Environmental Science: Environmental Chemistry

Abstract

Request for single articles

370 KB  Full paper 6 downloads since July 2001

DOI: http://dx.doi.org/10.1065/ehs2001.06.009

The objective of this paper is to show, by means of simple statistical Multiple Linear Regression (MLR) models, that an air pollution and meteorological network initially designed for diagnosis purposes can be used to forecast hourly levels of O3, NO2, NO up to eight hours ahead of time. The network can be used as a prognostic tool at a given location if O3, NO2, NO and meteorological parameters are measured jointly.

Two groups of MLR models were built for one location in the Bilbao area (Spain) using current and past air pollution and meteorological data from the monitoring network of the zone. The first group of models were deterministic MLR models, whose coefficients were obtained with one year´s hourly data (1993) from the network and used to forecast the next 8760 cases of year 1994. The other groups were mixed deterministic-statistical models, whose coefficients were continuously recalculated with the most recent data available and used to forecast one single case. The performance of the two types of models was determined by comparison of hourly predictions and real observations for a one year period (1994) and was compared with the simplest prediction possible: persistence of levels.

The performance of the models has been evaluated using the set of statistical standard parameters included in the so-called Model
Validation Kit. The results show that both groups of models yield a significantly better prediction up to 8 hours ahead for O3, NO2,
and NO than the persistence of levels.

Deterministic models’ predictions for the three pollutants are better than those from the mixed deterministic-statistical in all cases except for the prediction of NO between three and eight hours ahead. The results obtained are as good as those obtained with much more sophisticated models and are much easier and cheaper to implement since the models can be easily calculated and run on a simple PC.