Using a Bayesian Network to Summarize Variability in Numerical Long-Term Simulations of a Meteorological–Marine System: Drift Climatology of Assumed Oil Spills in the North Sea

Abstract

Climate-related scientific analyses of meteorological–marine systems are often based on numerical long-term simulations at high spatial and temporal detail. Such comprehensive data sets require much resources and specific evaluation tools, which sometimes hampers their use within interdisciplinary projects. In the present study, we propose the use of a Bayesian network to represent simulated transports in the North Sea depending on variable external forcing in terms of conditional probabilities. Eliciting probability tables from multi-decadal numerical simulations ensures that all realistic weather and resulting sea state conditions are covered in agreement with the frequency of their occurrence. The probabilistic representation conveniently allows for conditioning numerical simulations on either external forcing (weather conditions) or observed transports. In the latter case, the Bayesian inversion formula becomes involved to transfer information in a direction opposite to causal dependencies encoded in the underlying mechanistic model. We show that simulated travel time distributions even allow for taking into account a substance’s specific half-life, although this was not an issue in the original passive tracer simulations.