GEOS-Chem v11-02 benchmark history

From Geos-chem
Revision as of 16:49, 29 November 2017 by Lizzie Lundgren (Talk | contribs) (v11-02b HP 1-year Rn-Pb-Be)

Jump to: navigation, search

On this page we have posted complete information about all benchmark simulations (both 1-month and 1-year) for GEOS-Chem v11-02.

1-year full-chemistry benchmarks

v11-02c-Run0

This 1-year benchmark simulation was approved by the GEOS-Chem Steering Committee on 21 Sep 2017.

Three GEOS-Chem model versions were compared to each other:

Color Model Version Met Type Year Updates affecting the benchmark simulation Annual Mean OH
[105 molec/cm3]
Black Observations        
Blue v11-02c-Run0 GEOS-FP,
72L, 4x5
2013

Updates introduced in v11-02c:

11.596
Green v11-02a-Run1 GEOS-FP,
72L, 4x5
2013

Updates introduced in v11-02a:

11.829
Red v11-01-public-release-Run0 GEOS-FP,
72L, 4x5
2013

Updates introduced in v11-01 public release:

12.001

The output plots for this run may be downloaded from:

ftp ftp.as.harvard.edu
cd gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02c/Run0/output
mget *

You may also view the PDF files online by pointing your browser to

http://ftp.as.harvard.edu/gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02c/Run0/output/

New plots were added this version for evaluating the two SOA schemes, including:

Please also view the following pages comparing this version to past 1-year benchmarks:

--Melissa Sulprizio (talk) 16:52, 15 September 2017 (UTC)

Comments about the 1-year benchmark v11-02c-Run0

Computing PM2.5 and AOD

Jeff Pierce wrote:

Can you tell us which SOA is being included into the PM2.5 and AOD calculations for the three different simulations? I assume the SOA scheme changes between these different simulations?

Melissa Sulprizio responded:

Following the recommendation of the Aerosol WG leaders, simple SOA is the default in all full-chemistry simulations for v11-02c and later. However, for the benchmark simulations we turn on both complex SOA and simple SOA so that we can continually evaluate the two schemes. That means, when a user downloads v11-02 and creates a run directory for the full-chemistry simulations (tropchem, standard w/ UCX, etc.), input.geos will automatically be set up to use simple SOA. The complex SOA option will only be turned on if a user creates a run directory for the “complexSOA”, “complexSOA_SVPOA”, or “benchmark” simulation. As I mentioned below, by default in aerosol_mod.F, when complex SOA is turned on the complex SOA species will be used in the calculation of PM2.5 and AOD over the simple SOA species to avoid double counting. This is only an issue in the benchmark simulations when both options are turned on – in all other simulations, it will be clear whether complex SOA or simple SOA species are used in the calculations based on the settings in input.geos.
Ocean ALD2

Dylan Millet wrote:

I missed it in the prior benchmark but it seems as part of the PAN update the ocean acetaldehyde source went into the model. That seems fine to me but it is a quite large global reactivity source (can someone tell me if the 66TgC for ALD2osr in the emission summary file is gross or net? I hope gross) so there needs to be traceability for that in the model narrative (e.g. if it's based on Millet et al. 2010 or on something else).

Melissa Sulprizio responded:

The ALD2 ocean source (ALD2osr) in the emissions summary file is gross. HEMCO handles the ALD2 ocean sink and we can’t currently output that as a diagnostic. This is similar to how we currently handle the ACET ocean source/sink. Here are comments from Christoph added during the implementation from HEMCO:
    !  (1 ) in HEMCO, Acetone sink from OCEAN is handled by drydep and the
    !       corresponding diagnostics. If needed, we can write a wrapper that
    !       explicitly calculates the Acetone ocean sink. (ckeller, 08/04/14)
Halogens

Dylan Millet wrote:

There are some large relative changes for many of the halogen species, which I guess is linked to OH/NOx feedbacks from the isoprene chemistry. It would be great if someone close to the halogen stuff can confirm that they're comfortable with the changes.

Mat Evans wrote:

So as Dylan says the halogens change significantly with these updated. I think most of these changes relate to the increases in the oceanic ALD2 source. ALD2 reacts pretty quickly with Br resulting in HBr. So BrOx concentrations tend to decrease and HBr increases.
This is useful as we currently don’t include the sea-salt debromination source. It seems likely that the additional oceanic oVOCs leads to a shorter BrOx lifetime and so might allow for an increase in the Br source consistent with the observational BrO constraint we’ll have to see in future releases.

--Melissa Sulprizio (talk) 14:53, 25 September 2017 (UTC)

v11-02a-Run1

This 1-year benchmark simulation was an unofficial benchmark to validate removal partitioning of NOx biomass burning emissions directly to PAN and HNO3 from v11-02a.

Three GEOS-Chem model versions were compared to each other:

Color Model Version Met Type Year Updates affecting the benchmark simulation Annual Mean OH
[105 molec/cm3]
Black Observations        
Blue v11-02a-Run1 GEOS-FP,
72L, 4x5
2013

Updates introduced in v11-02a:

11.829
Green v11-01-public-release-Run0 GEOS-FP,
72L, 4x5
2013

Updates introduced in v11-01 public release:

12.001
Red v11-01k-Run0 GEOS-FP,
72L, 4x5
2013

Updates introduced in v11-01h:

Updates introduced in v11-01i:

Updates introduced in v11-01j:

Updates introduced in v11-01k:

12.004

The output plots for this run may be downloaded from:

ftp ftp.as.harvard.edu
cd gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02a/Run1/output
mget *

You may also view the PDF files online by pointing your browser to

http://ftp.as.harvard.edu/gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02a/Run1/output/

Please also view the following pages comparing this version to past 1-year benchmarks:

--Melissa Sulprizio (talk) 16:37, 1 June 2017 (UTC)

v11-02a-Run0

This 1-year benchmark simulation was approved by the GEOS-Chem Steering Committee on 12 May 2017, with the recommendation to remove partitioning of NOx biomass burning emissions directly to PAN and HNO3.

Three GEOS-Chem model versions were compared to each other:

Color Model Version Met Type Year Updates affecting the benchmark simulation Annual Mean OH
[105 molec/cm3]
Black Observations        
Blue v11-02a-Run0 GEOS-FP,
72L, 4x5
2013

Updates introduced in v11-02a:

11.750
Green v11-01-public-release-Run0 GEOS-FP,
72L, 4x5
2013

Updates introduced in v11-01 public release:

12.001
Red v11-01k-Run0 GEOS-FP,
72L, 4x5
2013

Updates introduced in v11-01h:

Updates introduced in v11-01i:

Updates introduced in v11-01j:

Updates introduced in v11-01k:

12.004

The output plots for Run0 may be downloaded from:

ftp ftp.as.harvard.edu
cd gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02a/Run0/output
mget *

You may also view the PDF files online by pointing your browser to

http://ftp.as.harvard.edu/gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02a/Run0/output/

Please also view the following pages comparing this version to past 1-year benchmarks:

--Melissa Sulprizio (talk) 18:14, 10 May 2017 (UTC)

1-month benchmarks

v11-02c

This section includes the assessment form for the 1-month benchmark simulation of v11-02c.

Description
New features added into GEOS-Chem
Feature Submitted by
Features affecting the full-chemistry simulation:
Enhance default GEOS-Chem simple SOA Sal Farina (Colorado State)
Aerosols Working Group Chairs
Updates to isoprene and monoterpene chemistry Katie Travis (MIT)
Jenny Fisher (U. Wollongong)
Eloïse Marais (U. Birmingham)
Christopher Chan Miller (Harvard)
Kelvin Bates (Caltech)
Rebecca Schwantes(Caltech)
Add aqueous isoprene uptake to SOA scheme Eloïse Marais (U. Birmingham)
Carbon balance (fix C creation) Sarah Safieddine (MIT)
Fix bugs for EOH and MGLY following implementation of PAN updates in v11-02a Melissa Sulprizio (GCST)
Update HEMCO from v2.0.004 to v2.1.001 Christoph Keller (NASA GMAO)
Features not affecting the full-chemistry simulation:
HEMCO updates:
Christoph Keller (NASA GMAO)
Andy Jacobson (NOAA)
Paulo Tuccella (L'Aquila)
Fixes for several minor issues:
Chris Holmes (Florida State)
Chris Holmes (Florida State)
Jiawei Zhuang (Harvard)
Amanda Giang (MIT)
Katie Travis (MIT)
Katie Travis(MIT)
Fixes for the TOMAS simulation:
Sal Farina (Colorado State)
Jack Kodros (Colorado State)
Bob Yantosca (GCST)
Fix STE flux diagnostic and add to benchmark procedure Melissa Sulprizio (GCST)
Initial structural modifications for netCDF diagnostics:
  • Introduce Headers/State_Diag as a stub module (for now)
  • Add a registry object into State_Met, State_Diag, and State_Chm in order to obtain a pointer to any module variable (or slice) by looking up its name
  • Add new module Headers/registry_mod.F90 which contains derived types and routines for registering module variables.
Bob Yantosca (GCST)
Combine timestep settings in input.geos in a Timesteps menu Melissa Sulprizio (GCST)
Update CO data used in 1-year benchmark plots Jenny Fisher (U. Wollongong)
Version, resolution, met fields used: v11-02, GEOS-FP (72L), 4x5, July 2013
1-month benchmark finished on: Wed Sep 6 00:02:51 EDT 2017
Performance statistics:
  • Ran on 24 CPUs of holyjacob01.rc.fas.harvard.edu (Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50 GHz)
  • Wall time: 05:23:05
  • CPU time / wall time: 22.3511
  • % of ideal performance: 93.13%
  • Memory: 5.8683 GB
Compared to previous benchmark: v11-02b
This update will impact:
(select all that apply with boldface)
Advection, BL Mixing, Convection, Met Fields, Dry Dep, Wet Dep, Stratosphere, Anthro Emiss, Biogenic Emiss, Biomass Emiss, Photolysis, Chemistry, Other (please specify):
Unit test results may be viewed at: http://ftp.as.harvard.edu/gcgrid/geos-chem/1mo_benchmarks/v11-02/v11-02c/v11-02c.results.html
Plots may be viewed at: http://ftp.as.harvard.edu/gcgrid/geos-chem/1mo_benchmarks/v11-02/v11-02c/
Comparison with high performance option enabled:
(new in v11-02)
The GCHP benchmark is in progress. Results will be posted on this wiki page when they are available.
Metrics
Global mean OH from log file (% change): 12.3036776026986 x 105 molec/cm3 (-0.63 %)
Methyl chloroform lifetime (% change): 5.1288 years (1.03 %)
O3 STE flux (% change): 585.2005 Tg O3/year (-42.77%)
At the SURFACE, list all species that changed by 10% or more: NO, O3, PAN, ALK4, ISOP, HNO3, H2O2, MEK, ALD2, RCHO, MVK, MACR, NPMN, PPN, R4N2, PRPE, C3H8, CH2O, N2O5, HNO4, MP, DMS, SO2, SO4, NH3, NH4, NIT, NITs, Br2, Br, BrO, HOBr, HBr, BrNO2, BrNO3, MPN, ISOPND, ISOPNB, MOBA, PROPNN, HAC, GLYC, MVKN, MACRN, MAP, NO2, NO3, HNO2, BENZ, TOLU, XYLE, MTPA, LIMO, MTPO, TSOG0-3, TSOA0-4, ISOG1-3, ISOA1-3, ASOG1-3, ASOAN, ASOA1-3, EOH, MGLY, BrCl, HCl, Cl, ClO, HOCl, ClNO3, ClNO2, ClOO, OClO, Cl2, Cl2O2, OH, HO2
Comments on SURFACE differences: Below we summarize the notable changes caused by specific updates.

Enhance default GEOS-Chem simple SOA

  • The following species are new in this version and show up as large positive ratios: SOAP, SOAS. See this table for species descriptions.

Updates to isoprene and monoterpene chemistry, which includes aqueous isoprene uptake

  • The majority of the large differences in species concentrations are caused by these chemistry updates. These updates included adjusting reaction rates, modifying products, and adding heterogeneous chemistry and greatly impacted species like ISOPND, ISOPNB, PROPNN, HAC, GLYC, MVKN, MACRN, R4N2.
  • PMN is now split into NPMN and IPMN (PMN from non-isoprene and isoprene sources). This is shown as a large decrease in NPMN in the ratio plots.
  • Changes in MTPA and MTPO concentrations are caused by updates to reaction rates in the monoterpene chemistry originally introduced in v11-02a.
  • The large decrease in LIMO is caused by the addition of gas-phase reactions of limonene with OH, O3, and NO3. This decrease also contributes to large decreases in SOA products (TSOA/G*) produced by oxidation of LIMO.
  • The decrease in ISOG2 and ISOA2 may be explained by the removal of the ISOP lost to reaction with NO3. This pathway was removed from the SOA code in the aqueous isoprene uptake updates.
  • The following species are new in this version and show up as large positive ratios: GLYX, ACTA, HPALD, DHDN, ETHLN, HCOOH, IEPOXA, IEPOXB, IEPOXD, ISN1, RIPA, RIPB, RIPD, IMAE, SOAIE, SOAME, SOAGX, SOAMG, LVOC, LVOCOA, ISN1OG, ISN1OA, MONITS, MONITU, HONIT, IONITA, MONITA, INDIOL, IPMN, HC187. See this table for species descriptions.

Carbon balance (fix C creation)

  • The large decrease in R4N2 globally is caused by (1) replacing R4N2 by PROPNN in the ISNOOB + NO3 and ISNOOB + NO reactions and (2) removing R4N2 as product from the ATO2 + NO reaction.

Fix bugs for EOH and MGLY following implementation of PAN updates in v11-02a

  • A bug in the species definitions of EOH and MGLY meant the concentrations for these species were not getting properly updated after gas-phase chemistry. This is now fixed and results in large % differences in the EOH and MGLY species concentrations

Update HEMCO from v2.0.004 to v2.1.001

  • This update contributes to >10% changes in some species north of 70°N.
  • Species such as DST1-4, SALA, and SALC have small (<10%) but structured changes at the surface. These changes are caused by the bug fixes, I/O updates, and regridding updates added in HEMCO v2.1.
At 500 hPa, list all species that changed by 10% or more: NO, PAN, ALK4, ISOP, HNO3, H2O2, MEK, ALD2, RCHO, MVK, MACR, NPMN, PPN, R4N2, PRPE, CH2O, N2O5, HNO4, MP, DMS, SO2, NH3, NIT, NITs, Br2, Br, BrO, HOBr, HBr, BrNO2, BrNO3, MPN, ISOPND, ISOPNB, MOBA, PROPNN, HAC, GLYC, MVKN, MACRN, MAP, NO2, NO3, HNO2, BENZ, TOLU, XYLE, MTPA, LIMO, MTPO, TSOG1, TSOG3, TSOA1, TSOA3, TSOA0, ISOG1-3, ISOA1-3, EOH, MGLY, BrCl, Cl, ClO, HOCl, ClNO3, ClNO2, ClOO, OClO, Cl2, Cl2O2, OH, HO2
Comments on 500 hPa differences: See comments on surface differences
In the ZONAL MEAN differences, list all species that changed by 10% or more: NO, PAN, ISOP, HNO3, H2O2, ACET, MEK, ALD2, RCHO, MVK, MACR, NPMN, PPN, R4N2, PRPE, CH2O, N2O5, HNO4, MP, DMS, SO2, NH3, NIT, NITs, Br2, Br, BrO, HOBr, HBr, BrNO2, BrNO3, MPN, ISOPND, ISOPNB, MOBA, PROPNN, HAC, GLYC, MVKN, MACRN, MAP, NO2, NO3, HNO2, TOLU, XYLE, MTPA, LIMP, MTPO, TSOG1, TSOA1, TSOA3, ISOG2, ISOA2, EOH, MGLY, BrCl, HCl, Cl, ClO, HOCl, ClNO3, ClNO2, ClOO, OClO, Cl2, Cl2O2, OH, HO2
Comments on ZONAL MEAN differences: See comments on surface differences
In the EMISSION RATIO maps, list all species that changed by 10% or more:
  • Anthropogenic emissions: CO, NH3, NO, SO2, SO4
  • Ship emissions: CO, SO2
Comments on EMISSION RATIO differences: The band of positive differences ~70°N over Europe is caused by the HEMCO update from v2.0.004 to v2.1.001. Christoph Keller wrote:
I added an update to the regridding routines (regrid_a2a_mod.F90) to support non-global grids as well as regridding of missing values. Grid boxes with missing values are now ignored. The old code did not support missing values and would assign a value of 0.0 to all grid boxes with no input value. For regional emission inventories such as EMEP, this meant that all values outside of the EMEP domain were treated as zero's and the regridding did essentially dilute the emissions at the grid boundaries. The new code now ignores the grid boxes outside of the boundaries, which I find a better approach (and which explains the increase at 70N - right where the EMEP domain ends).
Additional or summary comments: Following an initial 1-month benchmark, the GCSC advised we restore the Henry's law constant for H2O2 wet deposition. This value was changed with the updates to isoprene and monoterpene chemistry. Daniel Jacob wrote:
For wetdep of H2O2 we should restore the old Henry’s law constant of 8.3E4exp[7400(1/T – 1/298)] because as Dylan points out that’s the physical value. For drydep of H2O2 we can keep the value of 5E7 as parameterized by Nguyen to fit his drydep data.
Approval
Requires further investigation:
  • Diagnostic issues will be resolved in the 1-year benchmark. This involves adding simple SOA to the PM2.5 diagnostic and fixing total SOA concentrations that are too high (likely from calling OASAVE too often). Both of these issues are manifested in the ND42 diagnostic output.
  • Open item is tracking of SOA in complex scheme and avoiding double-counting. This will be discussed in the Aerosol WG and will be resolved at a later date.
Approved by: Jeff Pierce, Eloise Marais, Jenny Fisher, Katie Travis, Daniel Jacob
Date of approval: 07 Sep 2017

--Melissa Sulprizio (talk) 17:43, 6 September 2017 (UTC)

v11-02b

This section includes the assessment form for the 1-month benchmark simulation of v11-02b.
New in this version: See also the assessment form for the 1-month benchmark of v11-02b with high performance option (GCHP).

Description
New features added into GEOS-Chem
Feature Submitted by
Features not affecting the full-chemistry simulation:
Source code updates for compatibility with high performance option (GCHP) Seb Eastham (Harvard)
Lizzie Lundgren (GCST)
Mike Long (GCST)
Jiawei Zhuang (Harvard)
Bob Yantosca (GCST)
Bug fixes for diagnostics:
Ilya Stanevic (Toronto)
Lee Murray (Rochester)
Write initial and final Ox mass to file when using the tagged O3 simulation Bob Yantosca (GCST)
Convert CO2 emissions units using dry pressure used in advection Meemong Lee (JPL)
Add QFED emissions for 2014-2016 Christoph Keller (NASA GMAO)
Version, resolution, met fields used: v11-02, GEOS-FP (72L), 4x5, July 2013
1-month benchmark finished on: Thurs June 07 23:08:02 EDT 2017
Performance statistics:
  • Ran on 24 CPUs of holyjacob01.rc.fas.harvard.edu (Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50 GHz)
  • Wall time: 04:25:40
  • CPU time / wall time: 22.5723
  • % of ideal performance: 94.05%
  • Memory: 5.0852 GB
Compared to previous benchmark: v11-02a
This update will impact:
(select all that apply with boldface)
Advection, BL Mixing, Convection, Met Fields, Dry Dep, Wet Dep, Stratosphere, Anthro Emiss, Biogenic Emiss, Biomass Emiss, Photolysis, Chemistry, Other (please specify):
Unit test results may be viewed at: http://ftp.as.harvard.edu/gcgrid/geos-chem/1mo_benchmarks/v11-02/v11-02b/v11-02b.results.html
Plots may be viewed at: http://ftp.as.harvard.edu/gcgrid/geos-chem/1mo_benchmarks/v11-02/v11-02b/
Comparison with high performance option enabled:
(new in this version)

We now include a plots subdirectory called gchp_comparison containing 1-month standard simulation output comparisons using the standard "classic" capabilities (GCC) and the high performance option (GCHP). GCHP features the same science as GCC but operates on a cubed-sphere grid and is parallelized using a message-passing interface (MPI) implementation. GCHP improves upon GCC by (1) enabling more accurate transport, and (2) providing efficient scaling making finer resolution global simulations possible.

There is a separate benchmark approval form for the comparison of GCHP and GCC. See the Assessment form for the 1-month benchmark of v11-02b with high performance option.

Metrics
Global mean OH (from log file): 12.3818817518568 x 105 molec/cm3
Methyl chloroform lifetime: 5.0764 years
Did either of these change by more than 5%? No, both values are identical to values of the previous version.
At the SURFACE, list all species that changed by 10% or more: none
Comments on SURFACE differences: none
At 500 hPa, list all species that changed by 10% or more: none
Comments on 500 hPa differences: No differences.
In the ZONAL MEAN differences, list all species that changed by 10% or more: none
Comments on ZONAL MEAN differences: No differences.
In the EMISSION RATIO maps, list all species that changed by 10% or more: none
Comments on EMISSION RATIO differences: none
Additional or summary comments: As expected, merging in the modifications for GCHP has zero impact on the GEOS-Chem "classic" standard simulation.
Approval
Requires further investigation: No
Approved by: Lizzie Lundgren (GCST)
Date of approval: 13 June 2017

--Lizzie Lundgren (talk) 18:54, 13 June 2017 (UTC)

v11-02a

Here is the assessment form for 1-month benchmark simulation v11-02a.

Description
New features added into GEOS-Chem
Feature Submitted by
Features affecting the full-chemistry simulation:
Update chemistry rate constants based on JPL Publication 15-10 Barron Henderson (US EPA),
Mat Evans (U. York), &
Oxidants and Chemistry WG
Fixes to correct ALK4 lumping issue Barron Henderson (US EPA)
PAN updates, including: Emily Fischer (CSU)
Monthly mean NEI2011 emissions GCST &
Katie Travis (Harvard)
Bug fixes in the GEOS-Chem sulfate module:
Prasad Kasibhatla (Duke)
Prasad Kasibhatla (Duke)
Qianjie Chen (UW)
Viral Shah (UW)
Fix bug in dry deposition aerodynamic resistance Brian Boys (Dalhousie)
Fix acetone parameterization in hcox_seaflux_mod.F90 GCST
Bug fix in cos(SZA) for start of timestep Lizzie Lundgren (GCST)
Include TOA pressure when calculating dry pressure edges Seb Eastham (Harvard)
Features not affecting the full-chemistry simulation:
Implement ISORROPIA v2.0 as a Fortran module Seb Eastham (Harvard) &
GCST
Updates to the HEMCO emissions component:
Christoph Keller (NASA GMAO)
Seb Eastham (Harvard)
Jessica Morena (Dalhousie)
Brian Boys (Dalhousie) & GCST
Updates to gain computational speedup:
Mike Long (Harvard)
Bob Yantosca (GCST)
Fixed typo in INIT_WINDOW Bob Yantosca (GCST)
netCDF file I/O updates:
Chris Holmes (Florida State)
Andy Jacobson (NOAA)
GCST
Makefile and build sequence updates:
Jiawei Zhang (Harvard)
GCST
Bug fixes for running UCX in ESMF environment Christoph Keller (NASA GMAO)
Bug fixes for diagnostics:
Aaron van Donkelaar (Dalhousie)
Jenny Fisher (U. Wollongong)
GCST
GCST
Jenny Fisher (U. Wollongong)
Chris Holmes (Florida State)
Removal of obsolete variables: GCST
Version, resolution, met fields used: v11-02, GEOS-FP (72L), 4x5, July 2013
1-month benchmark finished on: Sun Apr 16 01:37:36 EDT 2017
Performance statistics:
  • Ran on 24 CPUs of holyjacob01.rc.fas.harvard.edu (Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50 GHz)
  • Wall time: 04:22:54
  • CPU time / wall time: 22.5896
  • % of ideal performance: 94.12%
  • Memory: 5.2038 GB
Compared to previous benchmark: v11-01 public release
This update will impact:
(select all that apply with boldface)
Advection, BL Mixing, Convection, Met Fields, Dry Dep, Wet Dep, Stratosphere, Anthro Emiss, Biogenic Emiss, Biomass Emiss, Photolysis, Chemistry, Other (please specify):
Unit test results may be viewed at: http://ftp.as.harvard.edu/gcgrid/geos-chem/1mo_benchmarks/v11-02/v11-02a/v11-02a.results.html
Plots may be viewed at: http://ftp.as.harvard.edu/gcgrid/geos-chem/1mo_benchmarks/v11-02/v11-02a/
Metrics
Global mean OH (from log file): 12.3128108012973 x 105 molec/cm3
Methyl chloroform lifetime: 5.1159 years
Did either of these change by more than 5%? No. Mean OH changed by -2.57% and MCF lifetime changed by 2.94%. These changes were primarily caused by two updates:
  • The JPL 15-10 update changed mean OH by -1.03% and MCF lifetime by 0.58%.
  • The PAN updates changed mean OH by -1.62% and MCF lifetime by 2.41%
At the SURFACE, list all species that changed by 10% or more: NO, O3, PAN, CO, ALK4, ISOP, HNO3, H2O2, ACET, MEK, ALD2, RCHO, MVK, MACR, PMN, PPN, R4N2, PRPE, C3H8, CH2O, C2H6, N2O5, HNO4, MP, DMS, SO2, SO4, SO4s, MSA, NH3, NH4, NIT, NITs, BCPI, OCPI, BCPO, OCPO, Br2, Br, BrO, HOBr, HBr, BrNO2, BrNO3, CHBr3, MPN, ISOPND, ISOPNB, MOBA, PROPNN, HAC, GLYC, MVKN, MACRN, RIP, IEPOX, MAP, NO2, NO3, HNO2, BrCl, HCl, Cl, ClO, HOCl, ClNO3, ClOO, OClO, Cl2, Cl2O2, MTPA, LIMO, MTPO, TSOG1, TSOG2, TSOG2, TSOG0, TSOA1, TSOA2, TSOA3, TSOA0, ISOG1, ISOG2, ISOG3, ISOA1, ISOA2, ISOA3, BENZ, TOLU, XYLE, ASOG1, ASOG2, ASO3G3, ASOAN, ASOA1, ASOA2, ASOA3, OH, HO2
Comments on SURFACE differences:

Below we summarize the notable changes caused by specific updates.

Update chemistry rate constants based on JPL Publication 15-10

  • Notable changes caused by this update:
    • >10% increase in MVK, PRPE, N2O5, Cl2O2
    • >10% decrease in MACR, PMN, CHBr3
  • Matt Evans wrote
    The only major changes [from JPL 15-10] are to do with the MVK / MACR + OH changes we expected, changes in the PRPE +OH and changes in the N2O5 / HO2NO2 formation / loss rates. Unless I’ve missed something I think we are happy with all of this.

PAN updates

Monthly mean NEI2011 emissions

  • This update caused numerical noise differences over the US in NO, ALK4, RCHO, N2O5, DMS, SO2, NH3, NIT, BCPO, OCPO, NO2, NO3, HNO2, BENZ, TOLU, XYLE.

Bug fixes in the GEOS-Chem sulfate module:

  • The bug fix in SEASALT_CHEM caused a decrease in SO4s and NITs globally.
  • The bug fix in CHEM_NIT did not cause any changes that amount to >10%.
  • The fix for sulfate production in HET_DROP_CHEM caused an increase in SO2, NH4, and NITs over the oceans and a decrease in SO4, SO4s, NH3, and NITs over the oceans
  • The bug fixes in sulfate chemistry routines caused changes in SO2, SO4, SO4s, NH3, NH4, NIT, and NITs.
    • Viral Shah wrote:
      The effects of my proposed fixes are a bit more pronounced during the NH winter. The plots show significant increases in sulfate concentrations over large areas of the NH. The SO4 increases are accompanied by decreases in NIT in some areas. The increase in SO4 in the cold areas of NH is mostly because of the correction to HNO3 effective Henry's law coefficient that leads to higher cloud water pH (between 4 and 5) and higher production by in-cloud oxidation by O3. The lower SO4 concentrations in the SH are probably due to the fix to ∆H/R parameter. I expect the effects [of the LWC units fix] to be relatively small.

Fix bug in dry deposition aerodynamic resistance

  • This update caused slight decreases in many species at the surface. Species changed by >10% from this update include HNO3, H2O2, N2O5, and NIT.

The following updates impacted species concentrations slightly, but the changes did not amount to >10%.

At 500 hPa, list all species that changed by 10% or more: NO, PAN, CO, ALK4, ISOP, HNO3, H2O2, ACET, MEK, ALD2, RCHO, MVK, MACR, PMN, PPN, R4N2, PRPE, C3H8, CH2O, C2H6, N2O5, HNO4, MP, DMS, SO2, SO4, SO4s, MSA, NH3, NH4, NIT, NITs, BCPI, OCPI, BCPO, OCPO, Br2, Br, BrO, HOBr, HBr, BrNO2, BrNO3, CHBr3, MPN, ISOPND, ISOPNB, MOBA, PROPNN, HAC, GLYC, MVKN, MACRN, RIP, IEPOX, MAP, NO2, NO3, HNO2, BrCl, Cl, ClO, HOCl, ClNO3, ClOO, OClO, Cl2, Cl2O2, MTPA, LIMO, MTPO, TSOG1, TSOG2, TSOG2, TSOG0, TSOA1, TSOA2, TSOA3, TSOA0, ISOG1, ISOG2, ISOG3, ISOA1, ISOA2, ISOA3, BENZ, TOLU, XYLE, ASOG1, ASOG2, ASO3G3, ASOAN, ASOA1, ASOA2, ASOA3, OH, HO2
Comments on 500 hPa differences: See comments on surface differences.
In the ZONAL MEAN differences, list all species that changed by 10% or more: NO, PAN, ALK4, ISOP, HNO3, H2O2, MEK, ALD2, RCHO, MVK, MACR, PMN, PPN, PRPE, C3H8, CH2O, N2O5, HNO4, MP, DMS, SO2, SO4, SO4s, NH3, NH4, NIT, NITs, BCPI, OCPI, BCPO, OCPO, Br2, Br, BrO, HOBr, HBr, BrNO2, CHBr3, MPN, ISOPND, ISOPNB, MOBA, HAC, GLYC, MVKN, MACRN, RIP, IEPOX, MAP, NO2, NO3, HNO2, BrCl, Cl, ClO, HOCl, ClNO3, ClOO, OClO, Cl2, Cl2O2, MTPA, LIMO, MTPO, TSOG1, TSOG2, TSOG3, TSOG0, TSOA1, TSOA2, TSOA3, TSOA0, ISOG1, ISOG2, ISOG3, ISOA1, ISOA2, ISOA3, BENZ, TOLU, XYLE, ASOG1, ASOG2, ASOG3, ASOAN, ASOA1, ASOA2, ASOA3, OH, HO2
Comments on ZONAL MEAN differences: See comments on surface differences.
In the EMISSION RATIO maps, list all species that changed by 10% or more:
  • NO: Anthropogenic+biofuel, biomass burning, and fertilizer
Comments on EMISSION RATIO differences:
  • The differences in NO emissions can be explained by the following updates:
    • NO biomass burning emissions decrease because NO is partitioned directly to 40% PAN and 20% HNO3.
    • Changes in NO anthropogenic emissions are isolated to the oceans, indicating a change in ship emissions. Changes in ship emissions may result from slight changes in NO or NO2 concentrations.
    • Differences in fertilizer emissions are caused by changes in NOx concentrations.
  • Anthropogenic emissions over US change slightly (<10%) because of the change from hourly NEI2011 emissions to monthly mean NEI2011 emissions. This impacts the following species: ALD2, ALK4, BC, C2H6, C3H8, CH2O, CO, MEK, NH3, NO, OC, PRPE, SO2, SO4.
  • Slight (<10%) differences can be seen in the biomass burning emission difference plots. This is numerical noise caused by injecting biomass burning emissions at different levels (65% to the boundary layer, 35% to the free troposphere).
  • ACET ocean source decreases slightly (<10%) because of the fix for acetone parameterization in hcox_seaflux_mod.F90.
Additional or summary comments: Removal of biomass burning updates

After the 1-month benchmark for v11-02a, the GCSC decided we should remove the code updates for vertically distributed biomass burning emissions (see this wiki post for more details). In addition, after the 1-year benchmark v11-02a-Run0, the GCSC asked that we comment out the partitioning of NOx biomass burning emissions directly to PAN and HNO3 (see this wiki post for more details). An unofficial 1-month benchmark was run for v11-02a to evaluate the model changes with these two biomass burning updates removed. The plots for that run may be viewed at:

http://ftp.as.harvard.edu/gcgrid/geos-chem/1mo_benchmarks/v11-02/v11-02a_final/

Results from our v11-02a benchmark using the GNU Fortran Compiler

In order to evaluate the performance of GEOS-Chem using the the free and open source GNU Fortran compiler, we performed an additional 1-month benchmark for v11-02a. For this additional benchamark, we used GNU Fortran v6.2.0 to compile GEOS-Chem instead of our usual Intel Fortran Compiler version 11.1.069.

For more details, see this post on our GNU Fortran compiler wiki page.

Summary: The benchmark using GNU Fortran yielded essentially identical results to the benchmark using Intel Fortran. This is very encouraging, as it will allow GEOS-Chem development to take place on computational platforms that do not have proprietary compilers (such as Intel Fortran or PGI Fortran), which can be prohibitively expensive to purchase.

Approval
Requires further investigation: Yes, a 1-year benchmark simulation
Approved by: Emily Fischer, Prasad Kasibhatla, Brian Boys, Daniel Jacob
Date of approval: 24 Apr 2017

--Melissa Sulprizio (talk) 19:27, 20 April 2017 (UTC)

1-year Rn-Pb-Be benchmarks

v11-02b-RnPbBe

Two 1-year Rn-Pb-Be simulations were performed using GEOS-Chem v11-02b. The simulations utilized 4° x 5° GEOS-FP met fields for the year 2013, with a 4-year spinup. For comparison of the Pb-210 and Be-7 budgets to previous versions, please see the following posts on the Rn-Pb-Be simulation wiki page:

  1. Budget of Pb210 from 1-year benchmark simulations
  2. Budget of Be7 from 1-year benchmark simulations

Please note that we include a passive species to test for mass conservation in GEOS-Chem transport. The passive tracer continues to show that mass is not conserved when the non-local PBL mixing scheme (VDIFF) is turned on. We first discovered this issue in v11-01 and continue to complete the 1-year Rn-Pb-Be simulations with and without non-local PBL mixing to track this issue.

You may view the benchmark plots for the simulation by pointing your browser to the following links.

Using the non-local PBL mixing scheme (VDIFF):
http://ftp.as.harvard.edu/gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02b/RnPbBePasv/RnPbBePasv_VDIFF/output/

Using the full PBL mixing scheme (TURBDAY):
http://ftp.as.harvard.edu/gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02b/RnPbBePasv/RnPbBePasv_TURBDAY/output/

--Lizzie Lundgren (talk) 16:49, 24 July 2017 (UTC)

High Performance Option benchmarks

v11-02b HP 1-month

This section includes the assessment form for the 1-month benchmark simulation GEOS-Chem v11-02b with high performance v11-02b. GCHP features the same science as GCC but operates on a cubed-sphere grid and is parallelized using a message-passing interface (MPI) implementation. GCHP improves upon GCC by (1) enabling more accurate transport, and (2) providing efficient scaling making finer resolution global simulations possible.
See also the assessment form for the 1-month v11-02b benchmark simulation using the standard "classic" capability.

Description
New features added:

As this is the first formal GCHP benchmark, new features include the high performance model capability in its entirety. This includes all recent feature updates included in GCHP v11-02b as well as updates in v1.0.0 that were introduced over many years and are too numerous to list.

GEOS-Chem version, GCHP version tag: v11-02, hp_v1.1.0
Meteorology fields: GEOS-FP (72L), 2x2.5, July 2013

NOTE: Input 2x2.5 met-fields are regridded to cubed-sphere grid resolution c24 (approximately equivalent to 4x5) within GCHP. In contrast, the GCC simulations in this comparison use input 4x5 met-fields.

Cubed-sphere resolution: c24
1-month benchmark finished on: Thurs June 12 22:01:08 EDT 2017
Performance statistics:
  • Ran on 24 CPUs of holyjacob01.rc.fas.harvard.edu (Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50 GHz)
  • Wall time: 03:33:49
  • CPU time / wall time: 23.0338
  • % of ideal performance: 95.97%
  • Memory: 5.0852 GB
Compared to benchmark: GEOS-Chem v11-02b with "classic" capability
Additional comments:

We compared GEOS-Chem with and without the high performance option (GCHP and GCC respectively) for two different scenarios:

  1. Out-of-the-box
  2. Custom

The out-of-the-box comparison uses as-is code and run directory settings for both GCHP and GCC. Each simulation uses restart files regridded from higher resolution 2x2.5. GCHP uses input 2x2.5 met-fields which are regridded to cubed-sphere grid resolution c24 (approximately equivalent to 4x5) within the model. In contrast, the GCC simulations in this comparison use input 4x5 met-fields. GCC is compiled with netcdf diagnostics turned on and binary diagnostics turned off. GCHP output data are regridded from c24 (cubed-sphere grid) to 1x1.25 and GCC output data are regridded from 4x5 to 1x1.25 for comparison plotting.

The custom comparison includes additional modifications to the GCC code base and run directory to remove certain remaining known differences between GCHP and GCC. These include the following changes to GCC:

  1. Turn off OTD-LIS factors and set OTD-LIS scaling to 0.5568042545
  2. Turn off all 2D initial mixing ratios except for stratospheric H2O
  3. Turn off usage of the HEMCO restart file
  4. Force cos(SZA) at timestep mid-point to equal cos(SZA) calculated at timestep start

NOTE: Non-concentration diagnostics are not yet implemented in GCHP. The v11-02b HP benchmark therefore includes only comparisons of species concentrations. A complete set of diagnostics will be available pending implementation of netCDF diagnostics in GEOS-Chem (in progress).

Plots may be viewed at: http://ftp.as.harvard.edu/gcgrid/geos-chem/1mo_benchmarks/v11-02/v11-02b/gchp_comparisons/
Metrics
Global mean OH: Diagnostic not yet implemented. See note in additional comments section above.
Methyl chloroform lifetime: Diagnostic not yet implemented. See note in additional comments section above.
Did either of these change by more than 5%? N/A
At the SURFACE, list all species that changed by 10% or more:
  • Out-of-the-box: All species changed by 10% or more except CH2Br2, CH3Br, N2O, OCS, CH4, CCl4, CH3Cl, CH3CCl3, CFC113, CFC114, CFC115, HCFC141b, HCFC142b, CFC11, CFC12, HCFC22, H1211, H1301, and H2404.
  • Custom: All species except those listed above and HCFC123.
Comments on SURFACE differences:

Sources of general differences include the following:

  1. Many species are impacted by different met-fields: GCC uses 4x5 lat/lon while GCHP uses 2x2.5 regridded to cubed-sphere c24 (approximately 4x5). Differences due to regridding also extends to other inputs such as emissions.
  2. Differences are greatest along coastlines for many species. Gradients are generally greater along coastlines and steep gradients impart the strongest differences in regridding. Differences in land cover will also impact the differences along coastlines.
  3. Large fractional differences occur where concentrations are very low and can be attributed to numerical noise.
  4. All differences are further amplified by regridding GCHP output from cubed-sphere resolution to lat/lon for comparison with GCC. Characteristics of output regridding differences include banded patterns of alternating positive and negative differences.
  5. Several large differences between out-of-the-box GCC and GCHP are not present or reduced in the custom run comparison due to turning off initial mixing ratios in the GCC custom run. This includes HCFC123 and differences in chlorinated species away from the poles (e.g. BrCl, ClNO2, and Cl2). Removal of the initial mixing ratios is slated for inclusion in GCC v11-02e.
  • Differences specific to particular species are as follows:
  1. Numerical noise differences in NH3 and NIT are due to ISORROPIA.
  2. Differences in dust species (DST1, DST2, DST3, DST4) are especially impacted by the effects of differences in met fields. Dust tuning factors and emissions patterns are both dependent on met fields. Furthermore, edge cases where dust is subject to small local emissions based on met conditions are very sensitive to regridding, whereas consistent emissions locations (e.g. over the Middle East) will produce dust in similar quantities on any grid.
  3. Differences in isoprene (ISOP) and related biogenic species (e.g. MEK, MVK, MACR, PMN, and PRPE) are also especially impacted by differences in met fields. The MEGAN biogenic emissions module is sensitive to grid resolution because emissions are estimated based on several met fields.
  4. Differences in NO are influenced by differences in PARANOX ship emissions, soil NOx emissions, and fertilizer NOx emissions, all of which use NOx concentrations as input. This will also be affected by resolution-specific lightning factors.
  5. There are notable differences in chlorinated species (e.g. BrCl, ClNO2, Cl2) at the poles which are due to the differences in vertical transport between lat/lon and cubed-sphere grids, specifically since GCC estimates faster vertical mixing at the poles.
At 500 hPa, list all species that changed by 10% or more: All species changed by 10% or more except O3, CO, ACET, and all species listed above for the surface.
Comments on 500 hPa differences: See comments on surface differences above.
In the ZONAL MEAN differences, list all species that changed by 10% or more: All species changed by 10% or more.
Comments on ZONAL MEAN differences:
  • See comments on surface differences above.
  • H2O is consistently lower in GCHP in the stratosphere. This could be the result of differences in stratosphere/troposphere exchange, differences in convection, or GMAO mass scaling used for mass conservation. These would also all affect stratospheric CFC115.
Additional or summary comments:

Overall, the largest fractional differences between GCC and GCHP occur where concentrations are very low. Differences between GCHP and GCC are primarily due to the effects of the following:

  1. Different met fields (2x2.5 lat/lon regridded to cubed-sphere in GCHP versus 4x5 lat/lon in GCC)
  2. Higher-order vertical transport in GCHP due to a more recent version of tpcore
  3. Transport on a cubed-sphere grid versus a cartesian grid
  4. Regridding GCHP inputs and outputs to and from the cubed-sphere grid

The concentration differences we see are inevitable given the fundamental differences between GEOS-Chem with and without high performance capability.

Approval
Requires further investigation: Yes, a 1-year benchmark simulation
Approved by: Daniel Jacob
Date of approval: 16 June 2017

--Lizzie Lundgren (talk) 18:05, 13 June 2017 (UTC)

v11-02b HP 1-year Standard

v11-02b-HP-Run1

This 1-year benchmark simulation is an unofficial benchmark to help validate GEOS-Chem with the high performance option enabled. It uses offline lightning and dust emissions developed to minimize differences due to lightning and dust and provide a better understanding of differences observed in v11-02b-HP-Run0.

We compared two GEOS-Chem model versions for this 1-year benchmark:

Color Model Version Met Type Year Updates affecting the benchmark simulation Annual Mean OH
[105 molec/cm3]
Green v11-02b-HP-Run0 GEOS-FP,
72L, 2° x 2.5° regridded online to c24
2013 Updates introduced in GCHP v11-02b:
  • The high performance model capability in its entirety.
  • Differences from v11-02b-HP-Run0:
    • v11-02b-HP-Run1 differs from v11-02b-HP-Run0 by using offline dust and lightning emissions developed by Seb Eastham (Harvard) to reduce error associated with lightning and dust. This serves to better understand the sources of differences observed in v11-02b-HP-Run0.
    • Lightning and dust emissions are pre-calculated by storing the hourly average flux from a previous 4x5 simulation using GEOS-FP meteorology. Lightning NO mass flux is stored as an hourly 3D field at 4x5 and dust emissions are stored as four hourly 2D fields, one for each of the four dust tracers. Since GCHP conservatively regrids the emissions rates to the C24 grid, using these 4x5 emissions ensures that the temporal and spatial distributions are well matched albeit with some diffusion due to the regridding process.
    • See also the resolution dependency of emissions open issue description on the GEOS-Chem HP v11-02 wiki page.
N/A
Red v11-02a-Run1 GEOS-FP,
72L, 4° x 5°
2013 Updates introduced in /GEOS-Chem v11-02a: 11.829
Black Observations        

The output plots for Run1 may be downloaded from:

ftp ftp.as.harvard.edu
cd gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02b-HP/Run1/output
mget *

You may also view the PDF files online by pointing your browser to:

http://ftp.as.harvard.edu/gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02b-HP/Run1/output/

--Lizzie Lundgren (talk) 16:40, 27 July 2017 (UTC)

v11-02b-HP-Run0

This 1-year benchmark simulation is the first official benchmark to validate GEOS-Chem with the high performance option enabled.

We compared two GEOS-Chem model versions for this 1-year benchmark:

Color Model Version Met Type Year Updates affecting the benchmark simulation Annual Mean OH
[105 molec/cm3]
Green v11-02b-HP-Run0 GEOS-FP,
72L, 2° x 2.5° regridded online to c24
2013 Updates introduced in GCHP v11-02b:
  • The high performance model capability in its entirety. As discussed in the v11-02b HP 1-month benchmark assessment form, notable features of GEOS-Chem HP that result in expected differences with the GCC benchmark results include:
    • Use of different met fields, specifically 2° x 2.5° lat/lon regridded to cubed-sphere in GCHP versus 4° x 5° lat/lon in GCC
    • Higher-order vertical transport in GCHP due to a more recent version of tpcore
    • Transport on a cubed-sphere grid versus a cartesian grid
    • Regridding GCHP inputs and outputs to and from the cubed-sphere grid
    • Lack of availability of OTD-LIS redistribution grids for cubed-sphere resulting in lightning NOX distribution differences. This is part of a larger issue of the resolution dependency of emissions.
  • There are no updates introduced in v11-02b that impact the full-chemistry simulation.
N/A
Red v11-02a-Run1 GEOS-FP,
72L, 4° x 5°
2013 Updates introduced in GEOS-Chem v11-02a: 11.829
Black Observations        

The output plots for Run0 may be downloaded from:

ftp ftp.as.harvard.edu
cd gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02b-HP/Run0/output
mget *

You may also view the PDF files online by pointing your browser to:

http://ftp.as.harvard.edu/gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02b-HP/Run0/output/

Please also view the following pages comparing this version to past 1-year benchmarks:

--Lizzie Lundgren (talk) 15:30, 12 July 2017 (UTC)

v11-02b HP 1-year Rn-Pb-Be

This 1-year benchmark simulation is the first official benchmark to validate GEOS-Chem with the high performance option enabled.

We compared 1-year Rn-Pb-Be simulation output of GEOS-Chem v11-02b with "classic" capability (GCC) and GEOS-Chem v11-02b with the high performance option (GCHP) for this 1-year benchmark:

Versions Compared Internal Resolution Met Source Met Resolution Year Notes
GEOS-Chem Classic v11-02b 72L, 4° x 5° GEOS-FP 72L, 4° x 5° 2013 See also the Rn-Pb-Be 1-year benchmark for GEOS-Chem v11-02b which compares v11-02b Rn-Pb-Be results with previous version v11-01i
GCHP v11-02b 72L, c24 (cubed-sphere) GEOS-FP 72L, 2° x 2.5° regridded online to c24 2013 Notable features of v11-02b HP that result in expected differences with the v11-02b Rn-Pb-Be simulation include:
  • Use of different met fields, specifically 2° x 2.5° lat/lon regridded to cubed-sphere in GCHP versus 4° x 5° lat/lon in GCC
  • Higher-order vertical transport in GCHP due to a more recent version of tpcore
  • Transport on a cubed-sphere grid versus a cartesian grid
  • Regridding GCHP inputs and outputs to and from the cubed-sphere grid

You may view the benchmark comparison plots for the simulation by pointing your browser to the following links.

Using the non-local PBL mixing scheme (VDIFF):
http://ftp.as.harvard.edu/gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02b-HP/RnPbBePasv/VDIFF/output/

Using the full PBL mixing scheme (TURBDAY):
http://ftp.as.harvard.edu/gcgrid/geos-chem/1yr_benchmarks/v11-02/v11-02b-HP/RnPbBePasv/TURBDAY/output/

Please note that this benchmark does not include comparisons of Pb-210 and Be-7 budgets or passive tracer mass conservation because the required diagnostics are not yet output by GCHP.

--Lizzie Lundgren (talk) 17:57, 25 July 2017 (UTC)