Local and urban scale modeling - Air Pollution Modeling and Its Application - page 60

here was the GEM-MACH North American 15 km/7.5 min forecast grid (GEM-
MACH15), with the same eight species as AURAMS but only two bins (fine and
coarse modes) in order to maintain computational efficiency. It should be noted
that GEM-MACH is currently under development; the model results shown here
are taken from an intermediate developmental version of the model (version
AGY).
The 42 and 15 km AURAMS simulations were compared to North American
monitoring data for hourly PM 2.5 and O 3 (Airnow network) for the 3 months simu-
lated including the simulation period. The 2.5 km AURAMS simulations were
compared to detailed observations made during individual flights and CRUISER
drives during the measurement intensive.
GEM-MACH15 simulations were compared to monitoring networks for hourly
PM 2.5 and O 3 for the time period simulated (see Table 1) . The 15 km AURAMS and
15 km GEM-MACH results for PM 2.5 and O 3 give an approximation of the relative
performance of the two models (different start and end times were employed for
operational reasons, and the extent of the model domains differ). The numbers thus
allow a rough comparison of model performance; further work will be required for
a more robust comparison.

Table 1. Comparison of AURAMS42 km, AURAMS15 km, GEM-MACH15 kmAGY to
observations during the BAQS-Met period

Model Species Corr. coeff Mean bias RMS error

AURAMS-42 km O 3 /PM 2.5 0.47/0.42 17.5/7.7

14.0/−1.0

AURAMS-15 km O 3 /PM 2.5 0.48/0.57 6.7/0.2 13.1/8.4

GEM-MACH15 O 3 /PM 2.5 0.61/0.18 −1.7/0.9 13.1/24.8

O 3 units ppbv, PM 2.5 units μg/m 3

GEM-MACH15 performance is better than AURAMS for ozone, with higher
correlation coefficient, lower magnitude biases, and the same or lower RMSE.
AURAMS has better performance than this version of GEM-MACH15 for PM 2.5 .
The latter errors relate to the two-bin approach used in GEM-MACH15; sub-binning
was used to improve mass transfer between bins, but further investigation is req-
uired to determine the cause of the lower correlation coefficient and higher RMS
values in GEM-MACH than in AURAMS15 km.
The Harrow super-site ozone time series ( Fig. 1) is in accord with Table 1; the
time series show several instances (e.g. 6/10/2007 8:00, 6/11/2007 6:00, 6/13/2007
17:00, 6/18/2007 16:00, 6/20/2007 9:00, 6/20/2007 18:00) where the GEM-MACH
maximum or minimum ozone prediction is much closer to observations than that
of AURAMS15. A significant GEM-MACH15 over-prediction occurs at 26/0:00
and 26/22:00.
The PM time series at Harrow (Fig. 2 ) is also in accord with Table 1. PM 2.5 SO 4
is given as an example, and the GEM-MACH values are considerably higher than