[ccsm-scientists] Bug in treatment of time-varying CO2 in CCSM

[William Collins]

Dear SSC, WG co-chairs, CCSM scientists, and CCSM programmers:

CSEG and CMS software engineers discovered a bug in the code used to implement 
the 1%CO2 runs late Friday Feb. 20. They were in the process of formally 
incorporating the 1%CO2 code into CAM. For the last several years, it's been 
a piece of code that gets handed off between folks doing CO2 runs and it has 
not been a part of CAM. This bug dates back to the period when CCSM2 was 
being developed and evaluated.

In their review of the code, the SEs recognized that we are ramping
co2vmr, the volume mixing ratio of CO2, but not co2mmr, the mass mixing ratio. 
This problem is somewhat more general -- the mass mixing ratio was computed
from the volume mixing ratio once, during initialization of the model.  In 
general, experiments in which the volume mixing ratio was a function of time
and the mass mixing ratio was not synchronously updated may suffer from this 
defect.

All the 1%CO2 runs and the branches from those runs at 2xCO2 and 4xCO2 for 
CCSM3 are corrupted. We have killed all the 1%CO2 runs and the branches until 
the problem is fixed. The SEs are working on a solution, and we believe we 
will be able to restart these runs early this week.

That's the bad news.

The good news is that this bug does not affect the ongoing SOM runs, 1870 
controls, or 1990 controls.  It may also help to explain why the climate 
sensitivities estimated from the 1%CO2 and SOM integrations of CCSM2 could
not be reconciled.

The CSEG and CMS SEs certainly deserve our thanks for identifying this bug 
before it propagated further.

The details of the bug are as follows.  The radiation code, as originally 
written, uses co2vmr and co2mmr for calculating three effects:

1. The greenhouse effect of CO2 (co2vmr);

2. The absorption of near-IR sunlight (co2mmr);

3. The overlap of infrared extinction between CO2 and CH4, N2O, CFC-11, and 
CFC-12 (co2mmr).

Due to the bug, the change in the first effect as CO2 increases was included,
but the changes in the second and third effects were omitted.  Fortunately, 
the 1st effect is dominant, but the 2nd and 3rd effects should not be 
neglected.

We have simulated the changes in TOA and surface LW fluxes and in the 
shortwave atmospheric absorption under mid-latitude summer (MLS) conditions 
for 2xCO2 and 1xCO2. In one simulation, both co2vmr and co2mmr double, as 
they should; in the second, just co2vmr doubles. No clouds are included.

The changes in the fluxes from doubling CO2, without and with the bug, are:

2xCO2 - 1xCO2:      
                          
  dflnt = -2.89          
  dflns = -1.57          
                          
  d(fsnt - fsns) = 0.097 

2xCO2 - 1xCO2 + BUG:

  dflnt = -2.46
  dflns = -0.65

  d(fsnt - fsns) = 0

This shows, for example, that the bug results in an underestimate of the 
change in outgoing longwave from doubling CO2 of 0.43 W/m^2, an error
of 15% for clear-sky MLS conditions.  The global annual-mean biases for 
all-sky conditions may vary.  Based upon previous experience, the biases
for all-sky conditions should be 20% smaller than the biases for clear-sky 
conditions.

Thanks to everyone who contributed to finding, analyzing, and solving this 
problem.

Bill