International risk of food insecurity and mass mortality in a runaway global warming scenario

By Vasco Grilo🔸 @ 2023-09-02T07:28 (+15)

This is a linkpost to https://www.sciencedirect.com/science/article/pii/S0016328723000770

This is a linkpost for International risk of food insecurity and mass mortality in a runaway global warming scenario by C.E. Richards, H.L. Gauch, and J.M. Allwood, published on June 2023. The abstract, factors over and underestimating starvation, and my quick overall take are below.

Abstract

Climate and agriculture have played an interconnected role in the rise and fall of historical civilizations. Our modern food system, based on open-environment production and globalised supply chains, is vulnerable to a litany of abiotic and biotic stressors exacerbated by anthropogenic climate change. Despite this evidence, greenhouse gas emissions continue to rise. Current trajectories suggest global warming of ∼2.0–4.9 °C by 2100, however, a worst-case emissions scenario with rapid combustion of all available fossil fuels could cause a rise of ∼12 °C. Even if emissions decline, unprecedented atmospheric CO₂-e concentrations risk triggering tipping points in climate system feedbacks that may see global warming exceed 8 °C. Yet, such speculative ‘runaway global warming’ has received minimal attention compared to mainstream low- to mid-range scenarios. This study builds on The Limits to Growth to provide new insights into the international risk of mass mortality due to food insecurity based on a higher-resolution illustration of World3’s ‘runaway global warming’ scenario (∼8–12 °C+). Our simulation indicates rapid decline in food production and unequal distribution of ∼6 billion deaths due to starvation by 2100 [or is it 5 billion^[1]?]. We highlight the importance of including high-resolution simulations of high-range global warming in climate change impact modelling to make well-informed decisions about climate change mitigation, resilience and adaptation.

Factors overestimating starvation

Holding land use constant^[2]:

We sourced grid cell data on land use classifications for the base year (2020)—noting that we hold land use constant in preparing this dataset as crop land growth is accounted for separately—from Chen et al. (2020).

Not including all sources of food:

This approach—as adopted by other studies (e.g., Hasegawa et al. (2021)) for computational simplicity, and where crops constitute ∼90 % of human food energy intake (Su et al., 2017)—accounts for conventional crop-based agriculture but does not include aquaculture, animal-based agriculture (though feed crops are included), nor unconventional ‘future foods’ (Parodi et al., 2018), and thus likely underestimates total food produced by a given country.

Assuming constant cropland area:

Secondly, crop land growth does not evolve in response to climate-impacted crop yields, and thus adaptation to the impacts of climate change on food production through agricultural land expansion were likely underestimated.

Not incorporating learning feedbacks nor exploring societal adaptation:

For instance, while the food production input data does incorporate some conventional technological development for higher-crop yields in SSP5-RCP8.5, our simulation does not account for advances in unconventional ‘future food’ production systems, nor polycentric food networks, described by Tzachor et al. (2021). As such, we likely overestimate the risk of starvation in this regard.

Factors underestimating starvation

Only accounting for energy requirements:

We used custom units of energy-equivalent-grain-weight-at-harvest to standardise food production and food demand quantities in simulation calculations. Although accounting for production of some 128 crops, this approach—as adopted in other studies (e.g., Molotoks et al., 2021) for computational simplicity—accounts only for energy but not for other nutritional contents. Dietary diversity—i.e. not only energy from carbohydrates, proteins and fats but also non-energy aspects of these macro-nutrients, such as amino acids, essential fatty acids and fibres, as well as micro-nutrients, including vitamins and minerals such as iron, zinc and calcium—is essential to food security, and thus our approach likely underestimates the potential for malnutrition.

Accelerating both climate change and technological development instead of just climate change:

Firstly, the raw data on crop yields in SSP5-RCP8.5 incorporates both climate change (positive and negative depending on region) and technological development effects (positive) and therefore could not be treated in isolation in this study. As such, both effects were accelerated in modelling the artificial ‘runaway global warming’ scenario, and thus the net impacts of climate change on food production were likely underestimated.

Not incorporating socioeconomic growth shock feedbacks:

For instance, while we account for population loss feedback in absolute terms, such a loss of population, and the potential turmoil that may follow a largescale starvation event, would likely have an impact of stunting the GDP growth of a nation compared to the business-as-usual growth adopted in the model, which would in turn feedback to further exacerbate the risk of starvation. As such, we likely underestimate the risk of starvation in this regard.

My quick overall take

Most studies analysing the impact of climate change focus on low levels of warming, so I am glad this article looked into more extreme scenarios, as asked in Kemp et al. 2022.

On the other hand, 5 billion deaths given a global warming of 10 ºC (= (8 + 12)/2) by 2100 seems very pessimistic neglecting AI risk. Xia 2022 estimated 5.08 billion people without food given a cropland cooling of 16 ºC in 1 year, corresponding to a global cooling of 8 ºC^[3], and little adaptation^[4]. Moreover, I think 1 ºC of cooling is usually worse for yields than 1 ºC of warming. So it would be quite surprising if a temperature change 1.25 times (= 10/8) as large over 80 times as much time (2020 to 2100) were similarly bad.

I believe holding land use and cropland area constant, and not incorporating learning feedbacks nor exploring societal adaptation for such a long period of 80 years implies deaths have been majorly overestimated. I guess by more than a factor of 10. It is also unclear to me whether CO${}_2$ fertilization was taken into account, as I did not find "fertilization" in the article nor its supplementary material^[5].

1. ^{^}

   From section 5.1 (emphasis mine):

   Our results then show a stark divergence in risk of starvation as ‘runaway global warming’ results in mass mortality of ∼5 billion people (i.e., ∼68 % of the current global population) by 2100.

2. ^{^}

   Also from the supplementary material (emphasis mine):

   The food_production_growth_cropyield input dataset contains a SSP5-RCP85 based growth profile for the crop yield (i.e. climate and technology) related component of food production growth, in units of annual average percentage, for each country in yearly increments from 2020 to 2099. It was derived from high resolution GIS data on climate impacted crop yields based on scenario SSP5-RCP85 from (Iizumi et al. 2017, 2020) and current land use from (Chen et al. 2020), as follows.

3. ^{^}

   See Fig. 3a of Toon 2014. The cropland cooling is larger than the global cooling because this accounts for the air temperature over oceans, which does not change as much.

4. ^{^}

   See 3rd paragraph of the Discussion, or discussion on LessWrong.

5. ^{^}

   I have asked the corresponding author 14 days ago, but have not heard back.