Satellite-Based Assessment of Snow Dynamics and Climatic Drivers in the Changbai Mountain Region (2001–2022)

Abstract

Changbai Mountain is located in China’s northeastern seasonal stable snow zone and is a high-latitude water tower. The changes in snow cover have a great influence on the hydrological process and ecological balance. This study quantitatively analyzed the spatio-temporal variation in snow cover in the Changbai Mountain region and its driving factors based on Moderate Resolution Imaging Spectroradiometer (MODIS) data. To improve the accuracy of snow cover analysis, a simple cloud removal algorithm was applied, and the locally optimal NDSI threshold was investigated. The results showed that the snow-covered area (SCA) in the Changbai Mountain region exhibited strong seasonality, with the largest SCA found in January. The SCA during the winter season showed an insignificant increasing trend ($83.88{\mathrm{km}}^2$) from 2001 to 2022. The variability in SCA observed from November to the following March has progressively decreased in recent years. The snow cover days (SCD) showed high spatial variation, with areas with decreased and increased SCD mainly found in the southern and northern regions, respectively. It was also revealed that temperature is the primary hydrometeorological factor influencing the snow variation in the study domain, particularly during the spring season or in high-elevation areas. The examined large-scale teleconnection indices showed a relatively weak correlation with SCA, but they may partially explain the abnormally low snow cover phenomenon in the winter of 2018–2019.

1. Introduction

Snow is a critical component of the Earth’s surface and a key indicator of climate change. With unique physical properties such as high reflectivity, high latent heat of phase change, and low thermal conductivity, snow significantly impacts global water and energy cycles. Climate change has greatly altered the spatio-temporal variation of snow, leading to more frequent and severe snow-dominated floods and droughts [1,2]. In Northeast China, snowmelt runoff accounts for approximately 10% to 27% of the total annual runoff [2], impacting water supply, irrigation, and food production. Understanding the variation of snow cover and related phenology is important for water management and hazard prevention.

Traditional in situ snow measurements are labor-costly and present challenges in providing spatio-temporal continuous data, particularly over inaccessible areas. With advancements in remote sensing technology, satellite-based snow monitoring overcomes the limitation of in situ measurements and provides continuous snow data over complex terrains. Thus, it has become increasingly used for snow studies [3,4,5,6,7,8]. Owing to the optical properties of snow, optical remote sensing provides global snow cover data with high accuracy, particularly in open and homogeneous areas under cloud-free conditions [9]. The Normalized Difference Snow Index (NDSI), calculated as the ratio of the difference in reflectance between visible and short-wave infrared bands, is often used to extract snow cover for optical satellite data. Among existing optical snow products [10,11,12,13,14,15,16], such as data from the Advanced Very High Resolution Radiometer (AVHRR) [10], the Visible Infrared Imaging Radiometer Suite (VIIRS) [15], and the Moderate Resolution Imaging Spectroradiometer (MODIS) [3], MODIS snow cover products from the Terra and Aqua satellites have become one of the primary data sources for snow research due to their advantages of a long time series (available since 2000), high spatiotemporal resolution (500 m and daily), and global coverage.

MODIS daily snow cover products have been widely utilized in studies on snow cover variation [17,18], snow phenology [19], and hydrological applications [20,21]. Previous studies have compared MODIS snow products with other satellite-based data and field observations [22,23,24] and revealed that MODIS snow products possess high accuracy under clear sky conditions [25,26]. However, extensive cloud coverage limits MODIS applications [3,27,28]. To address this limitation, many attempts have been conducted, such as combining optical and microwave sensors [13,29,30,31], synthesizing daily snow images from two MODIS platforms (Terra and Aqua) [32], and applying statistical methods [33,34,35] to remove clouds from snow images.

The Changbai Mountain region, located in the northeastern part of China’s three major seasonal stable snow zones, serves as an important ecological barrier for Northeast China (Figure 1) [36]. The Changbai Mountain is a typical high-latitude water tower, and it is the source of the Second Songhua River, Tumen River, and Yalu River [37]. It supplies abundant water sources for natural and anthropogenic water demands. As a key ecological function zone, the region’s vegetation ranges from temperate broad-leaved forests to tundra, showcasing a typical vertical zonation pattern [38]. The response of snow variation to climate change in the Changbai Mountain region greatly influences regional ecological balance, hydrological processes, and water regulation [37,39,40,41]. However, the spatio-temporal variation of snow and its driving factors in the Changbai Mountain region have not been well investigated.

Chen et al. [41] analyzed the snowpack variations at different altitudes in the Changbai Mountains from 1960 to 2018 based on 54 ground-based stations. They found that the snow cover duration exhibited an increasing trend from 1960 to 1980, followed by a subsequent decrease at all altitudes. Additionally, they found earlier melting of the spring snowpack in the 2010s than in the 1960s. Based on Landsat 8 satellite images, Leng [42] found that the snow-covered area (SCA) in the Changbai Nature Reserve decreased dramatically from 2014 to 2023, particularly in the northwestern and central regions, as well as the edges of Changbai Mountain. More studies on snow variation in the Changbai Mountain region focused on exploring the impact of snow parameters (e.g., SCA, SCD, and snow depth) on vegetation [40,43,44,45]. For example, Zong et al. [45] investigated the linkage between snow cover and shrub distribution changes in the alpine tundra of Changbai Mountain based on multi-source satellite data from 1965 to 2019. They found that the spring snow cover decreased in most of the alpine tundra at elevations below 1950 m but increased at high elevations above 2300 m and around an elevation of 2000 m. Chang et al. [43] utilized MODIS daily snow cover datasets to explore the relationship between snow cover phenology and forest spring phenology in the Changbai Mountains and revealed prolonged snow cover duration (0.43 day/year) and earlier snow cover end day (−0.1 day/year) from 2001 to 2020.

Despite these studies being conducted over the Changbai Mountain regions, most of them only analyzed one single snow parameter, and a comprehensive analysis of snow parameters in the Changbai Mountain regions is still lacking. Chen et al. [41] focused on the snowpack variation analysis, but the results were based on measurements from snow stations, which have a limited footprint to represent the spatial characteristics. Satellite-based studies provided spatial continuous monitoring of snow cover, but they usually extract snow cover by comparing the Normalized Difference Snow Index (NDSI) with a threshold of 0.4 based on a global reference [45,46] or a threshold of 0.1 based on the China reference [40,43,47]. A more appropriate NDSI threshold for identifying snow-covered pixels over the Changbai Mountain region is needed for a more accurate capture of snow-cover conditions. Additionally, the driving factors of snow variation in the Changbai Mountain region, particularly the linkage between snow cover characteristics and atmospheric teleconnections (e.g., El Niño-Southern Oscillation, Arctic Oscillation, etc.) in the study domain remains unclear.

Thus, the purpose of this study was to assess snow dynamics in the Changbai Mountain region over the past two decades and explore the linkage between snow cover and climate indices. To enhance the accuracy of snow analysis, a simple cloud removal algorithm was used to mitigate cloud impacts, and the optimal NDSI threshold for the Changbai Mountain regions was investigated via evaluation against ground-based measurements. The overall objectives of this study were: (1) to quantify the spatiotemporal variation of snow-covered area (SCA) and snow cover days (SCD) in the Changbai Mountain region from 2001 to 2022; (2) to unveil the impact of hydrometeorological factors (temperature, precipitation, and evaporation) and large-scale atmospheric teleconnections on snow variations in the study domain.

2. Study Area

The Changbai Mountain region, as defined by the National Ecological Functional Area Planning and the China Biodiversity Conservation Strategy and Action Plan (2011–2030) [48], is located between 40°48′ to 44°39′N and 125°12′ to 131°17′E, with an area of approximately $1.05\times {10}^5{\mathrm{km}}^2$ (Figure 1). The study domain is located in the eastern and central parts of Jilin province, covering approximately 55% of the province. The elevation of the Changbai Mountain region ranges from 4 to 2683 m, and the high-elevation regions are primarily located in the southern Changbai Mountain region and parts of the central and eastern areas. The east terrain is complex, featuring a mixture of hills, mountains, plateaus, and valleys, while the western part is characterized by lower elevation and predominantly flat terrain.

The study area has a temperate continental climate, characterized by distinct seasons with warm to hot summers and cold winters. The annual minimum temperature is approximately −28 °C in January, and the maximum temperature can reach 30 °C in July [49]. Precipitation mainly falls during the summer season from June to September, accounting for about 60–70% of the annual precipitation [50]. According to the atmospheric reanalysis data from the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2), the annual average temperature ranges from −5.52 °C to 3.03 °C, and the annual precipitation varies from 731 to 1223 mm. Snowfall in the study area predominantly occurs from November to the following March, with a maximum snow depth of around 100 cm [41] and a maximum daily snowfall exceeding 20 mm [51]. Snowmelt runoff is the primary source of spring water resources in the area, typically lasting from March to April. Spring floods often occur when temperatures rise rapidly during the spring thaw [8].

3. Data and Methodology

3.1. Data

3.1.1. MODIS Data

In this study, snow cover data from 1 September 2001 to 31 August 2023 (represented as hydrological years from 2001 to 2022) was acquired from the MODIS Terra Daily Level 3 Global 500 m Grid, Version 6 (MOD10A1.006) product, and accessible via the National Snow & Ice Data Center (NSIDC) (https://nsidc.org/data/mod10a1/versions/61, accessed on 25 January 2025). MODIS detects snow cover based on the Normalized Difference Snow Index (NDSI), calculated as the ratio of the reflectance difference in the visible and short-wave infrared bands:

$$NDSI={\displaystyle \frac{b_4-b_6}{b_4+b_6}}$$

(1)

where b₄ is MODIS bands 4 ($0.54–0.56\mathsf{\mu }\mathrm{m}$), and b₆ is MODIS bands 6 ($1.62–1.65\mathsf{\mu }\mathrm{m}$).

A pixel with NDSI between 0 and 1 is considered to have snow presence, and a pixel with $\mathit{NDSI}<0$ is considered a snow-free land surface. It is also noted that pixels with greater than zero NDSI values may be caused by some other surface features, such as salt pans and cloud-contaminated pixels at the edge of clouds [15]. Therefore, a series of data screens was applied to the raw NDSI data to alleviate snow commission error [52]. This study used the MOD10A1 NDSI snow cover product, with data screens applied and quality assessment conducted for snow characteristic analysis in the Changbai Mountain area. NDSI values between 0 and 1 (i.e., 0 to 100 in the MOD10A1 NDSI snow cover data, as they are stored as 8-bit unsigned integers) are used to indicate the presence of snow in a pixel.

In this study, the MODIS data were reprojected and bilinearly interpolated to the 500 m resolution grid in the universal transverse Mercator projection with the WGS-84 ellipsoid for a more precise calculation of the snow-covered area. To enhance the accuracy of snow assessment and mitigate the impact of clouds, the MODIS NDSI values were converted to the fractional snow cover (FSC) based on an empirical relationship, and a cloud removal algorithm was applied (more details in Section 3.2.1). Additionally, a locally optimal NDSI threshold for identifying binary snow cover data was investigated in this study (Section 3.2.2).

3.1.2. MERRA-2 Data

The atmospheric reanalysis data Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2) was obtained from the National Aeronautics and Space Administration (NASA)’s Goddard Modeling and Assimilation Office (GMAO), and the data were used to assess the impact of hydrometeorological factors on snow characteristics in the Changbai Mountain region. MERRA-2’s assimilation system is the fifth-generation Goddard Earth Observing System Model (GEOS-5), with data available from 1980 to the present, having a spatial resolution of 0.5° × 0.625° [53]. The MERRA-2 hourly time-averaged two-dimensional data collection (M2T2NXFLX_5.12.4), which includes estimates of precipitation, temperature, and evaporation, was used in this study. The data are available for download at NASA’s Goddard Earth Sciences Data and Information Services (https://earthdata.nasa.gov/gesdisc, accessed on 25 January 2025). For consistency, the hourly MERRA-2 data were aggregated to a daily scale corresponding to the MODIS data. In terms of the difference in spatial resolution between the MODIS and MERRA-2 data, MERRA-2 estimates of precipitation, temperature, and evaporation for each MOIDS pixel were assigned based on the nearest neighbor principle.

3.1.3. Ground-Based Snow Depth Data

A GNSS-based snow depth dataset over Northern China (GSnow-CHINA v1.0) provides historical snow depth estimates from 80 ground-based GNSS stations across northern China [54]. The dataset spans from 2013 to 2022, covering 25–55°N and 70–140°E. It has temporal resolutions of 24, 12, and 2 h (for some stations) and represents an area of approximately $1000{\mathrm{m}}^2$ around each station. The data are available at the National Tibetan Plateau/Third Pole Environment Data Center (https://data.tpdc.ac.cn/zh-hans/data/87726e28-570e-4dbf-ba9e-e856f8abd50e, accessed on 25 January 2025). It was found that eleven stations were located in the study domain. After quality control and removing stations providing data records for less than two years, eventually, a total of six stations were used in this study (Figure 1). To be consistent with the MODIS daily data, the 24 h snow depth data (i.e., daily data) from GSnow-CHINA (abbreviated as GNSS data afterward) were used as the ground reference to find the optimal NDSI threshold in the Changbai Mountain region.

3.2. Methodology

3.2.1. Cloud Removal Algorithm

To overcome the constraints of cloud cover on the application of MODIS snow cover products, this study implemented a cloud removal method based on cubic spline interpolation proposed by Tang et al. [55]. This method has demonstrated its capability to improve snow detection accuracy in the Tibetan Plateau. Similar to Tang et al. [55], the MODIS NDSI snow cover data were first divided into cloudy and cloud-free categories. The cloud-free NDSI values were converted to fractional snow cover area (FSC; i.e., percentage of snow cover of the pixel) based on the empirical relationship established from Landsat data using Equation (2) following Salomonson et al. [56]. For pixels that were identified as cloud (250), missing data (200), no decision (237), detector saturated (254), and fill (255) in the NDSI snow cover product, they were all merged into the cloudy category. The FSC value for each cloudy pixel was interpolated based on the cubic spline interpolation method using FSC values on cloud-free days.

$$FSC=1.45\times NDSI-0.01$$

(2)

The cubic spline interpolation aims to find a piecewise continuous curve $S\left(x\right)$ that fits a series of points $(x_i,y_i)$ for $i\in [0,n]$. The curve $S\left(x\right)$ can be represented as a cubic polynomial $s_i\left(x\right)$ for each subinterval $[x_i,x_{i+1}]$:

$$s_i\left(x\right)=a_i+b_i\left(x-x_i\right)+c_i{\left(x-x_i\right)}^2+d_i{\left(x-x_i\right)}^{3}x\in \left[x_i,x_{i+1}\right]$$

(3)

where i represents the numbering of an interval, and $a_i$, $b_i$, $c_i$, and $d_i$ are the coefficients of the cubic polynomial. The spline $S\left(x\right)$ needs to fulfill certain conditions to ensure smoothness, which include: (1) $S\left(x\right)$ is continuous at each $x_i$ (i.e., $s_i\left(x_i\right)=y_i$ and $s_i\left(x_{i+1}\right)=y_{i+1}$), and (2) the first and second derivatives of $S\left(x\right)$ are continuous at each $x_i$ (i.e., $s_{i-1}^{\prime }\left(x_i\right)=s_i^{\prime }\left(x_i\right)$ and $s_{i-1}^{\left(2\right)}\left(x_i\right)=s_i^{\left(2\right)}\left(x_i\right)$). To fix the spline interpolation, the second derivative of $S\left(x\right)$ at the endpoints is often set to zero as a natural boundary condition [57]. Finally, the set of n equations and all the unknown coefficients can be solved through cubic spline interpolation.

In this study, cubic spline interpolation was conducted for each cloudy pixel in each hydrological year, with September 1st as the first day. The day of the year is used as x, and the corresponding FSC value is used as y. The sets of points $(x_i,y_i)$ were formed by the pairs of the day of the year and FSC value on cloud-free days in the corresponding hydrological year. For the interpolated FSC value, if it is larger than 1, it is reassigned as 1, and if it is less than zero, it is reassigned as zero. The cloud gap-filled FSC data were then used for identifying snow-covered and snow-free pixels, which were subsequently used for snow characteristics analysis.

3.2.2. NDSI/FSC Threshold Selection

Previous studies [46,58] found that pixels entirely covered with snow usually have high NDSI values and can be spectrally distinguished from clouds. Meanwhile, NDSI decreases for pixels that are fractional mixtures with other surface features, such as under forest canopies [59]. Dozier [60] revealed that pixels with over 50% covered by snow yielded NDSI values larger than 0.4. Considering the surface features, viewing conditions, and cloud impacts, a fixed NDSI threshold of 0.4 is often regarded as a global reference and has been widely used for binary snow cover mapping [3,52]. However, an increasing number of studies have revealed that the optimal NDSI threshold varies with the study domain [52]. Based on daily snow depth data from over 200 stations within China, Zhang et al. [47] recommended using an NDSI threshold of 0.1 for identifying snow cover across China. Wang et al. [61] found that using a threshold of 0.1 for non-forest areas in Northeast China improved the accuracy of the binary snow cover data. Similarly, Wang et al. [59] set an NDSI threshold of 0.1 to identify cloud cover. Therefore, this study further investigated the optimal NDSI threshold in the Changbai Mountain region based on finding the NDSI threshold that provides the highest accuracy. A range of NDSI thresholds from 0 to 0.4 in increments of 0.01 were used to determine if the pixel is covered by snow for the MODIS data, and the results were evaluated against ground-based GNSS data. It should be noted that the MODIS NDSI snow cover data were converted to FSC for cloud gap-filling in this study. Therefore, different NDSI thresholds were also converted to FSC thresholds based on Equation (2) and subsequently applied to the cloud gap-filled FSC data to generate a binary snow product. Considering the potential errors in measurements and the retrieval algorithm in the GNSS data, a snow depth threshold ranging from 0 to 5 cm with 1 cm increments was tested to identify the presence of snow for GNSS. Afterward, the F1 score, which is a harmonic mean of the precision and recall scores, was calculated to evaluate the accuracy of the threshold selection, as shown in Equation (4).

$$\begin{array}{cc}\hfill & precision={\displaystyle \frac{TP}{TP+FP}}\hfill \\ \hfill & recall={\displaystyle \frac{TP}{TP+FN}}\hfill \\ \hfill & F1=2\times {\displaystyle \frac{\mathit{precision}\times \mathit{recall}}{\mathit{precision}+\mathit{recall}}}\hfill \end{array}$$

(4)

where TP is the true positive (observed snow correctly detected by MODIS), FP is the false positive (snow detected by MODIS but observed by GNSS), and FN is the false negative (snow wrongly detected by MODIS but observed by GNSS).

Finally, the NDSI (FSC) threshold providing the highest F1 score was selected as the optimal threshold and used for snow cover identification and analysis in the Changbai Mountain region.

3.2.3. Evaluation of Snow Parameters

Following the definitions of snow parameters in previous studies [62,63], four parameters were used in this study to evaluate the spatio-temporal variations of snow in the Changbai Mountain region, which included Snow-Covered Area (SCA), Snow Cover Days (SCD), First Snow Cover Day (FSD), and the Last Snow Cover Day (LSD). The definitions of the examined snow parameters are summarized in

Table 1. The snow parameters used in this study.

Snow Parameters	Definition
Snow-Covered Area (SCA)	The area covered by snow in the study domain.
Snow Cover Day (SCD)	The number of days a pixel is snow-covered (i.e., FSC > threshold) in a hydrological year.
First Snow Cover Day (FSD)	The first day of the initial 5-day consecutive snow cover period in a hydrological year for a pixel.
Last Snow Cover Day (LSD)	The last day of the final 5-day consecutive snow cover period in a hydrological year for a pixel.

.

In this study, the SCA values at daily and monthly scales were calculated and evaluated. The daily SCA is defined as the area of pixels covered by snow on a given day, and the monthly SCA is calculated as the average of daily snow areas in each month. Considering the strong seasonality of snow, the winter SCA, which represents the average daily SCA during the winter season from December to February (DJF), was particularly investigated. Regarding the SCD, the regional average SCD is computed as the average SCD values of all pixels in the study domain.

The non-parametric Mann–Kendall (MK) test and the Theil–Sen median method were used to determine the significance and magnitude of the long-term trend of the snow characteristics in the Changbai Mountain region. The MK test does not require samples to follow a specific distribution and is not affected by outliers, making it suitable for non-normally distributed data in hydrology and meteorology [64,65]. The MK test provides the standard value Zs (Z statistic), UFk (forward statistic), and UBk (backward statistic) for significance evaluation. A Zs > 0 indicates an increasing trend, while Zs < 0 indicates a decreasing trend. A significance level of 5% was used in this study, corresponding to a comparison of the absolute value of Zs with 1.96. UFk represents the cumulative statistic calculated in the forward sequence of the data, and it is used to detect increasing or decreasing trends. UBk denotes the cumulative statistic derived from the backward sequence of the data, which helps to confirm the results of UFk as well as locate the possible change points. When the UFk and UBk intersect within the range of the significance threshold, the intersection point may indicate the potential onset of the trend change. If either UFk or UBk exceeds the significance threshold (e.g., ±1.96), the trend is considered statistically significant. Additionally, the Pearson correlation coefficient was calculated between snow parameters and climate and hydrometeorological factors to analyze the linkage between them.

4. Results

The spatio-temporal variation of snow cover in the study area was examined over 22 hydrological years, from 1 September 2001 to 31 August 2023. A hydrological year is defined as from 1 September to 31 August of the following year. Through a systematic evaluation, an NDSI threshold of 0.09 (corresponding to an FSC threshold of approximately 0.12) was used to identify the snow-covered pixels in the Changbai Mountain region (more details in Section 5). Furthermore, the influence of hydrometeorological factors and climate variables on snow characteristics in the study area is analyzed in this section.

4.1. Analysis of SCA Variation

4.1.1. Variations in Monthly SCA

Figure 2 illustrates the spatial distribution of the multi-year average FSC for each month. The results showed that the snow season in the study domain spans from November to the following April. Snow accumulation starts in November, and the snow melts quickly in March. During the winter season (DJF), a large portion of the study domain exhibited high FSC values, suggesting extensive and stable snow cover during winter. It is also found that snow exists in the northern and southern mountain regions for a longer period, with relatively high FSC values found in October and April. The seasonal differences in FSC are more evident in low-elevation regions (northern and central areas), with FSC values larger than 0.6 during the winter season and close to zero in other seasons.

By applying the selected FSC threshold to daily FSC data, the daily SCA was calculated and averaged to the monthly SCA. Figure 3a presents the multi-year averaged monthly SCA and the corresponding trends in SCA from 2001 to 2022 for each month. The SCA showed an evident seasonal variation, with SCA increasing from November to December and decreasing afterward. The percentage of area covered by snow is larger than 35% from November to the following March, and the peak value of approximately 75% was observed in January. Regarding the inter-annual trend of SCA in each month from 2001 to 2022, both November and December exhibited an increasing trend, particularly for the November snow (0.48%/year, corresponding to about $503.25{\mathrm{km}}^2/\mathrm{year}$), suggesting the expansion of SCA in November and December in recent years. It was also noticed that January through March exhibited decreasing trends of SCA, with January showing the largest decreasing trend of 0.21%/year (corresponding to approximately $220.17{\mathrm{km}}^2/\mathrm{year}$). Such changes in SCA trend also indicate reduced temporal variations in SCA, with the monthly differences in SCA in the study area gradually diminishing. However, it should be noted that most monthly SCA trends are not statistically significant at a level of 5%, except for August and October (Figure 3a). Considering that MODIS only provides observations from 2001, more investigation on the long-term trends in SCA in the Changbai Mountain area with an extended period from multi-data sources would enhance the reliability of our findings in future studies.

Additionally, a scatterplot of the SCA trend and SCA was generated to investigate the potential relationships between them (Figure 3b). K-means clustering analysis was employed for data segmentation, and three primary clusters were identified, including the percentage of SCA over the study domain, predominantly between 0 and 27.2% with a minimal annual trend (the red area in Figure 4b), within the 27.2–55.4% range and a significant increasing trend (the green area), and above 55.4% with most points exhibiting decreasing trends (the purple area). A close investigation found that the points identified as the first group with relatively low SCA values were mainly from April to September, which is coincident with seasonal variations in snow cover. The points that exhibited the significant increasing trend (i.e., the second cluster) were mainly from the beginning (mid-November to the end of November) and end (mid-March to the end of March) of the snow period, with an average trend of 0.72%/year (approximately $754.88{\mathrm{km}}^2/\mathrm{year}$). The third cluster with the largest SCA and slightly decreasing trend was found during the mid-snow season from the end of November to early March.

The clustering analysis revealed opposite long-term trends in SCA for the second (increasing trend at the beginning and end of the snow period) and third (decreasing trend in the mid-snow season) clusters. This finding is consistent with the monthly SCA analysis (Figure 3a), suggesting that the temporal difference in SCA during the snow period from November to the following March gradually diminished in recent years.

4.1.2. Variations in Winter SCA

Considering the seasonality of snow, the variation in snow cover during the winter season was investigated. Figure 4a shows the proportion of pixels with average FSC during the winter season (DJF) located in the groups of 0.0–0.2, 0.2–0.4, 0.4–0.6, 0.6–0.8, and 0.8–1.0 in each hydrological year. It was found that the percentages of pixels with high FSC values varied substantially in the winters of different hydrological years. There were a small proportion of pixels (less than 0.5%) with FSC values greater than 0.8 in the winters of 2001–2002, 2003–2004, 2007–2008, and 2018–2019, while during the consecutive winters from 2009 to 2017, a certain number of pixels exhibited FSC > 0.8. In the winter of 2018–2019, the percentage of areas showing FSC values less than 0.2 reached the highest level during the study period (75.06%), and the percentage for FSC values greater than 0.8 was the second-lowest (0.14%) on record.

The winter SCA was calculated by applying the selected FSC (Section 3.2.2) and averaging the daily SCA from December to January for each hydrological year, as shown in Figure 4b. The results revealed that the winter SCA was relatively large during 2009–2016, with the percentage of the SCA exceeding 69.93%. The winters for the years 2004–2015 and 2013–2014 experienced the largest snow cover, with the ratio of snow-covered pixels exceeding 80% of the study domain. Due to the large percentage of pixels with low FSC values, the winter SCA in 2018–2019 was exceptionally low, with a snow-covered percentage of approximately 30.72% ($3.22\times {10}^4{\mathrm{km}}^2$). The winter SCA for the other years were all larger than 50% ($5.24\times {10}^4{\mathrm{km}}^2$) but below 80%. To represent the dispersion of inter-annual winter SCA, the coefficient of variation (Cv) was analyzed. It was found that the Cv values calculated based on the winter SCA of all years (Cv₁ = 0.16) and by removing the winter SCA for 2018–2019 (Cv₂ = 0.11) both fell in the moderate range, between 0.1 and 0.3. Without considering the extremely low SCA in the 2018–2019 winter, the Cv value exhibited a significant decrease, suggesting a non-negligible deviation of SCA condition in the 2018–2019 winter. More discussion of the low SCA in the 2018–2019 winter is provided in Section 5.

The long-term trend analysis based on the Theil–Sen median method revealed an increasing rate of $83.88{\mathrm{km}}^2/\mathrm{year}$ (corresponding to 0.08%/year) in winter SCA from 2001 to 2022 (Figure 4c). However, the monotonic increasing trend was not significant at a significance level of 5% (Zs = 0.28 < 1.96) according to the MK test. The change-point detection analysis showed that the UFk value remained between −1.96 and 1.96, with no intersection with Ubk, indicating that the annual winter SCA in the study area did not exhibit a significant change in trend or abrupt shift at the 5% significance level.

4.2. Analysis of SCD Variation

4.2.1. Temporal Variations in SCD

The SCD for each pixel and hydrological year was calculated, as shown in Figure 5. During the study period, area-average SCDs were generally between 50 and 150 days but occasionally exceeded 150 days in some years, such as 2009 and 2012. It was also observed that the average SCDs from 2009 to 2016 were relatively longer, with the largest area-average SCD of 164.08 days in 2009 (Figure 5b). The hydrological year of 2018–2019 exhibited the fewest average SCD (68.55 days), followed by 2001–2002 (90.73 days). The SCD showed consistent inter-annual variation with winter SCA, showing a high correlation coefficient of 0.88 between the two parameters. Similar to the findings in the winter SCA analysis (Section 4.1.2), an extremely low SCD for 2018–2019 was detected. The Cv values using all SCD data and after excluding the SCD value in 2018–2019 were 0.18 and 0.15, respectively, suggesting that the deviation in SCD during 2018–2019 was less pronounced than that of winter SCA.

The trend analysis revealed that the area-average SCD reduced with a rate of −0.29 days/year from 2001 to 2022 (Figure 5c), but the slope was not significant at the level of 5%. A more detailed investigation of the first and last snow days was conducted. The results showed that the FSD typically occurred between October and November, and the LSD fell between March and April in the Changbai Mountain region. From 2001 to 2022, the study domain experienced a delayed FSD and an earlier LSD, with a trend of 0.09 days/year and −0.18 days/year, respectively. Notably, the snow period in 2018–2019 experienced a late FSD (around November 10th) and early LSD (around March 28th), corresponding to the abnormal snow cover characteristics observed in that year. Such findings for the FSD and LSD aligned well with the overall change in SCD.

4.2.2. Spatial Variations in SCD

Figure 6a depicts the spatial distribution of multi-year average SCD from 2001 to 2022. Most areas were covered by snow for approximately 50–150 days. Regions with higher elevation, such as the southern and northern mountainous regions, experienced a longer SCD, exceeding 150 days and even over 200 days. The trend in SCD for each grid was also calculated, as presented in Figure 6b. It was found that approximately 46.98% of the area exhibited a decreasing trend, and 38.16% exhibited an increasing trend. The areas with increasing trends were mainly located in the eastern and northern parts of the study area, while decreasing trends were detected in the southern areas.

A statistical test was conducted to evaluate whether the trend in SCD was significant at a level of significance of 5%. The results showed that only approximately 2.14% of the study area exhibited a statistically significant trend in SCD, including 0.19% for an increasing trend and 1.94% for a decreasing trend. A close investigation revealed that the significant decreasing trend was mainly found in relatively high-elevation regions (defined as R1, with an average elevation of 959.51 m), and the significant increasing trend was found in regions with lower elevations (defined as R2, with an average elevation of 540.96 m). To unveil the underlying reasons for the opposite trends at different elevations, area-averaged precipitation and temperature from MERRA-2 in corresponding regions (i.e., R1 and R2) were acquired, as shown in Figure 6c,d. It was found that both regions experienced an increase in temperature and precipitation from 2001 to 2022. However, the increasing rate in temperature in R1 (0.022 °C/year) was larger than in R2 (0.018 °C). On the contrary, R2 exhibited a greater increasing trend (7.05 mm/year) in precipitation than R1 (2.76 mm/year). A possible explanation for the opposite SCD trends in R1 and R2 may be differences in the predominant driving factor at different elevations. In the lower-elevation region (R2), the precipitation had a larger impact on the snow relative to the temperature. Therefore, the larger increasing trend in precipitation dominated the snow characteristics and led to increased SCD. Conversely, temperature played a more important role in the snow in the high-elevation region (R1); thus, the large increasing trend of temperature leads to reductions in SCD in R1, despite the increasing trend in precipitation.

Figure 7 further explores the relationship between elevation and SCD, as well as elevation and SCD trends. It was observed that the SCD generally increased with elevation, as anticipated, while the variation pattern in SCD trend versus elevation was more complex. Figure 7b shows the positive area-average SCD trends at relatively low (<900 m) and high elevation (>2100 m) ranges, while for elevations between 900 and 2100, negative SCD trends were observed. As elevation increased, the average SCD decreased from positive values to negative and reached the largest negative trend of −0.77 days/year for the elevation range of 1200–1500 m. For regions above 1500 m, the average SCD increased with the increase in elevation.

4.3. Influence of Hydrometeorological Factors

To explore the intrinsic connections between snow and hydrometeorological factors in the study domain, the impacts of temperature (T), precipitation (P), and evaporation (E) data on SCD and winter SCA were examined. The cross-correlation analysis (Figure 8) showed that the temperature exhibited the strongest correlation with snow characteristics in the study area. The correlation coefficients between temperature and snow characteristics all exceeded 0.46 in both the spring (MAM) and winter (DJF) seasons. Meanwhile, precipitation showed a weaker positive correlation with snow characteristics (e.g., ${\mathrm{R}}_{Pspr-SCD}$ = 0.50). Evaporation had a minimal correlation with snow characteristics, with correlation coefficients below 0.3. These relationships between hydrometeorological factors and snow characteristics align with previous studies in other snow-dominated regions, such as the Tibetan Plateau [66].

The Pearson correlation coefficients between hydrometeorological variables and snow characteristics are notably influenced by season. Spring temperature correlated more strongly with SCA and SCD than winter temperature (${\mathrm{R}}_{Tspr-SCA}$ = −0.52, ${\mathrm{R}}_{Tspr-SCD}$ = −0.70). Spring precipitation also showed a slightly stronger correlation with snow characteristics relative to winter precipitation (${\mathrm{R}}_{Pspr-SCA}$ = 0.38, ${\mathrm{R}}_{Pwin-SCA}$ = 0.39). In terms of evaporation, the winter evaporation exhibited a negative correlation with SCA and SCD, while the spring evaporation provided a positive correlation. This discrepancy can be attributed to the physical properties of snow that increased snow cover and reduced evaporation in the winter, while snow melting in the spring alters the surface cover, leading to increased evaporation.

Additionally, it was found that SCD exhibited a stronger correlation with hydrometeorological factors relative to winter SCA. This may be related to the differences in the definition of parameters, as SCD was counted using all days in the hydrological year, while winter SCA reflects the average daily SCA during the winter season (DJF). Thus, SCD more effectively captures interannual variations in snow characteristics than snow areas. Conversely, winter evaporation demonstrates the opposite relationship because the snow area better represents changes in winter snow coverage in the study area.

To further investigate the relative importance of the six hydrometeorological factors on snow characteristics, a principal component analysis (PCA) was conducted, and the results are shown in

Table 2. Principal component (PC) table for six hydrometeorological factors including Pspr, Tspr, Espr, Pwin, Twin, and Ewin.

PC Number	Eigenvalue ($\mathit{\lambda }$)	Contribution Rate (%)	Accumulative Contribution Rate (%)
1	2.51	36.43	36.43
2	1.95	30.32	66.76
3	0.95	15.00	81.76
4	0.43	11.24	92.99
5	0.11	6.17	99.17
6	0.04	0.83	100.00

. The cumulative contribution rate of the first four principal components (PCs) reached 92.99%; thus, these four components were selected for analysis. From

Table 3. The first four principal components of Pspr, Tspr, Espr, Pwin, Twin, and Ewin.

PC Number	PC1	PC2	PC3	PC4
Pspr	−0.48	−0.12	0.23	0.78
Tspr	0.53	0.27	−0.06	0.17
Espr	−0.35	0.59	−0.28	0.14
Pwin	−0.45	0.08	−0.55	−0.54
Twin	0.39	0.23	0.75	0.22
Ewin	−0.07	0.72	0.00	−0.07

, it can be seen that spring temperature and precipitation coefficients have large absolute values in PC1, indicating that these are the main determining factors, corresponding to the cold and dry characteristics of the temperate continental monsoon climate in the study area. The second principal component highlights evaporation and temperature as significant factors. The first two principal components suggest that the influence of hydrometeorological factors on snow characteristics is ranked as temperature > precipitation > evaporation. However, PCA aims to explain the total variance among variables through linear combinations rather than focusing solely on simple correlations. Therefore, although evaporation exhibits a weaker direct correlation with individual snow characteristics (such as winter SCA or SCD), the high contribution rate of PC2 (30.32%) and the large absolute coefficients of evaporation and temperature indicate that the synergistic effects of evaporation and temperature influence changes in snow cover.

4.4. Influence of Climatic Factors

To explore the potential driving factors of the seasonal to inter-annual variations in snow characteristics in the Changbai Mountain region, the linkage between large-scale atmospheric teleconnection indices and SCA is analyzed in this section. Considering the effects of various climatic indices in Northeast China, four teleconnection indices, including the North Atlantic Oscillation (NAO), Arctic Oscillation (AO), NINO3.4 (El Nino), and PDO (Pacific Decadal Oscillation), were examined in the study. All of the climate indices data were provided by the National Oceanic and Atmospheric Administration (NOAA) Climate Prediction Center, and monthly data are available from (https://psl.noaa.gov/data/climateindices/list/, accessed on 25 January 2025).

Similar to the correlation analysis for hydrometeorological factors in Section 4.3, comparisons were made for anomalies between selected climate indices and SCA, as illustrated in Figure 9. The anomalies were calculated by subtracting the mean seasonal cycle (i.e., multi-year average values for each month) from the original data. On the monthly scale, the SCA anomalies showed better consistency with AO and NAO, with the peaks and troughs being well-matched. This aligns with theoretical expectations: when both are in a negative phase, extreme cold weather is more likely to occur, while in a positive phase, the climate tends to be relatively warm. However, the variations in NINO3.4 and PDO had weak agreements with SCA anomalies, exhibiting several contradictions to theoretical predictions, especially for NINO3.4 and PDO. When the 3-month moving average of NINO3.4 exceeds 0.50 and persists for 5 months, it is identified as an El Niño event, which typically causes warmer winters in the study area. However, this contrasts with the winter of 2010, where the SCA anomaly exceeded 15%. Nonetheless, some influence was observed during the winter of 2019. In general, when the PDO is in a positive phase, temperatures in Northeast China tend to be lower, making cold weather more likely. Conversely, when the PDO is in a negative phase, the weather is generally warmer. Both the PDO and NINO3.4 had some influence on the winter of 2019 (a warm winter). Overall, the results indicate that snow characteristics in the study area are primarily influenced by AO and NAO, while NINO3.4 and PDO have a negligible overall impact.

The analysis of inter-annual variations in snow in Section 4.1.2 and Section 4.2.1 revealed that the Changbai Mountain region experienced notably low SCA and short SCD in the 2018–2019 hydrological year, which may be related to the combined effects of multiple climatic factors. In the winter of 2018–2019, both the NAO and AO exhibited a positive phase. The coupling of these two climatic factors contributed to warmer conditions in the study area, while the larger positive phases of PDO and NINO3.4 also played a role in the occurrence of the warm winter that year.

5. Discussion

The selection of the NDSI threshold significantly impacts snow detection and characteristic analysis. In this study, an NDSI value of 0.09 (corresponding to an FSC of 0.12) was used as the threshold for identifying snow pixels using snow data derived from GNSS station networks as the ground truth (more results in the Supplementary Materials). The selected NDSI threshold is similar to the previous findings of Wang et al. [61], who examined the effects of different thresholds on snow accuracy using snow depth data from meteorological stations. They found that in non-forest areas, an NDSI threshold of 0.1 yielded the highest accuracy, while in forested areas, a threshold of 0 was optimal. In this study, the use of 0.09 as a threshold is justified, with precision rates exceeding 75% at all except one station and recall rates above 80% for all stations. Nonetheless, considering the limited number of stations, indirect measurement of snow depth from GNSS stations, and the variability of the optimal threshold for different land surface features, more exploration on refining NDSI threshold selection in the Changbai Mountain region based on varying topographical conditions with a denser station network is required to further enhance snow extraction accuracy.

This study revealed that the Changbai Mountain region experienced an insignificant (at a level of 5%) increasing trend in SCA from 2001 to 2022, with a trend of $83.88{\mathrm{km}}^2/\mathrm{year}$ (corresponding to a percentage of 0.08%/year). The divergence of SCA in different months was reduced, with increased SCA in the early snow period in November and decreased SCA during the middle and end of the snow seasons from January to April. The decreased SCA during the spring season was also noted in previous studies. Zong et al. [45] investigated the spring snow cover variation in the alpine tundra of the Changbai Mountains from 1965 to 2019 with multi-source satellite. They found that the region experienced a significant increase in temperature, leading to decreased snow cover in most regions below 1950 m and in the range of 2050–2250 m. In terms of SCD, this study found an insignificant decreasing trend in area-average SCD from 2001 to 2022, with a rate of −0.29 days/year. The regions where SCD decreased over time are mostly located in the south, which is consistent with the findings of Chang et al. [43]. It is also revealed that variations in SCD and SCD trends were highly influenced by elevation, with positive SCD trends found in regions with relatively low (<900 m) and high (>2100 m) elevations. A similar finding was reported by Chen et al. [41], that the snow cover duration increased with time from the 1980s to the 2010s when the elevation was lower than 600 m.

The anomalous low SCA and SCD for the hydrological year of 2018–2019 in the Changbai Mountain region was noticed in this study. Temperature analysis indicated that the temperature anomalies for 2018–2019 were 1.74 °C and 1.25 °C during the winter and spring seasons, respectively, which are both among the highest records in the last 22 years. The extremes in climatic factors such as AO, NAO, and NINO3.4 were evident. The China Climate Bulletin [67] also reported a winter temperature anomaly of approximately 1.2 °C for Jilin Province in 2018–2019, suggesting that the northeastern region of China was significantly affected by severe and extreme climatic anomalies in 2019, leading to increased disaster areas and severity [68].

The exploration of the dominant driving factor for snow changes showed that temperature was more important than precipitation in high-elevation regions, with an average elevation of approximately 960 m, while for lower-elevation regions (average elevation of 540 m), precipitation dominated the snow variation over temperature. These findings are consistent with many previous studies on snow-dominated mountain regions. For example, Huang et al. [19] found that the influence of temperature on SCD increases with elevation in the Tibetan Plateau. Chen et al. [41] investigated the influence of temperature and precipitation on snow depth in the Changbai Mountain region and found that temperature has a greater impact on snow depth than precipitation above 600 m from November to April. However, for lower-elevation areas (below 600 m), temperature and precipitation exhibited greater effects from November to January and February to April, respectively.

Although the overall trend of snow characteristics in the Changbai Mountain region did not exhibit significant changes, changes in snow characteristics, such as the winter SCA and shifts in the beginning time of snowmelt, pose challenges for water resource management in the region. Abnormal changes in snow characteristics (e.g., the 2018–2019 hydrological year) could disrupt the stability of the local ecosystem, lead to reduced crop yields, and cause issues such as drought due to the decrease in spring snowmelt. Moreover, since spring temperature influences snow characteristics more than winter temperature, monitoring trends in spring temperature will be crucial for predicting changes in snow characteristics, ultimately aiding in water resource management and agricultural planning.

6. Conclusions

Changbai Mountain is an important water tower in Northeast China. Previous studies on snow analysis in the Changbai Mountain region have mainly focused on the influence of snow on vegetation phenology. A systematic analysis of snow variation characteristics and the driving factors based on spatiotemporal continuous data over the Changbai Mountain region is lacking. This study quantitatively investigated the variability of SCA and SCD at different spatial and temporal scales in the Changbai Mountain region based on MOD10A1 NDSI snow cover daily data from September 2001 to August 2023 (represented as hydrological years 2001 to 2022).

The SCA analysis showed that the snow cover in the Changbai Mountain region is characterized by marked seasonality, with SCA expanding quickly from November, reaching a peak in January, and shrinking rapidly from March. The SCA during the winter season (DJF) exhibited an increasing trend from 2001 to 2022, with a rate of $83.88{\mathrm{km}}^2/\mathrm{year}$ (about 0.08%/year). However, the trend is not significant at a level of 5%. It is also noted that the differences in SCA from November to the following March diminished progressively over the years. In terms of SCD, the area-averaged SCD decreased with an insignificant rate of −0.29 days/year. The decreasing SCD trend was mainly observed in the southern region, while the northern region exhibited an overall increasing trend. Further exploration revealed that significant decreasing and increasing trends were mainly observed at high-elevation and low-elevation regions, respectively, which may be attributed to the different dominant factors at different elevations. In the low-elevation region, precipitation has a greater impact on snow than temperature; thus, increased precipitation leads to extended SCD. Meanwhile, in the high-elevation region, the impact of temperature exceeded precipitation, and increased temperature led to shortened SCD. Further investigation on the correlation between hydrometeorological factors and snow suggested that the spring temperature is a key factor influencing SCA and SCD. Evaporation yielded the lowest correlation with snow variations in the study domain. Regarding the linkage between snow and large-scale atmospheric teleconnections, all the examined indices showed a relatively weak correlation with SCA. However, the combined effects of the natural variability of climate may contribute to the low snow cover phenomenon in the winter of 2018–2019 to some extent.

Despite many findings in this study, there are some limitations that could be further explored in future work. This study employed a simple cloud removal algorithm based on cubic spline interpolation to fill cloud gaps for more accurate snow cover assessments. Considering various cloud removal algorithms proposed for MODIS snow products, the performance of different algorithms in the Changbai Mountain region should be compared and discussed. Moreover, MERRA-2 estimates were used in this study to investigate the potential driving factors of snow variation. However, the coarse spatial resolution of MERRA-2 may limit its capability to capture the highly varied precipitation and temperature in the Changbai Mountain region. Additionally, only large-scale climate indices were considered in this study, and the influence of smaller-scale weather systems on the snow characteristics remains unclear. More exploration with extended study periods and finer-spatial-resolution data could further enhance our understanding of climate change on snow characteristics in the Changbai Mountain region.