Assessing the impact of land cover map resolution and geolocation accuracy on evapotranspiration simulations by a land surface model

The land surface models used in numerical weather forecasts and hydrological applications rely on the accuracy of land cover maps available from satellite remote sensing to simulate the energy and water balance at the surface of the Earth. While the impact of classification accuracy on land surface simulations has already been reported, little attention has been paid on the consequences of land cover map uncertainty driven by geolocation accuracy. This impact on the estimated evapotranspiration (ET) from the land surface model H-TESSEL at spatial resolutions ranging from 1 to 30 km is evaluated here, making use of land cover maps at two different initial spatial resolutions, 300 and 1200 m, derived from the GlobCover global product. Geolocation uncertainty affects the land cover maps aggregated at different resolution (from 1.2 to 30 km). However, the effect decreases towards the coarsest resolutions. In addition, aggregated land cover maps are less affected by geolocation errors when the finest original resolution (300 m) is used. The maximum possible effect on ET is quantified over a heterogeneous/transition area in Europe. The result shows an impact up to 10% at 1.2-km resolution to less than 1% at 10-km resolution, at daily timescale, stressing the importance of such issues for kilometre scale applications of land surface models. The use of the highest initial land cover resolution (300 m) reduces by a factor 3 the impact of geolocation on estimated ET at 1.2- to 3-km scale. This study, therefore, stresses on the importance of a careful choice of the land cover map before carrying on land surface model simulations at the kilometre scale.