We investigated the potential capability of the random forest (RF) machine learning (ML) model to estimate snow depth in this work. Four combinations composed of critical predictor variables were used to train the RF model. Then, we utilized three validation datasets from out-of-bag (OOB) samples, a temporal subset, and a spatiotemporal subset to verify the fitted RF algorithms. The results indicated the following: (1) the accuracy of the RF model is greatly influenced by geographic location, elevation, and land cover fractions; (2) however, the redundant predictor variables (if highly correlated) slightly affect the RF model; and (3) the fitted RF algorithms perform better on temporal than spatial scales, with unbiased root-mean-square errors (RMSEs) of ∼4.4 and ∼7.3 cm, respectively. Finally, we used the fitted RF2 algorithm to retrieve a consistent 32-year daily snow depth dataset from 1987 to 2018. This product was evaluated against the independent station observations during the period 1987–2018. The mean unbiased RMSE and bias were 7.1 and −0.05 cm, respectively, indicating better performance than that of the former snow depth dataset (8.4 and −1.20 cm) from the Environmental and Ecological Science Data Center for West China (WESTDC). Although the RF product was superior to the WESTDC dataset, it still underestimated deep snow cover (>20 cm), with biases of −10.4, −8.9, and −34.1 cm for northeast China (NEC), northern Xinjiang (XJ), and the Qinghai–Tibetan Plateau (QTP), respectively. Additionally, the long-term snow depth datasets (station observations, RF estimates, and WESTDC product) were analyzed in terms of temporal and spatial variations over China. On a temporal scale, the ground truth snow depth presented a significant increasing trend from 1987 to 2018, especially in NEC. However, the RF and WESTDC products displayed no significant changing trends except on the QTP. The WESTDC product presented a significant decreasing trend on the QTP, with a correlation coefficient of −0.55, whereas there were no significant trends for ground truth observations and the RF product. For the spatial characteristics, similar trend patterns were observed for RF and WESTDC products over China. These characteristics presented significant decreasing trends in most areas and a significant increasing trend in central NEC.

Accurate long-term daily cloud-gap-filled fractional snow cover products are essential for climate change and snow hydrological studies in the Asian Water Tower (AWT) region, but existing Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover products are not sufficient. In this study, the multiple-endmember spectral mixture analysis algorithm based on automatic endmember extraction (MESMA-AGE) and the multistep spatiotemporal interpolation algorithm (MSTI) are used to produce the MODIS daily cloud-gap-filled fractional snow cover product over the AWT region (AWT MODIS FSC). The AWT MODIS FSC products have a spatial resolution of 0.005° and span from 2000 to 2022. The 2745 scenes of Landsat-8 images are used for the areal-scale accuracy assessment. The fractional snow cover accuracy metrics, including the coefficient of determination (R²), root mean squared error (RMSE) and mean absolute error (MAE), are 0.80, 0.16 and 0.10, respectively. The binarized identification accuracy metrics, including overall accuracy (OA), producer's accuracy (PA) and user's accuracy (UA), are 95.17 %, 97.34 % and 97.59 %, respectively. Snow depth data observed at 175 meteorological stations are used to evaluate accuracy at the point scale, yielding the following accuracy metrics: an OA of 93.26 %, a PA of 84.41 %, a UA of 82.14 % and a Cohen kappa (CK) value of 0.79. Snow depth observations from meteorological stations are also used to assess the fractional snow cover resulting from different weather conditions, with an OA of 95.36 % (88.96 %), a PA of 87.75 % (82.26 %), a UA of 86.86 % (78.86 %) and a CK of 0.84 (0.72) under the MODIS clear-sky observations (spatiotemporal reconstruction based on the MSTI algorithm). The AWT MODIS FSC product can provide quantitative spatial distribution information on snowpacks for mountain hydrological models, land surface models and numerical weather prediction in the Asian Water Tower region. This dataset is freely available from the National Tibetan Plateau Data Center at https://doi.org/10.11888/Cryos.tpdc.272503 (Jiang et al., 2022) or from the Zenodo platform at https://doi.org/10.5281/zenodo.10005826 (Jiang et al., 2023a).

Launched on 15 November 2017, China’s FengYun-3D (FY-3D) has taken over prime operational weather service from the aging FengYun-3B (FY-3B). Rather than directly implementing an FY-3B operational snow depth retrieval algorithm on FY-3D, we investigated this and four other well-known snow depth algorithms with respect to regional uncertainties in China. Applicable to various passive microwave sensors, these four snow depth algorithms are the Environmental and Ecological Science Data Centre of Western China (WESTDC) algorithm, the Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E) algorithm, the Chang algorithm, and the Foster algorithm. Among these algorithms, validation results indicate that FY-3B and WESTDC perform better than the others. However, these two algorithms often result in considerable underestimation for deep snowpack (greater than 20 cm), while the other three persistently overestimate snow depth, probably because of their poor representation of snowpack characteristics in China. To overcome the retrieval errors that occur under deep snowpack conditions without sacrificing performance under relatively thin snowpack conditions, we developed an empirical snow depth retrieval algorithm suite for the FY-3D satellite. Independent evaluation using weather station observations in 2014 and 2015 demonstrates that the FY-3D snow depth algorithm’s root mean square error (RMSE) and bias are 6.6 cm and 0.2 cm, respectively, and it has advantages over other similar algorithms.

Snow depth estimation with passive microwave (PM) remote sensing is challenged by spatial variations in the Earth’s surface, e.g., snow metamorphism, land cover types, and topography. Thus, traditional static snow depth retrieval algorithms cannot capture snow thickness well. In this study, we present a new operational retrieval algorithm, hereafter referred to as the pixel-based method (0.25° × 0.25° grid-level), to provide more accurate and nearly real-time snow depth estimates. First, the reference snow depth was retrieved using a previously proposed model in which a microwave snow emission model was coupled with a machine learning (ML) approach. In this process, an effective grain size (effGS) value was optimized by utilizing the snow microwave emission model, and then the nonlinear relationship between snow depth and multiple predictive variables, e.g., effGS, longitude, elevation, and brightness temperature (Tb) gradients, was established with the ML technique to retrieve reference snow depth data. To select a robust and well-performing ML approach, we compared the performance of widely used support vector regression (SVR), artificial neural network (ANN) and random forest (RF) algorithms over China. The results show that the three ML models performed similarly in snow depth estimation, which was attributed to the inclusion of effGS in the training samples. In this study, the RF model was used to retrieve the snow depth reference dataset due to its slightly stronger robustness according to our comparison of results. Second, the pixel-based algorithm was built based on the retrieved reference snow depth dataset and satellite Tb observations (18.7 GHz and 36.5 GHz) from Advanced Microwave Scanning Radiometer 2 (AMSR2) during the 2012–2020 period. For the pixel-based algorithm, the fitting coefficients were achieved dynamically pixel by pixel, making it superior to the traditional static methods. Third, the built pixel-based algorithm was verified using ground-based observations and was compared to the AMSR2, GlobSnow-v3.0, and ERA5-land products during the 2012–2020 period. The pixel-based algorithm exhibited an overall unbiased root mean square error (unRMSE) and R² of 5.8 cm and 0.65, respectively, outperforming GlobSnow-v3.0, with unRMSE and R² values of 9.2 cm and 0.22, AMSR2, with unRMSE and R² values of 18.5 cm and 0.13, and ERA5-land, with unRMSE and R² values of 10.5 cm and 0.33, respectively. However, the pixel-based algorithm estimates were still challenged by the complex terrain, e.g., the unRMSE was up to 17.4 cm near the Tien Shan Mountains. The proposed pixel-based algorithm in this study is a simple and operational method that can retrieve accurate snow depths based solely on spaceborne PM data in comparatively flat areas.

Daily snow-covered area retrieval using the imagery in solar reflective bands often encounters extensive data gaps caused by cloud obscuration. With the inception of geostationary satellites carrying advanced multispectral imagers of high temporal resolution, such as Japan’s geostationary weather satellite Himawari–8, considerable progress can now be made towards spatially-complete estimation of daily snow-covered area. We developed a dynamic snow index (normalized difference snow index for vegetation-free background and normalized difference forest–snow index for vegetation background) fractional snow cover estimation method using Himawari–8 Advanced Himawari Imager (AHI) observations of the Tibetan Plateau. This method estimates fractional snow cover with the pixel-by-pixel linear relationship of snow index observations acquired under snow-free and snow-covered conditions. To achieve reliable snow-covered area mapping with minimal cloud contamination, the daily fractional snow cover can be represented as the composite of the high temporal resolution fractional snow cover estimates during daytime. The comparison against reference fractional snow cover data from Landsat–8 Operational Land Imager (OLI) showed that the root–mean–square error (RMSE) of the Himawari–8 AHI fractional snow cover ranged from 0.07 to 0.16, and that the coefficient of determination (R²) reached 0.81–0.96. Results from the 2015/2016 and 2016/2017 winters indicated that the daily composite of Himawari–8 observations obtained a 14% cloud percentage over the Tibetan Plateau, which was less than the cloud percentage (27%) from the combination of Moderate Resolution Imaging Spectroradiometer (MODIS) onboard Terra and Aqua.

Snow density varies spatially, temporally, and vertically within the snowpack and is the key to converting snow depth to snow water equivalent. While previous studies have demonstrated the feasibility of retrieving snow density using a multiple-angle L-band radiometer in theory and in ground-based radiometer experiments, this technique has not yet been applied to satellites. In this study, the snow density was retrieved using the Soil Moisture Ocean Salinity (SMOS) satellite radiometer observations at 43 stations in Quebec, Canada. We used a one-layer snow radiative transfer model and added a $\tau -\omega$ vegetation model over the snow to consider the forest influence. We developed an objective method to estimate the forest parameters ($\tau$, $\omega$) and soil roughness ($S_D$) from SMOS measurements during the snow-free period and applied them to estimate snow density. Prior knowledge of soil permittivity was used in the entire process, which was calculated from the Global Land Data Assimilation System (GLDAS) soil simulations using a frozen soil dielectric model. Results showed that the retrieved snow density had an overall root-mean-squared error (RMSE) of 83 kg/m³ for all stations, with a mean bias of 9.4 kg/m³. The RMSE can be further reduced if an artificial tuning of three predetermined parameters ($\tau$, $\omega$, and $S_D$) is allowed to reduce systematic biases at some stations. The remote sensing retrieved snow density outperforms the reanalysis snow density from GLDAS in terms of bias and temporal variation characteristics.