Comparison of b.e30_beta02.BLTHIST.ne30_t232.104 with the observation-based temperature record

[islasimpson]

This is a bit of a closer look at the observation-based temperature record and how it compares with b.e30_beta02.BLTHIST.ne30_t232.104.

Previously I had made a plot of global mean surface temperature where I had compared the BEST data with the simulations and BEST was sitting nicely within the LENS2 spread. Here's that figure again as a reminder:

For the above figure I used the globally averaged surface temperature product provided by BEST (https://berkeley-earth-temperature.s3.us-west-1.amazonaws.com/Global/Land_and_Ocean_complete.txt).

However, now looking a bit more closely at trends from 1900 to 1950, the following maps show three different observation-based products (BEST, GISTEMP, HADCRUT). From these maps we get the sense that LENS2 isn't actually warming enough compared to the historical record.

The gray regions are where the dataset has years that are NaN's between 1900-1950 resulting in the trend also being NaN. I'm not sure what BEST does with these NaN areas when they make their globally averaged product. But if we focus in on the Northern Hemisphere where we have better observational coverage, there are indications that LENS2 might not be warming enough compared to the observation-based data. The following figure now shows time series only averaged over 30N-60N.

Indeed, for 30N-60N, 104 looks a lot better than LENS2. Maybe things are actually not as bad as they appeared in the first globally averaged plot.

So, now I'm wondering whether we can really trust the BEST global average product. Below is a sanity check just to make sure I'm not making any mistakes with my plotting of the BEST globally averaged product. The first figure is taken from the BEST website. The second figure is me plotting BEST in the same way i.e., anomalies relative to 1850 to 1900 and showing in black the annual timeseries and in red 30 year running means, which is I think what's being shown in the figure from the BEST website and then we have LENS2 and 104 for context. I think my BEST time series looks very similar to the one from the website, so I don't think I'm doing anything wrong with the plotting.

BEST website figure:

My figure:

[islasimpson]

Now trying to understand what's going on with the BEST globally averaged product a little better. In the figure below, I've taken the BEST globally averaged product in blue and compared it with the BEST lat-lon product and taking the global average of that assuming zero anomaly at each location and time where it has NaN's i.e., those regions don't contribute to the globally averaged anomaly (in orange). These two look pretty similar. So, it seems like assuming zero anomaly every where we have NaN's in the BEST data gives us something fairly similar to the BEST globally averaged product

So, then we can sample LENS2 and 104 in a similar way to BEST i.e., set the anomaly to zero at every location and time where the BEST data has NaN's to see how that kind of masking changes the picture. That beings the obs and 104 a little closer to each other

Below is a series of plots looking at different latitude bands in this way.

104 is warming more than obs in the Southern Ocean hemisphere (30S-90S but who knows whether we can trust the observations there. Both 104 and LENS2 warm more than the obs in the Arctic as well.

But I think the 30N-60N plot above is indicating that 104 is actually doing something a lot better when it comes to the aerosol forced response. We may have had compensating biases in LENS2 with the NH mid-latitudes not warming enough, but the SH and Arctic warming more to compensate. Now we're going better in the NH mid-latitudes and those other issues mean that overall we're warming more in the global average than LENS2. Then there are observational issues. This preprint also suggests there are issues in the observational record during this time. They propose a new dataset (in red) in the figure below. Not sure it would solve our issues for the global mean comparison, but it's clear we have some observational uncertainties.

Overall, maybe we don't actually have as big of a problem as it appears from the global mean temperature timeseries.

[islasimpson]

Just summarizing what was discussed in the project meeting today. Maybe setting the anomalies to zero where there are NaN's in the observational record as done above isn't really appropriate. Instead, here's a comparison with three different 200 member ensembles of observational surface temperature with the members aiming to represent the uncertainties associated with the missing data.

Then also with an 1850-1900 baseline

We're still perhaps too warm in the global average. But given the improvements we see in the NH mid-latitudes discussed above, we're also doing something more correct.

More warming in the Southern Ocean and more warming in the Arctic are leading to more warming overall. It may be something to revisit, but given the big improvements in the Northern Hemisphere mid-latitudes, for now we perhaps shouldn't be searching for something being wrong with the aerosol impacts.