How to judge whether CAM finishes spinning up

[punch]

Hi everyone!
I run F_2000 in CESM1.2.1 for 30 years and I want to know whether it finishes spinning up. So, I calculate FSNT-FLNT (global) to judge. But I don't think it tends to be stable. The figure is placed below.
So, does anyone know the method I used is right or whether there is a better way to judge?

 

[punch]

Thank you for any helpful advice in advanced.

 

[brianpm]

I usually look at globally averaged time series like you have done. Whether the values you show are "stable," I think is debatable. Even in F_2000 there will be some interannual variability, and I think that is what you are seeing here. Other variables I tend to look at are: TMQ and PRECT (and/or precipitation minus evaporation), and TREFHT. Since SST/ice are specified in this case, you are really asking whether the land surface is spun up. The atmosphere will spin up very quickly compared to any of the surface models.... I've done some analysis of aquaplanets and it seems like even the slower fields "spin up" within 2 years (Medeiros et al. DOI: 10.1002/2015MS000593, I think).

So tests you could do would be to look at whether there is a trend either by taking the total trend from the beginning or a "running" trend of fixed length. You could similarly ask whether the variance is changing in "running" chunks. In either case, it looks like you are probably "spun up" in the sense that the variations that you see are representative of the internal noise of the system.

 

[punch]

I usually look at globally averaged time series like you have done. Whether the values you show are "stable," I think is debatable. Even in F_2000 there will be some interannual variability, and I think that is what you are seeing here. Other variables I tend to look at are: TMQ and PRECT (and/or precipitation minus evaporation), and TREFHT. Since SST/ice are specified in this case, you are really asking whether the land surface is spun up. The atmosphere will spin up very quickly compared to any of the surface models.... I've done some analysis of aquaplanets and it seems like even the slower fields "spin up" within 2 years (Medeiros et al. DOI: 10.1002/2015MS000593, I think).

So tests you could do would be to look at whether there is a trend either by taking the total trend from the beginning or a "running" trend of fixed length. You could similarly ask whether the variance is changing in "running" chunks. In either case, it looks like you are probably "spun up" in the sense that the variations that you see are representative of the internal noise of the system.

Thanks for your reply! Your reply really solves my confusion a lot. But I still have something that I don't understand. Does the interannual variability come from the different initial atmosphere condition, although the seasonal SST is prescribed? Thanks again!

 

[brianpm]

No, the interannual variability should be independent from the initial condition. The initial condition can be perturbed to produce different realizations of the case. Such realizations will differ from each other in the details, but the atmosphere pretty much forgets the initial condition after a few months (the land initial condition can have a longer impact). Let's say 10 realizations differing only in the initial condition are each run for 20 years. The statistical features of the 10 realizations are all expected to be the same, including interannual variability. Ultimately, this internal variability develops because of the chaotic nature of the system, but since it is the same system with the same forcing, we expect the internal variability to be statistically the same for each realization. This is exactly the methodology used for the CESM1 large ensemble (and similar to the method for the CESM2 large ensemble).

One caveat is that some kinds of systems can have multiple equilibria. In some of those cases, the realizations of the system can end up in two (or more) states that are persistent. In climate one might think of snowball earth scenarios versus non-glaciated condition, and an example study is here: Climate Determinism Revisited: Multiple Equilibria in a Complex Climate Model

 

[ohmpawat]

Hi everyone!
I run F_2000 in CESM1.2.1 for 30 years and I want to know whether it finishes spinning up. So, I calculate FSNT-FLNT (global) to judge. But I don't think it tends to be stable. The figure is placed below.
So, does anyone know the method I used is right or whether there is a better way to judge?

Hi, I have encountered the similar question. Could you tell me the picture is the result of calculating the average of time and lat/lon or calculating the average of time and the sum of lat/lon. Thanks in advance!

 

[punch]

Hi, I have encountered the similar question. Could you tell me the picture is the result of calculating the average of time and lat/lon or calculating the average of time and the sum of lat/lon. Thanks in advance!

It's the average of time and lat/lon.

 

[ohmpawat]

It's the average of time and lat/lon.

I simulated a year and calculated the average of lat/lon and time. The result of FSNT-FLNT is negative and it is more than 20W/m2. Have you met the similar situation?

 

[punch]

I simulated a year and calculated the average of lat/lon and time. The result of FSNT-FLNT is negative and it is more than 20W/m2. Have you met the similar situation?

Is the data you used is monthly or daily data?

 

[ohmpawat]

Is the data you used is monthly or daily data?

The history file? It is monthly .

 

[punch]

I simulated a year and calculated the average of lat/lon and time. The result of FSNT-FLNT is negative and it is more than 20W/m2. Have you met the similar situation?

Maybe you should use the weights to calculate the average of lat/lon.

 

[ohmpawat]

Maybe you should use the weights to calculate the average of lat/lon.

I didn't consider the weights. But I only see the gw(lat) without lon.

 

[brianpm]

It is very important to do area-weighted averaging (nb Retraction Note: A 10 per cent increase in global land evapotranspiration from 2003 to 2019 - Nature).

You can use gw(lat) or calculate the weights using cos(lat). The weight in the longitude direction is uniform.

If you use python with numpy and xarray, you can use the built-in broadcasting:

avg = data.weighted(np.cos(np.radians(data.lat))).mean(dim=("lat","lon"))

 

[ohmpawat]

It is very important to do area-weighted averaging (nb Retraction Note: A 10 per cent increase in global land evapotranspiration from 2003 to 2019 - Nature).

You can use gw(lat) or calculate the weights using cos(lat). The weight in the longitude direction is uniform.

If you use python with numpy and xarray, you can use the built-in broadcasting:

avg = data.weighted(np.cos(np.radians(data.lat))).mean(dim=("lat","lon"))

Thanks a lot! I will try it!