PART 0

WhoAmI ?

McGill University | Mila - Quebec AI Inst.

Sunbird AI

Benjamin Akera

Climate Change

Need net-zero greenhouse gas emissions by 2050 (IPCC 2018)

• Across energy, transport, buildings., industry, agriculture etc.

How does ML fit into this picture? 

Impacts felt globally

Disproportionate impacts on most disadvantaged populations

Takeaways and how to get involved

The state of climate Change

Temperature change

Human Activity as a Driving Factor

Approaches to Addressing Climate Action

01 - 

02 - 

03 - 

04 - 

05 - 

Has induced major changes in climate

• Climate = "average weather"
• Extreme heatwaves, precipitation, droughts, hurricanes

Earth has already warmed over 1 º C compared to pre-industrial period

Due to Excess greenhouse gas (GHG) -- emissions from human activities 

• E.g carbon dioxide, methane, nitrous oxide

Axes of Action

Climate Science: Understanding and predicting climate change 

Mitigation: Reducing or preventing greenhouse gas emissions

Adaptation: Responding to the effects of a changing climate

Reducing or preventing GHG Emissions

Sectors

Energy supply 

Transportation

Buildings

Industry

Agriculture

Forestry

.

.

.

Rising Temperatures

Changing Rain Patterns

Rising sea levels

Ocean acidification

Climate Impacts

Drougts & Heatwaves

More intense storms & flooding

More Frequent wildfires

Loss of Ecosystem services 

Downstream Effects

Biodiversity Loss

Spread of disease vectors & pests

Fig. adapted from Kris Sankaran

Opportunities for ML in climate action

Considerations for Research and Deployment

Is ML a help or a hinderance? 

Take Aways and How to get involved

01 - 

02 - 

03 - 

04 - 

01 - Distilling Raw Data into Actionable information

03 - Optimizing Complex Systems

04 - Improving Predictions

02 - Accelerating Scientific Discovery

05 - Approximating Time-intensive simulations

see also: https://www.climatechange.ai/summaries

Usecase - Glacier Mapping with Satellite Data

Mapping deforestation and carbon stock [M]

Open buildings John Quinn et al.

Role: Suggesting experiments in order to speed up the design process

Relevant ML Areas:  Generative Models, Active Learning, Reinforcement Learning, Graph Neural Networks

👨‍🔬⚗️🧪💊gg

Role: Forecasts and timeseries predictions

Relevant ML Areas:  Timeseries analysis, Computer Vision, Bayesian methods

Examples

- Nowcasting for Solar/wind power

- Forecasting electricity demand

- Predicting crop yield from satellite imagery

- Bird species distribution Modelling 

Role: Improving efficient operation of complex, automated systems

Relevant ML Areas:  Optimization, Control, Reinforcement Learning

Examples

- Controlling heating/cooling systems

- Optimizing rail and multimodel transport 

- Demand response in electrical grids 

Role: Accelerating time-intensive, often physics-based simulations

Relevant ML Areas:  physics informed ML, Computer Vision, Interpretability, Causal ML

Examples

- Superresolution of predictions from climate models 

- Simulating portions of car aerodynamics

- Speeding up planning models for electric grids

ML is not a silver bullet

High-impact applications are not always flashy

Interdisciplinary collaboration is key

Sophisticated algorithms can be required but aren't always

Equity Consideration

Sophisticated algorithms can be required but aren't always

Any innovations should be shaped by needs of the relevant applications

Physics/Engineering constraits or robustness guarantees

Interpretability or causality

Uncertainty quantification

Generalization e.g across geographies, under non-stationarity, or with limited data

Consider the full pathway to impact, including

Data, simulators, Evaluation metrics 

Responsible AI & Climate considerations

Stakeholder engagement

Takeaways and how to get involved

Introduction to climate change

Opportunities for ML in climate action 

Considerations for research and deployment 

Climatechangeai.org

Credit to Dr. David Rolnick, Dr. Priya Donti & CCAI from whom key points in this presentation are derived.