Resolve time-stepping issue in ELMO when used within WRF-CMAQ and MPAS-CMAQ #181
Conversation
On branch elmo_time_step modified: CCTM/src/driver/ELMO_DATA.F modified: CCTM/src/driver/ELMO_PROC.F modified: CCTM/src/driver/driver.F modified: CCTM/src/init/initscen.F
bnmurphy
force-pushed
the
elmo_time_stepping
branch
from
ec643e2
to
0c9369a
Compare
|
@bnmurphy - did you want me to try and replicate or test this issue and bug fix? |
|
@lizadams - if it's not too much trouble. If it fits within your workflow to do a WRF-CMAQ test on your system, that would be helpful. But if not, it would also be good to verify that the offline CMAQ is still working on a non-EPA system. |
|
@lizadams , the issue can be reproduced by simply running two-way WRF-CMAQ. With the current version of the code, all variables in AELMO are low by roughly a factor of 8-10, though the factor depends on the specific setup and calling frequency. The problem is most easily seen by comparing a variable like PMF_SI to ASIJ from ACONC since these should be a 1:1 match, or by looking at meteorological variables like TSURF and PRES in AELMO. After applying the bug fix, AELMO variables have the correct order of magnitude and are consistent between AELMO and ACONC for 1:1 matched variables like PMF_SI or PMF_TI (ATIJ in ACONC), except for numerical noise that's also present when comparing these variables between AELMO and ACONC for the offline model. @bnmurphy , @fisidi , please add or correct anything I might have missed or misstated. |
|
I created VERDI plots showing ATIJ[ACONC]-PMF_TI[AELMO] differences for 4 comparisons. WRF-CMAQ before fix These plots indicate that the fix for WRF-CMAQ was successful and did not impact the CMAQ model output. |
Contact:
Ben Murphy, Center for Environmental Measurement and Modeling, US EPA
Priority: High
Description and/or issue being addressed: ELMO gives erroneous results in WRF-CMAQ and when using met inputs not aligned with hour time steps.
Impact on results: This PR resolves the problem by adjusting the algorithm for identifying when the initial time step is hit for the simulation and synchronization cycles.
Issue:
This Pull Request addresses issue #179
Significance: This model bug fix should be adopted as soon as possible for any user running WRF-CMAQ or using met inputs with time steps not aligned with hourly structure. Simulations for offline CMAQ with hourly met inputs are unaffected.
Tests conducted:
The code has been checked for the 2016 Southeast benchmark domain using 5 min met inputs. It's also been checked on the 2018 NE benchmark with default inputs. Only intel in optimized mode has been run. Reviewers within EPA have run this code for the 12 km CONUS case using WRF-CMAQ and the differences in results between the ELMO and CONC files are of the same magnitude as numerical noise.