GBRT-Based Estimation of Terrestrial Latent Heat Flux in the Haihe River Basin from Satellite and Reanalysis Datasets

Abstract

An accurate and spatially continuous estimation of terrestrial latent heat flux (LE) is fundamental and crucial for the rational utilization of water resources in the Haihe River Basin (HRB). However, the sparsity of flux observation sites hinders the accurate characterization of spatiotemporal LE patterns over the HRB. In this study, we estimated the daily LE across the HRB using the gradient boosting regression tree (GBRT) from global land surface satellite NDVI data, reanalysis data and eddy covariance data. Compared with the random forests (RF) and extra tree regressor (ETR) methods, the GBRT obtains the best results, with R2 = 0.86 and root mean square error (RMSE = 18.1 W/m2. Then, we applied the GBRT algorithm to map the average annual terrestrial LE of the HRB from 2016 to 2018 with a spatial resolution of 0.05°. When compared with the Global Land Surface Satellite (GLASS) and Moderate Resolution Imaging Spectroradiometer (MODIS) LE products, the difference between the terrestrial LE estimated by the GBRT algorithm and the GLASS and MODIS products was less than 20 W/m2 in most areas; thus, the GBRT algorithm was reliable and reasonable for estimating the long-term LE estimation over the HRB.

Keywords

• latent heat flux
• gradient boosting regression tree
• machine learning method
• eddy covariance
• Haihe River Basin