The normalized difference snow index (NDSI) is the most popular snow detection index. Due to cloud cover, it is difficult to produce complete and gap-free NDSI datasets. In this study, a spatial and temporal adaptive gap-filling method (STAGFM) is developed, whereby a weighted cloud-free similar pixel function is established for NDSI prediction. Cloud-covered NDSI gaps are filled by combining daily MOD10A1 and MYD10A1, and adjacent temporal composite (ATC) is applied. STAGFM is implemented with long-time interval data to completely recover NDSI gaps. Moderate Resolution Imaging Spectroradiometer (MODIS) NDSI product data (from 1 November 2017 to 31 March 2018) of Northeast China are chosen as an example, and daily cloud-free NDSI time series over this period are produced. The method effectiveness was validated by cloud assumption and snow depth (SD) data, the results show that STAGFM completely removes clouds and achieves an average correlation coefficient (r), root-mean-square error (RMSE) and mean absolute error (MAE) of 0.95, 0.08 and 0.06, respectively.

Snow cover is an important water source and even an Essential Climate Variable (ECV) as defined by the World Meteorological Organization (WMO). Assessing snow phenology and its driving factors in Northeast China will help with comprehensively understanding the role of snow cover in regional water cycle and climate change. This study presents spatiotemporal variations in snow phenology and the relative importance of potential drivers, including climate, geography, and the normalized difference vegetation index (NDVI), based on the MODIS snow products across Northeast China from 2001 to 2018. The results indicated that the snow cover days (SCD), snow cover onset dates (SCOD) and snow cover end dates (SCED) all showed obvious latitudinal distribution characteristics. As the latitude gradually increases, SCD becomes longer, SCOD advances and SCED delays. Overall, there is a growing tendency in SCD and a delayed trend in SCED across time. The variations in snow phenology were driven by mean temperature, followed by latitude, while precipitation, aspect and slope all had little effect on the SCD, SCOD and SCED. With decreasing temperature, the SCD and SCED showed upward trends. The mean temperature has negatively correlation with SCD and SCED and positively correlation with SCOD. With increasing latitude, the change rate of the SCD, SCOD and SCED in the whole Northeast China were 10.20 d/degree, −3.82 d/degree and 5.41 d/degree, respectively, and the change rate of snow phenology in forested areas was lower than that in nonforested areas. At the same latitude, the snow phenology for different underlying surfaces varied greatly. The correlations between the snow phenology and NDVI were mainly positive, but weak correlations accounted for a large proportion.

Northeast China is one of the primary snow-covered regions, and its forest coverage is over 40%. Forest snow identification is usually a challenging problem, and the SNOMAP algorithm tends to underestimate the amount of snow cover in forest regions for the lower normalized difference snow index. In this article, an improved method of the snow-cover identification based on the Landsat operational land imager is proposed. One improvement includes using the normalized difference forest snow index (NDFSI) to discriminate between snow-covered and snow-free forests. The threshold value of the NDFSI in different forest types is set according to the normalized difference vegetation index. On the other hand, the sun elevation is very low in winter in Northeast China with high latitude; as a result, the snow in shadow areas is usually classified as liquid water for its low near-infrared reflectance in the current SNOMAP algorithm. Then, another improvement is introducing the land surface temperature, which is retrieved from the thermal infrared band to distinguish liquid water from snow in shadow areas. We applied this improved method to evaluate forest areas in the Daxinganling, Xiaoxinganling, and Changbai Mountain areas in different seasons. The total classification accuracy reached 97.5%, and the pixels that introduce omission error and commission error were mainly distributed in areas of dense forest shadows. This improved method retains the computational simplicity and effectiveness of the SNOMAP algorithm in nonforest areas and improves the underestimation of snow cover in forest regions and shadow areas.

Seasonal snow cover is closely related to regional climate and hydrological processes. In this study, Moderate Resolution Imaging Spectroradiometer (MODIS) daily snow cover products from 2001 to 2018 were applied to analyze the snow cover variation in northern Xinjiang, China. As cloud obscuration causes significant spatiotemporal discontinuities in the binary snow cover extent (SCE), we propose a conditional probability interpolation method based on a space-time cube (STCPI) to remove clouds completely after combining Terra and Aqua data. First, the conditional probability that the central pixel and every neighboring pixel in a space-time cube of 5 × 5 × 5 with the same snow condition is counted. Then the snow probability of the cloud pixels reclassified as snow is calculated based on the space-time cube. Finally, the snow condition of the cloud pixels can be recovered by snow probability. The validation experiments with the cloud assumption indicate that STCPI can remove clouds completely and achieve an overall accuracy of 97.44% under different cloud fractions. The generated daily cloud-free MODIS SCE products have a high agreement with the Landsat–8 OLI image, for which the overall accuracy is 90.34%. The snow cover variation in northern Xinjiang, China, from 2001 to 2018 was investigated based on the snow cover area (SCA) and snow cover days (SCD). The results show that the interannual change of SCA gradually decreases as the elevation increases, and the SCD and elevation have a positive correlation. Furthermore, the interannual SCD variation shows that the area of increase is higher than that of decrease during the 18 years.

It is promising to improve Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover mapping by training effective machine learning models. However, considering the strong spatiotemporal heterogeneity of snow cover, the representativeness of training samples becomes a critical issue to ensure the model effectiveness over large areas and long terms. To deal with this issue, we propose the $m$ -day dynamic training strategy. The core of this strategy is to divide a long-term snow cover mapping task into multiple short-term tasks with consecutive $m$ days ( $m$ is greater or equal to the number of days that higher resolution remote sensing images used as equivalent ground references cover the study area once). It can ensure the spatial representativeness of training samples by utilizing all available equivalent ground references data, and reduce the issue caused by snow cover changing over time. We apply this strategy to random forest (RF) for both binary snow cover mapping (BSC) and fractional snow cover mapping (FSC) using Landsat-8 data as equivalent ground reference and validate its effect on the Tibetan Plateau. Results show that this strategy achieves higher accuracies than other training strategies, with $F1$ of 0.9596 for BSC mapping, $R$ and mean square error (MSE) of 0.9037 and 0.0260 for FSC mapping, respectively. In addition, spatiotemporal analysis of results further demonstrates that this strategy holds advantages for snow cover mapping in areas of complex terrain and in periods when snow cover changes rapidly. In conclusion, this strategy can effectively ensure the representativeness of training samples generated from Landsat-8 data and thus improve MODIS snow cover mapping through machine learning.

Reliable cloud masks in Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover products have a high potential to improve the retrieval of snow properties. However, cloud–snow confusion is a popular problem in MODIS snow cover products, especially in boreal forest areas. A large amount of forest snow is misclassified as clouds because of the low normalized difference snow index (NDSI), and excessive cloud masks limit the application of snow products. In addition, ice clouds are easily misclassified as snow due to their similar spectral characteristics, which leads to snow commission errors. In this paper, we quantitatively evaluated the cloud–snow confusion in Northeast China and found that snow-covered forests and transition zones from snow-covered to snow-free areas are prone to being misclassified as clouds, while clouds are less likely to be misclassified as snow. A temporal-sequence cloud–snow-distinguishing algorithm based on the high-frequency observation characteristics of the Himawarri-8 geostationary meteorological satellite is proposed. In the temporal-sequence images acquired from that satellite, the NDSI variance in cloud pixels should be greater than that of snow because clouds vary over time, while snow is relatively stable. In the MODIS snow cover products, the cloud pixels with NDSI variance lower than a threshold are identified as cloud-free areas and attributed their raw NDSI value, while the snow pixels with NDSI variance greater than the threshold are marked as clouds. We applied this method to MOD10A1 C6 in Northeast China. The results showed that the excessive cloud masks were greatly eliminated, and the new cloud mask was in good agreement with the real cloud distribution. At the same time, some possible ice clouds which had been misclassified as snow for their spectral characteristics similar to those of snow were identified correctly.