A machine learning-based analysis of actual evaporation predictors across different land covers in high-elevation drylands
Loading...
Date
2025
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Accurate estimation of actual evaporation (E) is essential for sustainable ecosystem management in arid, highelevation regions, where water losses are dominated by evaporation processes. In such landscapes, characterized by endorheic basins and heterogeneous land covers including salt flats, vegetated surfaces, bare soils with shallow groundwater, and saline lagoons, E results from complex interactions among energy availability, water availability, and atmospheric dynamics. While recent research has highlighted machine learning (ML) algorithms as promising tools to predict E, the extent to which environmental predictors vary across land cover types remains underexplored. This study applies the Random Forest (RF) algorithm to identify the key predictors of E across evaporitic land covers in the Chilean Altiplano, using evaporation fluxes and environmental variables collected during multiple field campaigns. Predictor importance is assessed through the Shapley Additive Explanation (SHAP) method. Results reveal that the most significant predictors of E vary substantially with land cover: energy limitation governs salt flats and vegetated surfaces; bare soils are influenced by both energy and water availability; and mechanical turbulence dominates over open water. We further demonstrate that accurate E predictions can be achieved using a reduced set of variables, four for land surfaces and three for water bodies, supporting the development of cost-effective monitoring networks focused on key environmental predictors. These findings improve our understanding of land-atmosphere interactions in arid, high-elevation ecosystems and offer a scalable, data-driven approach to enhance water resource management in high-elevation drylands worldwide.
Description
Keywords
Evapotranspiration, Random forest, Shapley additive explanation (SHAP), Chilean Altiplano, Arid regions