An artificial neural network approach to estimate evapotranspiration from remote sensing and AmeriFlux data

Abstract

Listen

Translate article

A simple and accurate method to estimate evapotranspiration (ET) is essential for dynamic monitoring of the Earth system at a large scale. In this paper, we developed an artificial neural network (ANN) model forced by remote sensing and AmeriFlux data to estimate ET. First, the ANN was trained with ET measurements made at 13 AmeriFlux sites and land surface products derived from satellite remotely sensed data (normalized difference vegetation index, land surface temperature and surface net radiation) for the period 2002–2006. ET estimated with the ANN was then validated by ET observed at five AmeriFlux sites during the same period. The validation sites covered five different vegetation types and were not involved in the ANN training. The coefficient of determination (R2) value for comparison between estimated and measured ET was 0.77, the root-mean-square error was 0.62 mm/d, and the mean residual was − 0.28. The simple model developed in this paper captured the seasonal and interannual variation features of ET on the whole. However, the accuracy of estimated ET depended on the vegetation types, among which estimated ET showed the best result for deciduous broadleaf forest compared to the other four vegetation types.