A global multitask deep learning soil moisture model for disaster management
Abstract
Soil moisture plays an important role in linking surface water and groundwater on Earth. It is essential for forecasting agricultural drought, floods, and pests. However, soil moisture monitoring is sparsely distributed globally, especially in Africa and Asia. We trained a multitask deep learning model based on satellite grid data and in-situ networks. The model's performance was evaluated based on temporal, spatial cross-validation, and continental cross-validation. The model showed very high performance and achieved a global median correlation of 0.84 and 0.79 in temporal and spatial holdout tests, respectively. The model also provides daily soil moisture datasets at 9 km resolution (5 cm depth) from 2002 to 2021 on a global scale. It is beneficial in agricultural disaster management as this model is being integrated into an application for agricultural assistance in Africa.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H12N0850L