Datalore
Collaborative data science platform for teams
Can AI Fix Climate Change? Perspective of a Data Nerd
This is a guest blog post by Thu Vu.
Climate change is a frustrating topic. Politicians are not committed to doing anything meaningful about it. And most people like you and me feel powerless and don’t really know what to do to help. Climate change is happening, and it’s likely accelerating.
We seem to be living in a world where every summer is warmer than the last. As a millennial, sometimes I seriously wonder whether it’s fair to bring children into this world if they are doomed to suffer from the future climate apocalypse.
However, an important point is missing in mainstream media narratives.
I’ve collected some of the most important data points around climate change, and in this blog post, I’ll share with you the status quo through the lens of a data nerd. We’ll also debunk some misconceptions about climate change, explain why there is hope, and how AI, among other technologies, can help us tackle climate challenges (or make them worse).
Watch the video below to see my research, or simply read on for the text version of it.
Disclaimer: This post was written by an external contributor, and published on their behalf.
Make sure to also check this data report in Datalore, where I’ve collected the relevant data and analyzed it.
The unfolding climate story
A warming world
Last year, 2023, was the warmest year on record since 1850, making it an unusual year. In fact, this is the first time that the annual average temperature has exceeded the pre-industrial baseline period by more than 1.5 °C, as some datasets suggest. If you’re unfamiliar with this jargon, the pre-industrial period is the reference period from 1880 to 1900.
You might argue that it could just be an anomaly, or that El Niño was responsible for this uncommonly hot year. But from what we can see on the following graph, it looks like a very solid upward trend in the average temperature across the last four decades.
Under the Paris Agreement, many countries have set an aspirational goal of limiting long-term global warming to no more than 1.5 °C (2.7 °F). That target is based on the state of the climate averaged over many years, so a single year exceeding 1.5 °C isn’t automatically considered as breaching this target. However, this is a stark warning sign of how close the overall climate system has come to exceeding this Paris Agreement goal.
As humans continue to pump more and more carbon dioxide into the atmosphere, it’s likely that climate warming will regularly exceed 1.5 °C over the next decade.
As you can see on the graph below, the CO2 level in the atmosphere never rose above the dashed line at any point during the past million years. And this red spike protruding above the line represents the past 70 years. I know that I, for one, certainly wouldn’t want to sit on it.
The rise in CO2 emissions over the past century can be mostly attributed to fossil fuels use and industry, or in other words, human activities.
According to a report of the Intergovernmental Panel on Climate Change, the nationally determined contributions (NDCs) committed by 2030 show the temperature will increase by 1.5 °C in the first half of the 2030s, and if that’s the case then we may even struggle to keep the temperature increase below 2.0 °C by the end of the twenty-first century.
But why do scientists have to make such a fuss about a 1.5–2 °C change in the global temperature? We just need to turn on the air conditioner, right? Well, it’s a little bit more serious than that.
With increased temperature, hot places will get hotter, rainy places rainier, and the risks and strength of extreme weather events will increase significantly.
At 1.5 °C warming, about 14% of the Earth’s population will be exposed to severe heat waves at least once every five years, while at 2 °C warming, that number jumps to 37%. Extreme heat waves will also become widespread. At 2 °C warming, deadly heat waves may occur annually in some countries.
Warming beyond 2 °C will make all of these extremes even more extreme (source), including frequent hurricanes, droughts, and wildfires. More ecosystems will be put under major pressure. Some simply won’t survive.
For humans, that also means we may not be able to produce enough food to eat. A large number of people will migrate, and this could destabilize the nation-state.
The melting ice and rising seas
Putting extreme weather to one side, there’s an equally big problem: melting ice at the poles. This can be monitored using satellite data and ground observations.
Since the beginning of the twenty-first century, the Antarctic and Greenland ice sheets have decreased in mass.
The Antarctic and Greenland ice sheets account for more than 99% of the Earth’s freshwater ice. If they both melted completely, they would raise sea level by an estimated 67.4 meters (223 feet).
So, are we doomed? Is the climate apocalypse inevitable? What about artificial intelligence? Can it save us from ourselves? Now, let’s talk about whether AI can help us fight climate change.
How AI is helping to tackle climate change
2023 and 2024 have brought huge leaps in the development of artificial intelligence, with GPT4, Gemini, and many open-source language models becoming available to the public.
Although AI cannot draft new climate policies and enforce them (yet), or rather, politicians won’t let it, you might be asking, what can AI actually do to help us tackle many climate challenges?
AI, at the most basic level, can help us better understand what’s going on and stop denying the problems. The main uses of AI I’ve seen in my research fall into these categories: monitoring, predicting, and optimizing.
Toward real-time monitoring of greenhouse gasses (GHGs)
Climate Trace is an initiative that uses AI and machine learning to calculate GHG emissions on a global scale, with the goal of moving toward real-time precision. It offers a powerful, free, and independent overview of greenhouse emissions by location and source. Looking at the south of Vietnam, my home country, I’m not surprised to see the huge number of emissions from oil and gas fields. I can also see a concentration of emissions around Ho Chi Minh City Airport. You can also download this data from the Climate Trace website for your own research. It’s totally mind-blowing!
AI maps icebergs 10,000 times faster than humans
In addition, researchers (source) have been using neural network models to quickly and accurately map and monitor the extent of Antarctic icebergs in satellite images. This is important when it comes to quantifying how much meltwater they release into the ocean.
This task is often challenging because from a satellite image, icebergs, sea ice, and clouds all appear white, making it hard to pick out actual icebergs. The neural network model handles this task much more accurately and efficiently.
Mapping deforestation with AI
Similarly, machine learning models have also helped map forest coverage and the impact of deforestation.
Using AI to recycle more waste
AI is also making waste management more efficient. Waste is a big producer of methane and is responsible for a significant number of CO2 emissions.
Using a machine learning system for object detection, a startup tracked 32 billion waste items across 67 waste categories in 2022. The company identified 86 tons of material on average that could be recovered but is being sent to landfill. Large supermarkets around the world are also using AI to predict demands and thereby reduce waste.
AI is cleaning up the ocean
In the Netherlands, an environmental organization called The Ocean Cleanup is using AI and other technologies to help clear plastic pollution from the ocean.
The neural network algorithm that detects objects is helping the organization create detailed maps of marine debris in remote locations. This marine waste can then be gathered and removed, which is more efficient than previous cleanup methods using trawlers and airplanes.
AI helps predict climate disasters
AI models have also helped us model and more accurately predict temperature and climate disasters. The AI model called GraphCast not only delivers 10-day weather predictions but also offers earlier warnings of extreme weather events. It can predict the tracks of cyclones, identify atmospheric rivers associated with flood risk, and predict the onset of extreme temperatures. This means it has the potential to save lives.
So far, I feel like AI has mostly been used as a machine learning tool, and I haven’t seen AI being used to create new solutions on its own. In the future, if we have artificial general intelligence, this kind of AI could potentially do so much more.
How AI is making things worse
However, some skeptics (source 1, source 2) think we shouldn’t be too romantic about AI saving the planet. They believe claims that artificial intelligence will help solve the climate crisis to be misguided.
Models like GPT4 and Gemini require a tremendous amount of energy to train and to run.
It is also challenging to calculate the carbon footprint of an AI model precisely because companies like OpenAI and Google don’t typically publish detailed specifications for their models.
In the most simple form, the carbon footprint of an AI model is equal to the energy needed to train the model plus the number of queries and how much energy each query would require. All of this will be multiplied by the energy efficiency of the hardware.
Footprint = (electrical energy train + queries × electrical energy inference) × CO2edatacenter/KWh,
where CO2edatacenter means CO2 efficiency of a data center.
Most companies spend much more energy serving an AI model (performing inference) than training it. In fact, it’s estimated that 90% of the energy is spent on serving. The electricity demands of AI mean a doubling of data centers is needed to help keep pace with the industry.
In the US, AI requires so much power that old coal plants are still needed to satisfy its energy demand. That’s quite terrifying.
Companies like OpenAI and Google don’t typically publish the environmental impact figures of their models. There is only speculation and estimations so far. I tend to think reporting the environmental impact should be regulated for these large AI models.
Why the past decades are different
To see if there are any upward trends in fossil fuels in recent years, we’ll take a closer look at energy consumption over the past 10 years.
Coal consumption has actually plummeted in rich countries like the US and the UK. It has also leveled off in upper-middle-income countries like China. These countries are also the largest consumers of coal. The same trend can be seen for oil and gas.
Given that fossil fuels – coal, oil, and gas – account for over 75% of global greenhouse gas emissions, this is a promising trend.
Also, technologies are making it cheaper and more attractive to produce renewable energy, such as wind and solar power. Electric cars have become the norm in many European countries as more people prefer environment-friendly solutions.
This is not to say that all is well, and the development of AI models hasn’t caused any visible harm to the environment. This can likely only be observed more clearly in the next few years.
Conclusion
AI could be a double-edged sword if we’re not careful. Right now, there isn’t enough concrete evidence to say the benefits outweigh the costs.
On the other hand, data has made it crystal clear how climate change is heading if we don’t do anything. But we’ve also seen no shortage of solutions and human ingenuity.
Many individuals, startups, and organizations are working to solve the different aspects of climate risks.
This makes me feel so positive because, if we want to make a change in the world, we first need to believe that change is possible and no problem is too big to solve.
As more younger people move into influential positions, they are increasingly prioritizing climate change and working on new solutions. Hopefully, in a few years, AI will help us take significant steps toward a better climate future, and the benefits will actually outweigh the costs.
If you are interested in finding out more, check out the data report in the Datalore notebook below.
Data sources
1. Average Global mean temperature (ºC) for 1850–1900
2. CO2 level in the atmosphere during the past 800,000 years
- All studies: https://www.ncei.noaa.gov/access/paleo-search/study/17975
- Revised Core CO2 data 800,000 – 55 before present (2001): Antarctic Ice Core Revised Composite and Individual Core CO2 Data
- For the 2002–2023 period, we need to use the modern instrumental data that’s available here: https://climate.nasa.gov/vital-signs/carbon-dioxide/?intent=121#:~:text=The
3. Global greenhouse gas emissions
4. Coal, oil, and gas consumption by country
- https://ourworldindata.org/fossil-fuels
- For coal consumption: https://ourworldindata.org/grapher/coal-consumption-per-capita?tab=chart&country=High-income+countries~Upper-middle-income+countries~Lower-middle-income+countries
5. Melting ice sheet: Antarctic and Greenland ice sheet mass balance 1992–2020 for IPCC AR6
- https://ramadda.data.bas.ac.uk/repository/entry/show?entryid=77b64c55-7166-4a06-9def-2e400398e452
- https://www.epa.gov/climate-indicators/climate-change-indicators-ice-sheets
6. Cost of training LLMs
- https://tinyml.substack.com/p/the-carbon-impact-of-large-language
- https://www.statista.com/statistics/1384418/co2-emissions-when-training-llm-models/
Subscribe to Datalore News and Updates
