The comparison of different methods of statistical prediction of diurnal dynamics in the ground ozone concentration
P.N. Antokhin, B.D. Belan, D.E. Savkin, G.N. Tolmachev
V.E. Zuev Institute of Atmospheric Optics of Siberian Branch of the Russian Academy of Science, 1, Academician Zuev square, Tomsk, 634021, Russia
Keywords: atmosphere, ozone, modeling, prediction
Abstract
Оn the basis of long series of observations obtained at TOR-station at the Tomsk Akademgorodok, an empirical model for prediction of average daily ozone concentrations is developed based on a multilayer neural network. A comparison with models based on multiple linear regression and autoregression was conducted. The method of neural network approach turned out to be the most successful among all others. It gives a possibility to describe 70% of the total variance and the average value of 50% of the variance of the standard deviation. In this case, the value of the mean square prediction error does not exceed the instrumental error of measurements.
|
