Time Scavengers

Tackling Climate Change with Machine Learning 

David Rolnick, Priya L. Donti, Lynn H. Kaack, Kelly Kochanski, Alexandre Lacoste, Kris Sankaran, Andrew Slavin Ross, Nikola Milojevic-Dupont, Natasha Jaques, Anna Waldman-Brown, Alexandra Luccioni, Tegan Maharaj, Evan D. Sherwin, S. Karthik Mukkavilli, Konrad P. Kording, Carla Gomes, Andrew Y. Ng, Demis Hassabis, John C. Platt, Felix Creutzig, Jennifer Chayes, Yoshua Bengio

Summarized by Samir, a first year masters’ student at Binghamton University State University of New York majoring in geosciences. He has experiences in designing and delivering effective solutions using programming skills and knowledge in geosciences, physics, and mathematics. Also, he is planning to dive into machine learning since he believes it is one of the most effective methods to tackle global issues! By the way, he is a big fan of basketball.

What data were used? One of the many datasets used in the study is a large-scale climate dataset for detecting, localizing, and analyzing extreme weather occurrences in a semi-supervised manner. The discoveries, assumptions, significantly important results, and models present in this study wouldn’t be possible without historical climate dataset, high-resolution satellite images, video, CO₂ emissions, remote sensing data.

Methods: The methods that were already used or might be potentially used in the future include but are not limited to remote sensing of emissions, precision agriculture, monitoring peatlands, managing forests, and carbon dioxide removal. It should be mentioned that there were more domains discussed in the original study, however this particular summary mainly focuses on implementation of machine learning on farms, forests, and carbon dioxide removal techniques to tackle climate change. Figure below provides the summary of methods and areas of implication mentioned above. The most impactful and interesting methods will be discussed in detail.

 

It might sound as a surprise; however, agriculture is responsible for 14% of the greenhouse gas emissions. Modern methods used result in massive, sequestered carbon release, such as in particular: tilling, which basically exposes topsoil to the air which is the reason behind release of locked carbon that was bound to soil. Also, as some agricultural techniques deplete soil nutrients, nitrogen-based fertilizers must be reintroduced. However, while some nitrogen is up taken by plants, the rest is being transform to nitrous oxide – a greenhouse gas 300 times more powerful than CO₂. 

Precision agriculture, which is the combination of tools and machine learning methods, can be used to make it possible for farmers to work on a large scale and not diminishing production as it happens when conventional methods are used. For instance, the hyperspectral camera-equipped robot can undertake mechanical weeding, targeted pesticide application, and insect vacuuming. Monitoring peatlands, one of the main sequestered carbon sources in the world, is essential. It is not only releasing carbon while decomposing but also is susceptible to fire. Therefore, identification and estimation of these carbon “stocks” through machine learning techniques plays an important role in potential fire risk assessment. 

One of the main points made in the study is even if all the emissions stop today, because of the carbon that is already in the atmosphere, the planet will still experience consequences of global warming. Direct CO₂ capture and consequent sequestration is a most reliable and promising solution. The main concept underlying direct air capture (DAC) is to blast air over CO₂ sorbents, and then employ heat-powered chemical processes to purify the CO₂ for sequestration. To optimize sorbent reusability and CO2 absorption while reducing energy consumption, machine learning might be utilized to speed up materials discovery, process engineering operations, such as corrosion-resistant components. 

Consequent step is to sequestrate carbon dioxide. Direct injection into geologic formations such as saline aquifers, which are analogous to oil and gas reservoirs, is the best-understood method of CO₂ sequestration. Machine learning can be utilized to find potential storage locations. Also, machine learning can contribute to the maintenance of active sequestration sites and monitor them in order to detect potential leaks CO₂ leaks.

Results: As for the results of the study, such as precision agriculture discussed in the previous section there was an actual implementation of camera equipped robots that can cover 5 acres each day and collect big datasets for continuous development using solar energy. It works for specific types of crops now, however there is a room for improvement to adapt machine learning algorithms to make them work in any kind of environment. 

In addition to this, to quantify the thickness of peat and measure the carbon store of tropical peatlands, machine learning was applied to characteristics collected from remote sensing data. Maps that are going to predict the risk of fire are expected to be developed in the nearest future using advanced machine learning techniques.

Regarding CO₂ sequestration, for more than two decades, a Norwegian oil firm has effectively sequestered CO₂ from an offshore natural gas field in a saline aquifer. Recently, machine learning approaches, as well as computer vision systems for emissions detection, have been utilized to monitor potential CO₂ leakage from wells and finding most reliable sites for the sequestration.

Why is this study important? This study gives an overview of how machine learning can be used to make a meaningful contribution in the fight against climate change, whether through effective engineering or research. Therefore, it provides valuable information and potential ideas for data scientists, machine learning enthusiast, investors, researchers that can be used to prevent catastrophic consequences.

The big picture: Climate change is a complex issue that requires a multidisciplinary approach to be solved. Greenhouse gases emissions are one of the main reasons behind global changes in temperatures, precipitation, ice glacier masses loss, and frequent fires. Mitigation of those gases requires fundamental changes in the number of sectors that includes transportation, construction, electricity systems, and industries. Unfortunately, the majority of the solutions are computationally expensive to be implemented due to big amounts of data, such as for example some climate models, where conventional statistical methods don’t work. Machine learning methods and techniques can be used to address those issues since they are less computationally expensive and more accurate. 

Citation: Rolnick, D., Bengio, Y., Chayes, J., Creutzig, F., Platt, J., Hassabis, D., Ng, A., Gomes, C., Kording, K., Mukkavilli, K., Sherwin, E., Maharaj, T., Luccioni, A., Brown, A. W., Jaques, N., Dupont, N., Ross, A. S., Sankaran, K., Lacoste, A., … Donti, P. L. (2019, November 5). Tackling Climate Change with Machine Learning. https://arxiv.org/. Retrieved December 15, 2021, from https://arxiv.org/pdf/1906.05433