Ensemble modeling of transport and dispersion simulations guided by machine learning hypotheses generation

In this article an approach is presented where machine learning classifiers are used to drive an ensemble modeling method of multiple atmospheric transport and dispersion simulations. The goal is to achieve a higher spread of the results with a lower number of ensemble simulations. Symbolic machine learning algorithms are used to define choices for the variation of meteorological input data, model parameters, model physics, based on their combined effects on the final dispersion calculations (i.e., construction of ensembles). The methodology uses an iterative approach with the aim to identify ensemble members leading to a more balanced distribution of results. thodology is tested using real meteorological data from Istanbul, Turkey, simulating atmospheric releases along the Bosphorus channel. In an extensive evaluation, different settings of the approach are compared in a series of experiments. The results indicate that the desired effect of more balanced results of the ensemble members can be achieved by the approach.