Evaluation of three satellite-based latent heat flux algorithms over forest ecosystems using eddy covariance data

Abstract

We have evaluated the performance of three satellite-based latent heat flux (LE) algorithms over forest ecosystems using observed data from 40 flux towers distributed across the world on all continents. These are the revised remote sensing-based Penman-Monteith LE (RRS-PM) algorithm, the modified satellite-based Priestley-Taylor LE (MS-PT) algorithm, and the semi-empirical Penman LE (UMD-SEMI) algorithm. Sensitivity analysis illustrates that both energy and vegetation terms has the highest sensitivity compared with other input variables. The validation results show that three algorithms demonstrate substantial differences in algorithm performance for estimating daily LE variations among five forest ecosystem biomes. Based on the average Nash-Sutcliffe efficiency and root-mean-squared error (RMSE), the MS-PT algorithm has high performance over both deciduous broadleaf forest (DBF) (0.81, 25.4 W/m(2)) and mixed forest (MF) (0.62, 25.3 W/m(2)) sites, the RRS-PM algorithm has high performance over evergreen broadleaf forest (EBF) (0.4, 28.1 W/m(2)) sites, and the UMD-SEMI algorithm has high performance over both deciduous needleleaf forest (DNF) (0.78, 17.1 W/m(2)) and evergreen needleleaf forest (ENF) (0.51, 28.1 W/m(2)) sites. Perhaps the lower uncertainties in the required forcing data for the MS-PT algorithm, the complicated algorithm structure for the RRS-PM algorithm, and the calibrated coefficients of the UMD-SEMI algorithm based on ground-measured data may explain these differences.

Spatial and decadal variations in satellite-based terrestrial evapotranspiration and drought over Inner Mongolia Autonomous Region of China during 1982–2009

“…However, the data-based ET algorithms require a lot of ground-observations data to train the model and there are limited training data available (Yang et al 2006;Jung et al 2010Jung et al , 2011. In contrast, the process-based ET algorithms require fewer training data but they have low performance in the grassland ecosystems in this region due to their complex parameterization schemes (Yao et al 2013(Yao et al , 2015aErshadi et al 2014).…”