New Deep-Learning Tool to Diagnose and Predict Flash Droughts

In recent decades, rapid-onset "flash droughts" (FD) have increasingly been recognized as a significant climate hazard. These rapid onset events can cause significant damage, inflicting substantial crop losses, elevating fire risk, and creating other economic damages. In recent years, research efforts have focused on understanding the mechanisms behind flash droughts, documenting their historical occurrences, and developing practical tools for diagnosis and prediction. Leveraging a newly developed soil moisture volatility index (SMVI), we have recently published a 30-year, high-resolution global FD (GFD) dataset. This dataset enables the identification and assessment of the severity of FD events globally. In our ongoing research, we utilize advanced machine learning (ML) algorithms, specifically clustering and classification techniques, to categorize the severity of GFD events. Furthermore, we are developing a sophisticated deep-learning model that integrates the GFD dataset with various hydrometeorological and geographical parameters, such as precipitation and land cover. This model aims to enhance our ability to diagnose and predict the severity and category of FD events. By providing insights into potential flash drought occurrences, this model will serve as a valuable tool for decision-making processes related to predicting, preventing, or mitigating extreme events associated with flash droughts, such as wildfires.