Early 20th century temperature trends over land

[islasimpson]

Here's a place to discuss whether we think early CESM3 historical simulations were doing something better in representing historical temperature trends than some of the recent simulations and whether we want to see whether getting back to what we had in earlier development runs should be a goal.

The following plots focus on historical runs with 104 and 121 from further back in CESM3 development and 271 and 276 from the recent set of simulations.

Below is a comparison of land averaged temperature timeseries. This showed 11-year running means and temperature is masked in a way that mimics where there is missing data in each observational dataset. For DCENT and BEST I'm showing anomalies relative to 1850 to 1880. For GISTEMP, I've not done that since it only extends back to 1880, but that won't matter for the 1880-1950 trends in subsequent plots and focus on the DCENT and BEST plots here...

LENS2 exhibited a lack of warming over the late 19th/early 20th century over land. In 104 and 121 we had previously noticed that this seemed to be improved. However, the more recent simulations 271 and 276 look more like LENS2 with less warming over land between 1850 and the mid 20th century.

Here are maps of trends in the annual mean from 1880-1950 in comparison to the three observation-based datasets. DCENT has a lot of missing data but somehow it still ends up being similar to the others in the land area average. Internal variability is going to have an influence on the trends locally, but I think the point from this figure is that, while 104 and 121 had more warming overall, it's not obvious that the look much more like the observations. For example, they show a lot more warming over Asia than the recent runs, but the observations don't have much warming over central Asia at least. They also show more warming over South Ameirca than the recent runs, but are actually showing more warming than observed there with the recent runs being closer. It kind of looks like thelack of warming over North America in all of the runs, is being opposed by more warming elsewhere in the earlier runs and that's not happening in the most recent runs.

Compared to BEST:

Compared to GISTEMP:

Compared to DCENT:

Here's the trend averaged over land regions compared with spread from the CESM2 large ensemble. The new runs are warming less than the older runs and are generally further from observed (although the large ensemble spread suggests large uncertainty).

Here are plots of the zonal mean trends. Perhaps ignore DCENT since the data is so sparse. For the others, over the Northern mid-latitudes the older runs look closer to obs, especially when you compare to BEST.

Here's a plot of the root mean squared difference between simulated trend and observed trend across all lons and lats. Whether the older runs have a lower RMSE than the newer runs varies depending on the dataset being compared with (for BEST the red point is hidden under the green).

Overall, I'm not really convinced this is something worth chasing after, given large observational uncertainties and the fact that while the old runs warm more in the land area average, it's not obvious in map form that they're really doing better. But glad to hear thoughts from others. @wwieder - a quick check on whether there's any obvious difference between these runs on the land side may still be helpful to decide how much to dig into this.