Actual evapotranspiration estimation over the Tuojiang River Basin based on a hybrid CNN-RF model

Yao Li,Wensheng Wang,Guoqing Wang,Qiaofeng Tan

DOI: https://doi.org/10.1016/j.jhydrol.2022.127788

IF: 6.4

2022-07-01

Journal of Hydrology

Abstract:Actual evapotranspiration (AET) is a crucial variable linking water cycle, energy balance, and carbon cycle. Accurate regional AET estimation is a prerequisite for investigating its spatio-temporal characteristics and is critical to climate change assessment and hydrological cycle. Eddy covariance (EC) measurements are usually regarded as the ground truth to verify the reliability of other AET estimation models. However, EC flux towers are usually distributed sparsely and cannot fully reflect the actual spatial features of AET. This study proposes a hybrid machine learning model, called the hybrid Convolutional Neural Network and Random Forests (CNN-RF) model, and applies it to the regional AET estimation over the Tuojiang River Basin. The results demonstrate the excellent performance of the CNN-RF model at the site level and the superiority of the CNN-RF model over hybrid Convolutional Neural Network and Support Vector Machine (CNN-SVM), CNN, SVM, and RF models. The AET modeled by the CNN-RF model shows the highest correlation (correlation coefficient: 0.97) with the target value. Moreover, the mean absolute error (MAE) and the root mean square error (RMSE) for the CNN-RF-based modeling are the lowest, only 4.02 mm and 6.29 mm, respectively. Additionally, the CNN-RF-estimated AET shows good agreement with the AET simulated by the SWAT over the Tuojiang River Basin, further illustrating that the CNN-RF model based on remotely sensed variables has a strong generalization ability in space. The AET shows a high correlation with the air temperature and the precipitation; the correlation coefficients are 0.925 and 0.79, respectively. The evolution characteristics of these two meteorological variables govern the overall rise in annual AET during 1981–2020 (1.59 mm/year) over Tuojiang River Basin. This result may yield some insights into the interaction between various meteorological factors in the context of global warming and provide evidence for the impact of climate change on water and energy cycles.

geosciences, multidisciplinary,water resources,engineering, civil