Comparing evapotranspiration estimations using crop model-data fusion and satellite data-based models with lysimetric observations: Implications for irrigation scheduling

Abstract

Irrigation scheduling relies on accurately estimating actual evapotranspiration (ET_a). However, achieving this goal remains challenging, with current trends attempting to integrate sensor data into biophysically-sound models. In this study, we used ET_a models that integrate satellite-derived data. Model outputs were compared with weighing lysimeter measurements collected at Bushland, Texas, focusing on the uncertainties associated with model estimations and lysimetric observations of ET_a. The data were for maize (Zea mays L) during the seasons 2013, 2016, and 2018, irrigated using a linear-move system applying 100 % and 75 % of ET_a, and a subsurface drip irrigation system applying 100 % ET_a. The models were CropSyst-W, a crop model integrating the normalized difference vegetation index to derive green canopy cover, and two remote sensing data-based energy balance models: EEFlux and OpenET. The average Willmott index of agreement (d, 0 = no agreement, 1 = perfect agreement) of 3 years and four lysimeters were 0.93 and 0.77 for CropSyst-W and EEFlux, respectively. OpenET estimates were only available in 2016 and 2018, with an average d of 0.89. The average normalized root mean square deviation was 0.31 and 0.47 for CropSyst-W and EEFlux and 0.32 for OpenET. Uncertainty factors affecting modeled and observed ET_a highlight the difficulty in defining model accuracy for adaptive irrigation scheduling, which is also affected by the spatial variability of ET_a and irrigation uniformity. Research on pathways for model improvement should continue, searching for practical solutions that integrate models and sensors and account for the limitations discussed here.