tm5 / Blog: Recent posts

Gradient test with coarsened mass fluxes:

~~~~
alpha J_bg J_obs J_tot DJ1 DJ2 DJ2/DJ1 1-DJ2/DJ1
0 4.09800000000e+03 1.62243225768e+03 5.72043225768e+03 3.34908136381e+04 0.00000000000e+00 0 1.00000000e+00
1 1.25921665110e+04 2.11797385352e+06 2.13056602003e+06 3.34908136381e+04 -2.12484558777e+06 -63.445625 6.44456245e+01
0.1 3.44033003739e+03 2.05143848727e+04 2.39547149101e+04 3.34908136381e+04 -1.82342826524e+05 -5.4445625 6.44456246e+00
0.01 4.01716213760e+03 1.58419562407e+03 5.60135776167e+03 3.34908136381e+04 1.19074496006e+04 0.35554375 6.44456246e-01
0.001 4.08976550510e+03 1.59933427536e+03 5.68909978046e+03 3.34908136381e+04 3.13324772131e+04 0.93555437 6.44456252e-02
0.0001 4.09717504342e+03 1.61992971625e+03 5.71710475968e+03 3.34908136381e+04 3.32749799744e+04 0.99355544 6.44456316e-03
1e-05 4.09791748927e+03 1.62218007610e+03 5.72009756537e+03 3.34908136381e+04 3.34692302501e+04 0.99935554 6.44456961e-04
1e-06 4.09799174878e+03 1.62240702024e+03 5.72039876902e+03 3.34908136381e+04 3.34886552801e+04 0.99993555 6.44462705e-05
1e-07 4.09799917488e+03 1.62242973374e+03 5.72042890862e+03 3.34908136381e+04 3.34905977434e+04 0.99999355 6.44638630e-06
1e-08 4.09799991749e+03 1.62243200528e+03 5.72043192277e+03 3.34908136381e+04 3.34907919751e+04 0.99999935 6.46835094e-07
1e-09 4.09799999175e+03 1.62243223244e+03 5.72043222418e+03 3.34908136381e+04 3.34908127115e+04 0.99999997 2.76659264e-08
1e-10 4.09799999917e+03 1.62243225515e+03 5.72043225433e+03 3.34908136381e+04 3.34907781507e+04 0.99999894 1.05961454e-06
~~~~... read more

I'm trying to split a forward run over multiple non-overlapping periods. First, check if the total emission during Jan 1 to Jul 15, 2008 is equal to the emission over the four shorter periods. All emissions below are in Tg CO/yr.

For Jan 1 to Jul 15 (196 days)

~~~~
glb300x200 natural 520.80
glb300x200 biomass burning 225.87
glb300x200 anthropogenic 461.70
~~~~... read more

For CO simulations, I am switching to Sander's latest CH4 fields. Along the way, I'm also modifying the way they are read. The first time they are read, they are coarsened according to the running region definition. During subsequent runs, the coarsened fields are read in. I will convert all of Sander's output fields to the new format, since that is likely the format that will be there going forward.... read more

After my presentation at the GMAC and CT meetings, there are a few correlation ideas that I would like to try out. They are:

postGMAC-01

17 FF regions. Within the US, nine regions broken up according to EIA classification. 500 km hybrid correlation. Land disequilibrium over 23 regions; 10 TRANSCOM regions, and N Am Boreal broken up into CT2013 regions. Regionally coherent correlation, error is 50% of net flux.... read more

In the future, I need to assimilate aircraft data, which are numerous (AMDAR tracks have 10,000 to 15,000 samples per hour). So we need to split the input and output by day or month. I will try this out with the CO+CO2 tracer combination. Without any split, this is the reference gradient test:

~~~~
alpha J_bg J_obs J_tot DJ1 DJ2 DJ2/DJ1 1-DJ2/DJ1
0 2.20800000000e+04 2.18215952442e+04 4.39015952442e+04 2.53570056323e+05 0.00000000000e+00 0 1.00000000e+00
0.1 1.88748350549e+04 3.88561223537e+04 5.77309574085e+04 2.53570056323e+05 -1.38293621644e+05 -0.54538625 1.54538625e+00
0.01 2.16453769801e+04 2.01123813788e+04 4.17577583589e+04 2.53570056323e+05 2.14383688526e+05 0.84546138 1.54538625e-01
0.001 2.20353966328e+04 2.16165471919e+04 4.36519438246e+04 2.53570056323e+05 2.49651419543e+05 0.98454614 1.54538625e-02
0.0001 2.20755282526e+04 2.18007491723e+04 4.38762774249e+04 2.53570056323e+05 2.53178192645e+05 0.99845461 1.54538625e-03
1e-05 2.20795527112e+04 2.18195072243e+04 4.38990599355e+04 2.53570056323e+05 2.53530869955e+05 0.99984546 1.54538627e-04
1e-06 2.20799552700e+04 2.18213864081e+04 4.39013416780e+04 2.53570056323e+05 2.53566137668e+05 0.99998455 1.54539346e-05
1e-07 2.20799955270e+04 2.18215743602e+04 4.39015698872e+04 2.53570056323e+05 2.53569664637e+05 0.99999846 1.54468371e-06
1e-08 2.20799995527e+04 2.18215931558e+04 4.39015927085e+04 2.53570056323e+05 2.53570024506e+05 0.99999987 1.25474873e-07
1e-09 2.20799999553e+04 2.18215950353e+04 4.39015949906e+04 2.53570056323e+05 2.53570040513e+05 0.99999994 6.23479134e-08
1e-10 2.20799999955e+04 2.18215952233e+04 4.39015952188e+04 2.53570056323e+05 2.53568723565e+05 0.99999474 5.25597506e-06
~~~~... read more

Tonatiuh gave us some MESSy OH output fields. I want to compare them vs our "standard" OH fields, as far as CO is concerned. With the standard OH fields, a run from 2010-01-01 to 2011-01-01 gives the following budget (se_asia zoom):

~~~~

---

Budget of tracers in region glb600x400 (Kg)
Tracer : CO CO2

---

Initial mass : 3.25635E+11 3.11579E+15
emitted : 1.82550E+12 1.55714E+13
chemistry : -1.84247E+12 2.89488E+12
update_p : 1.42229E+11 1.24265E+13
advection : -1.12610E-01 -4.90500E+03
deposition : -1.25376E+11 0.00000E+00
stratosphere : 0.00000E+00 0.00000E+00
Final mass : 3.25514E+11 3.14668E+15
Budget error : -2.35901E-01 -4.65550E+03

---

---

Budget of tracers in region sea300x200 (Kg)
Tracer : CO CO2

---

Initial mass : 3.85547E+10 3.00873E+14
emitted : 1.05720E+11 3.98783E+12
chemistry : -8.74943E+10 1.37470E+11
update_p : 1.31208E+11 8.30119E+12
advection : -1.42278E+11 -9.78494E+12
deposition : -7.20405E+09 0.00000E+00
stratosphere : 0.00000E+00 0.00000E+00
Final mass : 3.85054E+10 3.03515E+14
Budget error : -2.91756E-01 -7.79375E+01

---

---

Budget of tracers in region sea100x100 (Kg)
Tracer : CO CO2

---

Initial mass : 2.12912E+10 1.51761E+14
emitted : 2.94469E+11 8.08005E+12
chemistry : -1.40745E+11 2.21138E+11
update_p : 0.00000E+00 0.00000E+00
advection : -1.31252E+11 -7.08243E+12
deposition : -2.25154E+10 0.00000E+00
stratosphere : 0.00000E+00 0.00000E+00
Final mass : 2.12467E+10 1.52980E+14
Budget error : -2.44873E+01 3.31875E+01

---

~~~~... read more

Guillaume implemented a flexible reading of OH in his chemistry__4DVAR.F90, which defines one subroutine per OH format. These subroutines are called from chemistry_init. This is much cleaner than my existing code for CO+CO2. So I'm going to adapt his code for my CO+CO2 system. First, the budget for 2010 with my old code (to verify against once I modify code):

~~~~

---

Budget of tracers in region glb600x400 (Kg)
Tracer : CO CO2

---

Initial mass : 3.32167E+11 2.99212E+15
emitted : 1.82550E+12 1.55714E+13
chemistry : -1.84970E+12 2.90623E+12
update_p : 1.41927E+11 1.24270E+13
advection : 2.28638E-01 8.23000E+02
deposition : -1.25167E+11 0.00000E+00
stratosphere : 0.00000E+00 0.00000E+00
Final mass : 3.24728E+11 3.02302E+15
Budget error : 1.05408E-01 1.06350E+03

---

---

Budget of tracers in region sea300x200 (Kg)
Tracer : CO CO2

---

Initial mass : 3.89294E+10 2.88745E+14
emitted : 1.05720E+11 3.98783E+12
chemistry : -8.77634E+10 1.37893E+11
update_p : 1.31167E+11 8.30128E+12
advection : -1.42449E+11 -9.63987E+12
deposition : -7.19594E+09 0.00000E+00
stratosphere : 0.00000E+00 0.00000E+00
Final mass : 3.84076E+10 2.91532E+14
Budget error : -1.25877E-01 -5.00812E+02

---

---

Budget of tracers in region sea100x100 (Kg)
Tracer : CO CO2

---

Initial mass : 2.14392E+10 1.45622E+14
emitted : 2.94469E+11 8.08005E+12
chemistry : -1.40800E+11 2.21224E+11
update_p : 0.00000E+00 0.00000E+00
advection : -1.31412E+11 -6.98943E+12
deposition : -2.25017E+10 0.00000E+00
stratosphere : 0.00000E+00 0.00000E+00
Final mass : 2.11938E+10 1.46933E+14
Budget error : -2.45556E+01 -1.59688E+02

---

~~~~... read more

I've coded in my CO and CO2 inversions into a more general multi-tracer system. To test the multi-tracer system, at each step we will compare their inputs and outputs to the individual single-tracer systems.

• Are the CO and CO₂ emissions equal? Yes
• Are the CO and CO₂ point observations equal? Yes, except for station_id, which is expected to be different
• Are the RemoTeC and IASI observations equal? Yes
• Are the save_ files in the end equal? Yes
• Are the 3-hourly 3D mixing ratio fields equal? Yes
• Are the point and satellite output/track files equal? Yes
• Are the point and satellite departure files equal? Yes (CO₂ departures are different by ~10^-14, perhaps insignificant)
• Are the adjoint emissions equal? Yes (Again, different by 1 part in 10¹⁴, perhaps insignificant)
• Does the gradient test with two tracers converge? Yes
• Does 4DVAR give the same result as individual CO and CO₂ inversions? See next post

Good idea to post it. But please specify the machine you test on!

I'm going to test the run time for ten days of glb6x4+nam3x2+nam1x1 with coarsened meteo as a function of the number of threads, and archive the result here. Everything was run on a water-cooled ('normal') node of cartesius.

# of threads	Run time (over three runs)
1	0:03:52, 0:03:46, 0:03:48
2	0:02:49, 0:02:51, 0:02:51
3	0:02:28, 0:02:20, 0:02:19
4	0:02:02, 0:02:14, 0:02:11
5	0:02:05, 0:02:06, 0:02:05
6	0:02:02, 0:01:51, 0:02:03
7	0:01:48, 0:01:47, 0:01:48
8	0:01:44, 0:01:47, 0:01:45