Data loss from time series of pollutants in ambient air exhibiting seasonality: consequences and strategies for data prediction

Abstract

The effect of data loss on annual average concentrations of seasonal and non-seasonal pollutants in ambient air has been investigated. The bias caused to the true annual average has been shown to be significant for measurements of benzo[a]pyrene (BaP) in PM₁₀ (a highly seasonal pollutant) even when legislative requirements for data capture and time coverage are still met. In order to mitigate this bias, strategies to predict concentrations during periods of lost data have been tested. These have been based on fitting quadratic relationships to available data of BaP in PM₁₀ at individual monitoring stations on the UK PAH Monitoring Network. The annual average concentration values produced with and without the use of predicted data have been compared to the actual annual averages in the absence of data loss. The use of predicted data is a significant (but not universal) improvement at urban and rural monitoring stations where the data exhibit consistently good fits to the predicted quadratic model. At industrial stations, where the quadratic model fails, the use of predicted data shows no improvement, although the effect of lost data at these sites on the annual average is much less noticeable because of their lack of seasonality.