On Lessons Drawn: Prediction In Geomorphology

Abstract:

The 2000 AGU session on Prediction in Geomorphology was seen by some as a showdown among modeling approaches. For example, reductionist models with complete fidelity to Newtonian mechanics versus constructionist models developed from dynamics observed at the scale of emergent features. This is a thin basis for a robust discussion of geomorphic modeling. No model approach can apply to all time and space scales. Each approach has undeniable features and difficult challenges. For example, application of reductionist models to an apparently tractable river reach or hillslope can be dominated by emergent features such as bed forms, channel width, and valley slope that are difficult to predict from small-scale physics. Models defined at the scale of emergent features need mechanisms linking cause and effect that can be rigorously tested. Rather than a battle for best or true, the conversation in the AGU Monograph shifted toward more universal questions: Why do we model and to what purpose? How do we formulate and test models? How do we evaluate whether a model has predictive capability or any utility at all? Common themes that emerged included the need to unambiguously link cause and effect, the importance of rigorous and flexible model tests, and the utility of models in sharpening thought, organizing understanding, and expressing concepts in unambiguous and testable ways.

The purpose of this paper is to reiterate lessons learned from PIG and to evaluate their fate over the past two decades. A central question is whether the geomorphic processes of erosion, transport, and deposition can be defined at a scale sufficiently large to model landscapes and sufficiently small that they can be tested and parameterized using independent field or laboratory measurements. Widespread and routine application of numerical models from small to very large scales of space and time compels the question of whether the models are rigorously tested, particularly in the specification and prediction of emergent features and in the spatial and temporal averaging necessary to upscale models based on first physical principles.