Drought forecasts using a satellite-derived drought indicator through machine learning approach

Sumin Park¹, Daehyeon Han¹, Jungho Im^1*

1) Ulsan National Institute of Science and Technology, Ulsan, South Korea

in progress

Abstract

A drought occurs when water deficiency resulting from recording below-average precipitation is prolonged. The drought can prolonged continue for weeks, months or even years, which can have an influence on ecosystems. In order to reduce damage induced by droughts, the drought monitoring and forecasting are necessary. In this study, drought forecasts were conducted using Scaled Drought Condition Index (SDCI) generated from Moderate Resolution Imaging Spectroradiometer (MODIS) and Tropical Rainfall Measuring Mission (TRMM) products over the Korean Peninsula. Long Short Term Memory networks (LSTM) that is one of the Recurrent Neural Networks (RNN) were applied to forecast drought using SDCI considering different duration of precipitation.

Current state

Now we’re building ConvLSTM model to deal with the spatio-temporal forecasting model.