The second order ozone sensitivity coefficients to NOx and VOC is too high (e.g. -0.0099 ~ 0.4557 ppmV) from CMAQv5.4 HDDM. Second cross 2nd order sensitivity to NOx,VOC is even more higher.
The first order sensitivity looks fine.
I have yet to figure out why. I do wonder if some users have the same issue and how to fix it.
DDM3D options in CMAQ run script can be seen as follows CMAQ_run_script.txt (972 Bytes)
Please find sensitivity control file and snapshot of the first and second order sensitivity to NOx in the 4-km grid.
What values do you expect? You can try to calculate brute force 2nd order by running a +10% and a -10% simulation. That would give you something to compare against.
I ran CMAQv5.2-HDDM and CMAQv5.4-HDDM with the same inputs. I calculated the ozone design values (DVF) for one specific site using sensitivity coefficients from CMAQv5.2-HDDM and CMAQv5.4-HDDM, respectively, based on 10% reduction of NOx and VOC. Please see the results as below,
CMAQv5.2-HDDM gives more reasonable results than CMAQv5.4-HDDM does. Ozone design value changes from 75.30 ppb to 8297.79 ppb after reducing 10% VOC and NOx, it is definitely wrong. That is because the second order and cross second sensitivities from CMAQv5.4 are too high, You can see from second figure above that the maximum 2nd ozone sensitivity coefficient to NOx reaches 0.455756 ppm (455.756 ppb). It is much higher than the maximum 2nd ozone sensitivity coefficient to NOx 0.00283443ppm from CMAQv5.2_HDDM (see attached figure as follows). The average 2nd order are also too high compared with CMAQv5.2. According to the Taylor series expansion, the equation as below, is used to calculate concentration changes based on the first, second, and cross sensitivity coefficients.
I think CMAS needs to double check 2nd order calculations in CMAQv5.4. Something is wrong. I am wondering if we tested CMAQv5.4-HDDM with higher orders calculations. If yes, could you please provide the CMAQv5.4-HDD benchmark runs and inputs?
Thank you for the additional details. I will have to take some time to investigate this issue. Higher order coefficients were performing well during model development, so I will go back to those tests using parameter definitions closer to yours. We haven’t been creating and providing second order benchmark data in order to keep the examples more simple and manageable. But I will certainly check for abnormal behavior on the benchmark case and get back to you.
We were able to identify the issue in the code that was causing this problem. We will push through an official fix in a patch shorty. Until then, the required fix is very small and simple.
In the file DDM3D_CHEM.F around line 433, the following change needs to be made:
c PROD(S) = 0.0
PRD(S) = 0.0
The “PROD” reset to zero needs to be changed to the PRD reset to zero.
I apologize for any inconvenience, but this did slip through our testing at some point in the process.
Thank you very much to figure out the issue which makes 2nd order sensitivity coefficient too high. I will update the routine DDM3D_CHEM.F according to your post and run CMAQv5.4/HDDM again with teh same inputs. I will keep you updated on the new runs.
Certainly, Sergey. I have successfully switched to CMAQ v5.4.01. However, I haven’t had the opportunity to test HDDM yet. I plan to do so soon and see how it performs.
Thank you for reminding. I am currently utilizing CMAQ/DDM v5.4.03, which appears to be functioning effectively and has apparently addressed the problem of excessively high second-order sensitivities. Once the model run is completed, I will proceed to compute DVF and assess its reasonability at a later time.
