ENSEMBLE and AMET: Two systems and approaches to a harmonized, simplified and efficient facility for air quality models development and evaluation S. Galmarini a, * , R. Bianconi b , W. Appel c , E. Solazzo a , S. Mosca b , P. Grossi b , M. Moran d , K. Schere c , S.T. Rao c a Joint Research Centre, European Commission, ISPRA, Italy b Enviroware srl, Concorezzo (MB), Italy c Atmospheric Modelling and Analysis Division, U.S. Environmental Protection Agency, Research Triangle Park, NC, USA d Air Quality Research Division, Science and Technology Branch, Environment Canada, Toronto, Canada article info Article history: Received 31 May 2011 Received in revised form 27 July 2011 Accepted 3 August 2011 Keywords: Model evaluation Model community Monitoring information Information technology abstract The complexity of air quality modeling systems, air quality monitoring data make ad-hoc systems for model evaluation important aids to the modeling community. Among those are the ENSEMBLE system developed by the EC-Joint Research Center, and the AMET software developed by the US-EPA. These independent systems provide two examples of state of the art tools to support model evaluation. The two systems are described here mostly from the point of view of the support to air quality model users or developers rather than the technological point of view. While ENSEMBLE is a web based platform for model evaluation that allows the collection, share and treatment of model results as well as monitoring data, AMET is a standalone tool that works directly on single model data. The complementarity of the two approaches makes the two systems optimal for operational, diagnostic and probabilistic evaluations. ENSEMBLE and AMET have been extended in occasion of the AQMEII two-continent exercise and the new developments are described in this paper, together with those foreseen for the future. Ó 2011 Elsevier Ltd. All rights reserved. 1. Introduction In atmospheric dispersion and air quality modeling, the evalu- ation of the model performance has long been a topic of research activities and establishing best practices (e.g. Oreskes et al., 1994, Steyn and Galmarini, 2008). Before we tackle the model evaluation topic per-se, it is probably appropriate to give a couple of defini- tions. We will refer here to atmospheric dispersion models as the models that simulate the dispersion of a passive, decaying or reactive species released from a point source, be it a stack or a limited-area source. No limit exists to the scale of application which can extend as far as global. These are the models normally used for emergency preparedness and response applications. With air quality model we refer however to a model which deals with distributed sources of various chemical precursors and that treats the chemical transformations occurring in a volume of air that can range from the meso- to the global-scale in a three-dimensional domain. In the two cases the dynamic fields are acquired from meteorological models. The final application of these models is in the realm of support to air quality policy planning and regulation. The verification of the model capability to adhere to experi- mental evidence has occupied atmospheric scientists in recent decades. The complexity of the atmospheric system in terms of spatial and temporal variability, stochasticity and scales, makes the task of evaluating a model particularly complicated. The collection of experimental evidence, representative of space and time scales, which could be used for evaluation of a time- and space-averaged model result, has always constituted the first burden in the prac- tice of model evaluation. To date, operational networks of instru- ments provide organized, quality-checked information useful in this context for both atmospheric dispersion and air quality models. Routine monitoring data however are often insufficient to assess the performance of a model in depth, as they may only provide evidence of the general correspondence of model results with surface-based point measurements that are typical in such network data. These types of evaluations do not provide clear indications of the veracity of the modeled-chain of processes that leads to the model result. In other words, using a common expression, we would not know whether the right model results were obtained for the right reasons. The problem becomes particularly complicated in the case of predictions where data are not sufficiently available. This is the * Corresponding author. E-mail addresses: stefano.galmarini@jrc.it, stefano.galmarini@jrc.ec.europa.eu (S. Galmarini). Contents lists available at SciVerse ScienceDirect Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv 1352-2310/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.atmosenv.2011.08.076 Atmospheric Environment 53 (2012) 51e59