A proposed methodology for the assessment of arsenic, nickel, cadmium and lead levels in ambient air

Air quality assessment, required by the European Union (EU) Air Quality Directive, Directive 2008/50/EC, is part of the functions attributed to Environmental Management authorities. Based on the cost and time consumption associated with the experimental works required for the air quality assessment in relation to the EU-regulated metal and metalloids, other methods such as modelling or objective estimation arise as competitive alternatives when, in accordance with the Air Quality Directive, the levels of pollutants permit their use at a specific location. This work investigates the possibility of using statistical models based on Partial Least Squares Regression (PLSR) and Artificial Neural Networks (ANNs) to estimate the levels of arsenic (As), cadmium (Cd), nickel (Ni) and lead (Pb) in ambient air and their application for policy purposes. A methodology comprising the main steps that should be taken into consideration to prepare the input database, develop the model and evaluate their performance is proposed and applied to a case of study in Santander (Spain). It was observed that even though these approaches present some difficulties in estimating the individual sample concentrations, having an equivalent performance they can be considered valid for the estimation of the mean values – those to be compared with the limit/target values – fulfilling the uncertainty requirements in the context of the Air Quality Directive. Additionally, the influence of the consideration of input variables related to atmospheric stability on the performance of the studied statistical models has been determined. Although the consideration of these variables as additional inputs had no effect on As and Cd models, they did yield an improvement for Pb and Ni, especially with regard to ANN models.