Analyzing the uncertainty of potential evapotranspiration models in drought projections derived for a semi-arid watershed

Abstract

Potential evapotranspiration (PET) serves as a proxy for estimating actual evapotranspiration (AET) in hydrological model simulations and constitutes an input for drought analyses. Nonetheless, it is elusive how the inclusion of different PET models in a modeling chain, which encapsulates multiple general circulation models (GCMs) operating under varying emission scenarios, can affect drought projections. In this study, utilizing four GCMs, two representative concentration pathways (RCPs), and eleven widely used PET models, ensemble projections of the frequency of droughts in the Gordes, a semi-arid watershed located in Western Turkey, were derived for the period 2021–2050. The standardized precipitation evapotranspiration index (SPEI) was used to characterize meteorological drought, while the standardized runoff index (SRI) was preferred for inspecting hydrological drought. Using an analysis of variance decomposition, the contribution of each stage of the modeling chain to both meteorological and hydrological drought uncertainty was quantified. Results show that PET models expectedly produced large disparities in projected changes in the evapotranspiration regime. Even so, only the 25% uncertainty contribution of PET models to severe meteorological drought frequency can be deemed notable. Yet, their contribution to uncertainties in mild and moderate meteorological drought frequencies is rather marginal (≤ 5%) compared to what GCMs overwhelmingly do. It is also worth noting that SPEI and SRI respond differently to sources of uncertainty and that SRI suggests drought frequencies of significantly lower amounts compared to those of SPEI, possibly because the temperature dependence of AET, which SRI considers synthetically, is much less than that of PET.