Since 1995 a system for ozone forecast is operating in the State of Ba
varia. The forecast data are calculated by a neural network (PROZON).
Beside of PROZON two other forecast models - one applying regression e
quations, the second with a physical-chemical approach - have been inv
estigated. The physical-chemical approach proved to be inferior under
the aspect of accuracy of forecast. The results which have been gained
applying the three models during summer 1995 are presented and evalua
ted. The accuracy of the forecast applying regression equations respec
tively a neural network is satisfactory. The forecast of these models
is mainly influenced by the accuracy of temperature forecast.