A global model to explain drought occurrence and damage

Droughts are among the costliest natural hazards, ranked third after storms and floods globally. Furthermore, they often cause enormous indirect impacts such as famines. Understanding the occurrence of economic drought impacts is a challenging task due to their slow onset, vast spatial extent and long duration. Besides meteorological conditions, drought occurrence strongly depends upon local human water management interventions such as irrigation and water withdrawal altering vulnerability. Additionally, identifying drought vulnerable assets and their temporal development presents a major challenge as they strongly depend on the regional socio-economic structure. In order to attribute historical drought damage and to project future drought risk, a deeper understanding of changes in drought exposure, vulnerability, and the damage-intensity relationship is required. Previous damage functions neglect that intense physical drought conditions do not always translate into a damage event. Therefore, we develop a two-step approach that i) estimates the likelihood of event occurrence from the physical conditions and ii) establishes a damage-intensity relationship. We test the explanatory power of common drought indicators such as the standardized precipitation-evapotransporation index (SPEI), soil moisture, and low river flow to reconstruct historical time series of drought damage reported by EM-DAT and NatCatSERVICE, globally. The drought indicators are derived from the Inter-Sectoral Impact Model Intercomparison Project round 3a and vary in their modeling complexity. While SPEI is based on mere climate reanalysis data, soil moisture is derived from global hydrological models and low river flow from their output coupled with the hydrodynamic model CaMa-Flood. We find that the suitability of drought indicators for damage reconstruction varies regionally. While low river flow may be applied in Europe for damage reconstruction, SPEI and soil moisture are more reliable predictors for most world regions. The explanatory power of the model shows strong regional variations, depending also on the quality of observational data. Observed damage can be well reproduced in regions such as Latin America and South East-Asia, but the model fails to reproduce damage time series in North Africa and Central Asia. We show that both modeling steps are necessary to reproduce observed drought damage and that the likelihood of event occurrence as well as the damage ratio increase under more intense physical drought conditions. Omitting the likelihood-intensity relationship may lead to an overestimation of historical drought damage, which is used as a reference in attribution and projection studies. As reproducing observed damage is indispensable for sound attribution studies, the two-step approach may allow us to better account for non-linear changes in drought impacts under climate change.