Atmospheric Environment 40 (2006) 4935–4945 On joint deterministic grid modeling and sub-grid variability conceptual framework for model evaluation Jason Ching à , Jerold Herwehe, Jenise Swall Atmospheric Sciences Modeling Division, ARL, NOAA, RTP, NC, USA Received 15 March 2005; received in revised form 15 November 2005; accepted 7 December 2005 Abstract The general situation (but exemplified in urban areas), where a significant degree of sub-grid variability (SGV) exists in grid models poses problems when comparing grid-based air-quality modeling results with observations. Typically, grid models ignore or parameterize processes and features that are at their sub-grid scale. Also, observations may be obtained in an area where significant spatial variability in the concentration fields exists. Consequently, model results and observations cannot be expected to be equal. To address this issue, we suggest a framework that can provide for qualitative judgments on model performance based on comparing observations to the grid predictions and its SGV distribution. Further, we (a) explore some characteristics of SGV, (b) comment on the contributions to SGV and (c) examine the implications to the modeling results at coarse grid resolution using examples from fine scale grid modeling of the Community Multi-scale Air Quality (CMAQ) modeling system. r 2006 Elsevier Ltd. All rights reserved. Keywords: Neighborhood-scale models; CMAQ fine scale modeling; Sub-grid distributions; Sub-grid variability; Multiscale air-quality modeling 1. Introduction Comparison of meteorological and air-quality grid model simulations with point measurements is problematic. In statistical terms, this is considered a ‘‘change of support’’ problem in which inferences are made about differences between point-based measurements to model-predicted values that re- present volume average concentration (Gelfand et al., 2001). There is an extensive body of literature that recognizes and discusses sources of modeling uncertainties and methods for evaluating them. For example, Fine et al. (2003) review the issues associated with evaluation of model uncertainties (MU) in photochemical models. They explore the range of sensitivity, diagnostic and other useful studies used in performing uncertainty analyses. MU, in general, consists of contributions from all sources in varying degrees, to be both epistemic and aleatory (i.e., due to deterministic or stochastic causes, respectively) (Lohman et al., 2000). The classes of epistemic MU are those inclusive of model inputs such as emissions, meteorology, land-use, initial and boundary conditions, and imperfections in model formulations (parameterizations) of var- ious physical and chemical processes (NRC, 1991; Russell and Dennis, 2000). The aleatory classes of ARTICLE IN PRESS www.elsevier.com/locate/atmosenv 1352-2310/$ - see front matter r 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.atmosenv.2006.01.021 à Corresponding author. US E.P.A., (MD) E243-04 APMB, AMD, NERL, Research Triangle Park, NC 27711, USA. Tel.: +1 919 541 4801; fax: +1 919 541 1379. E-mail address: ching.jason@epa.gov (J. Ching).