The atmospheric ozone concentration in Seoul was forecasted using an artifi
cial neural network and spatiotemporal analysis. The artificial neural netw
ork was trained by using hourly pollutant and meteorological data that resu
lted in complex patterns of ozone formation. The finite-volume method was e
mployed in the spatiotemporal analysis in order to take into account the ef
fects of wind. Time horizons in the forecasts were 1-6 h and 16-21 h. The r
esulting predictions of ozone formation were compared to measured data. Fro
m the comparison, it was found that the neural network method gave reliable
accuracy within a limited prediction horizon.