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

• CALPUFF tended to overstate the maximum observed concentrations and understate the
plume widths at the 100 km receptor arc.
• The best performing CALPUFF configuration in terms of predicting the maximum observed
concentrations and plume width was when CALMET was run with MM5 data and surface
meteorological observations but no upper‐air meteorological observations.
• The CALPUFF CAL and AER turbulence dispersion options produced nearly identical results
and the performance of the CAL/AER turbulence versus PG dispersion options varied by
model configuration and statistical performance metric.
• The performance of CALPUFF/MMIF in predicting plume maximum concentrations and
plume widths was comparable or better than all of the CALPUFF/CALMET configurations,
except when CALMET used MM5 data and surface but no upper‐air meteorological
observations.
• The modeled plume centerline tended to be offset from the observed centerline location
by 0 to 14 degrees.
• Use of CALMET with just surface and upper‐air meteorological observations produced the
best CALPUFF plume centerline location performance, whereas use of just MM5 data with
no meteorological observations, either through CALMET or MMIF, produced the worst
plume centerline angular offset performance.
• Different CALMET configurations give the best CALPUFF performance for maximum
observed concentration (with MM5 and just surface and no upper‐air observations) versus
location of the plume centerline (no MM5 and both surface and upper‐air observations)
along the 100 km receptor arc. For Class I area LRT dispersion modeling it is important for
the model to estimate both the location and the magnitudes of concentrations.
The evaluation of the CALPUFF sensitivity tests for the 600 km arc of receptors included both
plume arrival, departure and residence time analysis as well as fitted Gaussian plume statistics.
The observed residence time of the tracer on the 600 km receptor arc was at least 12 hours.
Note that due to the presence of an unexpected low‐level jet, the tracer was observed at the
600 km receptor arc for the first sampling period. Thus, the observed 12 hour residence time is
a lower bound (i.e., the observed tracer could have arrived before the first sampling period).
The 1998 EPA CALPUFF evaluation study estimated tracer plume residence times of 14 and 13
hours, which compares favorably with the observed residence time (12 hours). However, the
1998 EPA study CALPUFF modeling had the tracer arriving at least 1 hour later and leaving 2‐3
hours later than observed, probably due to the inability of CALMET to simulate the low‐level
jet.
Most (~90%) of the current study CALPUFF sensitivity tests underestimated the observed tracer
residence time on the 600 km receptor arc by approximately a factor of two. The exception to
this was: (1) the BASEA_PG CALPUFF/CALMET sensitivity test (12 hours) that used just
meteorological observations in CALMET and the PG dispersion option in CALPUFF; and (2) the
CALPUFF/CALMET EXP2C series of experiments (residence time of 11‐13 hours) that used 36 km
MM5 data and CALMET run at 4 km resolution with no meteorological observations (NOOBS =
2). The remainder of the 28 CALPUFF sensitivity tests had tracer residence time on the 600 km
receptor arc of 4‐8 hours; that is, almost 90% of the CALPUFF sensitivity tests failed to
reproduce the good tracer residence time performance statistics from the 1998 EPA study.
For the 600 km receptor arc, the CALPUFF sensitivity test fitted Gaussian plume statistics were
very different than the 100 km receptor arc as follows:
9