These Mozilla awardees are building open-source AI tools that help, not harm, the planet

(MONDAY, FEBRUARY 5, 2024) - The newest cohort of the Mozilla Technology Fund will use open-source AI tools to track methane emissions, expose harmful mining operations, monitor air quality, and more.

Mozilla is announcing its 2024 awardees today. The winning projects span six countries: India, France, Kenya, Paraguay, Uganda, and the U.S. Mozilla is providing awards of up to $50,000 each and one year of mentorship and support to these awardees.

All projects in the 2024 cohort work at the intersection of open source, environmental justice, and AI, making a positive impact in ecosystems and human communities. They complement Mozilla’s broader work on environmental justice, like our Climate Commitments and participation in the Green Screen Coalition.

Launched in 2022, the Mozilla Technology Fund supports open-source technologists addressing the internet’s thorniest issues; past cohorts have increased transparency and mitigated bias in the AI ecosystem. The fund strengthens the community and sustainability of these open-source projects, and is aligned with Mozilla’s institution-wide Trustworthy AI funding principles.

Says Mehan Jayasuriya, Senior Program Officer at Mozilla: “How we choose to build and deploy AI systems has an outsized impact on the environment around us. These systems can contribute to environmental degradation — or help prevent it. They can harm the land that indigenous peoples call home — or help those communities push back against resource extraction and harmful land management practices. And AI systems can accelerate climate change — or help address it. These awardees are leading the way in demonstrating how open source AI can be used to help, not harm, the environment.”

"These awardees are leading the way in demonstrating how open source AI can be used to help, not harm, the environment.”

Mehan Jayasuriya, Senior Program Officer at Mozilla

Meet the awardees

MethaneMapper, by EyeClimate in the U.S.

MethaneMapper solves the problem of underreported methane emissions. When most oil and gas companies, coal plants, and factory farms ignore leaks in pipes and self-monitor their fields to measure the least carbon and methane possible, emissions accounting and regulation mean very little. MethaneMapper is an open-source, AI-powered hyperspectral imaging tool to detect methane emissions and trace them to their sources. The tool works because data does not come from handheld devices or even individual installed devices that can be easily fooled. The input data is overhead — airborne hyperspectral images of the target area. MethaneMapper is therefore far more accurate and works over a far wider geographic region than traditional monitoring.


Amazon Mining Watch, by the Rainforest Investigations Network, the Earth Genome, and Amazon Conservation in the U.S.

Amazon Mining Watch (AMW) is an open-source collaboration between the Pulitzer Center's Rainforest Investigations Network (RIN), Earth Genome, and Amazon Conservation, which stands at the nexus of pioneering technology, transparency, and environmental justice. AMW uses AI to search satellite imagery, identifying gold mines that threaten indigenous communities, cause deforestation, and endanger our planet's most crucial rainforest. These detections are exposed to the public through a user-friendly public website and GitHub repo. AMW is a beacon that exposes the covert and harmful mining operations menacing the Amazon and its indigenous peoples.


CodeCarbon, in France

CodeCarbon started in 2020 with a simple question: How can I measure the carbon footprint of my computer program? The team found some global data, like "computing currently represents roughly 0.5% of the world’s energy consumption," but nothing about the impacts of individual/organization-level impacts and where they stem from. At CodeCarbon, the team believes, as Niels Bohr said, that "Nothing exists until it is measured." So they found a way to estimate how much carbon they produce while running our code. They did this by creating a Python package that estimates hardware electricity power consumption (GPU + CPU + RAM) and applies it to the carbon intensity of the region where the computing is done to automatically calculate the energy consumption and carbon emissions of any piece of code.


Sand Mining Watch, by the University of California, Berkeley in the U.S. and India

Sand Mining Watch is a web-based sand mining and sand resource monitoring system, initially focusing on the river systems of India. The project’s goal is to build open-source, AI-based sand mine detection tools that make it possible to produce high-resolution, real-time maps of sand mining activity in river basins across the world. These tools and data can catalyze policy action, improve the monitoring and regulation of illegal mining activity, and help us characterize and better understand the socio-economic and environmental impacts of sand mining.


Zeus: Deep Learning Energy Measurement and Optimization, by Jae-Won Chung in the U.S.

Zeus is the current state-of-the-art in deep learning energy measurement and optimization. It has monitor components that allow users to measure GPU energy consumption and optimizer components that automatically optimize DNN or GPU knobs based on measurements from the monitor component. The team’s goal is to measure, understand, optimize, and expose the energy consumption of modern machine learning systems. They view energy consumption as an emerging, first-class metric in computer systems — and one that has been largely overlooked until now.


Public Utility Data Liberation, by the Catalyst Cooperative in the U.S.

The Public Utility Data Liberation (PUDL) project is an open-source data pipeline providing free, up to date, version-controlled and analysis-ready energy data products based on three decades of data reported to federal agencies. Its long-term vision is an open-source PUDL ecosystem that: 1) provides equitable access to high-quality energy data; 2) serves as a community repository of well-documented, reproducible data cleaning and analysis pipelines essential to energy system research and advocacy; 3) is accessible to users working with a variety of different tooling (e.g., Excel); and 4) is an inclusive and welcoming environment for users and contributors.


AI-Driven Air Quality Forecasting for Asunción, by Fernanda Carlés in Paraguay

Since 2022, Fernanda Carlés has been working alongside the National University of Asunción to develop machine learning models for predicting AQI levels for UNA's ten monitoring stations. This effort has resulted in high-performance machine learning models that offer air quality predictions for six- and 12-hour horizons with impressive accuracy rates of 91% and 86%, respectively. The goal of this project is to bring the predictive power of these machine learning models to the public. To achieve this, the project proposes the development of a user-friendly web application that offers real-time air quality forecasts with interactive visualizations and alerts. This application will utilize real-time data from existing monitoring stations in the Gran Asunción area, combined with validated machine learning models.


Leveraging AI for Environmental Justice and Environmental Impact Assessment for Advocacy on Nuclear Energy and Policy, by Center for Justice Governance & Environmental (ACTI) in Kenya

This team previously developed AI-based software products for separate projects in the Pacific (CCZ) and tropical north Atlantic Ocean (seafloor habitat classification and species detection and classification). For this project, they will refactor the two workflows into one open-source code base that is adapted for a working area in the Indian Ocean where the development of a new nuclear reactor has been proposed. The project seeks to utilize AI and the conservation community to produce a comprehensive ecological mapping that will assist the Uyombo community to protect their clean and healthy environment, their socioeconomic rights, and their cultural rights.


AIG SMS for Environmental Justice among the Mau Indigenous Community, by L. Lusike Mukhongo in the U.S. and Kenya

The AI-Generated (AIG) Short Message Texts (SMS) for Environmental Justice among the Mau Indigenous Community project is a groundbreaking effort to address pressing environmental issues faced by the Mau Indigenous Community in Kenya. The Mau Forest, one of East Africa's largest water catchment areas, is under threat from deforestation and illegal farming settlements, compounded by climate change, thereby directly impacting the livelihoods and well-being of the indigenous Ogiek community and the greater Mau community. The project’s goal is to advance the principles of environmental justice by adopting AI-generated Short Text Messaging among the indigenous community in the Mau forest that draws from their traditional knowledge of the conservation of forests and the preservation of vital water towers. The AI system will also have text-based prompts that allow users to query it and get results about a topic of interest.


Estimating and Validating PM2.5 Levels Using Satellite and Ground Observations in Selected African Cities for Environmental Justice, by AirQo in Uganda

This project will estimate particulate matter (PM2.5) levels from satellite observations based on Aerosol Optical Depth (AOD) for seven cities in seven African countries: Lagos, Accra, Nairobi, Yaounde, Bujumbura, Kisumu, and Gulu. It will employ varying ground monitoring resolutions using machine learning algorithms, and use available ground-based observations in the selected cities to validate the satellite estimates. The outcomes of the developed AI models will be deployed through user-friendly digital platforms and a mobile app which democratizes and empowers communities to access crucial air quality information. This will foster evidence-based, informed, and inclusive decision-making and strategic interventions for tackling local pollution challenges while achieving environmental justice.