Indicators of performance of dispersion models and their reference values

Abstract

INTRODUCTIONModern approaches to validation of dispersion models upon experimental data are usually based on calculation of different indicators of performance of the models (see for example Ohlesen, 1997). The aforementioned indicators can be considered as certain metrics invented for evaluating the measure of closeness of model predictions and measurements. The procedure of validation starts with generation of the model set of data called "predictions", which mimics a set of data of measurements called "measurements". Then one can use any kind of the concurrent analysis to evaluate the measure of closeness of these two sets or, in other words, to estimate the performance of the model in question.

Dispersion Modelling for Environmental Security: Principles and Their Application in the Russian Regulatory Guideline on Accidental Releases

“…Genikhovich and Filatova (2001) have demonstrated that even the axial ground-level concentrations (GLCs) from a point source are heavily contaminated with noise and that the standard deviations of their stochastical fluctuations are approximately equal to their mean values. It means that the errors of predictions of axial GLCs with a deterministic model resulting from the "noisiness" of data of measurements, E n , are equal at least about 100% (Genikhovich, 2003); correspondingly, the errors of prediction of concentrations at a given receptor point are of order of 1000% or even more. Let us compare these figures with other errors that could be expected in dispersion modelling.…”