Zhang et al. (2026) Daily seamless 30-m fractional snow cover mapping via an adaptive Time-Series approach

Identification

• Journal: International Journal of Applied Earth Observation and Geoinformation
• Year: 2026
• Date: 2026-01-06
• Authors: Cheng Zhang, Lingmei Jiang, Jinmei Pan, Jianwei Yang, Jian Wang, Zongyi Jin
• DOI: 10.1016/j.jag.2025.105068

Research Groups

• State Key Laboratory of Remote Sensing and Digital Earth, Faculty of Geographical Science, Beijing Normal University, Beijing 100875, China
• National Space Science Center, Chinese Academy of Sciences, Beijing 100190, China
• Research Institute of Forest Resource Information Techniques, Chinese Academy of Forestry, Beijing 100091, China
• State Key Laboratory of Earth Surface Processes and Disaster Risk Reduction, Faculty of Geographical Science, Beijing Normal University, Beijing 100875, China

Short Summary

This study introduces the Time-series-based Adaptive snow-Fraction Fusion (TAFF) framework to generate seamless daily 30-meter fractional snow cover (FSC) maps, effectively addressing data gaps caused by clouds and infrequent satellite revisits. TAFF demonstrates robust performance over the Qinghai-Tibet Plateau, achieving high spatial accuracy (R² = 0.76, RMSE = 19.58 %) and temporal fidelity (binary classification accuracy = 0.91).

Objective

• To develop a robust, computationally efficient, and automated framework (TAFF) for generating seamless daily 30-meter fractional snow cover (FSC) maps over large scales, specifically by synergistically leveraging multi-resolution time series to overcome data gaps from frequent cloud cover and limited satellite revisit cycles.

Study Configuration

• Spatial Scale: Qinghai-Tibet Plateau, with a target resolution of 30 meters. Processing was conducted on 1° × 1.2° tiles (approximately 13,700 square kilometers).
• Temporal Scale: Daily fractional snow cover mapping. Validation period: January 1, 2022, to March 5, 2023. Adaptive temporal windows (ranging from 4 to 27 days, depending on data availability and month) are used in the methodology.

Methodology and Data

• Models used:
  • Time-series-based Adaptive snow-Fraction Fusion (TAFF) framework:
    • TAFF-T (Temporal) Stage: Dual-path algorithm involving a snow stability assessment, followed by either a time-weighted average (for stable snow) or a pixel-level Ordinary Least Squares (OLS) regression (for rapidly-varying snow).
    • TAFF-D (Downscaled) Stage: Two-stage Random Forest (RF) model (classification and regression) for residual gap-filling and artifact correction.
    • Final TAFF Stage: Hierarchical fusion with direct high-resolution observations and final spatial/temporal gap-filling.
  • Fractional Snow Cover (FSC) retrieval: Multiple-Endmember Spectral Mixture Analysis with Automated Endmember Generation (MESMA-AGE).
  • Cloud masking: CFMask (Landsat 8/9), s2cloudless (Sentinel-2), and internal Quality Assurance (QA) bands (MODIS, VIIRS).
  • Spectral harmonization: Linear regression-based.
  • Benchmark algorithms for comparison: STARFM, ESTARFM, FSDAF, Fit-FC.
• Data sources:
  • Satellite Imagery (Surface Reflectance):
    • Fine-resolution (30 meters): Landsat 8 OLI, Landsat 9 OLI-2 (Collection 2 Level-2), Sentinel-2 MSI (Level-2A).
    • Coarse-resolution (500 meters): Terra and Aqua MODIS (Collection 6.1), S-NPP VIIRS (Collection 1).
  • In-situ Observations: Daily snow depth measurements from 46 meteorological stations operated by the China Meteorological Administration (CMA).
  • Ancillary Data: 10-meter ESA WorldCover product (for land cover stratification), static topographic parameters (elevation, slope, aspect), geospatial/temporal context (latitude, longitude, Day of Year (DOY)).
  • Processing Platform: Google Earth Engine (GEE).

Main Results

• The TAFF framework consistently achieved 100 % spatial coverage for daily 30-meter FSC maps, significantly outperforming benchmark spatiotemporal fusion algorithms (mean coverage: ESTARFM 57 %, STARFM 65 %, Fit-FC 86 %, FSDAF 93 %).
• The final 30-meter TAFF product demonstrated high spatial accuracy when validated against 215 independent Landsat 8 images, yielding a coefficient of determination (R²) of 0.76 and a root mean square error (RMSE) of 19.58 %.
• Validation against 46 in-situ snow depth stations showed excellent temporal fidelity for binary snow/no-snow classification, with an overall accuracy of 0.91, a specificity of 0.97, and a precision of 0.71.
• The framework is computationally efficient, processing a 1° × 1.2° tile (approximately 13,700 square kilometers) in about 147 seconds on a standard personal computer.
• Performance varied by land cover type, with strong results in Bare/sparse vegetation (R² = 0.71), Grassland (R² = 0.73), and Moss and lichen (R² = 0.74), but lower accuracy in Tree cover (R² = 0.33) and Permanent Snow and ice (R² = 0.42) due to inherent challenges in optical snow retrieval in these environments.

Contributions

• Introduces the Time-series-based Adaptive snow-Fraction Fusion (TAFF) framework, a novel and robust approach for generating seamless daily 30-meter fractional snow cover (FSC) maps.
• Proposes an innovative dual-path fusion strategy that adaptively responds to the physical state of the snowpack (stable vs. rapidly-varying), enhancing accuracy for dynamic snow conditions.
• Integrates gap-filling and downscaling into a unified, automated framework, eliminating the need for manual selection of cloud-free image pairs and making it highly resilient to data gaps from clouds and infrequent satellite revisits.
• Demonstrates superior operational robustness and accuracy compared to existing spatiotemporal fusion algorithms, particularly in producing spatially complete products under challenging data availability.
• Provides a computationally efficient and scalable solution suitable for large-scale, operational monitoring of dynamic snow cover.
• Establishes an extensible framework that can integrate future data sources (e.g., geostationary, microwave sensors) to further enhance high-resolution monitoring of rapidly changing land surface variables.

Funding

• National Natural Science Foundation of China (No. 42171317, 42090014, and 42571399)
• National Key Research and Development Program of China (No. 2022YFF0801302)

Citation

@article{Zhang2026Daily,
  author = {Zhang, Cheng and Jiang, Lingmei and Pan, Jinmei and Yang, Jianwei and Wang, Jian and Jin, Zongyi},
  title = {Daily seamless 30-m fractional snow cover mapping via an adaptive Time-Series approach},
  journal = {International Journal of Applied Earth Observation and Geoinformation},
  year = {2026},
  doi = {10.1016/j.jag.2025.105068},
  url = {https://doi.org/10.1016/j.jag.2025.105068}
}