CRC Report No. A-34 - Coordinating Research Council

April 2005
• Eliminating 20% of the species available to CMB (experiment 4) degraded performance,
especially when source profiles were poorly known. Eliminating species had a
substantial impact in Round 1 (blind analysis) and little impact in Round 4 (full profile
information).
• Changing the speciation profiles (experiment 3) improved or degraded CMB performance
depending upon whether the alternate profiles were a better or poorer fit with the profiles
used in CMB.
The last finding on the effects of alternate speciation profiles seems obvious and needs further
discussion. The alternate profiles used in experiment 3 included mobile source (gasoline)
profiles taken from other CMB studies. In Round 1, with minimal information, DRI obtained
more accurate gasoline contributions for experiment 3 than any other experiment, perhaps
because the experiment 3 profiles had internal characteristics that “work well” for CMB
analyses, such as lack of co-linearity. Round 4 with detailed profile information showed the
opposite result of poorer gasoline apportionments for experiment 3 than any other experiment,
likely because experiment 3 was the odd man out, i.e., it used profiles that differed from the
“known profiles.”
4.2 HOURLY SOURCE APPORTIONMENT RESULTS
The discussion of experiments 1-8 above focused on source contributions averaged over all
hours. However, analyses of ambient data often focus on morning (6-9 am) samples to minimize
effects of atmospheric dilution and chemical reaction. In contrast, afternoon (1-4 pm) samples
are expected to be chemically aged and so more difficult to analyze. Figure 4-3 shows the results
for experiment 1 in Round 4 at each receptor for three averaging times: 6-9 am, 1-4 pm and all
hours. The results for all hours were shown above (Figure 4-1c) and it was concluded that CMB
performed well for gasoline, solvents and biogenics at most receptors. Figure 4-3 shows that
there was more scatter in the CMB apportionments for gasoline, solvents and biogenics in 6-9
am samples than for all hours. Similarly, the CMB apportionments for 1-4 pm were more
scattered than for all hours, but were no worse than for 6-9 am. The degraded performance of
CMB for restricted time periods may be due to the reduction in sample size (averages over 12
samples rather than 96) in these experiments where only 4 days were simulated. This difficulty
could be avoided for real-world conditions if many more days are sampled (e.g., data from a
PAMS auto-GC site). However, the results obtained here show no evidence that focusing on 6-9
am samples will improve the accuracy of CMB. Correspondingly, analyzing afternoon samples
did not degrade the performance of CMB.
H:\crca34-receptor\report\Final\sec4.doc 4-10