Abstract

Efficient water resource management in paddy fields requires accurate estimation of actual evapotranspiration (ETa) at the small-scale farm level. In this study, two approaches were evaluated for estimating rice ETa in the paddy fields of Rasht using satellite images (Sentinel-2, Landsat 7 and 8): 1) ETa estimation using Leaf Area Index (LAI) values, and 2) ETa estimation through the crop coefficient (Kc) derived from LAI. To overcome the spatial resolution limitations of Sentinel-2 imagery, the S2DR3 deep learning model was utilized to enhance the spatial resolution from 10 m to 1 m. This model successfully reconstructed the NDVI and SAVI indices with coefficients of determination (R²) of 0.9979 and 0.9864, and normalized root mean square errors (NRMSE) of 2.8% and 9.1%, respectively. Evaluations indicated that in estimating LAI, the NDVI outperformed the SAVI, achieving an R² of 0.81, RMSE of 0.51, and MAE of 0.38, compared to RMSE and MAE values of 0.61 and 0.45 for SAVI. Finally, comparing the model outputs with lysimeter-measured data revealed that the Kc-based approach using NDVI exhibited higher agreement with field observations, achieving an average R² of 0.85. The error for this approach included an RMSE ranging from 0.40 to 0.58 mm/day, a MAE between 0.36 and 0.52 mm/day, and an NRMSE of 7% to 11%.