Featured Image: Machine Learning has proven itself to be an effective tool in interdisciplinary research, but how can it be useful in understanding climate change? CC BY-NC 4.0, via. Dean Long
Paper: Tackling Climate Change with Machine Learning (Chapter 8)
Authors: David Rolnick et al.
Machine Learning (ML) gives researchers extremely valuable ways of revealing patterns within enormous datasets, and making predictions. Climate change research is one of many fields that is beginning to explore ML approaches. There are three major areas of interest: (1) climate prediction/modeling, (2) assessing impacts, and (3) exploring solutions as we attempt to decarbonize energy production. Rolnick and his coworkers explored the merit of machine learning in climate research and where it can support scientists best. The authors also call for greater collaboration between researchers of different backgrounds to advance our understanding of such a complex issue.
Chapter 8 of their report, written by Kelly Kochanski, focuses on how ML can improve climate predictions. Cheaper and better technologies have given rise to an immense amount of recorded climate data ranging from temperature, wind and rainfall patterns, or energy being reflected off the Earth’s surface back into space. This data often provides vital information on the present and recent past of climate systems and how they interact with each other, highlighting geographical variations. However, forecasting future trends remains a far greater challenge. Kochanski identifies two scenarios where ML techniques are most effective at addressing certain challenges. Firstly, when large datasets cannot be analysed using traditional statistics and, secondly, when traditional statistical approaches work, but are too expensive to compute.
ML can help us to better understand many climate-related processes. This includes cloud cover, sea-level change, forest fires, and extreme weather events. Current climate models use complex equations to determine the influence of these variables on regional and global climate. Nevertheless, Kochanski points out in her work that there is still room for improvement. Neural networks offer one way to improve current models, but they have to be ‘trained’. This means, they need to be fed with existing problems and their correct solutions. In this way, neural networks might be able to find patterns which scientists cannot see. Cloud coverage is notoriously challenging to model compared with other variables. Now, Neural networks may help us to improve the accuracy and computation time of climate models including the effect of cloud coverage.
Another useful aspect of ML lies in its ability to combine multiple models into a more powerful ensemble. Random Forest modeling is identified as an ideal technique for this purpose and can, again, improve the accuracy of current models. This technique has shown promise when applied to rainfall and atmospheric energy balance data. An added benefit of ’Random Forests’ is that it allows scientists to work with lower-resolution data and is therefore more cost effective.
This paper provides a valuable reference guide and summary for modern, state-of-the-art approaches to understand climate change. While the report mentions that ML is no “silver bullet”, it is nonetheless a powerful tool for researchers to add to their toolkit and improve climate predictions. It will without doubt help scientists to overcome many challenges associated with climate data.
How Machine Learning Helps in the Fight Against Climate Change by Jordan Healey is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.