Use
of AI in Climate Justice: AlphaEarth Data for Indigenous Land
Protection

The satellite image of a forest looks the same whether the entity
capturing it intends to protect what it depicts or exploit it. This is the
central tension in AI’s emerging role in climate justice: the tools that can
support Indigenous land rights and environmental monitoring are the same
tools that can be used to identify, map, and ultimately open up territories
that communities have managed sustainably for generations. Whether AI serves
as a shield or a surveyor depends entirely on who controls the technology,
what questions they ask of it, and whether the communities whose lands are
being analysed have any meaningful say in the process.

Google DeepMind’s AlphaEarth initiative has generated significant
attention as a demonstration of AI’s potential for environmental monitoring
at scale. The platform applies machine learning to satellite imagery and
environmental sensor data to track deforestation, land degradation, wildfire
risk, and biodiversity change with a granularity and speed that was not
previously available. For climate scientists and conservation organisations,
the capability it represents is genuinely valuable. For Indigenous
communities and climate justice advocates, the question of who controls this
capability and on whose terms it is deployed is at least as important as what
the technology can see.

The Data Sovereignty Problem

The application of AI to Indigenous lands without the consent and
participation of those communities replicates a pattern that has
characterised extractive relationships between powerful institutions and
Indigenous peoples for centuries. The form changes, satellites and algorithms
instead of surveyors and maps, but the structure of the relationship does
not: knowledge is extracted from territories, processed by outside
institutions, and used to make decisions that affect communities who had no
voice in the process.

An arXiv study on
Indigenous data sovereignty and AI makes the case for applying CARE
principles, Collective benefit, Authority to control, Responsibility, and
Ethics, as the governance framework for AI applications involving Indigenous
communities and their territories. CARE principles place community consent
and benefit at the centre of data governance, in contrast to the FAIR
principles, Findability, Accessibility, Interoperability, Reusability, that
dominate much of the data science community’s thinking about research data.
The distinction is not merely philosophical: it determines whether AI applied
to Indigenous land is a tool that communities control or a tool applied to
communities without their agreement.

The Columbia
Climate School’s landscape assessment of AI for climate and nature
identifies the failure to integrate land custodianship values as a systematic
weakness of current AI conservation applications. When AI models reduce
Indigenous forest management to a set of pixels and carbon metrics, they
strip away the social, cultural, and spiritual dimensions of the relationship
between communities and their territories. Decisions based on those
stripped-down representations may be technically defensible but are missing
the information that matters most to the people most affected.

Where AI Can Genuinely Help

The risks of AI in the climate justice context should not obscure
the genuine potential. Indigenous communities managing territories across the
Amazon basin, central Africa, and Southeast Asia have successfully used
satellite monitoring and AI analysis to document illegal incursions,
deforestation events, and land grabs in real time, producing evidence that
has supported legal challenges and international advocacy.

The World
Economic Forum’s 2025 assessment of AI and climate justice
highlights examples where AI tools, deployed with genuine community consent
and control, have transformed the capacity of Indigenous land defenders to
document and respond to environmental threats. The difference between these
cases and the extractive model is not primarily technical; it is about
governance. When communities control the questions asked of the data, the
interpretation of the results, and the decisions taken on the basis of
analysis, AI becomes an extension of community agency rather than a
substitute for it.

The food security implications of this technology connect directly
to work explored in Algorithmic
Hunger. Indigenous land management systems have historically
maintained biodiversity and soil health at scales and over timescales that
industrial agriculture has rarely matched. AI tools that support those
systems, helping communities monitor land health, detect invasive species
early, and track the environmental changes that affect traditional food
sources, can make a genuine contribution to food security and ecological
resilience.

The “Underutilised Land” Problem

One of the most concerning potential uses of AI environmental
mapping is the identification of land characterised as underutilised or
underdeveloped. The Columbia
Justice Network’s analysis of AI and climate justice warns that AI
systems trained on economic productivity metrics can misclassify Indigenous
conservation practices as underuse, generating data that governments and
developers can then use to justify land acquisition under the guise of
development or climate action.

This is not a theoretical risk. In multiple jurisdictions,
governments have used conservation designations, sometimes supported by
scientific data, to displace Indigenous communities from lands they have
managed sustainably for generations. AI that provides more granular and
apparently authoritative data about land use patterns without integrating
Indigenous land governance systems into its analytical framework can amplify
these existing patterns rather than challenge them.

The invisible infrastructure concern raised in Invisible
Infrastructure is directly relevant here. The choices embedded in
how AI systems categorise land use, what counts as productive, what is
measured and what is not, are not neutral technical decisions. They reflect
assumptions about value that may be deeply at odds with the value systems of
the communities whose territories the AI is analysing. Making those
assumptions visible and subject to challenge is a prerequisite for AI climate
tools that serve justice rather than undermine it.

Toward Community-Controlled AI

The emerging best practice in AI applications for Indigenous
environmental monitoring centres on community control at every stage of the
process: data collection, model design, output interpretation, and
decision-making based on results. This is more resource-intensive than
deploying off-the-shelf AI tools on publicly available satellite imagery, and
it requires sustained relationships between technology providers and
communities that most current AI development cycles do not
accommodate.

Several organisations are working to bridge this gap. The Global Forest
Watch platform has developed community monitoring tools that allow Indigenous
land defenders to contribute their own observations alongside satellite data,
creating hybrid datasets that integrate scientific monitoring with
traditional ecological knowledge. Similar approaches are being developed for
marine territory monitoring and for the documentation of sacred sites at risk
from development.

The broader principle is that climate AI done well is not about
giving communities access to better tools. It is about building tools that
communities can direct toward their own defined goals, with their own
analytical frameworks, producing outputs that they interpret and act on. That
is a fundamentally different relationship between technology and community
than the one that characterises most current AI deployment in the
environmental sector. It is also the relationship that makes the technology
genuinely useful rather than merely impressive.

The linguistic dimension of this challenge is underappreciated.
Many Indigenous communities express their relationship to land in languages
that have no direct equivalent in the dominant languages in which AI systems
are trained. As explored in The
Forgotten Accent, AI systems trained predominantly on
majority-language data are systematically less capable when engaging with
minority language communities. For climate AI applied to Indigenous
territories, this means that the most ecologically sophisticated knowledge
about those territories, held in Indigenous languages through oral
traditions, may be entirely invisible to the AI systems that are supposed to
be helping protect them. Building climate AI that genuinely serves Indigenous
communities requires addressing the linguistic gap alongside the governance
one.

About the Author

Stuart Kerr is the Technology Correspondent for LiveAIWire,
covering artificial intelligence, ethics, and the ways technology is
reshaping everyday life.