Aiming at the problems of unstable accuracy and spatiotemporal discontinuity of existing snow water equivalent remote sensing products, a set of nested snow water equivalent product fusion algorithms based on deep learning theory was constructed, and a high-quality spatiotemporal sequence continuous Northern Hemisphere snow water equivalent data product was formed based on existing snow water equivalent products.
This dataset is based on the fusion of existing snow water equivalent data, including AMSR-E/AMSR2 SWE, GLDAS SWE, GlobSnow SWE, AMSR-E SWE, and GLDAS SWE.
The first step is to construct a snow water equivalent training dataset, which includes the snow water equivalent data used in this study AMSR-E/AMSR2 SWE、ERA-Interim SWE、MERRA-2 SWE、GLDAS SWE、GlobSnow SWE、ERA5_Land SWE。 In order to provide higher spatial resolution auxiliary information, MODIS snow cover area data with a resolution of 500 meters, geographic data (latitude and longitude), terrain data (altitude, slope, aspect), etc. are used as auxiliary training data, and ground observation snow water equivalent data is used as reference truth. On the basis of the training dataset, a snow water equivalent regression calculation model based on ridge regression model combined with LSTM is constructed. The ridge regression deep learning algorithm is used for the first learning, and the learning results are divided into two parts: significant and insignificant accuracy improvement. For the parts where the accuracy improvement is not significant, the advantages of LSTM algorithm in temporal dependence are adopted for further learning. On this basis, obtain high-quality spatiotemporal continuous snow water equivalent data.
| # | number | name | type |
| 1 | 2022YFF0711702-05 | National key R & D plan |
This work is licensed under
CC BY 4.0 (Creative Commons Attribution 4.0 International License).
CSTR:11738.11.NCDC.NIEER.DB6494.2024
10.12072/ncdc.nieer.db6494.2024
