Can AI be Used To Fight Climate Change

We invited three industry expert speakers using AI to battle climate change. During the hour long webinar, Anita Faul, Data Scientist at the British Antarctic Survey, Lauren Kuntz, CEO and Co-Founder of Gaiascope and Topher White, CEO and Founder of Rainforest Connections walked us through their business use applications of AI to fight the change in climate.

Below we have both a overview blog of the webinar & the full video presentation on youtube:

Anita started her talk with an explanation of the Thwaites Glacier, otherwise know as the ‘Doomsday Glacier’. This glacier is responsible for 4% of all sea level increase - if it were to melt completely, sea levels would rise by half a meter in total (hence the name). Therefore, Anita's objective at the Antarctic Survey is to identify icebergs efficiently and reliably in Synthetics Aperture Radar (SAR) satellite images to estimate ice loss. It was explained that satellites provide 2 images, horizontal polarization & vertical polarization. They are then looked at in regard to the pixel intensity values provided by the images. This then allows BAS to do a principal component analysis. Through this, the algorithm is then able to gather data and process clustering to show the values of 4 measured clusters - open water, sea ice, icebergs & 'ambiguous’. The algorithm is also able to decide the probability of masses belonging to the iceberg. This is not always completely reliable, there are instances when the clustering pixel values fail as the shading of the iceberg & surrounding sea ice differs from other iceberg data the algorithm relies on. This can usually be corrected with convoluting images, also known as CNNs.

You can watch from the start of Anita's presentation here.

You can also read more on the British Antarctic Survey & their use of AI here.

Lauren's talk began with the question, What is the energy transformation we need to see to have an impact on climate change? The use of Natural Gas, Crude Oil & Coal has increased dramatically, and we expect to see this continue to grow with the worlds ever growing energy demands. Lauren suggested that the realistic alternatives to these include carbon-free electric, alternative liquid fuels & a reduction in consumption, through which we would see decarbonization on the 'grid' or energy system. What does the grid look like today? The grid is based on top down control - Generators are dispatched to match the levels of demand - Inertia maintains stability - A lot of models are used to produce analysis, which can be quite cumbersome. What could a future grid look like? Relying solely on renewables creates volatility of generation & demand, as the amount of energy available is dictated by weather, through this, there is a reduction in predictability. This, however, creates both opportunities and questions for AI:

AI has the ability to solve 'the grid problem' as referenced by Lauren. Integration of intermittent generation with dispatchable resources through an increase in distributed generators. You do, however, need to think about transmission constraints - you could have 100's of thousands of ‘power plants’/sources of power which are adding into the grid, which will breed communication issues, storage control charge & discharge, as well as issues in demand response in appliance & more - Due to this, there is a need for a more controlled structure, which AI can provide.
AI can help with Grid design - We know moving forward there will be an increase in electrification, such as electric cars, so there needs to be an increase in the electric system as a whole. This then brings the issue of where you intelligently ‘sit’ your new assets. You guessed it, AI can help.
Reliability of the grid and maintaining safety is very important. It’s an issue AI needs to be able to do well in order for an algorithm to be effective - How do you ensure the physical feasibility of AI models? i.e. How do you make sure its never going try to dispatch an offline generator? High cost for an incorrect algorithm output - For example, if an algorithm on Instagram gets something wrong & shows an advert which is not of interest, it doesn’t have an impact on my life, but if AI got something wrong here, it could cost lives.
Due to the grid also being non-stationary, the underline network is constantly changing, with new generators being added etc, so it’s hard for AI to rely on any historical data, even in the short term the gird looks very different today than it did 2 years ago.

The solution then, is to combine AI & Physics based models, which is at the core of what Gaiascope does. How can this be achieved?

Create an AI Emulator: Train an AI model to reproduce whatever the physical model did.
Embedding physical simulation: You could use the physical laws embedded in a Neural Network, which guarantees that your solution obeys the laws of physics, however this method does not necessarily speed up processes, which you would want from implementing AI.
AI Structural Equivalence: if you can create structural similarity you can have more understanding of how models work, and can make changes as needed
Gaiascope proves the possibility & feasibility of a joint AI & Physical model in grid-scale applications.

The main challenge we face today is data quality and cleansing, Lauren went on to say. The key takeaway from Lauren's talk was the big opportunity for AI as the grid infrastructure shifts towards carbon free sources; there was however the caveat that we need to address the challenge of guaranteeing reliability and safety. That said, the strategies to embed physical constrains into AI models hold a lot of promise!

You can watch from the start of Lauren's talk here.

You can also see more on Gaiascope's technology here.

Topher White, started with a general overview of Rainforest Connection, an organisation focused on the concept that sound waves and audio detection can be used to protect the rainforest from devastating deforestation. Deforestation is the second largest cause of climate change, with illegal logging responsible for between 50-90% of all deforestation. Topher went on to explain that guardian devices can detect the sound of a chainsaw within a 3sq km radius, meaning that if we can translate this information to partners on the ground, who in turn can put a stop to this, is the equivalent of taking 3,000 cars off of the road per year in terms of carbon emissions. Topher is not doing this alone, however, as due to the lack of law enforcement or police, Rainforest Connection work with both volunteers on the ground and local tribes to combat this existential threat. Utilising both AI, sound recognition and 30 trained rangers, Rainforest Connection place guardian devices in dangerous areas, which, when connected to cellphone networks & using a CNN Models, they were able to pick out the sounds of chainsaws and logging equipment at levels which the human ear would not recognise.

Topher later explained that Rainforest Connection is soon launching a bioacoustics platform, also able to use AI to pick out rare occurrences in the wild, such as orcas gathering over selected time periods.

You can watch from the start of Topher's talk here.

You can also see more on the initiatives from Rainforest Connections here.

Top AI Resources - Directory for Remote Learning
10 Must-Read AI Books in 2020
13 ‘Must-Read’ Papers from AI Experts
Top AI & Data Science Podcasts
30 Influential Women Advancing AI in 2019
‘Must-Read’ AI Papers Suggested by Experts - Pt 2
30 Influential AI Presentations from 2019
AI Across the World: Top 10 Cities in AI 2020
Female Pioneers in Computer Science You May Not Know
10 Must-Read AI Books in 2020 - Part 2
Top Women in AI 2020 - Texas Edition
2020 University/College Rankings - Computer Science, Engineering & Technology
How Netflix uses AI to Predict Your Next Series Binge - 2020
Top 5 Technical AI Presentation Videos from January 2020
20 Free AI Courses & eBooks
5 Applications of GANs - Video Presentations You Need To See
250+ Directory of Influential Women Advancing AI in 2020
The Isolation Insight - Top 50 AI Articles, Papers & Videos from Q1
Reinforcement Learning 101 - Experts Explain
The 5 Most in Demand Programming Languages in 2020

Interested in reading more leading AI content from RE•WORK and our community of AI experts? See our most-read blogs below: