Application of a land - use regression model for calculation of the spatial pattern of annual NOx air concentrations at national scale: a case study for Poland

Land Use Regression (LUR) models are applied to derive spatial information on NOx air concentration for Poland, with 1km x 1km grid resolution for years 2005 and 2008. The LUR results are cross validated with the leave one out approach and compared with the output of the FRAME atmospheric transport model. The LUR and FRAME modelled spatial patterns of NOx concentrations are similar, with the grid-to-grid correlation coefficient at 0.89 in 2005 and 0.84 for 2008. The LUR approach showed higher concentrations than FRAME, with a mean grid-to-grid difference of 0.9 and 2.2 mu g for year 2005 and 2008. Missing emission sources may be the main reason for the better performance of the LUR models, compared to FRAME. Regression model can response to the gaps in emission inventory by adjusting the regression coefficients, while physical models show lower air concentrations resulting in larger errors. It should also be noticed that the cross validation approach, used for evaluation of the LUR models, and direct model-measurement comparison applied for error assessment in case of FRAME, are not directly comparable, with the latter being more demanding. (C) 2011 Published by Elsevier Ltd. Selection and peer-review under responsibility of Spatial Statistics 2011