A machine learning-based accuracy enhancement on EUMETSAT H-SAF H35 effective snow-covered area product


Kuter S., Bolat K., AKYÜREK S. Z.

Remote Sensing of Environment, cilt.272, 2022 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 272
  • Basım Tarihi: 2022
  • Doi Numarası: 10.1016/j.rse.2022.112947
  • Dergi Adı: Remote Sensing of Environment
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, PASCAL, Aerospace Database, Agricultural & Environmental Science Database, Applied Science & Technology Source, Aquatic Science & Fisheries Abstracts (ASFA), BIOSIS, Biotechnology Research Abstracts, CAB Abstracts, Communication Abstracts, Compendex, Computer & Applied Sciences, Environment Index, Geobase, INSPEC, Metadex, Pollution Abstracts, Veterinary Science Database, DIALNET, Civil Engineering Abstracts
  • Anahtar Kelimeler: Remote sensing of the cryosphere, Fractional snow cover mapping, Multivariate adaptive regression splines, Sentinel 2, ERA5-Land, MODIS, AVHRR, NORTHERN XINJIANG, CLOUD MASK, MODIS, VALIDATION, WATER, RETRIEVAL, CLIMATE, ALBEDO, DEPTH, MAPS
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

© 2022 Elsevier Inc.Snow is a major element of the cryosphere with significant impact on the Earth's water cycle and global energy budget. Acquiring consistent and long time series data on the spatial extent of snow cover doubtlessly plays a key role in our understanding and modeling of the current and future environmental dynamics. Remote sensing offers a powerful tool for continuous retrieval of snow cover information by utilizing snow's contrasting reflectance characteristics at optical wavelengths. The pre-operational H35 covers the Northern Hemisphere, and it is the successor of the operational Pan-European H12 daily fractional snow-covered area (fSCA) product at ~1 km. Both products are developed through the Satellite Application Facility on Support to Operational Hydrology and Water Management (H-SAF) project of EUMETSAT by exploiting AVHRR channels. This study is focused on developing an alternative fully data-driven H35 product with improved accuracy using a machine learning (ML)-based approach. Multivariate adaptive regression splines (MARS) algorithm is trained by using AVHRR reflectance data as well as the well-known snow and vegetation indices (i.e., NDSI and NDVI) to generate the new version of H35 fSCA product. The reference fSCA maps required for the training of MARS models are obtained from the higher resolution Sentinel 2 multispectral imagery. The MARS-based fSCA models are validated against an initial test dataset composed of 15 Sentinel 2 scenes over European Alps, Tatra Mountain Range, and Turkey. The final MARS-H35 product is then rigorously assessed over the whole Northern Hemisphere within a temporal domain spanning from Nov 2018 to Nov 2019. The quantitative testing process involves the use of reference data in both continuous and dichotomous scales: i) Sentinel 2 derived reference fSCA maps, ii) ERA5-Land snow depth data, iii) MODIS MOD10A1 NDSI snow cover data, and finally iv) in-situ snow depth data. Additionally, qualitative assessment is also performed by visually comparing MARS-H35/MODIS false-color and MARS-H35/Sentinel 2-derived reference fSCA image pairs over various geographic regions. The overall results indicate that: i) the proposed MARS-H35 fSCA product overperforms the original H35, and ii) it has higher capability in detecting the fine variations in the extent of snow cover, especially across the fringes of the slopes in complex mountainous terrains.