A Statistical Model For 1-hour- To 24-hour-ahead Prediction Of Hourly Ozone Concentrations At Ground Level In Singapore
Free (open access)
389 - 400
X. Liu, Y. Hwang, K. Yeo, J. Hosking, A. Barut, J. Singh & Y. Amemiya
A spatio-temporal statistical model is proposed for 1- to 24-hour-ahead prediction of hourly ozone concentrations. This is a joint work with the National Environmental Agency Singapore, and is Singapore’s first predictive model for ozone concentrations. Unlike many existing models which focus on either daily maximum or 8h average daylight ozone concentrations, the present work is concerned with the prediction of hourly ozone concentrations which are usually associated with higher variability. A recently proposed framework for spatiotemporal prediction is used to model ozone concentration data. The macroscale spatio-temporal variation of ozone concentrations is modeled by a linear function of five carefully constructed predictors, while the micro-scale variation is captured by a mean-zero spatio-temporally correlated random process. We show that this model also provides useful insights about the effects of some complex environmental processes on ozone concentration; this is indeed an attractive feature for any data-driven air quality model.
ozone, spatio-temporal statistics, random process.