A synoptic climatological approach to assess climatic impact on air quality in South-central Canada. Part I: Historical analysis

Automated synoptic weather typing and robust orthogonal stepwise regression analysis (via principal components analysis) were applied together to develop within-weather-type air pollution prediction models for a variety of pollutants (specifically, carbon monoxide - CO, nitrogen dioxide - NO2, ozone - O-3, sulphur dioxide - SO2, and suspended particles - SP) for the period 1974-2000 in south-central Canada. The SAS robust regression procedure was used to limit the influence of outliers on air pollution prediction algorithms. Six-hourly Environment Canada surface observed meteorological data and 6-hourly US National Centers for Environmental Prediction (NCEP) reanalysis data of various weather elements were used in the analysis. The models were developed using two-thirds of the total years for meteorological and air pollution data; the remaining one-third (randomly selected) was used for model validation. Robust stepwise regression analysis was performed to analytically determine the meteorological variables that might be used to predict air pollution concentrations. There was a significant correlation between observed daily mean air pollution concentrations and model predictions. About 20, 50, and 80% of the 80 prediction models across the study area possessed R-2 values >= 0.7, 0.6, and 0.5, respectively. The results of model validation were similar to those of model development, with slightly smaller model R-2 values.