Building Climate TRACE's Power Sector Intelligence: A 6-Year Partnership
Client: WattTime / Climate TRACE Coalition (Power Sector)
Engagement: Ongoing ML consulting partnership (2019–present, 6+ years)
The Mission-Critical Problem
The world cannot manage what it cannot measure. Before this project, there was no comprehensive, independent system to monitor carbon emissions from every large power plant on Earth. Governments and companies self-report emissions—or don’t report at all—creating a massive blind spot in global climate accountability.
WattTime, a nonprofit leading the power sector analysis for the Climate TRACE coalition, set out to solve this using satellite imagery. The challenge: build AI systems that could estimate power plant emissions from space across every satellite, sensor type, and geography worldwide—then scale it to track thousands of facilities in real-time.
Impact at Global Scale
Unprecedented transparency: First independent system to monitor emissions from all large power plants worldwide—presented at COP28 and COP29 by Al Gore and the Climate TRACE coalition
Global deployment: Now tracking 1000+ power plants globally through the power sector component of Climate TRACE’s public emissions inventory at climatetrace.org, enabling governments, investors, and journalists to make informed decisions about emissions and climate action
Scientific credibility: Peer-reviewed publications in Remote Sensing (MDPI, 2024), Climate Change AI Workshop (NeurIPS 2020), and IEEE IGARSS (2023) building trust through rigorous scientific validation
Policy and media impact:
- Climate TRACE featured in Time Magazine’s Best Inventions of 2020
- Database updates covered by Bloomberg, The New York Times, Washington Post, The Guardian, Axios, and others
- Gavin McCormick presented the work in a TED Talk (August 2021)
- Google.org recognized the solution as a breakthrough climate technology
Client Testimonial
“It’s clear to me Heather thinks like a scientist who is equipped with deep ML skills and a bit of software engineering, not like a software engineer who has deep ML skills and a bit of science. And I can’t emphasize enough how much more valuable that is to a deep tech startup.”
— Gavin McCormick, Executive Director, WattTime
Why It Worked
The long-term nature of this collaboration and its global impact speak to the depth of the partnership. WattTime needed more than ML engineering—they needed someone who could think like a scientist while building production systems. As a small nonprofit with limited budget, every consulting dollar had to deliver outsized value.
What made the difference:
- Built scientific credibility: Three peer-reviewed publications built trust with the climate science and policy communities
- Domain translation: Bridged the gap between remote sensing and deep learning, getting a new team focused quickly
- Avoided costly dead ends: Focused on data fundamentals over trendy architectures, compressing years of potential exploration into focused iteration
- Partnership stability: Delivered focused progress over 6+ years without scope creep or distractions
- Mission alignment: Understood that imperfect models deployed globally beat perfect models that never ship
The Technical Challenge
Building AI systems to monitor global power plant emissions from satellites required solving:
- Heterogeneous satellite data - Multiple satellites, sensors, wavelengths, and revisit schedules; images look fundamentally different from standard computer vision datasets
- Low spatial resolution and occlusion - Cloud cover and satellite timing constraints limit when and how clearly power plants can be observed
- Generalization across geographies - Models trained on plants with recorded generation or emissions (mostly developed nations) must work globally
- Resource constraints - Limited labeled data and need for efficient inference
What I Delivered
Enabled global emissions monitoring by developing AI-powered detection systems to identify visible operational signals: vapor plumes from cooling towers, mechanical cooling systems, and emissions control equipment—achieving reliable detection
Built scientific credibility through peer-reviewed publications that translated technical work into validated methodology
Drove global validation strategy - Established frameworks for model evaluation across geographies without reported emissions data, enabling confident deployment worldwide
Domain expertise and execution consistency - Bridged remote sensing and machine learning domains while maintaining focused progress over 6+ years of ongoing collaboration
Evidence & Recognition
Peer-reviewed publications:
- H.D. Couture, Estimating Carbon Dioxide Emissions from Power Plant Water Vapor Plumes Using Satellite Imagery and Machine Learning, Remote Sensing (MDPI), 2024
- H.D. Couture, Towards Tracking the Emissions of Every Power Plant on the Planet, Climate Change AI Workshop, NeurIPS 2020
- M. Hobbs, Inferring Carbon Dioxide Emissions from Power Plants Using Satellite Imagery and Machine Learning, IEEE IGARSS, 2023
Media coverage: The Guardian | Forbes | Science | Forbes | IEEE Spectrum | Quartz | Time Magazine
TED Talk - Gavin McCormick: “Tracking the whole world’s carbon emissions – with satellites and AI” (2021)
Talks & Videos: Duke Energy Data Analytics Symposium | Google Sustainability
Featured at COP - Al Gore and Climate TRACE unveil emissions inventory at COP28 | Al Gore and Gavin McCormick unveil data at COP29