Aguilera, Hector ， Laouchez, Marcel ， Kohfahl, Claus
2025-11-18 HYDROLOGICAL SCIENCES JOURNAL 2025 70(卷), 15(期), (2799-2813页)
This study investigates vapour flow control in dune sediments under dry, bare soil conditions using a high-precision weighing lysimeter located in southwest Spain. Over a 5-year period, the study applies machine learning models to predict four key target variables: outward vapour flow, inward vapour flow, net evaporation, and the log ratio of net evaporation to potential evaporation. Data were preprocessed using the adaptive window and adaptive threshold filter to remove noise. Recursive feature elimination was employed to select the most relevant predictor variables. The results demonstrate that machine learning models significantly enhance the accuracy of predictions for these variables compared to traditional linear regression models. The study provides valuable insights into the dynamics of vapour flow in semi-arid regions, emphasizing the importance of accurately predicting both evaporation and condensation processes to improve water resource management and understand soil-atmosphere interactions in such environments.