Controlling factors of evapotranspiration predictability under diverse climates with the effects of water storage change in the Budyko framework

The Budyko models (BM) have been extended in previous studies by incorporating water storage change (ΔS) (subtracting ΔS from precipitation) to estimate evapotranspiration (ET) under non-steady state conditions at scales finer than the climatological mean scale. However, a systematic assessment of the interannual ET predictability of the extended BM is still lacking, hence its validity and controlling factors of improvement (over the original BM) under globally diverse climates is not yet well understood. Based on a long-term (1984–2008) gridded water budget data set, we present a comparative analysis of annual ET predictability between the original BM (ET₁) and the extended BM considering ΔS (ET₂) in 32 global river basins to explore the sensitivity of climate factors and catchment hydrologic responses in determining ET predictability. Results show that the difference between ET₁ and ET₂ increases linearly with ΔS, with ET₂ < ET₁ (ET₂ > ET₁) when ΔS > 0 (ΔS < 0). When both ET₁ and ET₂ overestimate (underestimate) observed ET, the error in ET₂ is smaller than ET₁ when ΔS > 0 (ΔS < 0) for all 32 basins considered. When the error signs of ET₂ and ET₁ differ, however, the difference in the absolute magnitude of ET₂ and ET₁ errors (RE_diff) under extremely humid climates is determined by the difference between potential ET and ET, leading to comparable accuracy between ET₂ and ET₁. In contrast, under extremely arid climates, RE_diff is controlled by the combined influences of ΔS and R, resulting in more accurate ET₂ than ET₁ under the condition of the in-phase, positive-correlated relationship between ΔS and R.