The AVHRR Surface Reflectance Version 4 (AVHRR SR V4) is a Climate Data Record (CDR) product released by the National Oceanic and Atmospheric Administration (NOAA). This product is generated based on data from the Advanced Very High Resolution Radiometer (AVHRR) sensors onboard NOAA polar-orbiting meteorological satellites. Covering the period from 1981 to 2019, it has a daily temporal resolution and a spatial resolution of 5 km. After undergoing radiometric calibration, atmospheric correction, and cloud detection processes, the dataset serves as a fundamental source for generating long-term snow cover products.Landsat-5 TM data are obtained from the Thematic Mapper sensor onboard the Landsat-5 satellite, which was jointly operated by the United States Geological Survey (USGS) and the National Aeronautics and Space Administration (NASA). The dataset has a 16-day temporal resolution, 30 m spatial resolution, and covers the time period from 1984 to 2013.ERA5-Land land surface temperature (LST) data are sourced from the ERA5-Land reanalysis dataset produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) (Muñoz Sabater, 2019). This dataset provides continuous global coverage from 1981 to the present, with a spatial resolution of 0.1°.The digital elevation model (DEM) data are derived from the Shuttle Radar Topography Mission (SRTM) led by NASA, with a spatial resolution of 90 m. All the above datasets were accessed and utilized via the Google Earth Engine (GEE) cloud computing platform.Ground-based snow depth observations were provided by the China Meteorological Administration (CMA) through its national meteorological station network. A total of 191 meteorological stations were selected, with data spanning from 1981 to 2019, for validating the snow cover product.The China Long-Term Daily Snow Depth Dataset, developed by Che et al. (2008) and Dai et al. (2015), is based on observations from multiple passive microwave satellite sensors. The dataset has been processed using inter-sensor calibration techniques. Covering the period from 1979 to 2020, it provides daily snow depth information over China at a spatial resolution of 0.25°. This dataset, available from the National Tibetan Plateau Data Center (https://doi.org/10.11888/Geogra.tpdc.270194), was used in this study as a supplementary strategy for gap-filling.

 (1) Valid observations were selected using the quality control flags provided by AVHRR SR V4. Only pixels with valid values across all bands were retained for snow cover extraction, while invalid pixels were marked as missing values.(2) Based on the scheme proposed by Hori et al. (2017), the brightness temperature difference between BT37 and BT11 was optimized using Landsat-5 TM data as ground truth.(3) Using Landsat-5 TM data as reference, multispectral features of AVHRR over snow-covered and snow-free areas (including SR1, BT11, SR3/SR2, NDVI, and NDSI) were extracted. A three-level decision tree algorithm was then applied to determine the optimal combination of thresholds.(4) The China Long-Term Daily Snow Depth Dataset, land surface temperature (LST), and digital elevation model (DEM) were resampled or aggregated to 5 km resolution to match that of the AVHRR data. For missing areas in the preliminary records caused by clouds or invalid observations, a series of gap-filling techniques were applied, including hidden Markov random field (HMRF)-based interpolation and snow-depth interpolation. Postprocessing was conducted using land surface temperature and DEM data to eliminate falsely identified snow-covered pixels.

 (1) The product was evaluated using a confusion matrix and four accuracy metrics, including Overall Accuracy (OA), Producer’s Accuracy (PA), User’s Accuracy (UA), and the Kappa coefficient. The OA ranged between 80% and 90%, PA and UA ranged from 70% to 90%, and the Kappa coefficient ranged from 0.61 to 0.8.(2) Validation based on 38 years of CMA ground-based snow depth measurements from 191 meteorological stations showed that most stations had high OA values, generally between 80% and 90%. However, the PA, UA, and Kappa values were relatively low.(3) Further validation using nine Landsat-5 snow cover extent maps revealed an OA as high as 87.3%. The higher UA and lower PA indicate a slight tendency of the product to underestimate snow cover extent. The Kappa value was 0.695, which is close to that from ground-based validation (0.717).Therefore, from both the "point" perspective (ground observations) and the "area" perspective (Landsat-5 SCE maps), the accuracy of this product is reliable. Overall, this product shows great potential for applications in climate and related studies.