We are doing science for policy
The Joint Research Centre (JRC) is the European Commission's science and knowledge service which employs scientists to carry out research in order to provide independent scientific advice and support to EU policy.
The development of the procedure for air quality model benchmarking in the context of the Air Quality Directive 2008/50/EC (AQD) has been an on-going activity in the context of the FAIRMODE community. Central part of the studies was the definition of proper modelling quality indicators and criteria to be fulfilled in order to allow sufficient level of quality for a given model application under the AQD. The focus initially on applications related to air quality assessment has gradually been expanded to other applications, such as forecasting and planning. The main purpose of this Guidance Document is to explain and summarize the current concepts of the modelling quality objective methodology, elaborated in various papers and documents in the FAIRMODE community, addressing model applications for air quality assessment and forecast. Other goals of the Document are linked to presentation and explanation of templates for harmonized reporting of modelling results. Giving an overview of still open issues in the implementation of the presented methodology, the document aims at triggering further research and discussions. A core set of statistical indicators is defined using pairs of measurement-modelled data. The core set is the basis for the definition of a modelling quality indicator (MQI) and additional modelling performance indicators (MPI), which take into account the measurement uncertainty. The MQI describes the discrepancy between measurements and modelling results (linked to RMSE), normalized by measurement uncertainty and a scaling factor. The modelling quality objective (MQO) requires MQI to be less than or equal to 1. With an arbitrary selection of the scaling factor of 2, the fulfilment of the MQO means that the allowed deviation between modelled and measured concentrations is twice the measurement uncertainty. Expressions for the MQI calculation based on time series and yearly data are introduced. MPI refer to aspects of correlation, bias and standard deviation, applied to both the spatial and temporal dimensions. Similarly to the MQO for the MQI, modelling performance criteria (MPC) are defined for the MPI; they are necessary, but not sufficient criteria to determine whether the MQO is fulfilled. The MQO is required to be fulfilled at 90% of the stations, a criteria which is implicitly taken into account in the derivation of the MQI. The associated modelling uncertainty is formulated, showing that in case of MQO fulfilment the modelling uncertainty must not exceed 1.75 times the measurement one (with the scaling factor fixed to 2). A reporting template is presented and explained for hourly and yearly average data. In both cases there is a diagram and a table with summary statistics. In a separate section open issues are discussed and an overview of related publications and tools is provided. Finally, a chapter on modelling quality objectives for forecast models is introduced. In Annex 1, we discuss the measurement uncertainty which is expressed in terms of concentration and its associated uncertainty. The methodology for estimating the measurement uncertainty is overviewed and the parameters for its calculation for PM, NO2 and O3 are provided. An expression for the associated modelling uncertainty is also given.