Downscaling of coarse-grained geomorphological data

One of the basic limitations to the use of geomorphological maps is their coarse resolution relative to the needs of pure and applied geomorphological research. In response to this, attempts have been made to 'downscale' geomorphological information to finer spatial resolutions. However, the potential of statistical downscaling in geomorphology has been insufficiently examined. We downscaled four different periglacial features (wind deflation, palsa mire, earth hummock and sorted solifluction sheet) from a 100 ha grid to a I ha grid resolution utilizing two different techniques: point sampling (PSA) and direct (DA) approaches. We assessed the predictive accuracy of the models with the area under the curve (AUC) of a receiver operating characteristic plot using independent evaluation data. The PSA technique yielded encouraging results with a mean accuracy of 0-81, whereas the performance of DA was poorer. The predictive performance of the palsa mire and solifluction sheet models was excellent (AUC values from 0.89 to 0.96), whereas the AUC values of deflation and earth hummock models were lower (AUC = 0.57-0.81). The application of a point sampling approach as used here provides an efficient method to translate geomorphological information to finer resolution. However, further testing of the downscaling approaches is required before they can be applied to real-world situations. Copyright (c) 2007 John Wiley & Sons, Ltd.