|M.Sc Thesis||Department of Civil and Environmental Engineering|
|Supervisor:||Assoc. Prof. Broday David|
|Full Thesis text|
Air quality monitoring data are needed for air resources management by decision makers, and by public health officers in epidemiological studies. Clearly, only a minor part of the known air pollutants are monitored on a wide spatial and temporal scale. To overcome this gap indirect means for estimating ambient concentrations of harmful pollutants are needed.
Polycyclic aromatic hydrocarbons, in general, and benzo(a)pyrene (B[a]P), in particular, have received increased attention in recent years in air pollution studies due to their associated health effects. Yet, due to their low temporal and spatial measuring frequency they are good examples for pollutants managed without sufficient information.
This work presents data driven models for estimating daily and monthly ambient B[a]P concentrations, pertinent for small geographical scales and suitable for epidemiological studies. We demonstrate model development and validation using data from four monitoring stations in the Czech Republic, using the world richest B[a]P database available to date.
The presented models are used sequentially. First, a classification tree is used to distinguish between environmental conditions in which ambient B[a]P concentrations are expected to be above or below the measuring instrument detection limit (MIDL) Then, for cases in which the ambient concentrations are expected to be above the MIDL, B[a]P levels are estimated using a multivariate linear regression model. For cases in which the environmental conditions are expected to lead to concentrations below the MIDL, the B[a]P concentrations are set to a fixed value.
Models have been developed based on data from three monitoring stations. The performance of each model has been carefully evaluated following a 3-step evaluation procedure, including a complete internal-, external-, and temporal cross validation process. Model results suggest that the models’ performance is very good for estimating monthly mean ambient B[a]P concentrations, especially when the model is parameterized and used at the same locality. Spatial extrapolation resulted in a minor deterioration in the models' performance. Temporal extrapolation, i.e. using the model for periods with a dramatic change in the environmental conditions (increased emissions), reduced considerably the models' performance.
These findings have two important implications: (1) based on records from recent years, reliable estimations of ambient B[a]P concentrations in previous years can be obtained only if no major changes in key environmental parameters have occurred; (2) estimates of monthly means B[a]P concentrations are more reliable than daily means estimates, and may be used in retrospective environmental health studies.