GISTEMP Uncertainty Analysis
Observational Uncertainty Ensemble
This page provides information on the GISS Surface Temperature Analysis (GISTEMP) v4 uncertainty ensemble. We provide rationale for the ensemble, best practices for using the ensemble, and the relevant code for generating and analyzing the GISTEMPv4 uncertainty ensemble. The analysis is described in detail in the journal article "A NASA GISTEMPv4 Observational Uncertainty Ensemble" by Lenssen et al. (2024)ref.
The GISTEMP uncertainty contains 200 equally likely reconstructions of global historical monthly temperature anomalies on a 2°×2° grid from 1880-2020. Using this ensemble, new estimates of global annual uncertainty in the global temperature anomaly are calculated and are broadly consistent with previous estimates by GISTEMP and other analyses (Fig. 1).
Figure 1: A comparison of 95% confidence intervals for the global annual temperature anomaly as calculated by GISTEMP and various other estimates of global mean temperature.
Ensemble Downloads
For all downloads that request you to authorize, please enter userid "uncertainty" and password "analysis".
Download the key global, hemispheric, and zonal monthly mean series of 200-member ensemble (38 MiB ZIP archive)
Download the full 200-member ensemble (16 ZIP archives, each about 700 MiB)
Download gridded summary statistics of the 200-member ensemble (712 MiB ZIP archive)
Note: The netCDF data files were updated 2024-10-01 to fix a minor error in labeling the time dimension.
Best Practices for Using the GISTEMP Ensemble
Uncertainty ensembles of gridded products and key time series make including observational uncertainty in subsequent analyses simple. This study provides 200 possible realizations of global temperature anomalies on a 2°x2° grid for each month from 1880--2020 to be used in analyses that take in gridded historical surface temperature analyses and their input data. In addition, many studies use large-scale time series of historical temperature anomalies such as monthly or annual global and hemispheric mean series. Thus, we also provide 200-member ensembles of key large-scale series.
Given an analysis developed using the operational version of GISTEMP, only one additional analysis is required to fully incorporate observational uncertainty. A user of the ensemble will simply rerun their analysis on each member of the uncertainty ensemble, collecting the result of interest each time. When this process is complete, there will be 200 possible values of the result of interest, forming a distribution of the results given observational uncertainty. Then, results should be summarized using the mean or median estimate of the observational uncertainty distribution of results. The final uncertainty in the result of interest due to observational uncertainty can be described by the standard deviation or variance of the results if they are reasonably symmetric or as empirical confidence intervals.
If an analysis uses large-scale series such as the global mean temperature anomaly as input data, it is recommended that the 200-member large-scale time series ensembles be used instead of the gridded ensemble. This is because the time series ensembles account for sampling uncertainty in regions not estimated by GISTEMP (regions that show up as `NA` in gridded GISTEMP). Any other analysis should use the gridded ensemble as it provides the spatiotemporal structure of the estimated temperature anomaly field as well as the uncertainty.
It is particularly important to include observational uncertainty in historical surface temperature when analyzing regions with high uncertainty such as the polar regions as well as areas with lower forced signals, such as investigations over the tropics, the eastern tropical Pacific, and the Southern Ocean. However, uncertainty ensembles make the addition of observational uncertainty so easy that the effect of observational uncertainty should be explored in any analysis using historical surface temperature data.
Finally, all analyses that utilize one surface temperature product should confirm the results are consistent with multiple surface temperature analyses, particularly when utilizing SST observations in the early record and high latitude LSAT observations post-1960 as these estimates show the greatest spread in uncertainty between products. The GISTEMPv4 observational ensemble will join the HadCRUT5 ensemble in being publicly available through an official website. In addition, the NOAA GlobalTemp and DKRZ ensembles are currently available upon request. Together, these four historical surface temperature ensembles span the major methods and data sources used to estimate surface temperature and any conclusions that rely on surface temperature data should be robust to both observational uncertainty as well as the epistemic uncertainty as sampled by the four available uncertainty ensembles.
Code for Ensemble Generation
Note: The ensemble requires the 100 member GHCN-ERSST-GISTEMP ensemble. This is available in the intermediate data Zenodo repository (see below).
The code for the analysis is provided in the ensemble downloads (2.6 MB ZIP archive). The open-source software R is required to run. See the namelist for required libraries.
Zenodo Repositories of Published Ensemble
The results and raw data used in the Lenssen et al. (2024) analysis can be found on the Zenodo repository at doi:10.5281/zenodo.13343335. All intermediate steps, including the 100 member GHCN-ERSST-GISTEMP ensemble can be found on Zenodo at 10.5281/zenodo.13344579.
Previous GISTEMP Uncertainty Quantification
Lenssen et al. (2019)ref previously performed an uncertainty assessment of the GISTEMP analysis, focusing on quantifying uncertainty in large-scale annual mean temperature anomalies. That analysis, now deprecated in favor of this ensemble analysis, is described elsewhere.
References
Lenssen, N., G.A. Schmidt, M. Hendrickson, P. Jacobs, M. Menne, and R. Ruedy, 2024: A GISTEMPv4 observational uncertainty ensemble. J. Geophys. Res. Atmos., 129, no. 17, e2023JD040179, doi:10.1029/2023JD040179.
Lenssen, N., G. Schmidt, J. Hansen, M. Menne, A. Persin, R. Ruedy, and D. Zyss, 2019: Improvements in the GISTEMP uncertainty model. J. Geophys. Res. Atmos., 124, no. 12, 6307-6326, doi:10.1029/2018JD029522.
Note: PDF documents require the free Adobe Reader or compatible viewing software to be viewed.
