Evaluating MODIS cloud-free snow cover datasets using massive spatial benchmark data in the Tibetan Plateau

Yang Gao,Xuetao Wang,Naixia Mou,Yufeng Dai,Tao Che,Tandong Yao

DOI: https://doi.org/10.1016/j.scitotenv.2024.175245

2024-11-01

Abstract:Accurate snow cover data is crucial for understanding climate change, managing water resources, and calibrating models. The MODIS (Moderate-resolution Imaging Spectroradiometer) and its cloud-free snow cover datasets are widely used, but they have not been systematically evaluated due to different benchmark data and evaluation parameters. Conventional methods using station observations as a ground truth suffer from underrepresentation and mismatches in temporal and spatial scales. This study established a scale-matched spatial benchmark dataset, compiling from 18,433 Landsat series and 11,172 Sentinel-2 images over two decades, totaling ∼1.86 billion samples and ∼320 million snow samples. We evaluated seven MODIS cloud-free snow cover datasets for seasons, elevation zones, land covers and subregions using this benchmark data. For the clear-sky part, NIEER_MODIS_SCE (MODIS snow cover extent product over China) performs best due to its use of optimal NDSI thresholds suitable for each land use type. This highlights the importance of regional customization in snow mapping algorithms, and it can be further improved in spring, forests and zone 1 by combining it with M*D10A1GL06. For the cloud removed part, one-step integrated spatiotemporal cloud removal datasets perform better than any other approach does. The second-best dataset is obtained from a simple but effective single temporal cloud removal method using nearby time information. For the whole dataset, the best NIEER_MODIS_SCE has an overall accuracy of 0.82 and snow retrieval accuracy of 84.56 %. It performs excellently in most settings but weakest in forests thus requiring more efficient strategies. This research provides new perspectives and methods for objectively assessing MODIS snow cover products and other relevant datasets. These methods can be readily extended to other regions and adapted to future satellite missions. And such findings may guide the selection of more valid snow cover data and the developing of even better snow detecting strategies.