Geostatistical estimate of PM10 concentrations in Northern Italy: validation of kriging reconstructions with classical and flexible variogram models
Thursday 24th July 2008
Bonaventura, Luca; Castruccio, Stefano; Crippa, Paola; Lonati, Giovanni
The applicability of classical geostatistical tools to the reconstruction of PM10 concentration fields over the entire Po Valley has been assessed, based on a large dataset of daily PM10 data spanning the period 2003- 2006. The impact of data detrending by the median polish procedure and of the variogram model chosen for the geostatistical estimates have been investigated, by comparison of the results obtained with several isotropic variogram models as well as with anisotropic exible variogram models. The relative merits of the different approaches were evaluated by crossvalidating the resulting reconstructions and performing normality tests on the corresponding residuals. Although exponential and linear variograms yield reliable reconstructions in most of the cases, the analysis has highlighted significant seasonal and interannual variations in the basic features of the estimated concentration fields and residual correlation structure. As a consequence, none of the classical models is able to cope with all the different situations encountered, while the anisotropic exible variogram models appear to provide a more robust tool for automatic reconstruction of the PM10 concentration fields without expert user intervention.