Evaluating Remote Sensing-Based Evapotranspiration Estimation Using the UAS-METRIC Model in the Hot and Humid Climate of Tennessee

Abstract

Highlights First comprehensive evaluation of UAS-METRIC for both rainfed and irrigated crops in a hot and humid condition. Absence of dry bare soil and persistently non-stressed canopy within the UAS flight mission limited reliable internal calibration of UAS-METRIC. UAS-METRIC showed poor ETa estimation, likely due to the lack of an appropriate hot anchor pixel. ABSTRACT. Accurate estimation of crop evapotranspiration (ET a ) using the Unmanned Aerial System (UAS)-based Mapping Evapotranspiration at High Resolution with Internalized Calibration (METRIC) model in Tennessee's hot and humid climate is essential for producing geospatial ET a maps that support precision irrigation management. The overall goal of this study was to evaluate the performance of the UAS-METRIC model for both rainfed and irrigated crops in the hot and humid climate of Tennessee and assess the effectiveness of its internal calibration of the model, particularly the selection of hot and cold anchor pixels. The research was conducted during the 2024 and 2025 growing seasons at two sites: the TSU Research Farm in Ashland City (winter canola) and a commercial farm in Franklin (soybean and corn) Tennessee. UAS flights capturing multispectral and thermal imagery at these sites were performed at altitudes of 50 and 100 m, to generate high-resolution ET a maps at approximately 4 cm pixel -1 and 7 cm pixel -1 resolution, respectively. Ground-truth ET a was obtained by closing the soil water balance method, using an LI-710 evapotranspiration sensor, and modelling ET a with the pyfao56 Python package. Results showed that the UAS-based METRIC model has poor performance for evapotranspiration estimation across all crop types. Model performance was weak when validated against soil water balance measurements (R 2 = 0.21, RMSE = 2.93 mm/day, NSE = 0.16) and improved moderately with LI-710 data (R 2 = 0.41, RMSE = 1.08 mm/day, NSE = 0.25), pyfao56 estimates (R 2 = 0.47, RMSE = 1.45 mm/day, NSE = -0.21), and MOISTURE-app (R 2 = 0.52, RMSE = 1.32 mm/day, NSE = -0.89). The primary source of error in the UAS-METRIC model was improper anchor pixel selection. Persistent soil wetness in Tennessee, driven by frequent rainfall during the growing season, prevented the presence of true hot pixels representing dry, bare agricultural soil. In addition, as in commercial irrigated farms, the presence of a fully non–water-stressed crop canopy is not guaranteed; therefore, the cold and hot pixels identified through internal model calibration do not necessarily represent a fully irrigated or non-stressed canopy and dry bare agricultural soil. These limitations resulted in exaggerated surface temperature gradients and led to overestimation of ET a . These findings underscore the need for standardized artificial reference surfaces to ensure reliable UAS-METRIC calibration in hot and humid agricultural environments. Keywords: Actual evapotranspiration map, Hot and cold anchor pixel, LI-710 sensor, MOISTURE-app, Precision irrigation, pyfao56, UAS-METRIC energy balance model.