Quantifying thresholds for advancing impact-based drought assessment using classification and regression tree (CART) models

Abstract

Drought events can significantly impact water resources, agriculture, and socioeconomic sectors. Droughts are often quantified using numerous indices, which are widely used for forecasting, risk assessments, and spatiotemporal analysis. Quantifying the appropriate linkage between drought indices and their impact on hydrological and agricultural indicators will improve hazard communication with stakeholders and further advance impact-based forecasting tools. This study aims to quantify the thresholds associated with drought indices that can capture the intricate connection between droughts and impact-specific indicators. We investigated the performance of several drought indices, such as Palmer-based (e.g., PDSI, PMDI, and PHDI) and multiscale drought indices (e.g., SPI and SPEI) with the impact-specific indicators of hydrological (e.g., streamflow and reservoir level) and agricultural (soil moisture and crop yield) indicators. The 'threshold' values associated with drought indices with impact-specific indicators are quantified using the classification and regression tree (CART) algorithm. Our results suggest that the decision tree approach is suitable for identifying critical thresholds of drought indices associated with a range of hydrological and agricultural indicators relevant to the stakeholder's application to different sectors. The drought characteristics derived from selected indices broadly vary during extreme events. The results further indicate that the threshold associated with each drought index and the impact-specific indicator varies drastically within the same climate division. We argue that quantifying the drought indices thresholds can provide valuable information for impact-based monitoring and forecasting.