Hydrologic model assessment from automated spatial pattern comparison techniques

S. Wealands,R. Grayson,J. Walker

Abstract:Spatial patterns of hydrological response are observed from field studies and remote sensing, and predicted by distributed hydrologic models. However, techniques for using spatial patterns to assess model performance are presently very simple. Rather than make assessments based on the spatial patterns, spatially explicit predictions are generally assessed using point observations or integrated responses. The problem of comparing two spatial patterns is not unique to hydrology, and has been researched in many other disciplines (e.g. pattern recognition, image processing). Hydrologists currently rely on either simple quantitative comparisons or human perception to assess the spatial similarity of predicted and observed patterns. While visual methods work well at local scales and with simple patterns, larger scale and more complex patterns demand alternative techniques. Automation of new quantitative techniques should prove applicable to hydrological applications. This paper discusses various methods currently used for comparing spatial patterns in hydrology. A review of promising techniques from other disciplines with potential application to hydrologically relevant spatial patterns is then presented. Some preliminary assessment of these techniques is undertaken and future directions for this research are outlined.