An earth system model shows self-sustained melting of permafrost even if all man-made GHG emissions stop in 2020

The risk of points-of-no-return, which, once surpassed lock the world into new dynamics, have been discussed for decades. Recently, there have been warnings that some of these tipping points are coming closer and are too dangerous to be disregarded. In this paper we report that in the ESCIMO climate model the world is already past a point-of-no-return for global warming. In ESCIMO we observe self-sustained melting of the permafrost for hundreds of years, even if global society stops all emissions of man-made GHGs immediately. We encourage other model builders to explore our discovery in their (bigger) models, and report on their findings. The melting (in ESCIMO) is the result of a continuing self-sustained rise in the global temperature. This warming is the combined effect of three physical processes: (1) declining surface albedo (driven by melting of the Arctic ice cover), (2) increasing amounts of water vapour in the atmosphere (driven by higher temperatures), and (3) changes in the concentrations of the GHG in the atmosphere (driven by the absorption of CO2 in biomass and oceans, and emission of carbon (CH4 and CO2) from melting permafrost). This self-sustained, in the sense of no further GHG emissions, melting process (in ESCIMO) is a causally determined, physical process that evolves over time. It starts with the man-made warming up to the 1950s, leading to a rise in the amount of water vapour in the atmosphere—further lifting the temperature, causing increasing release of carbon from melting permafrost, and simultaneously a decline in the surface albedo as the ice and snow covers melts. To stop the self-sustained warming in ESCIMO, enormous amounts of CO2 have to be extracted from the atmosphere.

Sensitivity analysis of 14 randomly chosen uncertain parameters from the model. Sampled independently using Latin-Hypercube sampling from random uniform distributions with ranges of plus minus 10 % around their standard value for 200 sensitivity runs. For the parameters see Supplement Table 1.
In all runs, the self-sustaining melting of permafrost is maintained in the model. Sensitivity analysis of 14 randomly chosen uncertain parameters from the model. Sampled independently using Latin-Hypercube sampling from random uniform distributions with ranges of plus minus 10 % around their standard value for 200 sensitivity runs. For the parameters see Supplement Table 1.
In all runs, the self-sustaining melting of permafrost is maintained in the model.
The parameters for the sensitivity analysis of 14 randomly picked uncertain parameters from the model. Sampled independently using Latin-Hypercube sampling from random uniform distributions with ranges of plus and minus 10 % around their standard value for 200 sensitivity runs. Results for Scenario 1 are shown in Figure 3a and Supplement Figure 1, results for Scenario 2 are shown in Figure 3b and Supplement Figure 2.  Sensitivity to change in the fraction of carbon that is converted (by bacteria) from CH₄ to CO₂ before it leaves the melting permafrost. In the base run of ESCIMO all C being released from melting permafrost is released as CH₄. In these runs we uniformly sample this fraction from 0% (all C released as CO₂) to 100% (all C released as CH₄) for 500 runs. The yellow area in each graph shows 50% of the resulting runs, the green area extends this to show 75% of the resulting runs.
In all runs, the self-sustaining melting of permafrost is maintained in the model. Sensitivity to change in the fraction of carbon that is converted (by bacteria) from CH₄ to CO₂ before it leaves the melting permafrost. In the base run of ESCIMO all C being released from melting permafrost is released as CH₄. In these runs we uniformly sample this fraction from 0% (all C released as CO₂) to 100% (all C released as CH₄) for 500 runs. The yellow area in each graph shows 50% of the resulting runs, the green area extends this to show 75% of the resulting runs.
In all runs, the self-sustaining melting of permafrost is maintained in the model. For scenario 1 (a) and 2 (b) Sensitivity to change in the fraction of carbon that is converted (by bacteria) from CH₄ to CO₂ before it leaves the melting permafrost. In the base run of ESCIMO all C being released from thawing permafrost is released as CH₄. In these runs here we uniformly sample this fraction from 0% (all C released as CO₂) to 15% of all C released as CH₄ for 500 runs. The grey area in each graph shows 75% of the resulting runs.
In all runs, the self-sustaining melting of permafrost is maintained in the model. Sensitivity to change in the fraction of carbon that is converted (by bacteria) from CH₄ to CO₂ before it leaves the melting permafrost. In the base run of ESCIMO all C being released from melting permafrost is released as CH₄. In these runs here we uniformly sample this fraction from 0% (all C released as CO₂) to 15% of all C released as CH₄ for 500 runs. The yellow area in each graph shows 50% of the resulting runs, the green area extends this to show 75% of the resulting runs.
In all runs, the self-sustaining melting of permafrost is maintained in the model. Sensitivity to change in the fraction of carbon that is converted (by bacteria) from CH₄ to CO₂ before it leaves the melting permafrost. In the base run of ESCIMO all C being released from melting permafrost is released as CH₄. In these runs here we uniformly sample this fraction from 0% (all C released as CO₂) to 15% of all C released as CH₄ for 500 runs. The yellow area in each graph shows 50% of the resulting runs, the green area extends this to show 75% of the resulting runs.
In all runs, the self-sustaining melting of permafrost is maintained in the model. Sensitivity to change in the slope of the rate of melting of the permafrost that results from a given temperature. We run 500 sensitivity runs where we pick values within plus or minus 10% of the base run slope (which is 1) from a uniform random distribution. The yellow area in each graph shows 50% of the resulting runs, the green area extends this to show 75% of the resulting runs.
In all runs, the self-sustaining melting of permafrost is maintained in the model. Sensitivity to change in the slope of the rate of melting of the permafrost that results from a given temperature. We run 500 sensitivity runs where we pick values within plus or minus 10% of the base run slope (which is 1) from a uniform random distribution. The yellow area in each graph shows 50% of the resulting runs, the green area extends this to show 75% of the resulting runs.
In all runs, the self-sustaining melting of permafrost is maintained in the model.
Randers and Goluke, Supplement to "An earth system model shows self-sustained melting of permafrost ..." Page 12

Supplement Figure 10
Sensitivity to change in the slope of the additional blocking of long wave surface radiation that results from additional water vapour in the atmosphere. Since the relationship is not linear, we could not change the slope by a fixed percentage to generate the sensitivity runs. Instead we created changes in the relationship as follows: Blue dots represent the historical data we derived from calibrating the entire climate system to historical values of temperature, carbon and heat flows, albedo, etc.
The thick black line is the standard ESCIMO extension of history into in the region beyond what has been observed this far. The formula we use is a 3 rd order polynomial f(x) = -0.2842 * humidity³ + 1.8344 * humidity² -3.7148 * humidity + 2.4523, valid for the region of humidity from 2.1 to 2.5 g/kg.
The yellow lines show how the black line has been randomised for 50% of the sensitivity runs, the green lines show how the black line has been randomised for a further 75% of the sensitivity runs.
The relationship above reflects the lack of real world knowledge of the effect of water vapour, not a wellmixed GHG, on the blocking of heat transfer to space. Once we learn more about the real world relationship, we can incorporate the new knowledge into ESCIMO with relative ease. Sensitivity to change in the slope of the additional blocking of outgoing radiation that results from additional water vapour in the atmosphere. How we change the slope of this relationship is detailed in Supplement Figure  10, two pages back.
In all runs, the self-sustaining melting of permafrost is maintained in the model.
Randers and Goluke, Supplement to "An earth system model shows self-sustained melting of permafrost ..." Page 15

Supplement Figure 13
Another way to explore whether we have, in ESCIMO, passed the point-of-no-return in temperature rise is to run counterfactual experiments by cutting GHG emissions abruptly to zero at various points in the past.
The left panel above shows the cuts initiated at 10-year intervals from 1950 to 2030.
The right panel above shows the resulting global surface temperature difference to 1850 resulting from the various experiments. Sometime between 1960 and 1970 the trajectory takes on the characteristic selfsustaining pattern we see today. We also tested, in ESCIMO, whether removing carbon from the atmosphere is sufficient to avoid self-sustaining temperature rise.
The top left panel above shows the removal experiments, under scenario 1.
The top right panel above shows the resulting global surface temperature difference to 1850 resulting from the various experiments. It is possible, in ESCIMO, to avoid self-sustaining temperature rise if 1) enough GHGes are removed annually, at least 10 GtC/yr (black curve), and 2) if this removal effort continues at least until 2500. The reason that removal has to continue for so long is that the combination of reduced albedo, CH₄ release and water vapour, elevated due to the higher temperature, tries to keep the temperature high. GHG removal thus has to overcompensate, which can be seen in the middle left panel where CO₂ concentration falls below pre-industrial levels past the year 2050. We also tested, in ESCIMO, whether removing carbon from the atmosphere is sufficient to avoid self-sustaining temperature rise. The top left panel above shows the removal experiments, under scenario 1. It is the same as in Supplement Figure 14 a, but carbon is shown in GtCO₂e. ESCIMO runs in GtC throughout. Thus, when we calculate the effect of GHG molecules once they are in the atmosphere, we use the instantaneous effect. Since people are used to seeing summaries of all GHGs in GtCO₂e, we use, for display purposes when we show these summaries, the 100 year global warming potential method to convert gases to CO 2 e.
The top right panel above shows the resulting global surface temperature difference to 1850 resulting from the various experiments. It is possible, in ESCIMO, to avoid self-sustaining temperature rise if 1) enough GHGes are removed annually, at least 33 GtCO₂e/yr (black curve), and 2) if this removal effort continues at least until 2500.
The reason that removal has to continue for so long is that the combination of reduced albedo, CH₄ release and water vapour, elevated due to the higher temperature, tries to keep the temperature high. GHG removal thus has to overcompensate, which can be seen in the middle left panel where CO₂ concentration falls below pre-industrial levels past the year 2050.