A Snow Depth Downscaling Algorithm Based on Deep Learning Fusion of Enhanced Passive Microwave and Cloud-Free Optical Remote Sensing Data in China
High spatial resolution snow depth (SD) is crucial for hydrological, ecological, and disaster research. However, passive microwave SD product (10/25 km) is increasingly insufficient to meet contemporary requirements due to its coarse spatial resolution, particularly in heterogeneous alpine areas. In...
Gespeichert in:
Veröffentlicht in: | Remote sensing (Basel, Switzerland) Switzerland), 2024-12, Vol.16 (24), p.4756 |
---|---|
Hauptverfasser: | , , , , , , , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | High spatial resolution snow depth (SD) is crucial for hydrological, ecological, and disaster research. However, passive microwave SD product (10/25 km) is increasingly insufficient to meet contemporary requirements due to its coarse spatial resolution, particularly in heterogeneous alpine areas. In this study, we develop a superior SD downscaling algorithm based on the FT-Transformer (Feature Tokenizer + Transformer) model, termed FTSD. This algorithm fuses the latest calibrated enhanced resolution brightness temperature (CETB) (3.125/6.25 km) with daily cloud-free optical snow data (500 m), including snow cover fraction (SCF) and snow cover days (SCD). Developed and evaluated using 42,692 ground measurements across China from 2000 to 2020, FTSD demonstrated notable improvements in accuracy and spatial resolution of SD retrieval. Specifically, the RMSE of temporal and spatiotemporal independent validation for FTSD is 7.64 cm and 9.74 cm, respectively, indicating reliable generalizability and stability. Compared with the long-term series of SD in China (25 km, RMSE = 10.77 cm), FTSD (500 m, RMSE = 7.67 cm) provides superior accuracy, especially improved by 48% for deep snow (> 40 cm). Moreover, with the higher spatial resolution, FTSD effectively captures the SD’s spatial heterogeneity in the mountainous regions of China. When compared with downscaling algorithms utilizing the raw TB data and the traditional random forest model, the CETB data and FT-Transformer model optimize the RMSE by 10.08% and 4.84%, respectively, which demonstrates the superiority of FTSD regarding data sources and regression methods. Collectively, these results demonstrate that the innovative FTSD algorithm exhibits reliable performance for SD downscaling and has the potential to provide a robust data foundation for meteorological and environmental research. |
---|---|
ISSN: | 2072-4292 2072-4292 |
DOI: | 10.3390/rs16244756 |