Influence of modeling domain and meteorological forcing data on daily evapotranspiration estimates from a Shuttleworth-Wallace model using Sentinel-2 surface reflectance data

Sustainable use of available water resources in viticulture can be aided by frequent high-resolution information on vineyard water status. Recently, a new Shuttleworth-Wallace evapotranspiration (ET) model, which uses a contextual framework to determine dry and wet extremes from the Sentinel-2 surface reflectance data (SW-S2), showed promising results when tested over a GRAPEX (Grape Remote-sensing Atmospheric Profile and ET eXperiment) site in California. However, current knowledge on its applicability across the climate gradient in California and how the selections of modeling domain and meteorological data influence model outputs are limited. This study expands the evaluation of the SW-S2 model across multiple domains and meteorological inputs covering all three GRAPEX sites over the 2018-2020 growing seasons. In comparison with flux tower observations, the size of the modeling domain did not have a strong influence on model performance, although the model performed marginally better under a larger domain (yielding root mean square error within 1.03-1.11 mm d(-1) and mean biases within 2%). The source and quality of meteorological forcing data, in particular vapor pressure deficit (VPD) and wind speed (u), were found to have a strong influence on model output as indicated by the poor performance of the model with less accurate regional and coarse-scale gridded meteorological inputs. Results suggest that simple regression for local bias correction of VPD and u significantly improved model performance. Overall, this study supports future research aiming to merge outputs from more frequent spectral and less frequent thermal-based ET models and reduce latency in ET monitoring of California vineyards.

Automated summary

This study evaluated the applicability of the SW-S2 model in different climate gradients in California and assessed the impact of modeling domain and meteorological inputs on model outputs. The results showed that the size of the modeling domain had minimal influence on model performance, while the source and quality of meteorological data had a significant impact on model outputs. The study also suggested local bias correction as a method to improve model performance.