Calculation of pseudo PM2.5 annual mean concentrations in Europe based on annual mean PM10 concentrations and other supplementary data ETC/ACC Technical Paper 2010/9

11 Jan 2011

Iulian Petchesi

The two approaches are ‘Empirical Ensemble-based Virtual Sensing’ (EEVS), which is based on Artificial Neural Networks, and multiple linear regression (MLR), a standard variational technique. They give similar results, with the EEVS approach showing slightly lower RMSE than MLR. For all other metrics assessed there was no significant difference. Neither approach was found to fulfil the monitoring quality objectives of the air quality directive 2008/50/EC, but both would fulfil the requirements in the directive for modelling and indicative measurements.

From a practical application point of view the EEVS approach is opaque (no information is available on either the Neural Network models used or the parameters applied after training) and the results cannot be reproduced or understood by third parties. Unless this situation changes in the future, we recommend MLR as a suitable approach for future PM2.5 mapping for Europe.

Prepared by: Bruce Denby1, Giulio Gola2, Frank de Leeuw3, Peter de Smet3, Jan Horálek4
1Norwegian Institute of Air Research (NILU), Kjeller, Norway; 2Institute for Energy Technology (IFE), Halden, Norway; 3Netherlands Environmental Assessment Agency (PBL), Bilthoven, The Netherlands; 4Czech Hydrometeorological Institute (CHMI), Prague, Czech Republic

Published by: ETC/ACC, January 2011, 20 pp.

Map downloads:

All relevant ETC technical papers