I am using simulating PM2.5 concentration across the U.S. with CMAQV5.3.2. Although I managed to run CMAQ and got the simulated results, the model performance was not as good as I expected when comparing it with the observed PM concentration from the AQS network. I downloaded the met data from the CMAS data warehouse directly and used SMOKE to get the emission input from NEI2015. I did the simulation for the year 2015 and the scripts containing all my settings for CMAQ are attached (run_cctm_201507_12US1.csh (36.1 KB)
). Any suggestions on how to improve the model performance will be so appreciated.
I am trying to look into emission input, met input, and CMAQ settings, respectively. However, as I downloaded the met data from CMAS data warehouse directly and they are processed by MCIP, I am not sure how to compare those met files with any observations, like MADIS dataset. Also, is there any suggestion on how to set CMAQ to get a better model performance for the U.S.under grid 12US1?
Qian, there is not enough information here for people to be able to help you. What sort of performance (beyond “not as good as I expected”) did you get?
If emissions data are available from the CMAS data warehouse, you might try running using those data to see if you get different results. SMOKE is very complicated, with lots of opportunities for user errors that could lead to poor CCTM model performance.
Thanks for replying, @cgnolte. I compared the CMAQ output with the observed total PM2.5 concentration from AQS network (daily average concentration). When comparing all sites in the U.S and looking at a single month, the NMB is about 50% and the NME is greater than 60%, and r is below 0.5. I also downloaded CMAQ ready data from https://gaftp.epa.gov/air/emismod/2015/alpha/2015fd/cmaq_ready/ and ran CMAQ but the emission did not help reduce the bias or error a lot (3% reduction). Please let me know if any other detailed info you need.
Are your results consistently biased to one side or another (always too low or always too high)?
Is there benchmark output from this case that you can obtain from CMAS and compare against your results?
What are you using for boundary conditions?
.
The two plots are the daily average PM concentration during April and August, compared with all AQS sites in Texas. The boundary condition I used was processed (regrid) from CCTM_CONC_v53beta2_intel17.0_HEMIS_cb6r3m_ae7_kmtbr_m3dry_2016_quarterly_av.nc by BCON. I am running CMAQ for 2015 and I did not find any output for 2015 in the data warehouse. Do you have any output from 2015 for me to compare?
Hi Qian,
Here is Apr/August evaluation of daily average PM2.5 for TX for one of our recent 2015 simulations. There are several things different about this simulation compared to your set up, but hopefully this is a helpful reference case. Here is a link with some high level information on the 2015 modeling set up we used:
I just reran the benchmark case (07/01/2016 - 07/02/2016, 12SE1). I compared my output ([1]) with the output provided by CMAS ([2]) and the concentration difference seems not acceptable. Attached are two maps showing the concentration difference and some can be as high as 250% for O3. Does it indicate there was something wrong with my CMAQ? I did run the benchmark case when installing CMAQ in 2020 and I do not think the difference at that time was as big as today’s comparison. Do you have any suggestions on this issue?
Your plots did not come through. However, if your results are very different from the benchmark run, then yes, that is generally an indication of a problem with the simulation – that is the purpose of a benchmark. You might compare a less reactive species, such as CO.
It’s hard to know how to advise you without more information as to what you have done.
Are you using all benchmark data (meteorology, emissions and IC/BC)?
Is this a new system architecture you are running on, or have you successfully run CMAQ previously on this system?
What compiler and version are you using?
Did you modify either the build or run scripts, or the Emission Control Interface file?
Thanks for the suggestions, @cgnolte. I think the big difference was because I did not make biogenic emission online. After fixing this, the difference I got is shown the map below and I think it is reasonable.
So, there should not be any issue with my CMAQ. However, when I ran CMAQ again for August in 2016 with all the input data from CMAS data warehouse, expect for the initial condition files and compared the results with the AQS daily data by AMET, the r value is below 0.1. I have no idea about what is going on here. Do you have any suggestion on how to debug it? Attached is my run_script (run_cctm_201607_12US1.csh (35.7 KB)
).
