Publications

Amani, S; Shafizadeh-Moghadam, H; Morid, S (2025). Integrating sentinel-2 and sentinel-3 for actual evapotranspiration estimation across diverse climate zones using the sen-ET plugin and machine learning models. EARTH SCIENCE INFORMATICS, 18(4), 338.

Abstract
This study employed random forest (RF) and support vector machine (SVM) algorithms to estimate actual evapotranspiration (ETa) using data from Sentinel-2 and Sentinel-3 satellites. The primary objective was to propose efficient methodologies that rely on minimal input features while identifying crucial variables that influence ETa across various climates. Initially, ETa was computed for five distinct climatic regions in Iran using the two-source energy balance model_Priestley Taylor (TSEB_PT), and these results were utilized to calibrate the machine learning models. The findings revealed that the RF algorithm demonstrated superior performance in dry, Mediterranean, and humid regions, whereas the SVM algorithm was more accurate in semi-arid and very humid climates. The RF model identified land surface temperature (LST) and leaf area index (LAI) as the most important predictors. In contrast, the SVM model highlighted the significance of the soil-adjusted vegetation index (SAVI) and LAI. Additional influential variables included photosynthetically active radiation (PAR), elevation, normalized difference vegetation index (NDVI), and blue and red-edge spectral bands. Overall, LST emerged as the most impactful variable across all studied climates. Due to their simpler computations, reduced data requirements, and faster processing times, machine learning methods offer a viable alternative for predicting ETa.

DOI:
10.1007/s12145-025-01786-0

ISSN:
1865-0481