Geostatistical analysis of the temporal variability of ozone concentrations. Comparison between CHIMERE model and surface observations

Abstract

Deterministic models have become essential tools to forecast and map concentration fields of atmospheric pollutants like ozone. Those models are regularly updated and improved by incorporating recent theoretical developments and using more precise input data. Unavoidable differences with in situ measurements still remain, which need to be better understood. This study investigates those discrepancies in a geostatistical framework by comparing the temporal variability of ozone hourly surface concentrations simulated by a chemistry-transport model, CHIMERE, and measured across France. More than 200 rural and urban background monitoring sites are considered. The relationship between modelled and observed data is complex. Ozone concentrations evolve according to various time scales. CHIMERE correctly accounts for those different scales of variability but is usually unable to reproduce the exact magnitude of each temporal component. Such difficulty cannot be entirely attributed to the difference in spatial support between grid cell averages and punctual observations. As a result of this exploratory analysis, the common multivariate geostatistical model, known as the linear model of coregionalization, is used to describe the temporal variability of ozone hourly concentrations and the relationship between simulated and observed values at each observation point. The fitted parameters of the model can then be interpreted. Their distribution in space provides objective criteria to delimitate the areas where the chemistry-transport model is more or less reliable.