Northern Hemisphere summertime mid-latitude warm bias

[islasimpson]

When we look at JJA averaged near surface air temperature in the piControl's, we see a warm bias in the Northern Hemisphere mid-latitudes. This plot is comparing the piControl with 1979-2024, so even the piControl is warm compared to the present day. In CESM2 (left) we had a warm anomaly over the continents, but what seems to be new in CESM2 is also the warmth over the mid-latitude oceans in the NH summer.

We have initialized predictions, initialized May 1st for both the 194 (with clubb explicit diffusion) and 199 (without club explicit diffusion) configurations. These we can more easily compare with observations and the results are shown below, comparing to BEST surface temperature. The warming over the mid-latitude oceans is a bias relative to obs, as shown below for TREFHT.

Here are time series of the evolution of monthly averages over ocean and over land. Left is the actual time evolution and the right is the difference between the 194 and 199 configurations and CESM2 SMYLE. The middle row shows a comparison with TS instead of TREFHT, which would be the more appropriate comparison with BEST over the ocean since it's SST over the ocean but the conclusions remain the same whichever variable is used. It shows that for the biases over land CESM2 and the CESM3 development runs aren't that different. But over ocean, the mid-latitude summertime temperatures of CESM2 are closer to observed, with both the 194 and 199 configurations becoming too warm as we get into the summer. The uncertainty range on the right is based on selecting 200 different single member samples from the CESM2 SMYLE.

The plot below shows the terms in the surface energy balance over ocean in the mid-latitudes. The differences from CESM2 SMYLE are pretty similar between 194 and 199 although there is a systematic difference in the SHFLX between them.

Here is the difference in the terms on the same plot with sign conventions noted.

The dominant term in the difference between the CESM3 initializations and the CESM2 SMYLE is an increase in net downward shortwave over mid-latitude oceans and this is being compensated for by an increase in evaporation and an increase in net upward longwave and to a lesser extent sensible heat flux (presumably the increase in net longwave and sensible are a result of the ocean having warmed).

The difference in incoming shortwave is not a clear sky effect

This doesn't really seem to be related to a change in total cloud cover (left in the figure below) as that changes sign over the season. But there's a difference in the cloud ice and cloud liquid. CESM3 development simulations have more cloud ice and less cloud liquid.

We don't know that CESM2 didn't have compensating errors and that some of these things are improvements in CESM3, but we do know that in the end we end up with a bias in the surface temperature over the mid-latitude oceans.

[islasimpson]

One of the figures above got a bit messed up. Here it is again...

[adamrher]

CESM3 development simulations have less cloud ice and more cloud liquid.

Other way around, no?

[islasimpson]

Indeed, these seem useful. Are these all biases relative to CERES estimates? If so, it's reassuring that we can see that it's a bias in the SWCF and if we can improve it it wouldn't be getting the temperatures right for the wrong reasons.

[adamrher]

Are these all biases relative to CERES estimates?

yes

[islasimpson]

@adamrher run 199 is the one that I've shown in the analysis above which has diffusion off and I think 202 differs from 199 with sea salt tuning. Do you have the above plot for 199? Given the similarity of the temperatures between 199 and 198 I'm guessing 199 looks like 198. Maybe the sea salt tuning in 202 has already resolved some of this issue relative to 199.

[adamrher]

Hi Isla, I've added 199 to the above plots. 202 does have brighter clouds forcing, but I think this has more to do with the ocean changes in from 194/199 -> 198/202, although this is not very clear from the SST plots (added above; the 198 is cooler than 194, but 202 is warmer than 199 in some places, cooler in others). The lower seasalt emissions factor has a large impact on the clear sky fluxes, as this results in smaller burdens and less scattering of SW (see FSNTC plots, also above).

[islasimpson]

OK, thanks. I guess we can discuss tomorrow whether there's anything we want to do about this particular thing. Although it sounds like maybe we should wait to have a look at how the new ocean settings do in the initialized predictions.

[islasimpson]

Here's a look at how the summertime NH mid-latitude warm bias compares between May initializations using the original micro_mg_berg_eff_factor=1 and micro_mg_berg_eff_factor=0.1.

Reducing this factor to 0.1 has small impact on the NH mid-latitude warm bias - making it slightly cooler i.e., reducing the bias, but doesn't get us back to what we had in CESM2.

But, while the analysis above, suggested the dominant factor in the warm bias was an increase in the net downward shortwave (FSNS), this plot shows that the bergeron factor has a pretty big impact on FSNS and when it's set to 0.1, FSNS looks more like that in CESM2.

So I'm a little surprised that the temperature changes aren't more dramatic between the two. Above, we had seen that 194 had less cloud liquid and more cloud ice than CESM2. Changing the bergeron factor to 0.1, brings the cloud ice closer to what we had in CESM2 but gives us even more cloud ice than we had in CESM2.

Overall, I'm a bit confused and not sure how to interpret this. When we use the bergeron factor 0.1 we still have a bigger mid-latitude warm bias than we did in CESM2. But I'm surprised that changing the incoming shortwave didn't have that much of an impact on surface temperature.

[adamrher]

It looks like the reduction in FSNS in the berg = 0.1 run is being at least partially compensated by the other terms of the sfc energy budget. There is a reduction in LHFLX, and a reduction in FLNS, and the latter might be indicative of greater incoming LW, although cooler surface temperature would be expected to reduce LWUP. Greater incoming LW would be consistent with the increase in IWP (is the reduction in FLNSC or just FLNS?), which itself is strange since you would think reducing the bergeron efficiency should reduce ice production -- there must be some secondary effect occurring that facilitates ice growth.

I'd like to take a look at the vertical profiles of CLDLIQ and CLDICE in these experiments.

I'm curious about an intermediate reduction of the berg efficiency to 0.5, as perhaps the response would be more linear and actually reduce IWP.

[islasimpson]

Here's a comparison of FLNS and FLNSC. It looks like only a small part of the difference is coming from the clear sky (1 W/m2 out of about 4 W/m2). But this difference in FLNS looks like it's only really there in August. Not in the earlier months when FSNS has also changed quite a bit

[islasimpson]

@adamrher Here are some vertical profiles of CLDLIQ and CLDICE averaged over the ocean in the mid-latitudes

[islasimpson]

@wwieder

[islasimpson]

Adding some plots here over land (not with the bergeron parameter perturbation) but just comparing 199/194 and CESM2. Adding this given the discussion going on here about biases over land.

Firstly, the figures above show that the difference in the bias between CESM2 and CESM3 are much smaller over land than they are over ocean. At least this is true in the initialized predictions - it is less true in the free running simulations, which maybe suggests there are important land feedbacks acting on longer timescales that exacerbate the issue.

Pasted here again is the surface temperature over mid-latitude land. We look pretty good compared to ERA5, but are warm compared to BEST. Probably should trust BEST more. CESM3 is pretty similar to CESM2 until we get into July and August.

Here are the surface energy balance terms as they evolve in the May initialized predictions

Consistent with the surface temperatures not really being warmer in May and June, the difference in net downward shortwave (fsns) doesn't become positive until July and August. Compared to CESM2, we have a different partitioning of of the incoming energy. More latent heat flux in CESM3 and less sensible heat flux, which is a big confusing, because if anything that seems like it should make the land cooler in CESM3.

The conclusions that we had about there being more cloud ice and less cloud liquid over the oceans above also hold over land.

[islasimpson]

@wwieder