Choice of data scale: predicting resolution error in a regional evapotranspiration model

Abstract

Much attention is being given to the physics of surface heterogeneity in interscale, land-atmosphere process models. A model's physical sensitivity to heterogeneity, however, should not be developed without also simultaneously considering its spatial resolution sensitivity. The purpose of this paper is to determine the degree to which spatial resolution error in a regional evapotranspiration model is a function of the patterns present in the input parameter data layers. Predicted evapotranspiration (ET) maps were generated for a heterogeneous area in northeastern Connecticut at eight spatial resolutions. Input parameter maps included surface cover, albedo, canopy height, canopy resistance, red and near infra-red (NIR) reflectance and normalized difference vegetation index. Error in the mean and absolute point error for all maps were calculated. Absolute error was predicted using the spatial autocorrelation images of the original resolution maps. Several averaging methods differing with respect to the position of the spatial averaging operation in the sequence of procedures were intercompared. Results showed that error in the mean increases erratically as resolution coarsens for non-linear parameter combinations. Absolute error increases asymptotically and this shape is predicted well for input parameter maps. The shape of the ET error curve is not well predicted, but is quite similar to that of the input parameters. These results suggest that spatial patterns in input parameter maps are important in determining critical averaging scales for a regional process model.