Automated deep-learning-based soil moisture planning and forecast system for planning against natural disasters

Abstract:

Locust is a common natural disaster that causes economic and ecological damages worldwide. Locust swarms are the worst in Ethiopia, Somalia, and in 2020 and locusts are threatening the food security of 13 million people. Soil moisture can effectively determine where the locust swarms may lay eggs, and predicts the breed and spread of locusts, which reduces the threat of food security. First, we used the long short-term memory (LSTM) method to build a deep learning forecast model of soil moisture time series and used the Soil Moisture Active Passive (SMAP) satellite level-3 9km product, the Global Precipitation Mission (GPM), and the Global Land Data Assimilation System (GLDAS) as input data to train the model. We used the Global Forecast System (GFS) product to predict soil moisture in the next 7 days and got the correlation between soil moisture and the breed and spread of locust swarms. Then, in scenarios with different prediction days (1-7 days), we got the relationship between the prediction days and the model performance. Then, we compared the performance of the model with different satellite products combinations (e.g. GPM, GLDAS, GFS, ERA-Interm) under the same prediction days(n). Finally, the Google Earth Engine application was developed, which is automatically downloaded, processed, simulated, and updated to present near-real-time prediction of the soil moisture. On a related note, we also discuss the opportunity to use a similar system for landslides.