Tree-based explainable clustering for drought severity predictions in United States

Abstract

Climate change drives the environment to more extreme weather events. Increased air, land surface and canopy surface temperatures affect the industry of agriculture in different ways. Significant crop damages and losses are emerging and spreading throughout different regions, accompanied by water scarcity and imposed restrictions on farmers' water usage. The Eastern Mediterranean, Middle East, and North Africa (EMMENA) region is one of the most affected areas globally. The United States (US) developed a system for monitoring droughts in different counties and classifying them into six categories (i.e., no drought, abnormally dry, moderate drought, severe drought, extreme drought, and exceptional drought) based on the assigned drought score. To predict drought scores, Artificial Intelligence (AI) methodologies are applied to a dataset that combines meteorological variables from the NASA Langley Research Center with drought scores from the US drought monitor system. The main objective of this work is to propose a novel explainable AI technique based on unsupervised learning for drought severity predictions and raise the awareness for drought events in the wider EMMENA region.