Machine learning and complex information: How much of humanity has been affected by climate change?

A few years ago, I was working as a curriculum writer at a non-profit science education research company called TERC. The company has been around for a long time, but its central purpose, as I understand it, is to study how people teach and learn science of all sorts, with the goal of improving the process for both teachers and students. This means that while my job was to write lesson plans, readings, and so on, it was always as part of a larger research project. The difficulty with this sort of research is that if you’re trying to actually assess how well students understand the subject before and then after an attempt to teach it to them, you can’t just rely on an easily generated dataset like a multiple choice test. The best way to gauge a person’s understanding of a subject is to have them explain it, in their own words, to someone whose understanding is already good enough to assess the answer.

So how can you conduct data analysis on data that’s not in a simple numerical form?

You find a way to convert it.

For example, let’s take a basic question: What does the term “ecological mismatch” mean? Define it, and give an example.

For those who are unfamiliar, “ecological mismatch”, at least in the context of climate change, refers to a situation in which the seasonal patterns of different species that normally line up, cease to do so. For example, there are numerous bird species that breed in North America during the spring and summer, but fly south to South or Central America to avoid the cold winters. Why do they put in all the effort to make such a long trip? Why not just stay in the south? Because the explosion of insect and plant life in the northern spring provides an abundance of food far beyond the day to day in their more tropical “winter homes”.

The problem is that as the climate has warmed, spring has begun to come earlier to North America, and for those birds wintering around the equator, their evolved migratory instinct relies on Earth’s orbit around the sun, which is almost entirely unaffected by global warming. That means that their migration signal has stayed the same, but spring is coming earlier, so they arrive later in the season. The food supply that made this a successful behavior isn’t always there by the time they arrive. The timing is mismatched. This means the insects they’d normally eat have a population boom, as do the birds that don’t migrate as far. That in turn can affect plant populations, other insect populations, and so on.

So. We ask a class full of people to answer this question, and what we get is a mix of responses. Some are blank or completely wrong. Some get the definition mostly right, but the example wrong. Some get the example right and the definition wrong. Some get both right. Our goal, as researchers, is to convert this qualitative data into quantitative data, so that we can run it through equations, make graphs, and so on. One could simply go with “right” or “wrong”, but that’s going to give us an inaccurate picture. The students who are partly right do have some understanding of the subject. We could split it into three options – right, partially right, and wrong. That’s also not quite right, because it doesn’t tell us what they’re partly right about; so we split it into four – right, partly right (about the definition), partly right(about the example), and wrong. Now, with options 1, 2, 3, and 4, and a clear definition of each, we can go through everyone’s answer to the question, give it a number, and actually analyze the overall pattern of understanding.

And now we’re ready to teach the lesson.

Then, you give the same test after the lesson, break it down the same way, and compare the two to see how the overall level of understanding changed. Ideally, each student will be assigned a number so you can compare them to themselves, as well as looking at the group as a whole (the data should be anonymized as much as possible, both to protect people’s privacy in published data, and to prevent conscious or subconscious bias). There was also a long process of systematizing the instructions for evaluating these quizzes so that multiple qualified people would fairly reliably get the same results going through the same process. Remember: with research part of the goal is to ensure that strangers can reproduce what you did from your publication.

We generally didn’t do a quiz like this for a single lesson. The ideal was a full unit of about a week (longer if possible) to test what did or didn’t work, and the “after” quiz would be on the last day of the unit. Each quiz would have a mix of short- and long-answer questions, and once the framework for “coding” the tests was established, someone on the team would have to go through and code every test from every student, with checks on tough calls (you’d be amazed at how many ways there are to be almost right or partly wrong about a question like this), enter the numbers into a spreadsheet, and then we could start actually analyzing the data.

This is to answer a few relatively straightforward questions about how well students understand the subject matter we presented to them.

Now let’s get to the actual point of this article, and look at a different question – How much of the human population has been directly affected by climate change?

As before, we need to break the question down, so we know what answers we’re actually looking for. Obviously we need to define what it means to be affected by climate change. Going broadly, let’s say “forced to change behavior in some way (movement, spending, place of residence, etc.) by weather that would not have caused that change absent warming caused by humans”. That means we need to determine which weather phenomena count as “normal”, and which ones can be attributed to the rise in temperature. In many cases, that’s a matter not just of determining whether climate change influenced a given event, but how much of that event was due to higher global temperatures. Would the storm surge have breached the levees if sea levels were an inch lower? Would the storm have been as powerful if the planet was a degree cooler? Trying to figure this stuff out is very difficult and time-consuming, for those with the task of actually quantifying it.

And again, that’s just for one event, like a hurricane. We’re trying to see what patterns there are on a global scale, which means our best bet is to look at the answers that have already been given – the numerous publications on individual weather events, and how they affected people, and while all of these studies do have quantitative data, they’re often studying different things, using different methods, which means the numbers they get can also mean different things. Ideally, again, you would have a team of people who already know the subject matter very well, to analyze each publication and make sure the overall data analysis actually says what we think it does. If you’re being thorough, that means having a team examining over 100,000 studies of many different weather events, and generating a dataset that will give us results we can trust.

It’s a monumental amount of work – possibly more work than could reasonably be done by a group of experts.

So some folks trained computers to do it for them.

Increasing evidence suggests that climate change impacts are already observed around the world. Global environmental assessments face challenges to appraise the growing literature. Here we use the language model BERT to identify and classify studies on observed climate impacts, producing a comprehensive machine-learning-assisted evidence map. We estimate that 102,160 (64,958–164,274) publications document a broad range of observed impacts. By combining our spatially resolved database with grid-cell-level human-attributable changes in temperature and precipitation, we infer that attributable anthropogenic impacts may be occurring across 80% of the world’s land area, where 85% of the population reside. Our results reveal a substantial ‘attribution gap’ as robust levels of evidence for potentially attributable impacts are twice as prevalent in high-income than in low-income countries. While gaps remain on confidently attributabing climate impacts at the regional and sectoral level, this database illustrates the potential current impact of anthropogenic climate change across the globe.

The debut of the remarkable new word “attributabing” not withstanding, this approach to meta-analysis could end up being hugely useful in helping people in general keep up with the massive collective effort known as “science”. There are millions of scientific papers published every year, from people all over the planet. Our collective knowledge is such that it’s impossible for any person to read the current research on all but the tiniest fraction of it, and yet we need to have at least some general grasp of these issues if we want to have any control over our future as a species. Our capacity to understand what’s happening around us and respond accordingly is one of our greatest strengths as a species, but climate change is happening at a scale that’s a bit outside what we can easily wrap our minds around. That’s probably part of why we’ve gone so long without treating global warming as the crisis it is.

From a Washington Post article on the study:

In the United States, climate disasters have already caused at least 388 deaths and more than $100 billion in damage this year, according to analyses from The Washington Post and the National Oceanic and Atmospheric Administration.

Yet despite a pledge to halve emissions by the end of the decade, congressional Democrats are struggling to pass a pair of bills that would provide hundreds of billions of dollars for renewable energy, electric vehicles and programs that would help communities adapt to a changing climate.

The contrast between the scope of climate disasters and the scale of global ambition is top of mind for hundreds of protesters who have descended on Washington this week to demand an end to fossil fuel use.

“How can you say that we are in this climate emergency and be going around and saying we’re at this red point … and at the same time be giving away land for additional oil and gas infrastructure?” said Joye Braun, a community organizer with the Indigenous Environmental Network and a member of the Cheyenne River Sioux Tribe who rallied in Washington this week.

The activists, many of them from Indigenous communities that have been harmed by global warming, risked arrest as they remained on the sidewalk outside the White House after police ordered them to clear the area.

The new research in Nature adds to a growing body of evidence that climate change is already disrupting human life on a global scale. Scientists are increasingly able to attribute events like heat waves and hurricanes to human actions. In August, the U.N. Intergovernmental Panel on Climate Change devoted an entire chapter to the extreme weather consequences of a warming world.

The study’s conclusion that 85 percent of humanity is experiencing climate impacts may sound high. But it’s “probably an underestimation,” said Friederike Otto, a senior lecturer at the Grantham Institute for Climate Change and the Environment at Imperial College London, who was not involved in the study.

The study looked at average temperature and precipitation changes, rather than the most extreme impacts, for which Otto says there is even more evidence of climate change’s role.

“It is likely that nearly everyone in the world now experiences changes in extreme weather as a result of human greenhouse gas emissions,” she said.

The human toll of these events has become impossible to ignore. This summer, hundreds of people in the Pacific Northwest died after unprecedented heat baked the usually temperate region. More than 1 million people in Madagascar are at risk of starvation as a historic drought morphs into a climate-induced famine. Catastrophic flooding caused New Yorkers to drown in their own homes, while flash flooding has inundated refugee camps in South Sudan.

In a letter released Monday, some 450 organizations representing 45 million health-care workers called attention to the way rising temperatures have increased the risk of many health issues, including breathing problems, mental illness and insect-borne diseases. One of the papers analyzed for the Nature study, for example, found that deaths from heart disease had risen in areas experiencing hotter conditions.

“The climate crisis is the single biggest health threat facing humanity,” the health organizations’ letter said.

As Braun points out in the quote above, the feeble “climate response” plans of most governments become a sick joke when put next to continued expansion of fossil fuel extraction, and the unwillingness of groups like the Democratic Party to do more than just say they care about the issue. It’s hard to tell if it’s malice or delusion, but in either case, it’s a problem, and it feels like they’re hoping people will just continue to underestimate the scale of what’s happening. Things like this new research could help us make a more compelling case for the change we need, by making it harder to wave the problem away. That said, the resources going into understanding what’s happening around the world aren’t much better distributed than wealth has been in this era of colonialism and neoliberalism:

Yet in many of the places that stand to suffer most from climate change, Callaghan and his colleagues found a deficit of research on what temperature and precipitation shifts could mean for people’s daily lives. The researchers identified fewer than 10,000 studies looking at climate change’s effect on Africa, and about half as many focused on South America. By contrast, roughly 30,000 published papers examined climate impacts in North America.

In poorer countries, the researchers say, roughly a quarter of people live in areas where there have been few impact studies, despite strong evidence that they are experiencing changes in temperature and precipitation patterns. In wealthier countries, that figure stands at only 3 percent.

“But it indicates that we’re not studying enough,” Callaghan said, “not that there isn’t anything happening.”

Otto attributes this discrepancy, known as an “attribution gap,” to a lack of capacity and funding for research in poor countries, as well as researchers’ tendency to reflect the priorities of wealthy nations.

In South Sudan, for example, efforts to understand flooding have been stymied by conflict and the difficulty of collecting weather data in the world’s youngest country.

Liz Stephens, an associate professor in climate risks and resilience at the University of Reading, wrote in an email that the Global Flood Awareness System from the Copernicus Emergency Management Service is “notoriously bad” at forecasting flooding in the White Nile and Blue Nile river basins. Without good data, scientists can’t easily say what places are likely to be deluged or warn when a disaster is about to hit. Officials may be caught off guard by weather events. Vulnerable people are less able to get out of harm’s way.

South Sudanese officials say half a million people — about 4 percent of the country’s population — have been displaced by the floods.

But the “attribution gap” makes machine-learning-based analyses like Callaghan’s all the more valuable, Otto said. These programs can help identify climate impacts even in places where there are not enough scientists studying them.

“It seems a very useful way … to understand better what climate change is costing us today in a global way that is more bottom-up,” Otto said.

A September study in Nature found that 60 percent of Earth’s oil and fossil methane gas and 90 percent of coal must remain in the ground for the world to have a chance of limiting warming to 1.5 degrees Celsius (2.7 degrees Fahrenheit) — a threshold that scientists say would spare humanity the most disastrous climate impacts.

Increasingly, groups are calling on President Biden to restrict fossil fuel production outright.

On Wednesday, a coalition of more than 380 groups filed a legal petition demanding that the U.S. Army Corps of Engineers stop issuing permits for new fossil fuel infrastructure projects. Two days later, hundreds of scientists submitted an open letter asking Biden to do the same.

“The reality of our situation is now so dire that only a rapid phase-out of fossil fuel extraction and combustion can fend off the worst consequences of the climate crisis,” they wrote.

In response to Monday’s protests, however, American Petroleum Institute spokeswoman Megan Bloomgren said curbing the country’s energy options would harm the economy and national security. “American energy is produced under some of the highest environmental standards in the world,” she said.

In other words, they have no intention of changing course.

The difficulty in studying climate change in regions currently suffering from things like war or the effects of climate change, is one part of why it’s so important to stop the imperialist policies of the United States in particular, and wealthy countries in general – the pattern for the last century and beyond has been for powerful nations to subject the less powerful ones to debt, invasions, coups, assassinations, death squads, and more, all in the name of securing the “interests” of a tiny ruling class. This is what drives the obscenely high emissions of the U.S. war machine, and the overthrow of regimes – like that of the Brazilian Workers Party, or the Bolivian Movement for Socialism – that are committed to both eliminating poverty and finding a way for us to live that doesn’t destroy the ecosystems on which we rely.

We can’t adapt to what’s happening if we don’t know what’s happening, and if we’re still focused on narrowminded ideas like profit and nationalism, we won’t have the resources to study the problem, let alone prepare for what’s coming.

If we want to avoid an unprecedented tide of death, we’re going to need truly revolutionary change in our political and economic systems. The alternative could well be extinction.

Thank you for reading. If you find my work interesting, useful, or entertaining, please share it with others, and please consider joining the group of lovely people who support me at Life costs money, alas, and due to my immigration status in Ireland this is likely to be my only form of income for the foreseeable future, so if you are able to help out, I’d greatly appreciate it. The beauty of crowdfunding is that even as little as $1 per month ends up helping a great deal if enough people do it. You’d be supporting both my nonfiction and my science fiction writing, and you’d get early access to the fiction.


Leave a Reply

Your email address will not be published. Required fields are marked *