Assessment of non-parametric method for evapotranspiration estimation across extreme conditions

Abstract

Terrestrial evapotranspiration (ET) estimation faces constraint due to complicate parameterization. The non-parametric (NP) method offers an efficient alternative, but its effectiveness across diverse environments, particularly under extreme conditions, is uncertain. This study leverages data from 131 global flux stations across various climate and landcover types to assess the NP method's accuracy, focusing on extreme scenarios spanning from dry to wet and from hot to cold. Our results indicate substantial variability, with absolute errors ranging from −41.7 to 26.8 W/m² and relative errors between 0.4 % and 98.1 %. The NP method tends to overestimate ET in hot, dry conditions, and underestimate in cold, wet conditions, with performance declining under extremes. Cold climates exhibit the largest biases, with particularly severe underestimation in extreme wet (−54.5 W/m²) and cold (−69.2 W/m²) conditions. Closed shrublands demonstrate the largest overestimation (73.2 W/m²) in extreme hot conditions. Across stations, three error patterns are observed: (i) overestimation in dry-hot regions caused by a high difference between equilibrium and actual ET, (ii) underestimation in humid regions resulting from a low difference, and (iii) underestimation in cold regions arising from large surface-air temperature gradients, causing an excessive integral term resulting in overcorrection. These errors could be amplified under extreme conditions, reducing model performance and highlighting the NP method's sensitivity to hydroclimatic extremes while offering insights for improving its accuracy and robustness.