Climate crisis: Mapping where the earth will become uninhabitable

Climate (and other, e.g. land use) models try to simulate both the past and future. They represent good approximations, but not reality. Model results for historical time frames, when compared with the records of actual events like numbers of storms or maximum temperatures, don't correspond precisely. For this reason, they need to be aligned with historical observations. For future values, one additionally needs to decide on input parameters like expected CO₂ emissions (see What are RCP scenarios?). For both historical and future projections, usually averages across as many models as possible are used, so biases of individual models are balanced out (e.g. tendencies to over- or underestimate effects in certain regions). These models are continuously improved, so newer models (e.g. CMIP6-models, which are used for the new report of the Intergovernmental Panel on Climate Change) are more reliable than older ones.

To find out what effect future CO₂ emissions of humans will have on the climate, future projections are usually simulated for different scenarios. Scientists have largely settled on the so-called ‘Representative Concentration Pathways’, short RCP, scenarios over the past years. The most common ones are:

• RCP8.5: this scenario assumes that CO₂ emissions caused by humans continue to increase as they have so far, and that no counter-measures are implemented. By the year 2100, according to calculations from 2013, this would lead to global warming of around 4.3°C.
• RCP6.0: this scenario assumes CO₂ emissions that would cause global warming of about 2.8°C by 2100. Emissions would continue to increase until 2080, though a bit more slowly than in the RCP8.5 scenario, and then decline rapidly.
• RCP4.5: this one describes a scenario in which the planet would heat up by about 2.4°C by 2100 compared to pre-industrial levels. Emissions wouldn’t need to go down to zero, but would stabilise at lower levels than today (less than half) after 2080.
• RCP2.6: this scenario assumes the most drastic declines in worldwide CO₂ emissions. In the IPCC’s latest report, scientists expected global warming of about 1.6°C by 2100 for this scenario. However, this assumed that emissions would go down starting 2020, which didn’t happen. Additionally, this scenario also assumes net negative CO₂ emissions starting in about 50 years propelled by technologies which are still in development.

Researchers involved in the Climate Action Tracker expect global warming of about 2.7°C by 2100 (compared with pre-industrial levels) based on current policies and implemented measures by monitored states. If we assume that governments will additionally keep to all non-binding climate targets, the estimated warming levels amount to between 2.1°C and 2.4°C by 2100.

The height of the columns represents the number of people that, based on data from SEDAC / NASA / Columbia University , lived in the indicated areas in the year 2020. All points from the original grid data (2,5’ resolution, maximum distance between points around 5km) that fall inside the hexagonal base area of a column are added.

Data from Copernicus Climate Change Service / ECMWF includes maximum wet-bulb temperatures per day for different models. From this, we calculated the average number of days at 32°C or more across 18 climate models for the time periods 1990-2019 and 2070-2099 (RCP4.5). For each column visualised on the globe, the value of the grid cell (1.875° x 1.25° resolution) covering the midpoint of the base area was used. For columns in coastal areas, the average of all grid cells which the base area is touching was used. Since the original data is ‘bias-adjusted’, the values have already been synchronised with historical observations and no further conversions were necessary.

For present-day data, we used the World Resources Institute’s (WRI) water stress index for 50.000 sub-basins worldwide. For each visualised column, the ‘water stress’ value was extracted at the centre of its base area.
There are barely any projections about water stress by the end of the century. Upon request, data was made available to us from a paper by Stenzel et al., where the authors simulated the water stress index for 259200 raster points for the years 2006-2015 and 2090-2099 (RCP6.0) based on one climate and land use model each. From these projections, we calculated mean values for the change (in percent) for each column. The projected change in water stress is based on the assumption of moderate global population growth that levels off in the second half of the century (SSP2). For values for the year 2100, the previously calculated factors (change in percent) were multiplied with their respective water stress values from the WRI’s data.
Notes:

• We additionally visualised desert areas, see How did you calculate the data about desert regions?
• The model data we used is based on only one model. Averages across multiple models are more robust. The scientists looked at four different models, though, and decided on the model HadGEM2-ES because its values were somewhere in the middle from the selected models for most variables.
• For the model data, 10-year periods were used; the results are therefore less reliable than for the usual 30-year periods.

For desert regions, data from FAO and IIASA (United Nations’s Food and Agriculture Organization) about climate classification according to Köppen-Geiger was used. We took values from historical observations for the years 1981-2010 and data based on the climate model HadGEM2-ES and RCP6.0 scenario for the years 2070-2100, to match the water stress projections. For each column, the value of the closest raster point (5' x 5' resolution) to the base area’s centre was used. All columns that correspond to desert climate classifications (BWh or BWk, hot and cold deserts) are colored in the same hue as water stress index values of 80%.

In order to estimate which fraction of each column’s population would be subjected to flooding in their area, we used openly accessible data from a paper by Hooijer et al.. The authors include an elevation model of coastal areas and estimations for mean global ‘relative sea level rise‘ levels, which combine expected sea level rise and expected land surface subsistence. Assuming global warming between 2.5°C and 3°C by 2100, the expected relative sea level rise would be around 1 metre by 2100. We checked which areas from the elevation model (0.05° resolution) are lower than 1m above sea level, and calculated the fraction of the population who live in these areas per column. For present-day values, we did the same, but looked at areas that already lie below sea level.

Based on gridded data from Lange et al. / ISIMIP about areas affected by tropical cyclones, values for 30-year periods corresponding to warming levels between 2°C and 3°C (based on threshold values calculated by ISIMIP), and values for the years 1990-2019, were extracted. From the averages across these years and all four models, we calculated the change in percentage for each grid point. From another data set from Geiger et al., which contains information about actual storms between 1950 and 2015, we calculated the average number of tropical cyclones that hit each raster point location per year. For future values, the values derived from the second dataset were multiplied with the above-mentioned change percentage. For each column, the average of all raster points inside its base area were used.

Our data is based on various models. These models are global in their extent, but vary in their resolution and how well they capture small areas of land. Because of this, small island nations are not visible in some models. In such cases we marked the affected states with the message ‘unfortunately there is no data available for the selected country and variable’ in order to clearly distinguish them from countries that are not affected (‘in this region there are no areas affected by the selected variable’).

Here you can find the raw data out visualisations are based on. For a text-only version of this page, please click here.

Special thanks to Áine Kelly-Costello for proof-reading the English text version.