RealClimate

Climate science from climate scientists...

You are here: Home / Climate model projections compared to observations

Climate model projections compared to observations

Since we have been periodically posting updates (e.g. 2009, 2010, 2011, 2012, 2015, 2016 etc.) of model output comparisons to observations across a range of variables, we have now set up this page as a permanent placeholder for the most up-to-date comparisons. We include surface temperature projections from 1981, 1988, CMIP3, CMIP5, and CMIP6, and MSU satellite products from CMIP5 and CMIP6, SST and SSU from CMIP6, and we will update this on an annual basis, or as new observational products become available. For each comparison, we note the last update date, and where the comparison was first discussed.

This is a collation of opportunity. To be included, there needs to be a pre-exisiting archive of the processed model output for the historical period and projections, and at least one regularly-updated observational data source.

Note that Hausfather et al. (2020) made a more complete assessment of surface temperature projections from models from 1970 through to CMIP3 (discussed here).

Global mean surface temperature anomalies

Hansen et al (1981)

Hansen et al 1981 projections overlain with actual temperatures to 2023 that lie above the projections.

Original discussion (figure originally courtesy of Geert Jan van Oldenborgh, Hansen et al. (1981)). Observations are the GISTEMP LOTI annual figures and 5 year mean. Last updated: 15 Jan 2024.

Hansen et al (1988)

Time series from 1955 to 2023 of Hansen et al (1988) climate model hindcasts to 1984, and projections beyond (from three scenarios), compared to observed temperatures. The observed temperatures are just a little below Scenario B.

Original discussion (2007), Last discussion (2018). Scenarios from Hansen et al. (1988). Observations are the GISTEMP LOTI annual figures. Trends from 1984: GISTEMP: 0.21ºC/dec, Scenarios A, B, C: 0.33, 0.28, 0.15ºC/dec respectively (all 95% CI ~±0.02 or 0.03). Last updated: 16 Jan 2024.

CMIP3 (circa 2004)

Time series from 1979 of CMIP3 climate model hindcasts to 2000, and projections beyond, compared to observed temperatures. The long term trends in the models are a good fit to the actual temperatures.

Last discussion (2015). Model spread is the 95% envelope of global mean surface temperature anomalies from all individual CMIP3 simulations (using the SRES A1B projection post-2000). Observations (GISTEMP) are the standard quasi-global estimates of anomalies with no adjustment for spatial coverage or the use of SST instead of SAT over the open ocean. Last updated: 12 Jan 2024.

CMIP5 (circa 2011)

Time series from 1979 of CMIP5 climate model hindcasts to 2005, and projections beyond, compared to observed temperatures. The long term trends in the models are a good fit to the actual temperatures.
CMIP5 ensemble projections with observational record overlaid.

Last discussion (2015). Model spread is the 95% envelope of true global mean surface temperature anomalies from all CMIP5 historical simulations (using the RCP4.5 projection post-2005). Forcing adjustment is updated from Schmidt et al. (2014). Observations are the standard quasi-global estimates of anomalies with no adjustment for spatial coverage or the use of SST instead of SAT over the open ocean. Last updated: 15 Jan 2024.

Time series from 1979 of CMIP5 climate model hindcasts to 2005, and projections beyond, compared to observed temperatures (a blend of SAT/SST). The long term trends in the models are a good fit to the actual temperatures.

As above, but using the blended SST/SAT product from the CMIP5 models produced by Cowtan et al (2015) instead of the pure SAT field. Note that this makes about a 0.05ºC difference in 2022 (compared to a 0.1ºC difference estimated from the forcings adjustment above). Last updated: 15 Jan 2024.

CMIP6 (circa 2021)

The latest phase of CMIP has been ongoing since around 2019, and now has sufficient models to provide a basis for projections going forward. These models used observed boundary conditions (GHG levels, deforestation, solar, volcanoes etc.) up to 2014, and projections based on the Shared Socioeconomic Pathways (SSPs) from 2015 onwards. The same caveat with respect to the comparison to the blended SAT/SST observations stands as with CMIP5, but this is a relatively small effect (i.e. it’s expected to be around 0.05ºC in 2022).  Note however, that some CMIP6 models have climate sensitivities outside both the CMIP5 range and the range constrained by observations.  Thus in these figures, we plot the full mean (1 ensemble member per model) and the mean of a subset of the models that have a transient climate response (TCR) within the likely constrained range [1.4,2.2]ºC as assessed by IPCC AR6 (Hausfather et al., 2022).

Time series from 1979 of CMIP6 climate model hindcasts to 2014, and projections beyond, compared to observed temperatures. The long term trends in models screened for a likely range of sensitivity are a good fit to the actual temperatures.
Surface temperature changes in CMIP6 models, using a screened ensemble and a fuller ensemble (37 models).
Histogram of trends from 1979 to 2023 of CMIP6 climate model hindcasts to 2014, and projections beyond, compared to observed SAT products.

Model spread is the 95% envelope of surface air temperature anomalies using 37 model simulations from the historical runs and SSP2-4.5. TCR values used in the screen are Hausfather et al. (2022). Last updated: 3 Feb 2024.

Satellite-derived atmospheric temperatures

TMT (global and tropical means) (timeseries and trends)

Time series from 1979 of CMIP5 climate model hindcasts to 2005, and projections beyond, compared to observed mid-troposphere temperatures (TMT).
Time series from 1979 of CMIP5 climate model hindcasts to 2005, and projections beyond, compared to observed tropical mid-troposphere temperatures (TMT).
Histogram of trends from 1979 to 2023 of CMIP5 climate model hindcasts to 2005, and projections beyond, compared to observed mid-troposphere temperatures (TMT).
Histogram of trends from 1979 to 2023 of CMIP5 climate model hindcasts to 2005, and projections beyond, compared to observed tropical mid-troposphere temperatures (TMT).

Original discussion (Jan 2017). Mid-troposphere satellite products from UAH, RSS, NOAA STAR and UW (Po-Chedley et al, tropical values only – not yet updated to 2023). Model values use synthetic MSU/AMSU-TMT weightings, spread is 95% envelope of simulations. Last updated: 15 Jan 2024.

Corrected-TMT (Tropical means) (timeseries and trends)

Tropical TMT-Corrected temperature in CMIP6 models compared to observations from 1979 to 2023.
Histogram of tropical TMT-Corrected temperature trends in TCR-screened CMIP6 models compared to observations from 1979 to 2023.

Original discussion Mar 2023. The TMT-corrected diagnostic (where the TMT is adjusted by the lower stratospheric changes (TLS)) in a 24-model subsample of the CMIP6 database (Po-Chedley et al., 2021) compared to RSS v4, UAH v6 and NOAA STAR v5. Plots includes the full ensemble, and then a subset of models that have a TCR within the IPCC AR6 likely range. Model values use synthetic TMT-corrected weightings, spread is 95% envelope of simulations. Last updated: 15 Jan 2024.

TLT (timeseries and trends)

Histogram of TLT trends in CMIP6 models, showing the full ensemble and the TCR-screened subset, along with the the trends from RSS, UAH and NOAA STAR.

The Temperature of the Lower Troposphere (TLT) diagnostic in a 38-model subsample of the CMIP6 database (McKitrick and Christy, 2020) compared to RSS v4, UAH v6 and NOAA STAR v5. Plots includes the full ensemble, and then a subset of models that have a TCR within the IPCC AR6 likely range. Model values use synthetic TLT weightings, spread is 95% envelope of simulations. Last updated: 3 Feb 2024.

Global SST changes

Sea surface temperature trends in CMIP6 models compared to observations from 1979 to 2023. The TCR-screened simulations line up very nicely with the observations.
Histogram of 1979-2023 linear trends in global SST in CMIP6 models and observations. The TCR-screened models line up nicely with observed trends, but some models (e.g. CanESM5) are warming too fast.

Original discussion 3/19/2023. Global mean SST trends in CMIP6 compared to HadSST4 and ERSSTv5. As above, models are screened using the TCR criterion. The models are a 15-model subset from the U. Melbourne collation. Last updated 15 Jan 2024.

SSU trends

Time-series from 1979 to 2023 of three SSU channels showing comparison of observations and CMIP6 climate models.
Histogram of 1979-2023 linear trends in global SSU channels in CMIP6 models and observations.

Original discussion 5/21/23. Global mean SSU trends for three channels (NOAA-STAR v3.0) compared to a 9-member subset of the CMIP6 archive. Last updated 1/25/24.

If you have suggestions for additional comparisons, stylistic changes, clarifications etc., please leave a comment on the latest open thread. You can use these figures anywhere (with a citation and link back to RealClimate).

References

  1. Z. Hausfather, H.F. Drake, T. Abbott, and G.A. Schmidt, "Evaluating the Performance of Past Climate Model Projections", Geophysical Research Letters, vol. 47, 2020. http://dx.doi.org/10.1029/2019GL085378
  2. J. Hansen, D. Johnson, A. Lacis, S. Lebedeff, P. Lee, D. Rind, and G. Russell, "Climate Impact of Increasing Atmospheric Carbon Dioxide", Science, vol. 213, pp. 957-966, 1981. http://dx.doi.org/10.1126/science.213.4511.957
  3. J. Hansen, I. Fung, A. Lacis, D. Rind, S. Lebedeff, R. Ruedy, G. Russell, and P. Stone, "Global climate changes as forecast by Goddard Institute for Space Studies three‐dimensional model", Journal of Geophysical Research: Atmospheres, vol. 93, pp. 9341-9364, 1988. http://dx.doi.org/10.1029/JD093iD08p09341
  4. G.A. Schmidt, D.T. Shindell, and K. Tsigaridis, "Reconciling warming trends", Nature Geoscience, vol. 7, pp. 158-160, 2014. http://dx.doi.org/10.1038/ngeo2105
  5. K. Cowtan, Z. Hausfather, E. Hawkins, P. Jacobs, M.E. Mann, S.K. Miller, B.A. Steinman, M.B. Stolpe, and R.G. Way, "Robust comparison of climate models with observations using blended land air and ocean sea surface temperatures", Geophysical Research Letters, vol. 42, pp. 6526-6534, 2015. http://dx.doi.org/10.1002/2015GL064888
  6. Z. Hausfather, K. Marvel, G.A. Schmidt, J.W. Nielsen-Gammon, and M. Zelinka, "Climate simulations: recognize the ‘hot model’ problem", Nature, vol. 605, pp. 26-29, 2022. http://dx.doi.org/10.1038/d41586-022-01192-2