Resource information
Land surface models describe the exchange of heat, moisture and momentum between the land surface and the atmosphere. These models can be solved regionally using remote sensing measurements as input. Input variables which can be derived from remote sensing measurements are surface albedo, surface temperature and vegetation cover. A land surface model using those land surface characteristics is presented i.e. the Surface Energy Balance Index (SEBI) model. This model uses the observed temperature difference between the land surface and atmosphere as an indicator for evapotranspiration.
Spatially distributed land surface model results can be used as a boundary condition for numerical weather predicton models. The results should therefore be aggregated from the remote sensing pixel scale to the atmospheric model scale. However aggregated values will differ when derived from remote sensing data with different resolutions. This difference, the error due to aggregation is caused by two different aspects: land surface heterogeneity and non-linearity of the land surface model. Two approaches are presented to quantify the error due to aggregation: the linearization approach, where the land surface model is approximated by a Taylor expansion and a geometrical approach where the range of valid results for the land surface model is derived using a convex hull.
To measure the heterogeneity of land surfaces, the concept of length scale is introduced. The wavelet transform is being used to derive the length scale of the land surface characteristics. The wavelet variance derived from the Fast Wavelet Transform using the Haar wavelet is a good indicator for the variability of land surface characteristics at different spatial scales. For three different data sets the length scale of land surface characteristics have been derived: Barrax, Spain, the Jornada Experimental Range, USA and the Central Part of the Netherlands.
The two approaches for quantifying the error due to aggregation have been verified using the three data sets. The results obtained by the linearization show that aggregation error can indeed be estimated. For the three test sites the large scale error did not exceed 10 %. However the results based on the convex hull analysis show that the large scale error due to aggregation can be much larger than observed for the three test cases. Therefore low resolution remote sensing data cannot be used a priori as input for land surface models.
