Towards a probabilistic assessment of process-based, morphodynamic models

Abstract

Recent advances in the development of numerical, process-based models have led to remarkable performance in reproducing measured decadal morphodynamic developments. The advantage of this type of models is that they have a detailed output allowing for a close analysis of relevant processes. Drawback is that the output is associated with a high level of presumed uncertainty, because of the large number of processes involved and the high quality level of input data required.This study aims to explore possibilities to assess uncertainty levels associated with process-based morphodynamic modeling. In a probabilistic approach we consider the outcome of an ensemble of runs including variations of model input parameters and forcing schematizations. We propose to evaluate model performance by both a skill criterion (How well does the model reproduce observed patterns?), a confidence criterion (How sensitive are model results to uncertain input?) as well as a combination of these criteria. This methodology provides an objective assessment of the performance of process-based morphodynamic models. In addition, it can determine which input parameters cause largest uncertainty in the model outcome.The San Pablo Bay case study shows that 60% of the modeled volume meets the skill and confidence criteria for the depositional period (1856–1887) and 46% for the erosional period (1951–1983). Approximately 50% of the volume allocation meets the confidence criterion for a 30year morphodynamic forecast (1983–2013). Model results are sensitive to model input variations only to a limited extend. We attribute this to the high impact of the San Pablo Bay plan form and bathymetry. The forecast shows continuous erosion of the channel and on the northern shoals and a continuous accretion of the channel slopes, albeit more concentrated in the western part of the channel than in preceding decades.