I added an update to the end of the recent post Model-Data Comparison with Trend Maps: CMIP5 (IPCC AR5) Models vs New GISS Land-Ocean Temperature Index. That update included two graphs that showed the difference between the multi-model ensemble mean of the CMIP5-archived simulations of global surface temperatures and the GISS Land-Ocean Temperature Index (LOTI) data for the period of 1880 to 2012. The model mean is subtracted from the observations data in the graphs. The first graph in the update was the difference with the base years of 1961-1990, which were the base years for anomalies used by the IPCC in their model-data comparison in Figure 10.1 from the Second Order Draft of AR5. And the second graph, Figure 1, used the base years of 1880-2012 for anomalies. I also included the difference smoothed with a 5-year running-average filter to minimize the year-to-year variations. We’ll use the same base years (1880-2012) and smoothing for the other graphs in this post.
We’ll also present the differences between the models and the other two primary global surface temperature anomaly datasets: the product from the NCDC, and the HADCRUT4 data from the UKMO. I’ve also included comparison graphs of the model-data difference for the 3 datasets using the global data and a second comparison using the latitudes of 60S-60N, for those concerned that the model-data comparisons are biased by how the datasets and models account for the polar regions. Last, I was interested to see how much better the models performed when compared to the old GISS data (before the recent change in that dataset), so I’ve included a comparison of the model-data difference using the old and new GISS LOTI data.
Note: I have not included my normal discussion about the use of the model mean. If you have any questions, please see the post here.
MODEL-DATA DIFFERENCE WITH HADCRUT4 AND NCDC DATA
The plots of the model-data differences using NCDC and HADCRUT4 global surface (land plus sea) surface temperature are presented in Figures 2 and 3, respectively. The datasets share much of the same source data for land air surface temperatures and sea surface temperatures, so visually they present similar curves.
COMPARISONS OF THE DIFFERENCE WITH GISS LOTI, HADCRUT4 AND NCDC DATASETS
The model-data differences for the 3 datasets are shown in Figure 4, using global latitudes, 90S-90N. All data have been smoothed with 5-year running-average filters. It should come as no surprise that the GISS- and NCDC-based curves are so similar—they use the same sea surface temperature dataset, NOAA’s ERSST.v3b. On the other hand, the HADCRUT4 data use the recently updated HADSST3 dataset. There’s another reason for the differences between the HADCRUT4 curve and the others: missing data is not infilled in HADCRUT4, while it is infilled in the GISS and NCDC products.
One of the big differences between the datasets are how they handle the polar data. GISS infills Arctic temperatures by masking sea surface temperature data anywhere there has been seasonal sea ice and by extending land surface temperature data out over the Arctic Ocean. GISS uses the same tactic in the Southern Ocean surrounding Antarctica. On the other hand, HADCRUT4 and NCDC products do not mask sea surface temperatures and extend the land surface temperature data out over the oceans. GISS also includes more land surface air temperature data in Antarctica than the other two datasets. And there are differences between the ERSST.v3b data used by GISS and NCDC and the HADSST3 data used in HADCRUT4. There is very little observations-based sea surface temperature source data in the Southern Ocean surrounding Antarctica. The missing sea surface temperature data is infilled in the ERSST.v3b data, but with HADSST3 it is not.
The easiest way to account for all of those differences is to exclude the data north of 60N and south of 60S. See the comparison in Figure 5. The GISS and NCDC curves are now almost identical, while the HADCRUT4 model-data difference continues to diverge from the others. The models, as shown in Figure 5, perform a little better during recent years if the polar data is excluded—but not much better.
OLD AND NEW GISS LOTI DATA
GISS recently switched to ERSST.v3b sea surface temperature data from a combination of the Hadley Centre’s infilled sea surface temperature data (HADISST) for the period of 1880 to 1981 and the satellite-based Reynolds OI.v2 data from 1982 to present. I haven’t yet found an explanation by GISS for that switch, but the obvious differences are presented in the post A Look at the New (And Improved?) GISS Land-Ocean Temperature Index Data. The model-data differences using the old and new GISS LOTI data are shown in Figure 6. The models actually performed better against the old GISS data before the 1920s. But there’s less of a difference between the models and the new GISS data from 1982 to present, meaning the new GISS data runs a little warmer over the past 30 years without the satellite-based sea surface temperature data. That’s to be expected. The satellite-based data Reynolds OI.v2 sea surface temperature data have better coverage, especially in the high latitudes of the Southern Hemisphere and much less infilling as a result.
Presenting the differences between modeled and observed global surface temperatures is yet another way to show how poorly the climate models simulate global temperatures since 1880. The models cannot explain the observed cooling from 1880 to the 1910s, and they cannot explain the warming from the 1910s to the 1940s. Plotting the difference also helps to show that the divergence in recent decades, with the models simulating too much warming, started as far back as the early 1990s, when models overestimated the cooling from the volcanic aerosols associated with the eruption of Mount Pinatubo.
The GISS LOTI data and the outputs of the CMIP5-archived models are available through the KNMI Climate Explorer.