A temperature-based approach to predicting lost data from highly seasonal pollutant data sets.