An integrated approach to obtain high-precision regional root water uptake maps

Abstract

Root water uptake (RWU) is exceptionally difficult to determine in situ due to its unique nature of occurring underground. Accurately characterizing the spatial–temporal characteristics of RWU at a regional scale remains a significant challenge, and high-precision regional-scale RWU maps have not yet been reported. In this study, we develop a novel approach that integrates the Google Earth Engine (GEE) platform, mathematical models, and machine learning algorithms to generate high-precision regional maps of maize RWU in China’s primary maize production area. The GEE platform was used to extract essential parameters such as soil properties, leaf area index, and growth period. These variables, along with meteorological data from 122 agricultural meteorological stations, were used to calculate RWU at a point scale using the Hydrus-1D and AquaCrop models. Finally, we utilized Hydrus-calculated point-scale RWU fluxes, along with 31 covariates extracted from the GEE platform, in conjunction with the CatBoost algorithm to develop a machine-learning model for estimating regional-scale RWU. The accuracy of the developed machine-learning model was validated using statistical measures, such asa mean absolute error (0.0299 cm d−1), a root mean squared error (0.0398 cm d−1), and the correlation coefficient (0.9178). The model was then applied to generate high-precision (500 m × 500 m) regional-scale RWU maps from 2015 to 2019. Additionally, our analysis identified several parameters, including TempDew (i.e., 2-m dewpoint temperature), SurfaceSM (i.e., top layer soil moisture), TempMax (i.e., max 2-m air temperature), and Latitude, as key factors influencing regional-scale RWU. This approach provides a robust strategy for accurately generating regional-scale RWU maps, even with limited data. Furthermore, it can be extended to obtain other regional indices, offering valuable insights for various real-world applications.