change in PET (Wang et al., 2019), which was reduced due to the decrease of wind speed (Zhao and Ma, 2021). The reduction of PET was conducive to moisture and the growth of grassland. Dai (2013) indicated that, in the context of global warming, the semi-arid or arid area of the Qinghai-Tibet Plateau would be more severe droughts than the semi-humid or humid area. This was also the reason why the correlations between NDVI and PET for grassland vegetation in the western part of climate zone VIII and its time scale were lower than those in the eastern part (Figure 3A). The time scale of the NDVI response of grassland vegetation to PET in the Dzungarian Basin in climate zone VII was short-term, whereas the same responses in climate zones VI and VIII were medium-term (Figure 3C). Snow melt in the Dzungarian Basin may have offset the impact of high potential evaporation on vegetation in areas that have experienced climate changes (Duan et al., 2016). Forest vegetation in the study area was mainly distributed in the temperate monsoon climate zone (climate zone I and II), subtropical monsoon climate zone (climate zone IV) and tropical monsoon climate zone (V) (Figures 1A, B). Precipitation in climate zones IV and V was much higher than PET (Table 1), so the NDVI response of forest vegetation in the two climate zones to PET was lower than the same response in climate zones I and II, which was similar to Xu et al. (2018)’s findings. The Atlantic Ocean multi-decadal oscillation influences the Northern Hemisphere’s drought by regulating the ocean-atmosphere-land surface interaction processes and other large-scale circulations (Zhu et al., 2021). In climate zone I over the study time interval, the northern part of the Greater Xing’an Mountains was in the cold period of the multi-decadal oscillation, so drought in this area was effectively alleviated (Zhu et al., 2021). The growth of cropland vegetation in the northern part of climate zones VI, II and III mostly depended on irrigation (Yin et al., 2020), while large-scale irrigation and human activities have significantly changed PET (Han et al., 2009). The response time scale of NDVI for cropland vegetation to PET was longer than in other parts of the study area (Figure 7C).

P is a key factor leading to changes in NDVI, and the response of NDVI to P varies with vegetation type (Wang et al., 2001). There was a positive correlation between NDVI and P of grassland vegetation in almost all climatic regions (Figure 3B), which was consistent with the response relationship obtained by Nanzad et al. (2019). The Dzungarian Basin in climate zone VII was surrounded by mountains (Figure 1B), and this area had an arid and rainless temperate continental climate (Table 1). The grassland vegetation in this region was mostly fed by ice and snow melt water. Therefore, P had a relatively low impact on the NDVI of grassland vegetation in this area (Duan et al., 2016). Climate zone VI was at the junction of temperate continental and temperate monsoon climates (Wang et al., 2019), and the NDVI of grassland vegetation near the temperate monsoon climate with more precipitation showed a strong response to P. Climate zone VIII had an alpine plateau climate, which was located on the side of the moist airflow with less rainfall, resulting in a greater correlation between the NDVI and P for grassland vegetation in the region (Yang et al., 2017). The time scale of grassland vegetation NDVI response to P in densely distributed grassland areas in the study area was 1–2 months (Figure 3D). Because the densely distributed grassland areas were all in semi-arid areas, the water availability in semi-arid areas was strong, and precipitation was absorbed by the soil and used by plants quickly. Water availability is a decisive factor affecting the function of semi-arid ecosystems (He et al., 2021). According to Table 1, climate zones I and II were semi-humid regions, whereas climate zones IV and V were humid regions with abundant precipitation. The NDVI response of forest vegetation in climate zones I and II to P was stronger than the same response in climate zones IV and V (Figure 6B), which was consistent with the NDVI and P in temperate regions calculated by Schultz and Halpert (1993). This result was consistent with the greatest correlation. Studies have shown that the embolism resistance of forest vegetation during continuous dry periods determines its viability (McDowell et al., 2008), whereas the embolism resistance in areas with less rainfall is low, and precipitation is used rapidly. For this reason, the median time scale of forest vegetation NDVI response to P in climate zones I and II was only 2 months, much shorter than the same result in climate zones IV and V (6 months). We found that the median time scale of NDVI response of cropland vegetation to P in the north (climate zones II, III and VI) was only 2 months, whereas the same result in climate zone IV in the south was 5 months (Figure 8D). This showed that the response of cropland vegetation NDVI to P may be affected by human activities. For example, humans have used crop improvements and water-saving irrigation technologies to reduce the dependence of northern cropland vegetation on P (Deng et al., 2006). Due to the large distribution of water at the junction of the Sichuan Basin, the North China Plain and the Changjiang Downstream Plain (Figure 1B), the correlations between NDVI and P for cropland vegetation in these areas were generally less than 0.283 (Figure 8B).

With climate change, the increase in surface roughness in semi-arid areas has led to a decrease in wind speed, which in turn has caused a decrease in PET and promoted vegetation growth (Vautard et al., 2010). In the study area, the dense grassland vegetation in the semi-arid area (Figures 1A, B) had annual precipitation of 200 mm–400 mm (Table 1). This may be one of the reasons why NDVI for grassland vegetation was more dependent on PET than the other vegetation types (Figure 9A). The influence of P on the three types of vegetation studied in the study area was dominant (Figure 9B), which was consistent with the relationship between vegetation and precipitation described by Wang et al. (2003). The impacts of irrigation on cropland increases cropland PET (Qiu et al., 2008), which also produced the short-term time scale of NDVI and PET related to cropland vegetation that accounted for the largest proportion of the three types of vegetation in this study. The proportion of short-term time scales related to forest vegetation NDVI and P was the lowest among the three types of vegetation in this study (Figures 9C, D). Studies have shown that forest vegetation benefits from longer root systems that can absorb moisture from deep soils (Anderegg et al., 2015).