Using remote sensed actual evaporation to improve hydrological models

Abstract

In hydrological models evaporation is often still quite uncertain. Potential evaporation is used as input, but the modelling of plant stresses is not always accurate enough to describe the behaviour in reality. In situ measured actual evaporation data is rare, and doing measurements is time consuming and expensive. With the advance of satellite technology, remote sensing products modelling actual evaporation have been developed. To see if models can be easily improved using this actual evaporation data, it will be directly imposed. As remote sensing data product, the LSA SAF actual evaporation product will be used. As input it uses most importantly Meteosat-10 data. As other sources it has ERA-interim and ECOCLIMAP. Accuracy is generally high when comparing it to in situ measurements (R2=0.90 when comparing it to eddy covariance at Cabauw), although a correction factor might be necessary for some locations. Three models were tested, WALRUS, SIMGRO and FLEX. Two sites in the Netherlands were studied (Cabauw and the Hupsel Brook) and two catchments in Spain (Ubierna and Ulzama). WALRUS was applied to the Cabauw Polder and the Hupsel Brook, SIMGRO was applied to Cabauw, and FLEX was used in Spain. SIMGRO did not perform well, and modelled water stress during summers at Cabauw, while in situ measurements showed that water stress was not an issue. The performance was too low for LSA SAF evaporation to make a difference. WALRUS performed well at Cabauw, although there was no difference between the performance of the model when using actual evaporation instead of Makkink evaporation (NS=0.693 and NS=0.673 respectively), as the catchment does not suffer from water stress. In the Hupsel Brook water stress does occur, and the model performs slightly better when using actual evaporation instead of Makkink evaporation (NS=0.762 and NS=0.733 respectively). In Spain the uncertainty of the input data was high, and some corrections were necessary. For the Ubierna catchment, the performance was clearly better with the LSA SAF actual evaporation instead of the Makkink evaporation (log-NS=0.787 and log-NS=0.698). For the Ulzama catchment the model using LSA SAF actual evaporation also performed better than the Makkink evaporation (log-NS=0.769 and log-NS=0.708). These results show that the LSA SAF product can give a good representation of actual evaporation, and that directly using it in hydrological models can improve their performance if the vegetation in the catchment experiences water stress.