The Research Nest
Published in

The Research Nest

How To Tackle Climate Change With Machine Learning?

Building upon the major solutions based on a 97-page research paper

Photo by Bob Blob on Unsplash

Today, the effects of climate change are felt across the world, and the situation is alarming enough for us to not take it lightly. Global temperatures are rising, ice caps are melting and soon many places around the globe are set to become inhabitable.

The challenge of sustainable development presents a humongous scope for innovation. Machine learning techniques have previously been used to solve several diverse problems in multidisciplinary fields. Here, we explore the insights from the recently published collaborative research done by prominent scientists and engineers from 16 different organizations and universities, on how machine learning can be used to tackle climate change.

(Link to the full paper- https://arxiv.org/abs/1906.05433v1)

This article presents a concise summary and insights from the major approaches mentioned in it and my top picks among the plausible solutions expanding upon the important opportunities that can be leveraged by entrepreneurs and climate change activists. This is a global call to the machine learning community to act before it is too late. With that being said, lets quickly get started!

Broad Classification Of Solutions

Reducing emissions (the greenhouse gases (GHG))

The first section of solutions focus on something very direct, how can we reduce our GHG emissions.

The Energy Sector

  • By Using Low Carbon Sources To Generate Electricity Like Wind, Solar, Geothermal, etc.

Demand forecasting of electricity and related predictions of sources like sunlight, wind, etc can be done to optimize production accordingly. This can be achieved by using past power consumption datasets and weather information. You may refer to this article below for a detailed understanding.

  • By Developing New Materials For Better Energy Storage Or Harnessing

The discovery and testing of new materials can be a very slow and exhaustive process. Machine learning techniques can be in the automation of various subprocesses and to predict crystal structures, physical properties, in synthetic modeling of new materials, etc.

Beyond electrical systems, ML can also be used to fast track the creation of better materials, for example, those that can absorb CO2 more efficiently.

You may refer to the following paper, Materials discovery and design using machine learning for better understanding of this approach.

  • By Using Nuclear Energy More Effectively

ML algorithms can be used in cracks and fault detection using video or image data and other anomalies involved in operation by using various sensor data. This will make the plants more secure.

While completely shifting to renewable energies is a long term objective, one needs to consider reducing fossil fuel emissions in the current technologies until a significant transition in energy consumption is made.

Some solutions in that direction include early detection of leaks in pipelines and predictive maintenance of various systems, meaning, a prediction model can be used to give early warning before the systems go out of order, which may lead to greater emissions. Such a system can be put in various places from automobiles to power grids. You can refer to this presentation by Mathworks for more insights.

The Transportation Sector

Up to 25% of all CO2 emissions are estimated to come from the transport sector. So, how can we reduce the emissions here? Let us find out!

  • By Directly Reducing Transportation

There is a lot of scope for reducing the use of transport vehicles by optimizing the flow of vehicles. One can build solutions that can help in more efficient navigation, enable shared transport, etc, decreasing the overall usage and thereby reducing emissions.

  • By Building Autonomous Vehicles That Are Programmed To Be Energy Efficient

Self-driving cars are heavily powered by various deep learning methods, and they are already here. Building products or add ons along these lines will present an interesting solution. Another approach is by building technologies for electric vehicles with the same use case.

The Cities And Industries

  • Smart Buildings

One can think of a broad array of solutions to make a building ‘smart’ and energy efficient, thereby indirectly reducing emissions. Dynamic power management systems can be of some help. For example, the internal appliances can be automated to switch on or off based on the occupancy in the room. ML can also be used in the maintenance of the cooling and heating systems of the buildings. A good combination of IoT with ML can provide robust solutions.

  • Mitigating Industrial Emissions

Industries produce a lot of data, of their supply chain, their machines, production and many more, and where there is data, there can be some machine learning based solution. Unfortunately, most of this data may not be available in a structured format or in public domain. So, one may have to be a little careful in planning in this aspect. One of the key areas where ML can be used includes the optimization of supply chains. You may refer to this Forbes article for more insights.

  • Farms And Forests

Forest fires (man-made as well as natural) can be lethal to the environment as they release huge amounts of greenhouse gases into the atmosphere. Another intriguing thing to note is that the melting of permafrost can also release the trapped gases into the air. Miscellaneous activities like cattle farming release methane.

This sector provides a possibility of some interesting solutions-

  1. Risk prediction for large forest fires using relevant machine learning algorithms along with the weather and historic data.
  2. Early detection of fire deep in forests by applying ML to satellite imagery.
  3. Automated afforestation using drones, robots and other similar machines powered by machine learning for navigation and decision making. Such an ideal machine will help mass afforestation in a systematic manner within a short period of time.

Adapting To The Inevitable Climate Change

Earth’s climate has always been dynamic, changing drastically throughout its history, most of the mechanics behind which we don’t fully understand. It is high time, we start focusing on solutions to prepare for any unforeseen circumstances or doomsday scenarios that may happen far ahead in the future.

In fact, the actions we take now, this very moment, may decide if we still exist on this planet, in a thousand years from now.

ML can help us even in these aspects!

  • Predicting Climate

Predicting climate is equivalent to predicting the future and it by no means is an easy task. Today, we have very accurate models to do the same, but still not accurate enough. With loads of the Earth observation data from satellites and with help of extensive simulations by using supercomputers, creating even more efficient climate prediction models is not beyond reach. Remember that weather prediction is different from climate prediction.

  • Forecasting Extreme Events

The patterns of nature are never consistent throughout. There will always be some events happening all of a sudden or beyond our estimation. Special models can be trained just for this purpose, to forecast abnormalities in weather, or at the very least, estimate the possibility of abnormal weather conditions. We have all the data we need, it is just a matter of finding the hidden patterns inside it.

When it comes to climate, it becomes a lot more challenge due to the rareness of these events. We can predict that ‘something’ might happen in a large time scale, but we cannot be very specific as to when and if it will certainly happen or not. The reversing of earth’s magnetic poles every few hundreds of thousands of years in a good example. We do know that the poles reverse, but we can never be sure when it will happen again and how exactly it may affect the climate. More research needs to be carried out for better understanding.

Multiple startups and researchers have already been working on making models that can perform local forecasts. You can go through this detailed review for more information.

  • Ecosystem And Biodiversity Monitoring

Climate change has an influence on all life on earth. Monitoring the biodiversity and ecosystems across the planet is important to understand how it is affecting them and in turn how it will affect us. Such a large scale monitoring can be powered by ML. Satellite observations can be used to measure the forest covers, oceans, and other larger patches while real-time image recognition and sensor systems can be used to keep a track of biodiversity across the globe.

There are several parameters to monitor and complex situations to handle, but when broken down in multiple subproblems, this is something can be sustainably achieved, if not tomorrow, sometime in the far future.

You can refer to the article below for some additional information.

There are several other developments happening on this front, which can be explored in detail as per your interest.

As you can see, there are so many open problem statements to tackle and solutions to innovate upon.

So, How Can You, As A Machine Learning Enthusiast And A Climate Change Activist Do Something Tangible In This Aspect?

A roadmap illustrated from the paper.

Here are some tips, I have for you (apart from the road map above):

  • Pick up a very specific problem statement from the above-mentioned ones and do an in-depth literature review to understand its scope and possible solutions.
  • Spend considerable time in finding relevant datasets you need for your research. Procuring/Preparing clean datasets will help the research community as a whole.
  • Start work on a small scale. Even a simple prototype that completes a small task in a larger solution is a meaningful contribution. That might as well be the X-factor in saving humanity, you never know!
  • Find your investors properly. There are many governments, NGOs, and independent visionaries who are willing to fund research and developments in this direction.
  • Document your work well, so that other contributors can find and expand upon your work. Climate change cannot be tackled by a single person, nor in a single generation.
  • Focus on converting your projects into practical products. Wouldn’t it be amazing if you can make a livelihood out of saving the planet with your machine learning skills? It’s a win-win situation.

End Notes:

The original research paper on which this article was based on was co-authored by 22 researchers from 16 different universities and organizations across the globe including Harvard University, Mercator Research Institute on Global Commons and Climate Change, Massachusetts Institute of Technology, Stanford University, DeepMind, Google AI, Microsoft Research, and more!

It was a global wake up call to take action and us here at The Research Nest are up for the challenge. If you are interested in any collab or would like to showcase your work to us for consulting, feel free to reach out to us at the.research.nest@gmail.com.

Do clap if you found this insightful and useful. Also, don’t forget to share this with your fellow ML enthusiasts and innovators. Let’s save the planet together!

--

--

--

Empowering humanity with exclusive insights.

Recommended from Medium

Digital Image Processing

RL — Reinforcement Learning

Review — Focal Transformer: Focal Self-attention for Local-Global Interactions in Vision…

Review: STN — Spatial Transformer Network (Image Classification)

SinglePose and MultiPose Estimation papers Collection and Summary

AI Applications With Milvus

Enhancement of a low quality whole body bone scan image

[Paper] P3D: Pseudo-3D Residual Networks (Video Classification & Action Recognition)

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
XQ

XQ

Tech 👨‍💻 | Astronomy 🌠 | Life 🌱 | Poetry 🖊️ | Careers 👔

More from Medium

Denoiser: A noise detection and removal module

Advancements in Neural Machine Translation

Semi-supervised learning with autoencoders

Reproducing “Rainfall–runoff modelling using Long Short-Term Memory (LSTM) networks”