Historical and future Palmer Drought Severity Index with improved hydrological modeling

Abstract

With the ongoing climate warming, changes in drought and the adverse effects on water resources, food production and ecosystem functioning have been key research topics of ever-increasing interest. The Palmer Drought Severity Index (PDSI) is among the most widely used indicators for drought monitoring and research. However, the two-layer bucket water balance model embedded in the original PDSI model has been criticized for being over-simplified to accurately quantify the surface water balance and therefore raising uncertainties in the subsequent PDSI estimates (PDSIoriginal). Here we improve the water balance calculations in the PDSI model by using direct hydrological outputs from physically-based, more sophisticated global hydrological models (GHMs) participated in the Inter-Sectoral Impact Model Inter-Comparison Project (ISIMIP). Validation results show that the estimated runoff (Q) and evapotranspiration (ET) from ISIMIP GHMs perform much better than those from the original PDSI two-layer bucket model in capturing the long-term trend and monthly variabilities of Q and ET, especially in cold regions and relatively dry areas, using observed Q (at 2191 catchments) and an independent satellite-based ET product (the Global Land Evaporation Amsterdam Model, GLEAM; over the entire terrestrial environment) as the reference. In addition, the new PDSI estimates with improved hydrological modeling (PDSIISIMIP) exhibit a significantly stronger correlation with observed Q than PDSIoriginal in nearly all studied catchments, suggesting that PDSIISIMIP is superior to PDSIoriginal in capturing hydrological droughts. We further compare the long-term PDSI trends and changes in drought using PDSIoriginal and PDSIISIMIP under both historical climate (1900–2005) and future climate change scenarios (2006–2099). We find that PDSIoriginal and the PDSIoriginal-identified land areas under drought generally show a larger trend than those based on PDSIISIMIP. For future climate change scenarios, the PDSIoriginal-projected increasing trend of land proportion under drought is about two times larger than that assessed with PDSIISIMIP, implying that PDSIoriginal may largely overestimate future drought increases, as commonly done in existing studies. In this light, our approach of directly using hydrological outputs from physically-based, more sophisticated GHMs provide an effective, yet relatively simple approach to reduce uncertainties in PDSI estimates thereby achieving a better prediction of drought changes under warming.