Hydrological flash drought forecasting using meteorological flash drought indices and machine learning approaches - A case study in the Mississippi River Basin
Abstract
Flash droughts occur where high temperatures drastically increase the rate of evapotranspiration and regional rainfall is decreased, which reduces soil moisture and available surface water resources. Understanding and predicting flash droughts is critical for mitigating problems associated with reduced streamflow. This study uses meteorological flash drought indices and Long Short-Term Memory networks (LSTM) to predict 5-day (1 pentad), 10-day (2 pentad), and 15-day (3 pentad) hydrological flash drought indices. Two forecasting frameworks were proposed - individual and generalized LSTM-based forecasting. The LSTM was developed for each catchment in the first approach and a conventional LSTM was developed in the second approach. We examine the proposed frameworks over 197 catchments within the Mississippi River Basin from 1980-2014. The results of this study leverage the capabilities of machine learning for hydrological applications to predict flash drought conditions and better understand the propagation of meteorological droughts to hydrological drought events. Thus, this study will better inform decision makers and improve resilience of the Mississippi River Basin.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H46A..07B