Tackling Climate Change with Machine Learning
In this guest post, machine learning expert Fredrik Edin gives us all — data scientists and laypeople alike — a glimpse into how machine learning can be used to fight climate change.
By Fredrik Edin
Like many others, I have been asking myself what I can do personally to tackle the climate change question. What better way than doing what you are trained to do and applying that to climate issues? In my case, this is machine learning.
Machine learning (ML) is the use of algorithms that learn on their own from data provided to them, as opposed to us providing explicit rules.
For example, a non-ML algorithm for classifying a pet to be either a cat or a dog might be:
“If it weighs more than 7 kg -> cat, otherwise -> dog”. This is a rule-based classification. (Of course, more complex rules can increase accuracy.)
An example of an ML algorithm for classifying a picture to be either a cat or a dog might be to give the algorithm lots of pictures, along with answers, and tell it whether the picture represents a cat or a dog. The algorithm then tries to learn rules on its own in such a way that it can guess correctly on new images.
I know nothing more than the average Joe about the climate, so I decided to do some background research and came across this academic paper that from June 10, 2019, with the appropriate title, “Tackling Climate Change with Machine Learning”.
It is written by David Rolnick and has a bunch of people on its author list, including some machine learning superstars such as Andrew Ng and Yoshua Bengio, as well as experts from Harvard, MIT, Stanford, Google, Microsoft, and other well-known institutions. It’s a great reference, and I will describe it below.
Overview: a call for collaboration
This paper is not the average run-of-the-mill academic paper written by academics for academics. Rather, it is a call for collaboration among the following groups:
● Researchers and engineers
● Entrepreneurs and investors
● Corporate leaders
● Local and national governments
For this reason, it is not overly technical, but it is comprehensive. According to the authors the paper should serve to identify how machine learning skills can be useful.
For instance, the first table lists an overview of areas in which climate work and ML work can meet.
Table 2 is also interesting:
One thing I really like about the paper is that each area is marked by zero or more of three symbols:
These make it much easier for the uninitiated to make a choice about which problems to focus on.
The rest of the paper goes into a bit more detail about each individual dot in the table. Reading through it, it is possible to summarize a few conditions that need to be satisfied for a climate change mitigation through ML to happen. (I’ll get to those soon.)
My take-home message is that cross-discipline collaboration is imperative to the success of ML-aided emission reduction projects.
Reading the paper can be quite a mouthful. Reading it once is good, but primarily it serves as a good reference. If you’re a specialist in, for example, natural language processing, the tables above could help you focus on a few key areas for further immersion.
However, I suspect that many ML practitioners have broad rather than deep expertise. If so, I suggest you pick a topic or two and read those sections to get a feel for the problems, but don’t beat yourself up if the reading gets too heavy. Instead, go further into the deep dives once you have found a topic where you have the possibility to work.
How to find your climate change project
So how do you find which topic to focus on? The authors lay this out for you:
“For those who want to apply ML to climate change, we provide a roadmap:
• Learn. Identify how your skills may be useful — we hope this paper is a starting point.
• Collaborate. Find collaborators, who may be researchers, entrepreneurs, established companies, or policy makers. Every domain discussed here has experts who understand its opportunities and pitfalls, even if they do not necessarily understand ML.
• Listen. Listen to what your collaborators and other stakeholders say is needed. Groundbreaking technologies have an impact, but so do well-constructed solutions to mundane problems.
Deploy. Ensure that your work is deployed where its impact can be realized.”
I think this advice is extremely valuable. Climate change is a highly technical and complex field, and you really need to find good collaborators and listen to what they have to say. Also, be prepared to use the most appropriate approach for the problem at hand, even if it doesn’t involve ML. Remember that not every problem is a nail just because what you’re wielding is a hammer.
And how do you go about finding collaborators? Associated with the paper is a website, Climate Change AI. If you want to immerse yourself more, there is further study material here, well organized into academic fields.
In addition to that, the website also offers a well-organized forum. Just join it, look at the threads, and start reaching out to potential collaborators! For you readers who are not working with machine learning, joining the forum can help you get started with collaborating with machine learning experts to tackle climate change.
The forum was just launched on Oct. 30, 2019, so it is brand new. I expect it to grow fast, and you could be part of that.
As for myself, I have just reached out to a few people in the forum, taking my first steps in this field. Of course, anyone reading this post is welcome to reach out to me (details below), whether it be concerning specific projects or just general discussion. I would love for my next consulting project to be focused on mitigating climate change.
Conditions for successfully applying ML to climate change problems
Here are what I see as conditions that must be met for ML-based climate change projects to be successful. To me, the paper makes the point that we need interdisciplinary collaboration to maximize the chances of success.
- Technical solutions, including ML-based ones, need to be in place.
- Data need to exist and be available. However, collection of data through various kinds of sensors is sometimes problematic due to
- the associated cost,
- the impact of the sensors themselves on the climate,
- personal integrity that must not be compromised or at least needs to be weighed against the gains of the climate mitigation, and
- networks of sensors possibly being hijacked for other purposes (for example, by foreign powers).
- Financial and other major interests of companies, municipalities, and people in general must be aligned with lowered greenhouse gas emissions (for example, through taxation or other regulations).
- Once the financial incentives for reducing emissions are in place, their presence implies that companies increase profits by switching to low-emitting solutions. However, the increased profits will often lead to increased production of the emitting product, thereby counteracting the switch. Therefore, we see a reduction in overall emissions only if
- emission reductions per produced product are greater than the increased production motivated by the financial gains, or
- the demand curve is such that a reduction in price does not lead to a big enough sales volume boost to increase overall profit (a phenomenon well-known in environmental sciences, called Jevons paradox).
About the author, Fredrik Edin
A PhD in computational neuroscience, Fredrik Edin worked in the financial services and big pharma sectors with all kinds of quantitative analysis, as well as business development and management, before becoming an independent consultant focusing on ML.
Co-founder of Codon Consulting AB, a small ML consultancy in Sweden, and based in Taiwan, Fredrik believes in remote work and sees it as a small piece in the puzzle to reduce emissions. Fredrik has working-level proficiency in Swedish, English, Chinese, German, and Danish.
Contact Fredrik with questions or collaboration interests.
About We Don’t Have Time
We Don’t Have Time is the world’s largest social network for climate action. Together we are the solution to the climate crisis.
But we are running out of time.
Join us: www.wedonthavetime.org
