Identification of Flash Drought Events using Machine Learning Techniques

Date
Nov 6, 2021 2:00 PM — 3:00 PM
Location
Virtual: Google Meet
Speaker
Stuart Edris

Abstract

Drought is a climatological dry extreme that can have a number of impacts on water resources and agricultural output. These impacts can be exasperated further with flash droughts, which develop more rapidly (~1 month). To mitigate impacts, many studies have focused on being able to identify and predict droughts. In recent decades, machine learning (ML) has proved as a useful tool for drought identification and prediction, and ML applications using long-term metrics have found that random forests (RFs), support vector machines (SVMs), and artificial neural networks (ANNs) have performed best. However, recent studies have demonstrated that long-term drought metrics, such as the Palmer drought severity index (PDSI) and standardized precipitation index (SPI) cannot adequately identify the rapidly developing conditions that characterize flash drought. Thus, studies focused on flash drought have yielded a number of indices that accurately represent rapidly evolving conditions using precipitation, soil moisture, evapotranspiration (ET), and potential ET (PET). Furthermore, while much work has been conducted to investigate and identify flash drought events, limited work has been completed using ML techniques. Thus, this study aims to use the 42-year North American Regional Reanalysis (NARR) dataset to identify flash drought using ML techniques. In particular, this study aims to determine whether the previous methods (e.g., RFs, SVMs, and ANNs) retain their performance during rapidly changing environmental conditions on shorter timescales, and how best to represent flash drought in ML algorithms.

More

Presentation