Spatio-temporal analysis and interpolation of PM10 measurements in Europe for 2009 ETC/ACM Technical Paper 2012/8

08 Jan 2013

Iulian Petchesi

NOTE: Revised Version; released 27 March 2013

This study continues the work described in Gräler et al. (2012a) and investigates the potential of additional approaches for spatio-temporal kriging of daily mean PM10 concentrations as well as a re-assessment of the prediction quality for another year, i.e. 2009 next to 2005. All methods are applied to daily mean rural background PM10 concentrations across Europe for the year 2009. Additionally, temporal interval1 kriging approaches are investigated to assess their potential for analysing trends in yearly mean concentrations. Asymmetric spatio-temporal dependencies are also addressed.

The cross-validation statistics on an annual level are in general better for the 2009 data. Only the approaches not relying on a regression in 2009 outperform the spatio-temporal approaches in 2005 on a daily level. This study showed that the results from the cross-validation are not strictly robust against year-to-year variability for these years, but the results remain close. In especially, the pure spatial interpolation procedure to derive annual estimates from annual means could not repeatedly be outperformed. However, we believe that the spatio-temporal interpolation methods provide great potential for the production of PM10 concentration maps over Europe on a daily basis. The availability of daily EMEP model data may, however, limit the applicability of the regression based approaches.

Temporal interval kriging approaches did not perform very well in the cross-validation study, but in principle their ability to provide precise prediction error estimates is very valuable in uncertainty analysis. This becomes especially important when long term European wide trends are of interest. The temporal interval kriging approach provides valid kriging variances associated with the interval predictions where the alternative approach only provides a proxy of the variability of the predictions.

The study on asymmetric dependence patterns due to prevailing environmental impacts (e.g. wind) did not reveal a meaningful asymmetric pattern for the residuals. The raw data seem to be weakly asymmetric along the North-South axis.

Reference to R-scripts used for the calculations:

Prepared by: Benedikt Gräler, Mirjam Rehr, Lydia Gerharz, Edzer Pebesma (IfGI, Univ. Münster) under subcontract of the ETC/ACM Consortium.

Published by: ETC/ACM, March 2013 (revised version), 30 pp.