Hi there, Volcanologist here. Firstly, thanks for making the effort to look into this area and engage with it.
TLDR: I agree that extinction risk from volcanic risk is extremely extremely low*… (*on its own). But this post comes up with a single probability number, based on one new (and not yet established) paper based on playing about with climate model parameters. You could do a dive into another volcano-climate paper and get a completely different view of impacts e.g. https://iopscience.iop.org/article/10.1088/2752-5295/acee9f/meta. In fact, you could also see this same paper from a different side and say that it could increase the x-risk factor, by (depending on what aerosol size you pick and choose) heating the earth and driving rapid warming on top of Anthropogenic warming. Does this mean that we should not worry about volcanic eruptions as contributing to existential risk? My answer would be no.
To suppose that you can predict (and put a probability value) on what the societal reactions and subsequent decisions made, the interaction between other risks (AI, wars, pandemics etc) and geo-political landscape and actions would be when faced with the impacts of a large volcanic eruption (note that I say large and not super-eruptions- cause we’re not just talking about super eruptions here), seems overly reductionist and simplistic to me. For instance, When faced with the starvation of hundreds of millions of people (if not billions- though I think that study overestimates agricultural loss), as well as droughts and other extreme weather events (failure of monsoons etc), energy security, financial loss of tens of trillions etc, and knowing that often it’s the state of the world (e.g. vulnerability, inequality) and the geopolitical reactions to that risk, that really control whether the risk could be minimised through global governance cooperation and preparedness actions, or whether humanity’s reactions to such risks (e.g. banning exports) may worsen the initial risk considerably, as well as cascading risks. So for me, judging how volcanic eruptions contribute to civilisation collapse or x-risk needs to incorporate these factors, This post does not engage with this, and bases their numbers on one paper and has extrapolated using non-rigorous constriants.
Point about the paper itself:
I asked my climate science colleague about the paper and they had a lot of uncertainty on on how much trust we should place on this: “I would be wary about taking this 1.5C as ground truth...Lots of disagreement between our models, and lots of processes they don’t account for yet (e.g. halogen, water, injection height, time of eruptions, eruption time frame). The proxy record is always tricky as it’s a mix of forced response and natural variability, and not a full global picture. My understanding is also that if anything, tree ring underestimate peak cooling because they integrate the response a bit. So I don’t think they reliably support or don’t support this limit.”
“Models with interactive stratospheric aerosols have been telling us for ~15 years that aerosol growth would strongly limit the cooling after very large events. There might be disagreement on maximum aerosol radius/cooling because there is still a large uncertainty in stratospheric aerosol modelling (Clyne et al 2021)”
A few of my own thoughts about the same paper Vasco cites:
- We really don’t understand aerosol size at all, let alone for smaller eruptions (For example, In smaller eruptions, aerosol size change throughout eruptions and we don’t know why this is yet (link: https://acp.copernicus.org/articles/23/9725/2023/)), and so we have an even poorer grasp on aerosol size for bigger eruptions (they say as much in the paper- and as much as anything-the paper advocates for more experiment studies- I totally agree).
- Published proxy records record greater cooling than 1.5 degrees (mostly summer cooling and mostly based in Northern hemisphere- but that’s where the tree records are), e.g. the 540 eruption. So while it’s possible that 1.5 degrees is the limit, based on this smaller eruption exceeding that limit, it seems unlikely.
- Eruptions often do not erupt all their Sulfur in one go, and super eruptions especially likely last years and will be separated by breaks in activity, this periodicity mostly certainly affects the aerosol size- this doesn’t seem to be mentioned at all.
General points:
But your point about extinction risk being very low is mostly right, though you can do the same by diagnosing the statistics (super eruptions occur every 17,000 years or so, so it’s unlikely that humans would go extinct as we haven’t done previously 300,000 years). But there’s some more complicated caveats to this:
- We’ve never experienced one since human civilisations started ~12,000 years ago, the last one was 26,000 ago in New Zealand, and let alone since modern interconnected, globalised society.
- Since then we’ve globalised the food trade, and its susceptible to harvest fails if all grown in certain breadbaskets, and hydroclimate extreme events following eruptions make this far more likely (see paper cited above), not to mention trade choke points.
- Saying this the conclusion that volcanic eruptions cannot lead to human extinction* (*on its own). This is where it get a bit more complicated, and often analyses like this and Toby Ord’s analyses feel too reductionist and simplistic. For instance, many societal collapses in previous history (many coincide with eruptions), but they are rarely due to one factor alone. However if you were to take out say catastrophic volcanic eruptions, then they would likely not happen at all.
- This is also apparent when we look at major collapses, which don’t always correlate with the largest climatic impacts, so the idea that you approximate existential threat based on the magnitude of the effects seems a bit of a strawman case for these reasons. This effect is accentuated by the regional impacts to breadbaskets or physical blocking of trade routes via direct volcanic material- this means that location of the eruption is super important too, not just the magnitude. Vulnerability, reactions and other compounding risks (wars, inequality etc) turn out to be much more important. The logic of Ord et al to assess one risk in isolation is not realistic and the numbers he generates have very little meaning in the real world.
In general, explosive volcanic eruptions will never be the sole cause for extinction, but they could be very meaningful in increasing existential risk factor, so that once another big risk(s) comes along they have increased the vulnerability to that threat, where we might have otherwise come through it without the eruption.
Finally this quote it makes it hard for me to relate with this wing of the Effective Altruism community:
“So I would argue interventions to decrease deaths from supervolcanoes can only be competitive under an alternative worldview, like ones where the goal is boosting economic growth or decreasing disease burden.”
Irrespective of extinction risk—should we care (and do something) about that fact that large volcanic eruptions could threaten the lives and livelihoods of billions, and this is for a risk that has a 1 in 6 per century probability?
Perhaps I’m in an alternative/more isolated worldview- but I would say yes.
Thanks for the thorough comment, Michael! I strongly upvoted it.
Irrespective of extinction risk—should we care (and do something) about that fact that large volcanic eruptions could threaten the lives and livelihoods of billions, and this is for a risk that has a 1 in 6 per century probability?
Perhaps I’m in an alternative/more isolated worldview- but I would say yes.
There might be a misunderstanding. The world as whole should definitely care about volcanic risk. However, I do not think funders aligned with effective altruism, which have a high bar, should fund interventions to decrease it unless they decrease disease burden or boost economic growth more cost-effectively than GiveWell’s funds. Do you disagree with this? If not, are you suggesting there are interventions to decrease volcanic risk which are more cost-effective than GiveWell’s grants (there are a few examples here)? For reference, GiveWell’s top charities can save 1 life for each 5 k$.
I also wanted to clarify what you think will be the expected death toll from volcanic eruptions this century. Your claim that there is a probability of 1⁄6 of a volcanic eruption threatening the lives of billions seems very high to me, at least interpreting “theatening the lives” as having a high chance of death. Here are my reasons:
Your article (the one linked just above) says the probability this century of a volcanic eruption with volcanic explosivity index (VEI) of 7 or 8 is 1⁄6, and VEI 8 is 1 or 2 OOMs less likely than VEI 7, so I think you are essentially arguing for billions of lives threatened conditional on an eruption of VEI of 7.
The 1815 eruption of Mount Tambora was VEI 7, but caused 5 (= log10(3*10^9/(71*10^3))) orders of magnitude less deaths than 3 billion deaths (geometric mean between 1 and 10 billion deaths, which is the range compatible with “billions of deaths”). “The total figure of 71 000 makes the Tambora eruption the deadliest known in history. In the database of volcanic disasters compiled by Tanguy et al. (1998), which begins with Laki in 1783 and ends with Montserrat 1997, Tambora accounts for nearly 30% of the total number of deaths”.
71 k is probably an underestimate because it neglects global impacts. Maybe the total impact was 3 times the direct one, i.e. 213 k deaths, although it is hard to tell due to confounders (the Napoleonic Wars ended just before the eruption in 1815, so they could explain starvation in Europe).
Population was 1.05 G in 1815, so I guess the death rate was roughly 0.02 % (= 213*10^3/(1.05*10^9)), which is similar to the current death rate from protein-energy malnutrition in the worst off countries (search for “Central African Republic” here).
We have also had 200 years of economic growth since then, so we are now more resilient (agriculture is now a much smaller fraction of the economy), and the death rate for the same eruption would be lower.
To suppose that you can predict (and put a probability value) on what the societal reactions and subsequent decisions made, the interaction between other risks (AI, wars, pandemics etc) and geo-political landscape and actions would be when faced with the impacts of a large volcanic eruption (note that I say large and not super-eruptions- cause we’re not just talking about super eruptions here), seems overly reductionist and simplistic to me. For instance, When faced with the starvation of hundreds of millions of people (if not billions- though I think that study overestimates agricultural loss), as well as droughts and other extreme weather events (failure of monsoons etc), energy security, financial loss of tens of trillions etc, and knowing that often it’s the state of the world (e.g. vulnerability, inequality) and the geopolitical reactions to that risk, that really control whether the risk could be minimised through global governance cooperation and preparedness actions, or whether humanity’s reactions to such risks (e.g. banning exports) may worsen the initial risk considerably, as well as cascading risks. So for me, judging how volcanic eruptions contribute to civilisation collapse or x-risk needs to incorporate these factors, This post does not engage with this, and bases their numbers on one paper and has extrapolated using non-rigorous constriants.
These are great points! However, it is unclear to me whether they affect prioritisation across risks. As I say in the post:
Note supervolcanoes might have cascade effects which lead to civilisational collapse, which could increase longterm extinction risk while simultaneously having a negligible impact on the nearterm one I estimated. I do not explicitly assess this in the post, but I guess the nearterm annual risk of human extinction from supervolcanoes is a good proxy for the importance of decreasing volcanic risk from a longtermist perspective:
Supervolcanoes might have cascade effects, but so do other catastrophes.
Global civilisational collapse due to supervolcanoes seems very unlikely to me. There would be global impacts via volcanic ash, but I guess infrastructure loss would be more significant for a similarly likely nuclear war.
Even if a volcanic eruption causes a global civilisational collapse which eventually leads to extinction, I guess full recovery would be extremely likely. In contrast, an extinction caused by advanced AI would arguably not allow for a full recovery.
Are you suggesting the ratio between total existential risk and nearterm annual extinction risk is higher for supervolcanoes than for other risks? If not, nearterm annual extinction risk could be used as a proxy for importance.
- Published proxy records record greater cooling than 1.5 degrees (mostly summer cooling and mostly based in Northern hemisphere- but that’s where the tree records are), e.g. the 540 eruption. So while it’s possible that 1.5 degrees is the limit, based on this smaller eruption exceeding that limit, it seems unlikely.
Note McGraw 2024 does not quite argue for 1.5 ºC being the maximum global cooling. The abstract says (emphasis mine):
we estimate that global volcanic cooling is unlikely to exceed 1.5°C no matter how massive the stratospheric injection
1.5 ºC is the maximum global cooling given McGraw 2024’s preferred relationship between aerosol size and ejected mass of SO2, but it can be as much as 5 ºC for their most pessimistic curve (see Fig. 10b, which is the last one in my post). I use this level of cooling to estimate the extinction risk.
I asked my climate science colleague about the paper and they had a lot of uncertainty on on how much trust we should place on this: “I would be wary about taking this 1.5C as ground truth...Lots of disagreement between our models, and lots of processes they don’t account for yet (e.g. halogen, water, injection height, time of eruptions, eruption time frame). The proxy record is always tricky as it’s a mix of forced response and natural variability, and not a full global picture. My understanding is also that if anything, tree ring underestimate peak cooling because they integrate the response a bit. So I don’t think they reliably support or don’t support this limit.”
“Models with interactive stratospheric aerosols have been telling us for ~15 years that aerosol growth would strongly limit the cooling after very large events. There might be disagreement on maximum aerosol radius/cooling because there is still a large uncertainty in stratospheric aerosol modelling (Clyne et al 2021)”
As I said, 1.5 ºC is the maximum global cooling given McGraw 2024’s preferred relationship between aerosol size and ejected mass of SO2, but it can be as much as 5 ºC for their most pessimistic curve (see Fig. 10b, which is the last one in my post), and I use this level of cooling to estimate the extinction risk.
But this post comes up with a single probability number, based on one new (and not yet established) paper based on playing about with climate model parameters.
Maybe this is not too concerning? My mainline estimate is quite uncertain, but I think my conclusion that nearterm annual extinction risk is astronomically low is robust. McGraw 2024:
Reviews the past literature, and its model is able to reproduce and explain differences among past studies.
Covers past results well. This is illustrated in Fig. 3b, which I present below.
Could you clarify if/why the results of this paper go against those of McGraw 2024? The paper linked above focuses on droughts and floods instead of global cooling.
In fact, you could also see this same paper from a different side and say that it could increase the x-risk factor, by (depending on what aerosol size you pick and choose) heating the earth and driving rapid warming on top of Anthropogenic warming.
I estimated an annual probability of 3.38*10^-8 for an eruption causing a maximum surface cooling of at least 5 ºC. Following the same approach results in the same annual probability for an eruption causing a maximum surface warming of at least 2.5 ºC (eyeballed from Fig. 10b). Then such astronomically low chance would still have to be multiplied by the annual probability of rapid anthropogenic warming (assuming independence), which I guess is very low too. So I guess the overall likelihood of a double catastrophe involving a supervolcanic eruption on top of rapid anthropogenic warming would be so low for the risk to be negligible, in the sense I think the risk from supervolcanoes and rapid anthropogenic warming could be well approximated by the sum each risk considered alone. In any case, the maximum warming is just 2 ºC (see Fig. 10b), so the increase in the nearterm extinction risk would be very small.
But your point about extinction risk being very low is mostly right, though you can do the same by diagnosing the statistics (super eruptions occur every 17,000 years or so, so it’s unlikely that humans would go extinct as we haven’t done previously 300,000 years).
As I mention in the post, “1 M years is the typical lifespan of a mammal species”. However, note I estimated a much lower nearterm annual extinction risk from supervolcanoes of 3.38*10^-14. Do you have specific disagreements about my calculations?
- We’ve never experienced one since human civilisations started ~12,000 years ago, the last one was 26,000 ago in New Zealand, and let alone since modern interconnected, globalised society.
- Since then we’ve globalised the food trade, and its susceptible to harvest fails if all grown in certain breadbaskets, and hydroclimate extreme events following eruptions make this far more likely (see paper cited above), not to mention trade choke points.
A priori, I think we should expect humanity to be more resilient against famines today than in the past. Poverty is a major risk factor for famines, and there has been major economic growth since the dates you mention.
For instance, many societal collapses in previous history (many coincide with eruptions), but they are rarely due to one factor alone. However if you were to take out say catastrophic volcanic eruptions, then they would likely not happen at all.
The more factors are needed to cause extinction, the less likely it will tend to be, because it is then more conjunctive (even if many of the factors are correlated; one should be mindful of the conjunction fallacy). For example, a supereruption and nuclear war is less likely than any of these alone. It is also worth noting that global civilisational collapse is much less likely than local societal collapse (there are many examples of the latter, but none of the former).
The logic of Ord et al to assess one risk in isolation is not realistic and the numbers he generates have very little meaning in the real world.
I do not think Toby Ord’s guesses for the existential risk given in The Precipice strictly rely on assessing risks in isolation. There is a section named “Risk Factors” in chapter 6.
- We really don’t understand aerosol size at all, let alone for smaller eruptions (For example, In smaller eruptions, aerosol size change throughout eruptions and we don’t know why this is yet (link: https://acp.copernicus.org/articles/23/9725/2023/)), and so we have an even poorer grasp on aerosol size for bigger eruptions (they say as much in the paper- and as much as anything-the paper advocates for more experiment studies- I totally agree).
I agree we know little. Nevertheless, McGraw 2024 considers 4 different relationships between aerosol size and the ejected mass of SO2, and I estimate extinction risk based on the most pessimistic one. Do you think the range of predictions from McGraw 2024’s model (illustrated in Fig. 3b) is overly narrow?
- Eruptions often do not erupt all their Sulfur in one go, and super eruptions especially likely last years and will be separated by breaks in activity, this periodicity mostly certainly affects the aerosol size- this doesn’t seem to be mentioned at all.
Interesting point! Smaller eruptions have smaller aerosol size (see Fig. 10a, which is the penultimate in my post), and therefore cause more cooling per ejected mass of SO2. A priori, I assume multiple consecutive eruptions from the same volcano over a few years can be interpreted as eruptions from different volcanoes. So I guess that, for the same ejected mass of SO2, the greater the number of eruptions, the greater the cooling. This would contribute towards McGraw 2024 underestimating the cooling caused by supereruptions (although the effect would be small if there is usually just one large eruption). Nonetheless, the greater the number of eruptions, the more time there would be to adapt, so the net effect of your point on extinction risk is not clear to me.
In general, explosive volcanic eruptions will never be the sole cause for extinction, but they could be very meaningful in increasing existential risk factor, so that once another big risk(s) comes along they have increased the vulnerability to that threat, where we might have otherwise come through it without the eruption.
I appreciate volcanic eruptions can be a risk factor for extinction. At the same time, I would be mindful of the conjunction fallacy. The more factors are needed to cause extinction, the less likely it will tend to be, because it is then more conjunctive.
Once again, thanks for the comprehensive feedback!
Thanks for the reply- sorry not a lot of time to address all of these points.
I think in a simplistic world- one in which we react to such threats as you expect with resilience and preparedness, then yes the risk of Tambora sized eruption could be minimal* (*by this I mean raising food and energy prices and dramatically impacting/causing excess mortality of poorer/vulnerable nations).
Yes the world is more resilient and famines are better etc (note that a VEI 5 eruption in 1982 led to the droughts and thus famine in Ethiopia in 1983-5 that led to half a million dying), but a more complex world has inherent weaknesses and new vulnerabilities, see here and here
But as mentioned, the world is unpredictable and multiple factors normally converge at the same time- e.g. the consequences of an eruption now, versus the consequences of an eruption where China has become politically isolated- could dramatically change the impacts, our response and cascading risk (e..g wars from resource shortages/export bans).
You use a philsophical term to make that point that factors acting together is a conjuction fallacy, but I say this is exactly what has happened throughout history for smaller eruptions which have dramatically affected civilisations for centuries.
Nonetheless, the greater the number of eruptions, the more time there would be to adapt, so the net effect of your point on extinction risk is not clear to me.
Yes you can guess this is right-and that’s fine to have this opinion, but when speaking to agircultural professors and historical perspectives (see the Little Antique Ice age in 6th century), in fact the opposite is true, a short spike is something that can be absorbed if global trade is working, sustained periods of low growth is more impactful.
Perhaps this is my scientific background, but how you put a number on such uncertain factors seems deeply unscientific/unrigourous to me.
I think in a simplistic world- one in which we react to such threats as you expect with resilience and preparedness, then yes the risk of Tambora sized eruption could be minimal* (*by this I mean raising food and energy prices and dramatically impacting/causing excess mortality of poorer/vulnerable nations).
Yes the world is more resilient and famines are better etc (note that a VEI 5 eruption in 1982 led to the droughts and thus famine in Ethiopia in 1983-5 that led to half a million dying), but a more complex world has inherent weaknesses and new vulnerabilities, see here and here
I appreciate an increasingly complex world can have inherent weaknesses and new vulnerabilities (e.g. Bailey 2017 analyses the effects of interruptions at chokepoints in global food trade). However, we have to consider these effects together with ones pushing towards greater resilience to catastrophes, such as the benefits of having a world which can dedicate a greater fraction of its resources to food (including greater capacity for international assistance of low income countries). I assume one way of assessing the net effect is looking into how the death rate from protein-energy malnutrition has evolved over time. According to data from the Global Burden of Disease (GBD) study, it decreased 77.8 % (= 1 − 2.74/12.27) from 1990 to 2019 (see below). So I would say the world has become more resilient against small food shocks.
Large shocks happen less often, so the trend above is less informative, but a priori I would expect resilience to small shocks to be correlated with resilience to large shocks.
a VEI 5 eruption in 1982 led to the droughts and thus famine in Ethiopia in 1983-5 that led to half a million dying
I do not think this is a fair characterization of the risk linked to VEI 5 eruptions:
The 1983–1985 famine in Ethiopia was caused by “drought, the Ethiopian Civil War and military policies taken by the Ethiopian government”. Even if the drought was totally caused by a volcanic eruption of VEI 5 (wikipedia’s page does not mention any volcanic eruption; could you link to an article?), such eruption would not be responsible for all of the famine deaths.
Some of the famine deaths would have to be attributed to the other causes, such that the deaths attributed to each cause added up to the total death toll.
VEI 5 eruptions happen once every decade or so. As a result, a naive interpretation of your statement (I am not saying you would endorse it!) would say VEI 5 eruptions cause in expectation a famine with 0.5 M deaths every decade or so. However, the expected death toll would be much lower if the other causes (Ethiopian Civil War and military policies taken by the Ethiopian government) have a much lower probability.
This is why I mentioned in my last comment that:
The more factors are needed to cause extinction, the less likely it will tend to be, because it is then more conjunctive (even if many of the factors are correlated; one should be mindful of the conjunction fallacy).
Similarly, the more factors are needed to cause large famines (e.g. not only a VEI 5 eruption, but also civil war, and lack of effective international assistance), the less likely they will tend to be. This effect exists unless there is a perfect correlation between all the contributing causes (e.g. if a VEI 5 eruption always caused a civil war, and lack of effective international assistance).
Perhaps this is my scientific background, but how you put a number on such uncertain factors seems deeply unscientific/unrigourous to me.
I agree my analysis is far from rigorous. However:
It is hopefully better to have a non-rigorous analysis plus your thorough comment than the alternative of no analysis.
To illustrate, I could say the nearterm annual risk of human extinction from supervolcanoes is low, but then some readers may infer from this anything ranging from 0 to maybe 0.01 %. The latter would arguably imply volcanic risk receiving way more attention from the effective alstruism community than it currently does, whereas my actual best guess of 3.38*10^-14 does not.
So I think I had better provide my best guess for the risk. At the very least, this facilitates agreement/disagreement. If I had just mentioned the extinction risk was extremely low, you may not have provided such detailed and valuable feedback due to perceived agreement.
I also think researchers quite often implicitly put numbers on metrics. For example, you said:
Irrespective of extinction risk—should we care (and do something) about that fact that large volcanic eruptions could threaten the lives and livelihoods of billions, and this is for a risk that has a 1 in 6 per century probability?
Perhaps I’m in an alternative/more isolated worldview- but I would say yes.
It seems to me that, by answering yes to your question, you are implicitly arguing that there are interventions to decrease the expected mortality of volcanic eruptions whose cost to save a life is sufficiently low for such interventions to receive more attention (e.g. more research like yours on volcanic risk). Moreover, if by “we” you mean not just the world as a whole, but funders aligned with effective alstruism too (namely, Open Philanthropy), you may be implicitly claiming that such interventions can have a cost to save a life lower than 5 k$ (that of GiveWell’s top charities). I think it would be great if someone (maybe you and Lara!) published in a peer-reviewed journal a standard cost-effectiveness analysis of the interventions you think are being overlooked.
Hi there, Volcanologist here. Firstly, thanks for making the effort to look into this area and engage with it.
TLDR: I agree that extinction risk from volcanic risk is extremely extremely low*… (*on its own). But this post comes up with a single probability number, based on one new (and not yet established) paper based on playing about with climate model parameters. You could do a dive into another volcano-climate paper and get a completely different view of impacts e.g. https://iopscience.iop.org/article/10.1088/2752-5295/acee9f/meta. In fact, you could also see this same paper from a different side and say that it could increase the x-risk factor, by (depending on what aerosol size you pick and choose) heating the earth and driving rapid warming on top of Anthropogenic warming. Does this mean that we should not worry about volcanic eruptions as contributing to existential risk? My answer would be no.
To suppose that you can predict (and put a probability value) on what the societal reactions and subsequent decisions made, the interaction between other risks (AI, wars, pandemics etc) and geo-political landscape and actions would be when faced with the impacts of a large volcanic eruption (note that I say large and not super-eruptions- cause we’re not just talking about super eruptions here), seems overly reductionist and simplistic to me. For instance, When faced with the starvation of hundreds of millions of people (if not billions- though I think that study overestimates agricultural loss), as well as droughts and other extreme weather events (failure of monsoons etc), energy security, financial loss of tens of trillions etc, and knowing that often it’s the state of the world (e.g. vulnerability, inequality) and the geopolitical reactions to that risk, that really control whether the risk could be minimised through global governance cooperation and preparedness actions, or whether humanity’s reactions to such risks (e.g. banning exports) may worsen the initial risk considerably, as well as cascading risks. So for me, judging how volcanic eruptions contribute to civilisation collapse or x-risk needs to incorporate these factors, This post does not engage with this, and bases their numbers on one paper and has extrapolated using non-rigorous constriants.
Point about the paper itself:
I asked my climate science colleague about the paper and they had a lot of uncertainty on on how much trust we should place on this: “I would be wary about taking this 1.5C as ground truth...Lots of disagreement between our models, and lots of processes they don’t account for yet (e.g. halogen, water, injection height, time of eruptions, eruption time frame). The proxy record is always tricky as it’s a mix of forced response and natural variability, and not a full global picture. My understanding is also that if anything, tree ring underestimate peak cooling because they integrate the response a bit. So I don’t think they reliably support or don’t support this limit.”
“Models with interactive stratospheric aerosols have been telling us for ~15 years that aerosol growth would strongly limit the cooling after very large events. There might be disagreement on maximum aerosol radius/cooling because there is still a large uncertainty in stratospheric aerosol modelling (Clyne et al 2021)”
A few of my own thoughts about the same paper Vasco cites:
- We really don’t understand aerosol size at all, let alone for smaller eruptions (For example, In smaller eruptions, aerosol size change throughout eruptions and we don’t know why this is yet (link: https://acp.copernicus.org/articles/23/9725/2023/)), and so we have an even poorer grasp on aerosol size for bigger eruptions (they say as much in the paper- and as much as anything-the paper advocates for more experiment studies- I totally agree).
- Published proxy records record greater cooling than 1.5 degrees (mostly summer cooling and mostly based in Northern hemisphere- but that’s where the tree records are), e.g. the 540 eruption. So while it’s possible that 1.5 degrees is the limit, based on this smaller eruption exceeding that limit, it seems unlikely.
- Eruptions often do not erupt all their Sulfur in one go, and super eruptions especially likely last years and will be separated by breaks in activity, this periodicity mostly certainly affects the aerosol size- this doesn’t seem to be mentioned at all.
General points:
But your point about extinction risk being very low is mostly right, though you can do the same by diagnosing the statistics (super eruptions occur every 17,000 years or so, so it’s unlikely that humans would go extinct as we haven’t done previously 300,000 years). But there’s some more complicated caveats to this:
- We’ve never experienced one since human civilisations started ~12,000 years ago, the last one was 26,000 ago in New Zealand, and let alone since modern interconnected, globalised society.
- Since then we’ve globalised the food trade, and its susceptible to harvest fails if all grown in certain breadbaskets, and hydroclimate extreme events following eruptions make this far more likely (see paper cited above), not to mention trade choke points.
- Saying this the conclusion that volcanic eruptions cannot lead to human extinction* (*on its own). This is where it get a bit more complicated, and often analyses like this and Toby Ord’s analyses feel too reductionist and simplistic. For instance, many societal collapses in previous history (many coincide with eruptions), but they are rarely due to one factor alone. However if you were to take out say catastrophic volcanic eruptions, then they would likely not happen at all.
- This is also apparent when we look at major collapses, which don’t always correlate with the largest climatic impacts, so the idea that you approximate existential threat based on the magnitude of the effects seems a bit of a strawman case for these reasons. This effect is accentuated by the regional impacts to breadbaskets or physical blocking of trade routes via direct volcanic material- this means that location of the eruption is super important too, not just the magnitude. Vulnerability, reactions and other compounding risks (wars, inequality etc) turn out to be much more important. The logic of Ord et al to assess one risk in isolation is not realistic and the numbers he generates have very little meaning in the real world.
In general, explosive volcanic eruptions will never be the sole cause for extinction, but they could be very meaningful in increasing existential risk factor, so that once another big risk(s) comes along they have increased the vulnerability to that threat, where we might have otherwise come through it without the eruption.
Finally this quote it makes it hard for me to relate with this wing of the Effective Altruism community:
Irrespective of extinction risk—should we care (and do something) about that fact that large volcanic eruptions could threaten the lives and livelihoods of billions, and this is for a risk that has a 1 in 6 per century probability?
Perhaps I’m in an alternative/more isolated worldview- but I would say yes.
Thanks for the thorough comment, Michael! I strongly upvoted it.
There might be a misunderstanding. The world as whole should definitely care about volcanic risk. However, I do not think funders aligned with effective altruism, which have a high bar, should fund interventions to decrease it unless they decrease disease burden or boost economic growth more cost-effectively than GiveWell’s funds. Do you disagree with this? If not, are you suggesting there are interventions to decrease volcanic risk which are more cost-effective than GiveWell’s grants (there are a few examples here)? For reference, GiveWell’s top charities can save 1 life for each 5 k$.
I also wanted to clarify what you think will be the expected death toll from volcanic eruptions this century. Your claim that there is a probability of 1⁄6 of a volcanic eruption threatening the lives of billions seems very high to me, at least interpreting “theatening the lives” as having a high chance of death. Here are my reasons:
Your article (the one linked just above) says the probability this century of a volcanic eruption with volcanic explosivity index (VEI) of 7 or 8 is 1⁄6, and VEI 8 is 1 or 2 OOMs less likely than VEI 7, so I think you are essentially arguing for billions of lives threatened conditional on an eruption of VEI of 7.
The 1815 eruption of Mount Tambora was VEI 7, but caused 5 (= log10(3*10^9/(71*10^3))) orders of magnitude less deaths than 3 billion deaths (geometric mean between 1 and 10 billion deaths, which is the range compatible with “billions of deaths”). “The total figure of 71 000 makes the Tambora eruption the deadliest known in history. In the database of volcanic disasters compiled by Tanguy et al. (1998), which begins with Laki in 1783 and ends with Montserrat 1997, Tambora accounts for nearly 30% of the total number of deaths”.
71 k is probably an underestimate because it neglects global impacts. Maybe the total impact was 3 times the direct one, i.e. 213 k deaths, although it is hard to tell due to confounders (the Napoleonic Wars ended just before the eruption in 1815, so they could explain starvation in Europe).
Population was 1.05 G in 1815, so I guess the death rate was roughly 0.02 % (= 213*10^3/(1.05*10^9)), which is similar to the current death rate from protein-energy malnutrition in the worst off countries (search for “Central African Republic” here).
We have also had 200 years of economic growth since then, so we are now more resilient (agriculture is now a much smaller fraction of the economy), and the death rate for the same eruption would be lower.
These are great points! However, it is unclear to me whether they affect prioritisation across risks. As I say in the post:
Are you suggesting the ratio between total existential risk and nearterm annual extinction risk is higher for supervolcanoes than for other risks? If not, nearterm annual extinction risk could be used as a proxy for importance.
Note McGraw 2024 does not quite argue for 1.5 ºC being the maximum global cooling. The abstract says (emphasis mine):
1.5 ºC is the maximum global cooling given McGraw 2024’s preferred relationship between aerosol size and ejected mass of SO2, but it can be as much as 5 ºC for their most pessimistic curve (see Fig. 10b, which is the last one in my post). I use this level of cooling to estimate the extinction risk.
As I said, 1.5 ºC is the maximum global cooling given McGraw 2024’s preferred relationship between aerosol size and ejected mass of SO2, but it can be as much as 5 ºC for their most pessimistic curve (see Fig. 10b, which is the last one in my post), and I use this level of cooling to estimate the extinction risk.
Maybe this is not too concerning? My mainline estimate is quite uncertain, but I think my conclusion that nearterm annual extinction risk is astronomically low is robust. McGraw 2024:
Reviews the past literature, and its model is able to reproduce and explain differences among past studies.
Covers past results well. This is illustrated in Fig. 3b, which I present below.
Could you clarify if/why the results of this paper go against those of McGraw 2024? The paper linked above focuses on droughts and floods instead of global cooling.
I estimated an annual probability of 3.38*10^-8 for an eruption causing a maximum surface cooling of at least 5 ºC. Following the same approach results in the same annual probability for an eruption causing a maximum surface warming of at least 2.5 ºC (eyeballed from Fig. 10b). Then such astronomically low chance would still have to be multiplied by the annual probability of rapid anthropogenic warming (assuming independence), which I guess is very low too. So I guess the overall likelihood of a double catastrophe involving a supervolcanic eruption on top of rapid anthropogenic warming would be so low for the risk to be negligible, in the sense I think the risk from supervolcanoes and rapid anthropogenic warming could be well approximated by the sum each risk considered alone. In any case, the maximum warming is just 2 ºC (see Fig. 10b), so the increase in the nearterm extinction risk would be very small.
As I mention in the post, “1 M years is the typical lifespan of a mammal species”. However, note I estimated a much lower nearterm annual extinction risk from supervolcanoes of 3.38*10^-14. Do you have specific disagreements about my calculations?
A priori, I think we should expect humanity to be more resilient against famines today than in the past. Poverty is a major risk factor for famines, and there has been major economic growth since the dates you mention.
The more factors are needed to cause extinction, the less likely it will tend to be, because it is then more conjunctive (even if many of the factors are correlated; one should be mindful of the conjunction fallacy). For example, a supereruption and nuclear war is less likely than any of these alone. It is also worth noting that global civilisational collapse is much less likely than local societal collapse (there are many examples of the latter, but none of the former).
I do not think Toby Ord’s guesses for the existential risk given in The Precipice strictly rely on assessing risks in isolation. There is a section named “Risk Factors” in chapter 6.
I agree we know little. Nevertheless, McGraw 2024 considers 4 different relationships between aerosol size and the ejected mass of SO2, and I estimate extinction risk based on the most pessimistic one. Do you think the range of predictions from McGraw 2024’s model (illustrated in Fig. 3b) is overly narrow?
Interesting point! Smaller eruptions have smaller aerosol size (see Fig. 10a, which is the penultimate in my post), and therefore cause more cooling per ejected mass of SO2. A priori, I assume multiple consecutive eruptions from the same volcano over a few years can be interpreted as eruptions from different volcanoes. So I guess that, for the same ejected mass of SO2, the greater the number of eruptions, the greater the cooling. This would contribute towards McGraw 2024 underestimating the cooling caused by supereruptions (although the effect would be small if there is usually just one large eruption). Nonetheless, the greater the number of eruptions, the more time there would be to adapt, so the net effect of your point on extinction risk is not clear to me.
I appreciate volcanic eruptions can be a risk factor for extinction. At the same time, I would be mindful of the conjunction fallacy. The more factors are needed to cause extinction, the less likely it will tend to be, because it is then more conjunctive.
Once again, thanks for the comprehensive feedback!
Thanks for the reply- sorry not a lot of time to address all of these points.
I think in a simplistic world- one in which we react to such threats as you expect with resilience and preparedness, then yes the risk of Tambora sized eruption could be minimal* (*by this I mean raising food and energy prices and dramatically impacting/causing excess mortality of poorer/vulnerable nations).
Yes the world is more resilient and famines are better etc (note that a VEI 5 eruption in 1982 led to the droughts and thus famine in Ethiopia in 1983-5 that led to half a million dying), but a more complex world has inherent weaknesses and new vulnerabilities, see here and here
But as mentioned, the world is unpredictable and multiple factors normally converge at the same time- e.g. the consequences of an eruption now, versus the consequences of an eruption where China has become politically isolated- could dramatically change the impacts, our response and cascading risk (e..g wars from resource shortages/export bans).
You use a philsophical term to make that point that factors acting together is a conjuction fallacy, but I say this is exactly what has happened throughout history for smaller eruptions which have dramatically affected civilisations for centuries.
Yes you can guess this is right-and that’s fine to have this opinion, but when speaking to agircultural professors and historical perspectives (see the Little Antique Ice age in 6th century), in fact the opposite is true, a short spike is something that can be absorbed if global trade is working, sustained periods of low growth is more impactful.
Perhaps this is my scientific background, but how you put a number on such uncertain factors seems deeply unscientific/unrigourous to me.
Thanks for following up, Michael!
I appreciate an increasingly complex world can have inherent weaknesses and new vulnerabilities (e.g. Bailey 2017 analyses the effects of interruptions at chokepoints in global food trade). However, we have to consider these effects together with ones pushing towards greater resilience to catastrophes, such as the benefits of having a world which can dedicate a greater fraction of its resources to food (including greater capacity for international assistance of low income countries). I assume one way of assessing the net effect is looking into how the death rate from protein-energy malnutrition has evolved over time. According to data from the Global Burden of Disease (GBD) study, it decreased 77.8 % (= 1 − 2.74/12.27) from 1990 to 2019 (see below). So I would say the world has become more resilient against small food shocks.
Large shocks happen less often, so the trend above is less informative, but a priori I would expect resilience to small shocks to be correlated with resilience to large shocks.
I do not think this is a fair characterization of the risk linked to VEI 5 eruptions:
The 1983–1985 famine in Ethiopia was caused by “drought, the Ethiopian Civil War and military policies taken by the Ethiopian government”. Even if the drought was totally caused by a volcanic eruption of VEI 5 (wikipedia’s page does not mention any volcanic eruption; could you link to an article?), such eruption would not be responsible for all of the famine deaths.
Some of the famine deaths would have to be attributed to the other causes, such that the deaths attributed to each cause added up to the total death toll.
VEI 5 eruptions happen once every decade or so. As a result, a naive interpretation of your statement (I am not saying you would endorse it!) would say VEI 5 eruptions cause in expectation a famine with 0.5 M deaths every decade or so. However, the expected death toll would be much lower if the other causes (Ethiopian Civil War and military policies taken by the Ethiopian government) have a much lower probability.
This is why I mentioned in my last comment that:
Similarly, the more factors are needed to cause large famines (e.g. not only a VEI 5 eruption, but also civil war, and lack of effective international assistance), the less likely they will tend to be. This effect exists unless there is a perfect correlation between all the contributing causes (e.g. if a VEI 5 eruption always caused a civil war, and lack of effective international assistance).
I agree my analysis is far from rigorous. However:
It is hopefully better to have a non-rigorous analysis plus your thorough comment than the alternative of no analysis.
I quantify my estimates for the sake of reasoning transparency.
To illustrate, I could say the nearterm annual risk of human extinction from supervolcanoes is low, but then some readers may infer from this anything ranging from 0 to maybe 0.01 %. The latter would arguably imply volcanic risk receiving way more attention from the effective alstruism community than it currently does, whereas my actual best guess of 3.38*10^-14 does not.
So I think I had better provide my best guess for the risk. At the very least, this facilitates agreement/disagreement. If I had just mentioned the extinction risk was extremely low, you may not have provided such detailed and valuable feedback due to perceived agreement.
I also think researchers quite often implicitly put numbers on metrics. For example, you said:
It seems to me that, by answering yes to your question, you are implicitly arguing that there are interventions to decrease the expected mortality of volcanic eruptions whose cost to save a life is sufficiently low for such interventions to receive more attention (e.g. more research like yours on volcanic risk). Moreover, if by “we” you mean not just the world as a whole, but funders aligned with effective alstruism too (namely, Open Philanthropy), you may be implicitly claiming that such interventions can have a cost to save a life lower than 5 k$ (that of GiveWell’s top charities). I think it would be great if someone (maybe you and Lara!) published in a peer-reviewed journal a standard cost-effectiveness analysis of the interventions you think are being overlooked.