PARAMETER-INDUCED PREDICTIVE UNCERTAINTY IN PROCESS-BASED MODELING: APPLICATION OF MARKOV CHAIN MONTE CARLO

摘要

Predictions of process-based numerical models contain inherent uncertainties owing to model structural errors, such as poorly described or missing processes, to inaccuracies in the data, and to limited knowledge of the model’s free parameters. The aim of this paper is to show how parameter uncertainty in nearshore process modeling and associated uncertainty ranges in the model output can be estimated in an objective manner by means of Markov Chain Monte Carlo (MCMC). This work should be seen as a first step in quantifying (the relevant importance of) all uncertainty sources in process modeling, which may ultimately lead to improved process knowledge and system understanding by learning from and understanding model-data misfit. MCMC performs a thorough sampling of feasible parameter space to approximate the posterior distribution of the model’s parameters, given a specific model structure and data. The approach is illustrated with a one-dimensional, time-averaged alongshore current model based on the alongshore momentum equation and data collected at six instrumented positions across the inner subtidal bar at Egmond aan Zee (Netherlands).