Hi Sham,
I compared your evaluation results to a 12US1 2019 CMAQv5.3.2 simulation that we did for the EPA’s Air Quality Time Series (EQUATES) project. The EQUATES 2019 emissions are slightly different from what I think you are using and they did include fire emissions. Specifics on the model run (met, BCs, etc) can be found here https://doi.org/10.15139/S3/F2KJSK.
I noticed some similarities and some differences in our evaluation results for the NAPS sites 60438 and 61502 that may be helpful context for what you are seeing.
-
Our 2019 simulation also show a low NO2 bias in April, especially at the 60438 site near Toronto. (First 5 plots.) The large underestimation of NO2 at 60438 is consistent with the ozone overestimation at that site as it leads to less ozone titration.
-
When I compare the EQUATES timeseries plots to the timeseries from your post it looks like the diurnal patterns in your model results are shifted by a few hours compared to the observations. That is, the peaks and valleys of the model and observed values are offset, whereas we do not see this as much in our results. You may want to look at your post-processing workflow. Aligning the timing of these peaks and valleys should reduce your model bias, although clearly the model is still underestimating NO2, particularly the peaks.
-
We see this NO2 underestimation across different regions in our 12US1 domain in 2019 and in earlier years. For example the last 2 plots show monthly average NO2 at all NAPS sites and in the Northeast US. The negative bias is getting worse across time which has led us to hypothesize that the trends in our emissions inventories may be too steep in some sectors.
-
It is surprising that the NO2 values at 60438 in your 4km simulation are so much lower than the 12km model results. I would have thought that the 4km simulation in that urban area would have higher NO2 peaks than the 12km simulation since there is less spatial smoothing. Do you have a theory on why you are seeing this result?
I’m afraid I don’t have any more concrete suggestions but hopefully this reasures you that your model evaluation results are generally in line with other modeling efforts.