Time periods for which +1.5, +2, +3 and +4 degree Transient Global Warming Levels (GWLs) are reached (with respect to pre-industrial 1850-1900 mean value) are computed for CMIP5 and CMIP6 data using a 20-year moving window, building on the datasets listed in the corresponding CMIP5 and CMIP6 data sources.
The approach is similar to that described by e.g. Nikulin et al. (2018). The values provided in the GWL tables in this directory (CMIP5_Atlas_WarmingLevels.csv and CMIP6_Atlas_WarmingLevels.csv) correspond to the central year (n) of the 20-year window where the warming is first reached (the GWL period is thus calculated as [n-9, n+10]). Cells with 'NA' indicate that the GWL was not reached before (the central year) 2100. Cells with '9999' correspond to models with no available data for the particular scenario. The script provided for GWL calculation builds directly on the information available at the datasets-aggregated-regionally directory, in particular using the global values in the last column of the files ('tas_landsea' csv files, under the "world" heading).
The use of a 20-year moving window is selected to be consistent with 20-year time slices typically used for future projections: the near-term (2021-2040), mid-term (2041-2060) and long-term (2081-2100). However, the figures in the CMIPx_WarmingLevels_spread_scenario.pdf files compare the results for 20- and 30-year windows using the large CMIP5/CMIP6 ensembles.
The scripts folder also contains a script to produce plots of the GWL crossing time, with a flexible parameter setting. For instance, the following parameter set:
cmip <- "CMIP6"
gwl <- 2 # Possible values: 1.5, 2 , 3 or 4 (degrees)
exp <- "ssp370" # This is a CMIP-dependent parameter
window <- 20 # window width (in years) for centered moving average.produces the following plot:
See also the global-warming-levels notebook in the notebooks folder to learn the basics of GWL calculation.
A similar repository for GWL calculation is maintained by Mathias Hauser and provides similar information which allows double-checking the results. Small differences (1-year shifts) are attributable to different postprocessing methods and/or different members/versions.