Developing the Drought Risk Overview Product (DROP): Improving Flash Drought Forecasts and Early Warning Using Machine Learning and Extreme Value Theory Techniques

Drought is responsible for $327.7B in economic loss since 1980 in the United States and typically precedes, cooccurs, or initiates other hazards like wildfires or prolonged periods of intense heat. The compounding or cascading of these hazards routinely threatens the wellbeing and stability of communities. The economic impact from drought can be especially harsh when the drought begins rapidly in what is known as a flash drought. In this Phase I effort, CFD Research seeks to address the dynamics of flash droughts and improve the prediction and monitoring of their occurrence by developing an AI Convolutional Neural Network (CNN) and Long ShortTerm Memory (LSTM) applying deep learning and pattern recognition. An existing spatiotemporal tracking algorithm will be used to automatically categorize flash droughts from climate model and observational data into severity classes based on an adaptation of the Heat Severity and Coverage Index (HSCI) and Drought Severity and Coverage Index. Together, this will inform the development of DROP, the Drought Risk Overview Product. DROP is an application designed to add address the needs of decision makers by providing higher resolution spatial and temporal drought risk intelligence to better prepare for and monitor the development of flash drought.