Bayesian model averaging for emergency response atmospheric dispersion multimodel ensembles: Is it really better? How many data are needed? Are the weights portable?

Abstract

[1] In this paper, we investigate applicability of Bayesian model averaging (BMA) methodology to atmospheric dispersion multimodel ensemble system within the context of emergency response applications. The BMA method can be used both to evaluate model predictions and to combine model results using BMA weighing factors. We analyze time evolution of BMA weights and include a detailed quantitative comparison of different combinations of model results performed by the means of statistical indicators. The analysis allows us to identify similarities and differences among different combined models. Finally, we question the portability of BMA weights among various cases. From the analysis it follows that BMA can be applied in considered problems; however, the median of the model results also performs well and produces more conservative results.Citation: Potempski, S., S. Galmarini, A. Riccio, and G. Giunta (2010), Bayesian model averaging for emergency response atmospheric dispersion multimodel ensembles: Is it really better? How many data are needed? Are the weights portable?,

<i>Est modus in rebus</i>: analytical properties of multi-model ensembles

“…It should be also added that if individual models pdfs are known we can combine them using optimal weights to calculate ensemble pdf. This is in accordance with the general concept of applying the ensemble approach (Dabbert and Miller, 2000;Galmarini et al, 2004;Stull et al, 1997;Riccio et al, 2007;Potempski et al, 2009) to perform predictions in order to rely on stochastic paradigm rather than deterministic one.…”

<i>Est modus in rebus</i>: analytical properties of multi-model ensembles

“…This is in accordance with the general concept of applying the ensemble approach (Dabbert and Miller, 2000;Galmarini et al, 2004;Stull et al, 1997;Riccio et al, 2007;Potempski et al, 2009) to perform predictions in order to rely on stochastic paradigm rather than deterministic one.…”

A regional air quality forecasting system over Europe: the MACC-II daily ensemble production

“…The methods vary from very simple such as the average or the median to more elaborated such as weighted averages based on past scores, Bayesian models or spectral methods (e.g. Delle Monache et al, 2006;Riccio et al, 2007;Potempski et al, 2010;Galmarini et al, 2013).…”