Comparison of ET models over different land cover

The objective of this work is to compare various ET models based on a standard dataset. We selected 9 ET models for comparison, including three surface energy balance residual models (SEBS, TSEB-P and TSEB-S), four Penman-Monteith models (PM-Mu, PM-Yuan, PM-Sun and PM-SW), one Priestly-Taylor model (PT-Fi) and one semi-empirical statiacital model (ST). ET is evaluated using surface climate data from ground measurements as input. Remote sensing data including Ts, LAI and NDVI products from MODIS are used. Estimated ET is validated against 40 Fluxnet measurement sites across North United states and Europe. The sites land cover types include grassland, cropland, evergreen needle leaf forest, evergreen broadleaf forest, deciduous broadleaf forest, mixed forest, shrub land and savannas. Results show that ST model had a balanced performance with relative good precision over all the land cover types. PM-Sun has high R2 and low RMSE and bias over all land cover types. However, it overestimated high value and underestimated low value, mainly due to the overestimation of soil evaporation and underestimation of plant transpiration. The energy budget series models including SEBS, TSEB-P and TSEB-S have a bad performs on the forest land cover. PM-Mu and PM-Yuan underestimated ET obviously, resulting from the underestimation of soil evaporation.