Documentation of the Evaluation of CALPUFF and Other Long ...

FLEXPART 4 : The FLEXPART (Version 6.2; Siebert, 2006; Stohl et al., 2005 5 ) model is a
Lagrangian particle dispersion model. FLEXPART was originally designed for calculating
the long‐range and mesoscale dispersion of air pollutants from point sources, such as after
an accident in a nuclear power plant. In the meantime FLEXPART has evolved into a
comprehensive tool for atmospheric transport modeling and analysis
CAMx 6 : The Comprehensive Air‐quality Model with extensions (CAMx; ENVIRON, 2010) is
a photochemical grid model (PGM) that simulates inert or chemical reactive pollutants
from the local to continental scale. As a grid model, it simulates transport and dispersion
using finite difference techniques on a three‐dimensional array of grid cells.
CALGRID: The California Mesoscale Photochemical Grid Model (Yamartino, et al., 1989,
Scire et al., 1989; Earth Tech, 2005) is a PGM that simulates chemically reactive pollutants
from the local to regional scale. CALGRID was originally designed to utilize meteorological
fields produced by the CALMET meteorological processor (Scire et al., 2000a), but was
updated in 2006 to utilize meteorology and emissions in UAM format (Earth Tech, 2006).
The six LRT dispersion models represent two non‐steady‐state Gaussian puff models (CALPUFF
and SCIPUFF), two three‐dimensional particle dispersion models (HYSPLIT and FLEXPART) and
two three‐dimensional photochemical grid models (CAMx and CALGRID). HYSPLIT can also be
run in a puff and hybrid particle/puff modes, which was investigated in sensitivity tests. All six
LRT models were evaluated using the CAPTEX Release 3 and 5 field experiments and five of the
six models (except CALGRID) were evaluated using the ETEX field experiment database.
Evaluation Methodology
Two different model performance evaluation methodologies were utilized in this study. The
Irwin (1997) fitted Gaussian plume approach, as used in the EPA 1998 CALPUFF evaluation
study (EPA, 1998a), was used for the same two tracer test field experiments used in the 1998
EPA study (i.e., GP80 and SRL75). This was done to elucidate how updates to CALPUFF model
over the last decade have improved its performance. The second model evaluation approach
adopts the spatial, temporal and global statistical evaluation framework of ATMES‐II (Mosca et.
al., 1998; Draxler et al., 1998). The ATMES‐II uses statistical performance metrics of spatial,
scatter, bias, correlation and cumulative distribution to describe model performance. An
important finding of this study is that the fitted Gaussian plume model evaluation approach is
very limited and can be a poor indicator of LRT dispersion model performance, with the ATMES‐
II approach providing a more comprehensive assessment of LRT model performance.
Fitted Gaussian Plume Evaluation Approach
The fitted Gaussian plume evaluation approach fits a Gaussian plume across the observed and
predicted tracer concentrations along an arc of receptors at a specific downwind distance from
the tracer release site. The approach focuses on a LRT dispersion model’s ability to replicate
centerline concentrations and plume widths, modeled/observed plume centerline azimuth,
plume arrival time, and plume transit time across the arc. We used the fitted Gaussian plume
evaluation approach to evaluate CALPUFF for the GP80 and SRL75 tracer experiments where
the tracer concentrations were observed along arcs of receptors, as was done in the EPA 1998
CALPUFF evaluation study (EPA, 1998a).
4 http://transport.nilu.no/flexpart
5 http://www.atmos‐chem‐phys.net/5/2461/2005/acp‐5‐2461‐2005.html
6 http://www.camx.com/
4