Several global multi-year ET datasets, e.g. satellite-derived products, observation-driven land surface model simulations, reanalysis data products, or atmospheric water balance estimates, are currently available. However, a major constraint of these datasets is the difficulty of their validation. An inter-comparison of differently derived ET datasets, which is conducted in the framework of the LandFlux-EVAL initiative (www.iac.ethz.ch/research/url/LandFlux-EVAL), can partly stand in for a validation with in-situ measurements.
In the presented study, we compare and evaluate four different groups of datasets:
1. Observation-based datasets, where remote sensing data or in-situ measurements are combined to derive ET estimates using simple algorithms
2. Land-surface model (LSM) output, driven with observational data
3. Reanalysis data
4. IPCC AR4 global climate model output
The inter-comparison aims at evaluating spatial differences and uncertainties of the analyzed ET data products. We present a comparison of the single datasets on river basin scale for different seasons. The consistency of the spatial patterns is analyzed with a cluster analysis, and, for selected datasets, the triple collocation error method.
We find a similar, but relatively high (up to around 20% of their mean) uncertainty in the observation-based datasets, the reanalyses and the IPCC model simulation output. The LSMs show smaller uncertainties, which could be due to similar forcing data used for several models within this dataset group. The importance of the forcing data for the LSM output was further investigated with a cluster analysis, which shows strong spatial similarities between models from the same projects (e.g. from the Global Soil Wetness Project GSWP), i.e. driven with the same forcing data.
Even though a global validation of ET datasets with direct observations is not possible, this presentation shows possible avenues to approach this goal and reveals remaining large uncertainties in the quantification of this variable.