Some questions about SST sensitivity experiments with CAM

[Wenshuo]

Dear all,

I am currently running SST sensitivity experiments with CAM. After reviewing many papers on the details of experimental design, I still have some questions that I would like to ask:


1. How long should the spin-up period be in CAM simulations?
In different studies, the spin-up time ranges from a few months to 10 years. Since the atmosphere in CAM can typically respond to SST anomalies within just a few months (much faster compared with other components), why do many studies still use 5 to even 10 years for spin-up?


2. Time period for comparing control and sensitivity experiments.
Some studies compare the last 30 years (for decadal-scale analyses), while others only compare results after several months of integration (for intra-seasonal lag/response studies). Although this partly depends on the research timescale (seasonal vs. decadal), why is it that in some cases atmospheric anomalies emerge within just a few months, while in other studies decades of integration are still needed?


3. Influence of the background climate state in control experiments.
In control runs, the background climate is often taken either from a pre-industrial control (pi-control), around the year 2000, or tailored to the specific study period. How much influence does this choice of background state have on the experiment results?


4. Timing of applying SST anomalies.
Some studies impose SST anomalies at the very beginning of the sensitivity experiment, while others run the model for some time (spin-up) and then branch into control and sensitivity runs (hybrid runs) where SST anomalies are introduced. Which approach is more appropriate?


5. Methods for generating ensembles.
Most experiments use ensemble simulations to reduce uncertainty. The main approaches are: (1) perturbing the initial conditions slightly at the same start time (micro-ensemble), and (2) restarting from different years of the same control simulation (macro-ensemble). Which ensemble generation method is generally more recommended?

 

[aherring]

1. I'm assuming you're asking about prescribed SST/sea-ice experiments, as running w/ a fully coupled ocean model would require different spin-up methods. The troposphere responds to changes in boundary conditions fast, a few months is sufficient, and to be safe you could just consider the first year as the spin-up. Depending on your initial conditions, the control experiment might require a longer spin-up, up to 5 years if the stratosphere is way out of balance. I would suggest plotting a time-series of temperature at a level in the stratosphere to determine whether your control is spun up. Another component to consider is the land, which has a longer memory than the atmosphere. I would just check whether the control is sufficiently spun up.

2. 30 years is recommended (not including spin-up), again assuming you're using a prescribed SST/sea-ice configuration.

3. The background climate matters.

4. As I said in (1), if your control is out of balance, then make sure to spin it up and either (a) use the *.cam.i.* file from the end of the control to initialize a new "startup" run or (b) do a "branch" run from the end of the spun-up control.

5. The "micro-ensemble" is for quantifying uncertainty due to chaos and inter-annual variability in the atmosphere, whereas the "macro-ensemble" is uncertainty due to inter-decadal SST variability.

 

[Wenshuo]

1. I'm assuming you're asking about prescribed SST/sea-ice experiments, as running w/ a fully coupled ocean model would require different spin-up methods. The troposphere responds to changes in boundary conditions fast, a few months is sufficient, and to be safe you could just consider the first year as the spin-up. Depending on your initial conditions, the control experiment might require a longer spin-up, up to 5 years if the stratosphere is way out of balance. I would suggest plotting a time-series of temperature at a level in the stratosphere to determine whether your control is spun up. Another component to consider is the land, which has a longer memory than the atmosphere. I would just check whether the control is sufficiently spun up.

2. 30 years is recommended (not including spin-up), again assuming you're using a prescribed SST/sea-ice configuration.

3. The background climate matters.

4. As I said in (1), if your control is out of balance, then make sure to spin it up and either (a) use the *.cam.i.* file from the end of the control to initialize a new "startup" run or (b) do a "branch" run from the end of the spun-up control.

5. The "micro-ensemble" is for quantifying uncertainty due to chaos and inter-annual variability in the atmosphere, whereas the "macro-ensemble" is uncertainty due to inter-decadal SST variability.

Many thanks for your detailed response!