Snow Cover Phenology Change and Response to Climate in China during 2000–2020

: Snow cover phenology (SCP) is critical to the climate system. China has the most compre-hensive snow cover distribution in the middle and low latitudes and has shown dramatic changes over the past few decades. However, the spatiotemporal characteristics of SCP parameters and their sensitivity to meteorological factors (temperature and precipitation) under different conditions (altitude, snow cover classiﬁcation, or season) in China are insufﬁciently studied. Therefore, using improved daily MODIS cloud-gap-ﬁlled (CGF) snow-cover-extent (SCE) products, the spatiotemporal characteristics (distribution and variation) and respond to climate of snow cover area (SCA), snow cover start (SCS), snow cover melt (SCM), and snow cover days (SCD) are explored from 2000 to 2020. The results show that in the past 20 years, snow cover in China has demonstrated a trend of decreasing SCA, decreasing SCD, advancing SCS, and advancing SCM, with SCM advancing faster than SCS. The greatest snowfall occurs in January, mainly in northeastern China, northern Xinjiang, and the Tibet Plateau. Spatially, the slope of SCP was mainly within ± 0.5 day/year (d/y) Statistics indicated that the area proportion where SCD is signiﬁcantly reduced is greater than increased; SCD, SCS, and SCM are shortened or advanced in three snow-covered area classiﬁcations. Moreover, compared with precipitation, the signiﬁcantly correlated regions (6–47.2% more than precipitation) and correlation degree (1.23–8.33 times precipitation in signiﬁcantly correlated snow cover classiﬁcation) between temperature and SCP in different seasons are larger. For stable snow-covered areas (SSA), SCD are mainly affected by spring temperature below 1500 m and mainly by autumn temperature above 1500 m; the precipitation is more affected in autumn. The correlation of SCP with temperature and precipitation has obvious spatial and seasonal differences and shows characteristic variation with altitude. These results can provide important data support for climate prediction, hydrological research, and disaster warning.


Introduction
As one of the essential components of the cryosphere [1][2][3], due to its high albedo and low thermal conductivity, snow cover has a substantial impact on the climate system via regulating the surface energy budget, the hydrological cycle, and atmospheric circulation [4,5]. The sixth report of the Intergovernmental Panel on Climate Change (IPCC_AR6) stated that the average temperature in the past 10 years is 1.09 • C more than in the late 19th century and further warming has led to seasonal snow cover reduction and earlier snowmelt date.
(RMSE) is within 20 days and the mean absolute error (MAE) is within 8 days for all three snow cover phenology datasets. These improved MODIS CGF snow cover products are beneficial for us to obtain more accurate snow cover information in China.
The objective of this study is to explore the latest terrestrial snow cover phenology changes and respond to the climate in China during 2000-2020. Firstly, the high-precision NIEER CGF MODIS SCE product and the NIEER MODIS SCP product were used to conduct a spatial and temporal analysis of SCP in China during 2000-2020. Secondly, we divided snow cover types according to SCP parameters to explore the characteristics of SCP under different snow types. Finally, using meteorological data (temperature and precipitation) and topographic data, this study discusses the relationship between SCP and climate at different altitudes, snow-covered area types, and seasons. In-depth research and accurate snow cover analysis are expected to provide basic data for climate prediction; agricultural water resources utilization; and information services for warnings about disasters, such as floods and snow disasters.

Data
Two satellite-observed snow cover products based on MODIS and a reanalysis dataset were employed to explore the distribution and attribution of SCP in China.

Improved MODIS CGF Snow Cover Products
Improved MODIS CGF snow cover products are used as the basic dataset for SCP analysis, mainly including the NIEER CGF MODIS SCE dataset and the NIEER MODIS SCP dataset. The NIEER CGF MODIS SCE dataset provides basic data of SCA, and the NIEER MODIS SCP dataset provides basic data of SCD, SCS, and SCM. They are all from the National Cryosphere Desert Data Center. (http://www.ncdc.ac.cn, accessed on 26 November 2020 and 16 November 2021 respectively).

The NIEER CGF MODIS SCE Product
We identified snow cover as the true value using the NIEER CGF MODIS SCE product. The product improved the snow cover extraction algorithm of the aggregated MODIS product (aggregate the Terra-MODIS SCE and Aqua-MODIS SCE under clear skies to exclude some cloud gaps preliminarily) using surface reflectance data standards in forest and non-forest areas in China, respectively, based on the high-spatial-resolution cloud-free Landsat-5 TM/Landsat8 OLI image. Then, clouds were removed in two steps using Hidden Markov spatiotemporal modeling and snow depth data interpolation (derived from passive microwave remote sensing). Finally, the cloud-free product combined temperature data and water data to obtain a daily cloud-free snow cover product. The product has a spatial resolution of 500 m, a temporal resolution of 1 day, and a time range from 27 February 2000 to 31 December 2020. Product details for this dataset are shown in Table 1, including data categories (snow cover, land, water, and vacancy values), values, and remarks [34]. NIEER MODIS SCP products are calculated by hydrological year according to NIEER CGF MODIS SCE products. The dataset includes three SCP sub-datasets, which are the SCD, SCS, and SCM. Its time range is from 2000 to 2020, with a temporal resolution of 1 year and a spatial resolution of 500 m. The product details are shown in Table 2, including SCP parameters, value ranges, no data, water, and remarks.

Reanalysis Temperature and Precipitation Dataset
The ERA5-Land Monthly Averaged-ECMWF Climate Reanalysis dataset [38] provides a meteorological dataset to analyze snow cover distribution and change drivers. This dataset has been produced by replaying the land component of the ECMWF ERA5 climate reanalysis. Reanalysis combines model data with observations from across the world into a globally complete and consistent dataset using the laws of physics. This dataset includes 50 variables. Skin temperature and total precipitation are used as meteorological factors in this study. The spatial resolution is 0.1 • , and the time range is from January 1981 to March 2022. The product has been resampled into 500 m to match MODIS products using nearest neighbor resampling.

DEM Dataset
The digital elevation model dataset (DEM) is the fourth version dataset of the Shuttle Radar Topography Mission (SRTM) DEM products and is directly accessible on the Google Earth Engine (GEE) platform. The dataset has a resolution of 90 m and covers more than 80% of the world. This product mainly provides elevation data, which have been aggregated to 500 m to match the MODIS product [39].

Definition of Snow Cover Phenological Parameters
In the study, the hydrological year is defined as the 1st of September of one year to the 31st of August of the following year.
SCA is defined as the total area covered by snow in an area, as in Equation (1): where s is the area of a single pixel, which in this paper is 0.25 km 2 , and N is the sum of the number of pixels with snow in a single day. SCD is defined as the sum of snow days in a hydrological year, as in Equation (2): where t is the daily single-pixel information, which is 1 (snow) or 0 (no snow), and n is the total number of days in a hydrological year. SCS is defined as the first 5 consecutive days when a pixel is classified as snow if the pixel satisfies Equation (3): where i is day of year (DOY) i. SCM is defined as the last 5 consecutive days when snow occurs during each hydrological year if the pixel satisfies Equation (5): then, where m is DOY m. This is the easiest way to calculate SCS and SCM and would avoid the impact of ephemeral snow. The SCD, SCS, and SCM datasets are directly obtained from the NIEER MODIS SCP product [40].

Statistical Analyses
We calculated the average value and trend analysis of each pixel's snow cover phenological parameters in the past 20 years to explore the spatial distribution of snow cover in China.
The Theil-Sen Median method, also known as Sen's slope estimator, is a robust nonparametric statistical trend calculation method. This method is widely used in trend analysis of long time-series data to calculate the size of the trend, due to the advantages of high computational efficiency and insensitivity to outliers.
The Mann-Kendall (M-K) test, as a non-parametric trend test method, is widely used in the trend significance tests of long time series data. The M-K test does not require the measured values to obey the normal distribution and is not affected by missing values and outliers. In the M-K test, the positive Z value showed an upward trend, while the negative Z value showed a downward trend. We can also use the Z value to calculate the p value to describe the significance level accurately. The commonly used absolute value of z was equal to or greater than 1.28, 1.64, and 2.32, indicating that the values passed the 90%, 95%, and 99% significance tests, respectively. Generally, an evident trend is considered when the significance level is greater than 95% [41,42].

Snow-Covered Area Classification
Combining the average value and interannual variability (IV) of SCD [43], the regional scope and the spatiotemporal distribution of and variation in different snow-covered areas' SCP are explored to better understand the characteristics of snow cover in China. The types of snow cover areas in China are classified. The division rules are as follows: (1) stable snow-covered areas (SSA): SCD > 60 and IV < 0.4; (2) annual periodic unstable snow-covered areas (APA): 10 < SCD ≤ 60 and IV < 1.0 or SCD > 10 and 0.4 ≤ IV < 1.0; (3) non-annual periodic unstable snow-covered areas (NPA): 0 < SCD ≤ 10 and IV <3.0 or SCD > 0 and 1.0 ≤ IV < 3.0; and (4) snow-free areas (SFA): SCD = 0, where IV = δ/X, δ is the standard deviation of SCD, X is the annual mean of SCD.

Correlation between SCP and Climatic Factors
The Pearson evaluation method is a statistical method to measure the closeness of the relationship between two variables accurately. In the Pearson evaluation method, the Remote Sens. 2022, 14, 3936 6 of 22 correlation coefficient r indicates the degree of correlation, ranging from −1 to 1. The formula for this index is as follows: where xi is the SCP parameter of the DOY i; yi is the annual precipitation or average temperature of the DOY i year; x is the average value of the SCP for many years; y is the average value of the precipitation or temperature; and m is the number of years in the monitoring period. Positive values (r > 0) indicate facilitation, while negative values (r < 0) represent inhibition. The statistical confidence level of correlation was evaluated by performing the significance test. Among them, P < 0.1, P < 0.05, and P < 0.01 indicate that they passed the significance test of 90%, 95%, and 99% confidence, respectively [44].
In this study, the Pearson correlation coefficient was used to analyze snow cover driving factors (temperature and precipitation) and perform the significance test. To further explore the influence of altitude on snow cover changes, we analyzed the correlation between SCP and meteorological factors (temperature and precipitation) in the past 20 years in the whole of China, especially SSA, at different altitudes at 500 m intervals.

Temporal Variation Characteristics of SCP
The average SCA, temperature, and precipitation diurnal variation in China from 1 September 2000 to 31 August 2020 is shown in Figure 1a, which exhibits an annual periodic variation snow cover characteristic of "accumulation-ablation-accumulation" from 1 September 2000 to August 2020, but with temperature and precipitation trends opposite to snow cover. The SCA, SCD, temperature, and precipitation monthly variations are shown in Figure 1b. Snow began to increase in autumn (September-November) and entered the accumulation period and increased the fastest in October-November. In winter (December to February), the change was small and entered a stable period. The largest SCA (305.20 × 10 4 km 2 ) and the largest number of SCD (9.82 d) were reached in January. Autumn (March-May) entered the ablation period, and snow cover gradually decreased, with the fastest decrease in March-April. Compared with other seasons, there was only a small amount of snow in summer (June-August), and it was mainly distributed in some high-altitude areas. The temperature and precipitation trend are similar. They started to decrease from autumn. The temperature dropped the most from October to November, and it dropped below 0 • C in November. In winter, the precipitation maintained a relatively low state, with little change. The temperature reached 0 in January. In winter, temperature and precipitation were relatively low and less variable, with precipitation reaching its lowest in December and temperature in January. In spring, both temperature and precipitation began to rise, with the highest temperature rise from March to April, and it was higher than 0 • C in March. Summer temperatures and precipitation were high and peaked in July. Compared with precipitation, temperature and snow cover have completely opposite trends and have a greater impact. The changes in temperature and precipitation also better explain the reasons for the rapid increase and rapid melting of snow cover in China, which is also the reason why the snow cover period in China is mainly from November to March. However, the specific response of meteorological factors to snow cover needs more detailed analysis. However, these changes were not significant (P > 0.1). Although the definitions and slope of SCA and SCD are different, the changing trend is exactly the same and the advance rate of SCM is larger than that of SCS.
SCS, and SCM ( Figure 1c) during the hydrological years 2000/2001 to 2019/2020, the snow phenological parameters in China showed a fluctuating state of change, showing a decreasing trend in SCA (slop = −1.00 × 10 4 km 2 /y) and SCD (slop = −0.12 day/year (d/y)) and an advancing trend in SCS (slop = −0.22 d/y) and SCM (slop = −0.73 d/y). However, these changes were not significant (P > 0.1). Although the definitions and slope of SCA and SCD are different, the changing trend is exactly the same and the advance rate of SCM is larger than that of SCS.

Spatial Variation Characteristics of SCP
The average value and trend analysis of SCP were performed at each pixel to explore the spatial distribution pattern and changes in SCP. SCP spatiotemporal distribution and variations in CHN from 2000/2001 to 2019/2020 are shown in Figure 2. We observed high spatial heterogeneity of SCP in China. Figure 3 shows the percentage of SCD, SCS, and SCM slope in China.

Spatial Variation Characteristics of SCP
The average value and trend analysis of SCP were performed at each pixel to explore the spatial distribution pattern and changes in SCP. SCP spatiotemporal distribution and variations in CHN from 2000/2001 to 2019/2020 are shown in Figure 2. We observed high spatial heterogeneity of SCP in China. Figure 3 shows the percentage of SCD, SCS, and SCM slope in China.  35.51% of the areas, the changes in the SCD were not analyzed, and for 54.48% of the areas, the changes in the SCS and SCM were not discussed. In the past two decades, the slope of SCP parameters in China was mainly within ±0.5 d/y. In the statistics of the percentage of SCD, SCS, and SCM changes in the total area of the country, the decreased area (slope < 0) was 17.88%, 6.22%, and 11.02% more than the increased area (slope > 0), respectively. The SCD decreases were significant (P < 0.05) in 6.97% of the total area, mainly the Sanjiang Plain, and other scattered areas. Only in the northern and central parts of Inner Mongolia was SCS significantly delayed, accounting for 2.32% of the country's total area. China has a significant trend of early SCM in 5.15% of the regions, and only 1.39% of the regions have a significant trend of late SCM. These significant early trends were primarily distributed in the Greater Khingan Range, Altay, the northern part of the North China Plain, and the southern part of the Northeast Plain, and the significant late trends were distributed in the eastern part of the Himalayas and eastern Tarim Basin.

Snow-Covered Area Classification in China
According to the classification method of snow cover types in Section 3.   35.51% of the areas, the changes in the SCD were not analyzed, and for 54.48% of the areas, the changes in the SCS and SCM were not discussed. In the past two decades, the slope of SCP parameters in China was mainly within ±0.5 d/y. In the statistics of the percentage of SCD, SCS, and SCM changes in the total area of the country, the decreased area (slope < 0) was 17.88%, 6.22%, and 11.02% more than the increased area (slope > 0), respectively. The SCD decreases were significant (P < 0.05) in 6.97% of the total area, mainly located in the northern part of the Greater Khingan Range, the southern part of the Tianshan Mountains, the southern part of the Northeast Plain, most of Shanxi Province and other scattered areas; the area with a significant increase in SCD accounted for only 1.59%, mainly in Qinghai Province and the eastern part of the Himalayas. The regions with significant advancing SCS accounted for about 5.61% of China and were scattered at the junction of the Greater and Lesser Khingan Range, the Changbai Mountains, parts of the Sanjiang Plain, and other scattered areas. Only in the northern and central parts of Inner Mongolia was SCS significantly delayed, accounting for 2.32% of the country's total area. China has a significant trend of early SCM in 5.15% of the regions, and only 1.39% of the regions have a significant trend of late SCM. These significant early trends were primarily distributed in the Greater Khingan Range, Altay, the northern part of the North China Plain, and the southern part of the Northeast Plain, and the significant late trends were distributed in the eastern part of the Himalayas and eastern Tarim Basin.

Temporal and Spatial Distribution and Variation Characteristics of Different Snow-Covered Types' SCP in China 4.2.1. Snow-Covered Area Classification in China
According to the classification method of snow cover types in Section 3.2.2, snowcovered area types in China are classified by the interannual variability and average snow cover days (Figure 2a), as shown in

SCP in Different Snow-Covered Area Types
The changes in SCD, SCS, and SCM for snow cover types in China from 2000/2001 to 2019/2020 are shown in Table 3. All types of snow-covered areas in China have a trend of decreasing SCD and advancing SCS and SCM, and the changes are not significant, but there are slight differences between each type. In the stable snow-covered areas (SSA), the average number of SCD are 137.29 d, the average SCS is DOY 67.28 (7th November), and the average SCM is DOY 226.03 (13th April), decreasing or advancing at a rate of −0.16 d/y, −0.09 d/y, and −0.22 d/y, respectively. In the annual periodic unstable snow-covered areas (APA), the average number of SCD are 39.67 d/y, the slope is −0.17 d/y, the SCS is 85.99th (25th November), the change rate is −0.11 d/y, the average SCM is 148.91th (27th January), and the slope is −0.85 d/y. In the non-annual periodic unstable snow-covered areas (NPA), the average SCD, SCS, and SCM are 39.67 d/y, 85.99th (25th November), and 148.91th (27th January), respectively. Their slopes are −0.17 d/y, −0.11 d/y, and −0.85 d/y respectively.
Excluding the three types of snow cover outside the snow-free areas, the average value of snow cover phenological parameters shows that the SCD gradually decrease with SSA, APA, and NPA, but the snow cover of the SSA remains the longest; NPA have the earliest SCS and SCM, APA have the latest SCS, but SSA have the latest SCM. The slope of SCP shows that the change in SCD is similar among the three types of snow cover. SSA have the slightest slope of SCS and SCM, so the snow cover is relatively stable. The snow cover of APA has the most prominent change rate in SCM, and the snow cover of NPA

SCP in Different Snow-Covered Area Types
The changes in SCD, SCS, and SCM for snow cover types in China from 2000/2001 to 2019/2020 are shown in Table 3. All types of snow-covered areas in China have a trend of decreasing SCD and advancing SCS and SCM, and the changes are not significant, but there are slight differences between each type. In the stable snow-covered areas (SSA), the average number of SCD are 137.29 d, the average SCS is DOY 67.28 (7th November), and the average SCM is DOY 226.03 (13th April), decreasing or advancing at a rate of −0.16 d/y, −0.09 d/y, and −0.22 d/y, respectively. In the annual periodic unstable snow-covered areas (APA), the average number of SCD are 39.67 d/y, the slope is −0.17 d/y, the SCS is 85.99th (25th November), the change rate is −0.11 d/y, the average SCM is 148.91th (27th January), and the slope is −0.85 d/y. In the non-annual periodic unstable snow-covered areas (NPA), the average SCD, SCS, and SCM are 39.67 d/y, 85.99th (25th November), and 148.91th (27th January), respectively. Their slopes are −0.17 d/y, −0.11 d/y, and −0.85 d/y respectively. Excluding the three types of snow cover outside the snow-free areas, the average value of snow cover phenological parameters shows that the SCD gradually decrease with SSA, APA, and NPA, but the snow cover of the SSA remains the longest; NPA have the earliest SCS and SCM, APA have the latest SCS, but SSA have the latest SCM. The slope of SCP shows that the change in SCD is similar among the three types of snow cover. SSA have the slightest slope of SCS and SCM, so the snow cover is relatively stable. The snow cover of APA has the most prominent change rate in SCM, and the snow cover of NPA has the most significant slope of SCS.

Response of SCP to Meteorological Factors in China
The correlation spatial distribution between SCD and autumn, winter, and spring meteorological factors (temperature and precipitation) is shown in Figure 5. The spatial distribution of the correlation between SCS and autumn meteorological factors (temperature and precipitation) and that between SCM and spring meteorological factors (temperature and precipitation) are shown in Figure 6. The correlation statistical area percentages of Figures 5 and 6 are shown in Table 4.

Response of SCP to Meteorological Factors in China
The correlation spatial distribution between SCD and autumn, winter, and spring meteorological factors (temperature and precipitation) is shown in Figure 5. The spatial distribution of the correlation between SCS and autumn meteorological factors (temperature and precipitation) and that between SCM and spring meteorological factors (temperature and precipitation) are shown in Figure 6. The correlation statistical area percentages of Figures 5 and 6 are shown in Table 4.    1.

Response of SCD to meteorological factors in China
Since the SCD are calculated for the entire hydrological year, but the actual snow cover is less affected by the summer climate, when investigating the responses of SCD to climate change, we only considered the relationship between SCD and meteorological factors (temperature and precipitation) in autumn (September-November), winter (December-February), and spring (March-May) at the pixel level using the Pearson correlation. The results of the correlation calculation and significance test are shown in Figure 5. The SCD are 0 in 6.77% of China's regions, so the relationship between SCD and meteorological factors is not discussed in these regions.
In terms of the correlation of spatial distribution between SCD and temperature (Figure 5a,c,e) and the correlation statistical area percentage (Table 4), the response of SCD to temperature has obvious spatial differences and seasonality. The regions where SCD are negatively correlated with autumn temperature and spring temperature respectively account for 78.66% and 71.01% of the country's total area, and the regions with a significant negative correlation account for 18.26% and 22.29% of the country's total area, respectively. In spring and autumn, the significant negative correlation areas between temperature and SCD are similar, and they are mainly divided into Altay, the high-altitude mountains of the Qinghai-Tibet Plateau, the northern part of the North China Plain, the southern part of the Northeast Plain, and the northeastern part of the Inner Mongolia Plateau. In winter, the area with a negative correlation between SCD and temperature accounts for 89.07% of the country's total area, basically covering the snow-covered regions of the country, and the area with significant negative correlation accounts for 51.22% of the country's total area. We also observed a significant positive effect of temperature on the SCD, but these areas were less than 0.3% either in fall, winter, or spring, especially in winter (less than 0.1%).
The correlation also revealed that the precipitation might also control SCD in several areas (Figure 5b,d,f). Autumn precipitation has a significant positive effect on SCD in more regions (8.47%), mainly distributed in the Altai Mountains, the Tianshan Mountains, the southwest of the Northeast Plain, and some high-altitude mountains of the Qinghai-Tibet Plateau. Only 0.28% of the region's autumn precipitation has a significant negative effect. In winter, 2.46% of the regional SCD are significantly positively correlated with precipitation, located in parts of the Greater Khingan Range, Shanxi Province, and northern Tarim Basin, and 3.93% of the regional SCD are significantly negatively correlated with precipitation, located in parts of the Qinghai-Tibet Plateau, northern Turpan Basin, and the Junggar Basin. There is a significant positive correlation between SCD and spring precipitation mainly in the Greater Khingan Range, Shanxi Province (1.07%), the A-erh-chin Mountains, the Altai Mountains, and western Himalayas, and a significant negative correlation in central China and central and southern Qinghai-Tibet Plateau (3.00%).

Response of SCS and SCM to meteorological factors in China
Since SCS is mainly in autumn and SCM is mainly in spring in China, the relationship between SCS and autumn temperature and precipitation and between SCM and spring temperature and precipitation was calculated to explore the response of SCS and SCM to meteorological factors. As shown in Figure 6a,b, the negative correlation area between SCS and autumn temperature is 25.61% more than the positive correlation area, but the negative correlation area with autumn precipitation is 22.72% less than the positive correlation area. In particular, a significant positive correlation between the SCS and autumn temperature was found in 13.09% of China, mainly located in three snow-covered regions, and a significant negative correlation between the SCS and autumn temperature was found in 1.27% of China, mainly located in the east of the North China Plain and the middle and lower reaches of the Yangtze River Plain. About 1.25% of the regional SCS is significantly positively correlated with autumn precipitation, mainly in the three snow-covered regions, and 7.11% of the regions with a significant negative correlation between SCS and autumn precipitation are mainly distributed at the junction of Inner Mongolia and Gansu Province, the northwest of the Yunnan-Guizhou Plateau, and some sporadic areas. SCM was also controlled by spring temperature (Figure 6c) and precipitation (Figure 6d). The negative correlation area between SCM and spring temperature is 43.94% more than the positive correlation area, but the positive correlation area with spring precipitation is 5.94% more than the negative correlation area. The SCM was significantly negatively correlated with spring temperature in about 19.63% of China, primarily in most of the three snow-covered regions. However, a significantly positive correlation was only in 0.40% of the area. The areas with a significantly positive correlation between SCM and precipitation are mainly scattered in three snow-covered regions, accounting for 2.90% of China. In comparison, a negative relationship between SCM and temperature was observed in a similar area (2.05%) in the Sichuan Basin.

Response of SCP to Meteorological Factors in Different Snow-Covered Area Alassification
To further examine the spatial pattern of impacts of climate factors on SCD, SCS, and SCM, we also analyzed their relationships in different snow types (Table 5). For the whole of China, SSA, APA, and NPA, in addition to the positive effects of autumn and spring precipitation on SCD, temperature, and precipitation, have negative effects in spring, autumn, and winter. The effect of temperature on CHN and SSA is greater in spring, with a high correlation coefficient (−0.48 and −0.66, respectively), but in APA and NPA, the greater effect is in winter (correlation coefficients are -0.48 and −0.55, respectively). Except for SSA (the correlation coefficient is −0.37), the precipitation has the greatest correlation in winter, and the correlation coefficients are −0.5 (CHN), −0.38 (APA), and −0.37 (NPA). The correlation between SCS and autumn temperature was greater in SSA (r = 0.69) and APA (0.51), which were positive correlations, but showed a negative correlation in NPA. However, SCS and autumn precipitation showed a positive correlation in APA and a negative correlation in other regions. The correlation degree was similar, and the absolute value of the correlation coefficient was above 0.38. SCM and spring temperature are both negatively correlated, and SSA have the largest correlation (r = −0.75), while precipitation is less correlated, and the absolute value of the correlation coefficient is less than 0.2. SSA show a positive correlation, and other types show a negative correlation. Table 5. Correlation between snow cover days (SCD) and climate factors (average temperature and sum precipitation) in autumn, winter, and spring; correlation between snow cover start (SCS) and climate factors in autumn; and correlation between snow cover melt (SCM) and climate factors in spring in different snow-covered area classification.

Response of SCP to Meteorological Factors with the Elevation Variations
Areas with different elevations as a percentage (%) of the total area in China (CHN) or in the stable snow-covered areas (SSA) are shown in Table 6. The correlation between SCD and climate factors (average temperature and sum precipitation in autumn, winter, and spring), the correlation between SCS and climate factors in autumn, and the correlation between SCM and climate factors in spring at different altitudes in China (CHN) and in SSA are shown in Figure 7.
The DEM dataset indicating the Chinese area with an altitude below 0 m only accounts for 0.40% of the country (Table 6), mainly in the Junggar Basin and some southern coastal areas, where there is relatively little snow. The area of DEM above 6000 m only accounts for 0.11% of the total Chinese area and is mainly distributed in the Himalayas, the Kunlun Mountains, the Nyainqentanglha Mountains, etc., of the Qinghai-Tibet Plateau. These areas have little precipitation, low temperatures, a large number of SCD, early SCS, and late SCM. The overall average number of SCD in these areas can be as high as 310 days or more, and the changes are small, all of which are stable snow-covered areas, and some areas are even permanent snow-covered areas. Therefore, this study discusses the response of SCP to meteorological factors with the elevation variations mainly for DEM in the range of 0-6000 m.  Figure 7. Correlation between snow cover days (SCD) and climate factors (average temperature and sum precipitation in autumn, winter, and spring), correlation between snow cover start (SCS) and climate factors in autumn, and correlation between snow cover melt (SCM) and climate factors in spring at different altitudes in China (CHN), especially in the stable snow-covered areas (SSA).
The DEM dataset indicating the Chinese area with an altitude below 0 m only accounts for 0.40% of the country (Table 6), mainly in the Junggar Basin and some southern coastal areas, where there is relatively little snow. The area of DEM above 6000 m only accounts for 0.11% of the total Chinese area and is mainly distributed in the Himalayas, the Kunlun Mountains, the Nyainqentanglha Mountains, etc., of the Qinghai-Tibet Plateau. These areas have little precipitation, low temperatures, a large number of SCD, early Between 0 and 6000 m in elevation, the SCD and temperature in China are negatively correlated, mainly affected by winter temperature, followed by autumn temperature, and spring temperature has the smallest effect. Compared with temperature, SCD are relatively less affected by precipitation. The correlation between SCD and autumn precipitation is the largest, followed by winter precipitation, and the correlation is also the smallest in spring. At different altitudes, SCD are positively correlated with autumn precipitation, negatively correlated with winter temperature, and may have an opposite correlation (positive or negative correlation) with spring temperature at different altitudes. SCS was positively correlated with autumn temperature and negatively correlated with autumn precipitation and SCM was negatively correlated with spring temperature and had an opposite correlation with spring precipitation at different altitudes. Considering the large proportion of places with less snow cover in China, which may interfere with the results, we further studied the response of the SCP in the stable snow cover areas to the change in meteorological factors with altitude.
In SSA, SCD are basically negatively correlated with temperature and positively correlated with precipitation at different altitudes, but above 3000 m, SCD are negatively correlated with winter precipitation. SCD have a greater correlation with spring temperature below 1500 m in elevation and a greater correlation with autumn temperature above 1500 m. In terms of precipitation, SCD are more correlated with autumn precipitation than spring and autumn precipitation. The correlation between SCD and spring temperature gradually decreased with an increase in altitude, but it decreased slowly and tended to remain unchanged at 3000-6000 m. The correlation between SCD and the temperature in autumn increased gradually from 0 to 2500 m, then decreased gradually at 2500 m, and tended to remain stable until 4000 m. The correlation between SCD and winter temperature gradually decreased within 2500 m, then tended to remain unchanged, then increased slightly when greater than 3500 m, and then tended to remain unchanged. The correlation between SCD and spring precipitation is large between 2000 and 3000 m in elevation, and the correlation with autumn precipitation changes with altitude is similar to the correlation change trend between SCD and autumn temperature, but precipitation is positively correlated and temperature is negatively correlated. SCD are positively correlated with winter precipitation and negatively correlated above 4000 m in elevation. This is mainly because the winter temperature in high-altitude areas is too low and the precipitation temperature is higher than the surface, so the more the precipitation, the more the accumulation, and the fewer the snow days, showing a negative correlation. The SCD and winter precipitation are positively correlated between 1000 and 2000 m and negatively correlated after more than 4000 m. This is mainly because the winter temperature in high-altitude areas is too low and the precipitation temperature is higher than the surface, so the more the precipitation, the fewer the SCD, presenting a negative correlation.
There is a positive correlation between SCS and the temperature in autumn. The lower the temperature in autumn, the easier it is for water vapor to condense into ice crystals and reach the ground before it completely evaporates or melts, forming snowfall, and the earlier the SCS. SCS has a negative correlation with precipitation. The more correct the precipitation, the more water vapor in the air, so when encountering low temperature, it is easy for the water vapor to form snowfall. The correlation between SCS and temperature and that between SCS and precipitation both reach the maximum at 1000 m-1500 m and the minimum at 3500 m-4000 m. The snow cover is positively correlated with spring precipitation and negatively correlated with temperature. With the increase in altitude, the correlation between SCM and temperature gradually becomes smaller. The correlation between SCM and temperature changes little at altitudes less than 1000 m and greater than 3000 m but increases with altitude within 1000-3000 m; the correlation with precipitation shows a downward U shape, peaking at 2000-2500 m, with an inflection point at 1000-1500 m.

Comparison with Previous Studies and Explanation of Phenomena
The results obtained in the study are compared with those of previous studies, and the potential reasons for their consistency and inconsistency are discussed. The results show that the snow cover in China is widely distributed and uneven, with seasonality and high spatiotemporal heterogeneity. The main snow cover season in China is from November to March, and snow is mainly distributed in the Greater Khingan Range, the Lesser Khingan Mountains, the Altay Prefecture, the Tianshan Mountains, and the high-altitude areas of the Qinghai-Tibet Plateau. The SSA and APA are mainly distributed in Northeast China, northern Xinjiang, and the Qinghai-Tibet Plateau. This is basically the same as the spatial distribution of snow days and snow-covered classification obtained by Li [45] through the site data in the 1970s and the 1980s. Still, according to the snow type division result, we can see that the SSA have decreased significantly, which is mainly due to the changes in environmental factors, such as temperature and precipitation, in the 21st century. In the context of global warming, late SCS, early SCM, and shortened SCD would occur [46], with a more significant reduction in SCA. Furthermore, many previous studies have confirmed that both SCA and SCD in the high latitudes of the Northern Hemisphere (NH) have decreased significantly in recent years [47,48]. The analysis of this study shows that the overall SCD in China have decreased and the SCM advance is the same as the general trend, but the SCS also shows an advanced trend as a whole, and different regions also show different trends. This is related to the undulating and changing topography of China and the large spatial heterogeneity.
SCP is sensitive to changes in meteorological variables, especially temperature and precipitation [49][50][51]. This is mainly because precipitation can be directly or indirectly transformed into snow under certain temperature conditions. The decreased temperature could increase the probability of snowfall. When the temperature is lower than the freezing point, precipitation often occurs in the form of a snowfall [52]. Therefore, the increase in precipitation under the condition of temperature freezing point is beneficial to increasing and maintaining snow cover, resulting in more SCD, early SCS in autumn, and delayed SCM in spring. However, higher temperatures or increased precipitation at higher temperatures can delay snow formation or induce snowmelt, reduce SCD, delay SCS, and advance SCM.
There are different effects of temperature and precipitation on snow cover in different seasons. In autumn, the SCD are negatively correlated with temperature and positively correlated with precipitation, while SCS shows a completely opposite correlation. Snowfall is the solid form of precipitation and a necessary condition for forming a snow cover. After evaporation or sublimation, precipitation mostly exists in the form of water vapor in the atmosphere. When it is condensed, it forms ice crystals, which are then deposited to form falling snow. Therefore, the more the autumn precipitation, the more the number of SCD and the earlier the SCS; higher temperatures make snow more difficult to form or melt the snow quickly, resulting in fewer SCD and later SCS. The SCD are less affected by the winter precipitation, mainly because there is less precipitation in winter, leading to fewer water vapor sources. In addition, the surface temperature in winter is lower and the temperature of precipitation may be higher than that of snow, which promotes snow melting and reduces SCD. Thus, SCD are negatively correlated with winter precipitation. The precipitation and temperature in spring are similar to those in autumn, but because the temperature in autumn is gradually decreasing, the temperature in spring is gradually increasing, and the condensation of water vapor is more in autumn than in spring, so the precipitation in autumn has a greater impact on the SCD. Moreover, SCM is mainly affected by temperature and precipitation in spring. The higher the temperature, the earlier the SCM, and the more the precipitation, the later the SCM.
Elevation is a major factor in determining the role of temperature and precipitation in snowpack variability [53,54]. Our results also show that the effects of temperature and precipitation on SCP differ at different altitudes. Especially in SSA, SCD have an opposite effect with winter temperatures around 3000 m, mainly due to the relationship between surface temperatures and precipitation temperature at high altitudes. The effects of temperature and precipitation on SCP parameters vary with altitude, with similar trends but different magnitudes compared to previous studies [50], which may be caused by the different product accuracy, statistical area, and year we use.

Uncertainty Analysis
Compared with SCE indicated by the standard MODIS snow products, the improved MODIS CGF snow cover product's accuracies are obviously higher. However, because of the similarity of the optical properties of ice clouds to that of snow [29] and the inaccurate cloud mask provided by MOD09GA/MYD09GA products, the problem of cloud/snow confusion may contribute to the largest uncertainty in the new product [34]. Moreover, this product is a binary product. Although it has great advantages in large-scale snow cover spatial distribution and trend analysis and operation efficiency, the binary representation of single-pixel snow cover information will also bring certain uncertainty. Compared with other definitions of SCS and SCM, the definitions in this study have effectively solved the problem of regional instantaneous snow cover, but the regional snow cover in China is complex. Especially in some areas of the Qinghai-Tibet Plateau, the snow cover has a shorter existence time, less than 5 days, which will affect the accuracy of the snow cover. More appropriate definitions of SCS and SCM may be further explored in different regions.

Conclusions
This study analyzes the spatiotemporal characteristics of and change trends in SCP in China from 1 September 2000 to 31 August 2020, based on the NIEER CGF MODIS SCE product. The relationship between SCP, temperature, precipitation, and the influence of altitude are also investigated.
Snow cover distribution is extremely uneven and has apparent spatiotemporal heterogeneity in China. Snowfall mainly starts in autumn, reaches the maximum SCA and SCD in January, and gradually disappears at the end of spring and early summer, except for the perennial snow cover. In the past 20 years, the interannual variation in snow cover in China shows a trend of decreasing SCA, decreasing SCD, advancing SCS, and advancing SCM, and the advance slope of SCM is larger than that of SCS. The snow in China is mainly distributed in northeastern China (including Northeast China and west Inner Mongolia), northern Xinjiang, and the Tibet Plateau. The slope of SCP parameters in China was mainly within ±0.5 d/y. In the statistics of the percentage of SCD, SCS, and SCM changes in the total area of the country, the decreased area (slope < 0) is 17.88%, 6.22%, and 11.02% more than the increased area (slope > 0), respectively. All regions except SFA, SCD, SCS, and SCM are shortened or advanced.
Temperature should be of high concern in SCP analysis compared with precipitation. The correlation of SCP with temperature and precipitation has obvious spatial and seasonal differences. In the main snow cover area, SCD and temperature (spring, autumn, and winter) are negatively correlated in most areas and positively correlated with spring and autumn precipitation, and the areas are similar but opposite in winter. SCS was positively correlated with autumn temperature and negatively correlated with precipitation. SCM was negatively correlated with spring temperature and positively correlated with precipitation. The areas that pass the significance test are mainly in SSA and APA. For SSA, SCD are mainly affected by spring temperature below 1500 m in elevation and are mainly affected by autumn temperature above 1500 m. At different altitudes, SCD are greatly affected by autumn precipitation. At altitudes above and below 3000 m, SCD and winter precipitation show an opposite correlation.
This study explores SCP distribution, changes, and correlations with meteorological and topographic factors in China, providing better results and finer spatial resolution, which contributes to understanding the land surface warming over the past few decades of SCP changes and related climate prediction studies and provides primary support for mountain snow cover.