Accurately partitioning evapotranspiration (ET) of cropland into productive plant transpiration (T) and non-productive soil evaporation (E) is important for improving crop water use efficiency. Many methods, including machine learning methods, have been developed for ET partitioning. However, the applicability of machine learning models in cropland ET partitioning with diverse crop rotations is not clear. In this study, machine learning models are used to predict E, and T is obtained by calculating the difference between ET and E, leading to the derivation of the ratio of transpiration to evapotranspiration (T/ET). We evaluated six machine learning models (i.e., artificial neural networks (ANN), extremely randomized trees (ExtraTrees), gradient boosting decision tree (GBDT), light gradient boosting machine (LightGBM), random forest (RF), and extreme gradient boosting (XGBoost)) on partitioning ET at 16 cropland flux sites during the period from 2000 to 2020. The evaluation results showed that the XGBoost model had the best performance (R = 0.88, RMSE = 6.87 W/m², NSE = 0.77, and MAE = 3.41 W/m²) when considering the meteorological data, ecosystem sensible heat flux, ecosystem respiration, soil water content, and remote sensing vegetation indices as input variables. Due to the unavailability of observed E or T data at the 16 cropland sites, we used three other widely used ET partitioning methods to indirectly validate the accuracy of our ET partitioning results based on XGBoost. The results showed that our T estimation results were highly consistent with their T estimation results (R = 0.83–0.91). Moreover, based on the XGBoost model and the three other ET partitioning methods, we estimated the ratio of transpiration to evapotranspiration (T/ET) for different crops. On average, maize had the highest T/ET of 0.619 ± 0.119, followed by soybean (0.618 ± 0.085), winter wheat (0.614 ± 0.08), and sugar beet (0.611 ± 0.065). Lower T/ET was found for paddy rice (0.505 ± 0.055), winter barley (0.590 ± 0.058), potato (0.540 ± 0.088), and rapeseed (0.522 ± 0.107). These results suggest the machine learning models are easy and applicable for cropland T/ET estimation with different crop rotations and reveal obvious differences in water use among different crops, which is crucial for the sustainability of water resources and improvements in cropland water use efficiency.

Partitioning evapotranspiration (ET) into vegetation transpiration (T) and soil evaporation (E) is challenging, but it is key to improving the understanding of plant water use and changes in terrestrial ecosystems. Considering that the transpiration of vegetation at night is minimal and can be negligible, we established a machine learning model (i.e., extreme gradient boosting algorithm (XGBoost)) for soil evaporation estimation based on night-time evapotranspiration observation data from eddy covariance towers, remote sensing data, and meteorological reanalysis data. Daytime T was consequently calculated as the difference between the total evapotranspiration and predicted daytime soil evaporation. The soil evaporation estimation model was validated based on the remaining night-time ET data (i.e., model test dataset), the non-growing season ET data of the natural ecosystem, and ET data during the fallow periods of croplands. The validation results showed that XGBoost had a better performance in E estimation, with the average overall accuracy of NSE 0.657, R 0.806, and RMSE 11.344 W/m². The average annual T/ET of the examined ten ecosystems was 0.50 ± 0.08, with the highest value in deciduous broadleaf forests (0.68 ± 0.11), followed by mixed forests (0.61 ± 0.04), and the lowest in croplands (0.40 ± 0.08). We further examined the impact of the leaf area index (LAI) and vapor pressure deficit (VPD) on the variation in T/ET. Overall, at the interannual scale, LAI contributed 28% to the T/ET variation, while VPD had a small (5%) influence. On a seasonal scale, LAI also exerted a stronger impact (1~90%) on T/ET compared to VPD (1~77%). Our study suggests that the XGBoost machine learning model has good performance in ET partitioning, and this method is mainly data-driven without prior knowledge, which may provide a simple and valuable method in global ET partitioning and T/ET estimation.