Abstract
Evapotranspiration (ET) estimation models can be broadly classified as statistical or physical process based models. However, assuming the limitation of individual approaches, the integration of these two approaches has become a challenging task for ET simulation under varying surface and climatic conditions. To address this issue, a revised Penman-Monteith (PM) formula that uses a non-linear exponential Clausius-Clapeyron relationship was proposed in this study. The improved PM formula was further coupled into the loss function of the deep learning (DL) model, and subsequently, a hybrid DL model was formulated. The hybrid DL model with improved physical conceptualization considered the constraints of surface energy balance and turbulent diffusion processes in the ET simulation. The performance of the hybrid DL model was verified at 212 flux sites from the FLUXNET that contain ten types of underlying surfaces across the globe. The results revealed that as compared to the original DL model, the hybrid DL model improved the predictive capability of ET. The average root-mean-square-error (RMSE) and mean absolute percentage difference (MAPD) reduced by 12.1 W/m2 and 5.7%, respectively for latent heat flux (LE) simulation. Furthermore, the hybrid DL model also performed better than the original DL model in predicting the extreme events (such as ET under drought and heatwave conditions) which justifying its improved generalization capability. Sensitivity analysis outcomes showed that the vegetation parameters highest influence for ET simulations at the 212 flux sites, followed by soil parameters and meteorological parameters. The hybrid DL model was further applied to map the inter-seasonal distribution of global ET across twelve months of the year 2015 with five global ET products as the benchmark. Certainly, this research achieved the seamless integration of machine learning-based ET model and physical mechanism-based ET model and provided a new dimension for ET simulation. The hybrid DL model could be adopted to generate continuous ET datasets across regional and global scales.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.