Dynamic soil water stress function improves evapotranspiration estimation in areas with significant vegetation variability

Abstract

Accurate estimation of evapotranspiration (ET) is crucial for understanding water and carbon cycles, and water resource management. Direct ET measurements are expensive and technically challenging. A common indirect estimation method involves multiplying potential evapotranspiration (PET) by a soil water stress function (β). However, a soil water stress function with static parameters (β_S) was generally used in models, neglecting the variation in soil moisture availability under different atmospheric and vegetation conditions. Additionally, although various linear and nonlinear forms of β have been developed, comparisons of their performance in ET estimation using large observational datasets remain limited. In this study, we evaluate ET estimation using three widely used forms of β (linear, exponential and sigmoid) across 135 global eddy covariance sites, and propose a soil water stress function with dynamic parameters (β_D), which adjusts β based on variations in atmospheric demand (PET) and vegetation conditions (leaf area index, LAI). Subsequently, we compare the ET estimates from β_S and β_D with observed ET. Results show that the performance of ET estimation using the exponential function outperforms the linear and sigmoid functions, with mean NSE values of 0.38, 0.25, and 0.34, respectively. Furthermore, the β_D method (mean NSE = 0.58) significantly improves ET estimation accuracy compared to the β_S method (mean NSE = 0.33), particularly for vegetation types such as MF and DBF. Vegetation is the key determinant of the difference between β_D and β_S. Specifically, the β_S method tends to overestimate ET at low intensities and underestimate ET at high intensities, especially in regions with high vegetation variability. Our findings underscore the importance of incorporating dynamic characteristics of β, particularly in areas with significant vegetation changes.