Evaluating an NDVI-Driven, Locally Calibrated FAO-56 Evapotranspiration Model Against Global ET Products and In-Situ Measurements in Semi-Arid Agriculture

Abstract

Accurate estimation of evapotranspiration (ET) is fundamental for optimizing irrigation management in semi-arid regions, where water scarcity imposes severe constraints on agricultural productivity. The FAO-56 dual crop coefficient methodology provides a standardized framework for ET estimation; however, its fixed parameterization often fails to represent the spatial and temporal variability characteristic of heterogeneous cropping systems. To address these limitations, this study applies the Satellite Monitoring of Irrigation (SAMIR) model, a spatially distributed approach derived from the FAO-56 formulation that dynamically estimates basal crop coefficients (Kcb) from NDVI and explicitly accounts for vertical soil water redistribution. A data-fusion scheme combining Landsat and MODIS imagery was employed to produce daily NDVI maps at 30 m resolution, enabling high-resolution monitoring across an entire agricultural plain. Model performance was assessed by comparing ET estimates from a calibrated SAMIR configuration, the standard FAO-56 formulation, and three global satellite-based products (PML v2, WaPOR, and SSEBop) against in situ flux measurements at three contrasting sites within the Tensift Basin, Morocco: a drip-irrigated olive orchard (R3), a heterogeneous semi-arid landscape monitored by a large-aperture scintillometer (TAH-LAS), and a dense drip-irrigated wheat field (CHI-EC1). The calibrated SAMIR model consistently outperformed all other approaches, achieving monthly R² values of 0.50, 0.28, and 0.58 with corresponding RMSE of 0.85, 0.85, and 1.03 mm d⁻¹ at R3, TAH-LAS, and CHI-EC1, respectively. While the uncalibrated FAO-56 and PML v2 products exhibited moderate accuracy under certain conditions, WaPOR and SSEBop showed larger errors and lower correlations, including negative R² values and substantial PBIAS in sparse canopy environments. These findings demonstrate that spatially explicit, NDVI-driven modeling incorporating soil water dynamics and local calibration substantially improves ET estimation in semi-arid agricultural systems relative to both traditional FAO-56 approaches and existing global ET datasets.