Short-Term Forecasting of Nitrogen Dioxide (NO2) Levels Using a Hybrid Statistical and Air Mass History Modelling Approach

Abstract

A novel hybrid model has been developed to support the provision of real-time air quality forecasts. Statistical techniques have been applied in parallel with air mass history modelling to provide an efficient and accurate forecasting system with the ability to identify high NO2 events, which tend to be the episodes of most significance in Ireland. Air mass history modelling and k-means clustering are used to identify air mass types that lead to high NO2 levels in Ireland. Trajectory matching techniques allow data associated with these air masses to be partitioned during model development. Non-parametric regression (NPR) has been applied to describe nonlinear variations in concentration levels with wind speed, direction and season and produce a set of linearized factors which, together with other meteorological variables, are employed as inputs to a multiple linear regression. The model uses an innovative integrated approach to combine the NPR with the air mass history modelling results. On validation, a correlation coefficient of 0.75 was obtained, and 91 % of daily maximum (hourly averaged) NO2 predictions were within a factor of two of the measured value. High pollution events were well captured, as indicated by strong agreement between measured and modelled high percentile values. The model requires only simple input data, does not require an emission inventory and utilises very low computational resources. It represents an accurate and efficient means of producing real-time air quality forecasts and, when used in combination with forecaster experience, is a useful tool for identifying periods of poor air quality 24 h in advance. The hybrid approach outlined in this paper can easily be applied to produce high-quality forecasts of both NO2 and additional pollutants at new locations/countries where historical monitoring data are available.