Wildfires could be more effectively managed if artificial intelligence (AI) models incorporated Indigenous knowledge more fully for landscape use around the world, according to a recent study. 

The AI-based quantitative frameworks that are used to help predict and map the occurrence of fires should be designed to complement – rather than replace – the traditional and Indigenous knowledge (TIK) acquired through generations of observation, practice and cultural transmission. 

These are among the findings of the paper published by the Center for International Forestry Research and World Agroforestry (CIFOR-ICRAF) with the support of the Social Sciences and Humanities Research Council of Canada. 

Using Canada as a relevant case study, the paper examines scholars’ increasing interest in the integration of TIK with technological AI-based approaches to wildfire management. 

“Integrated fire management (IFM) is more relevant than ever to a wider population, often including Indigenous people and local communities,” says Isaac Rutenberg, the study’s author. “Better IFM systems based on AI analysis of large and diverse datasets are likely to be good for people, property and ecosystems.” 

Climate change and fires 

The importance of more effective fire management is increasingly obvious as climate change exacerbates the frequency, intensity and severity of wildfires, raising ecological, economic and public-safety risks.

Rutenberg, who participates in a CIFOR-ICRAF research programme at the intersection of artificial intelligence, agriculture and climate change, says AI-based systems could support integrated fire management, but only if they are developed in ways that respect Indigenous governance, participation and data sovereignty.

Last year, wildfires burned through almost 390 million hectares worldwide – an area nearly as large as the European Union – with over half of the total area being in  Africa, according to the United Nations Office for Disaster Risk Reduction (UNDRR).

Wildfires rank among the most economically destructive hazards globally, and the costs are rising. From 2014 to 2023, they caused an estimated USD 106 billion in economic losses and USD 74 billion in insured losses globally – far exceeding losses in the previous decade, according to a UNDRR report. 

While hundreds of people die from the direct effects of fires annually, scientists have estimated that wildfire smoke causes more than 1.5 million deaths globally each year.  

AI versus TIK 

Artificial intelligence – which performs tasks through symbolic reasoning, optimization, robotics, and machine learning – is often used to predict fire spread, to identify high-risk areas, to map burned areas, or to schedule prescribed burns. However, it can miss the cultural rules, spiritual context and communal governance framework that give this knowledge its full meaning, according to Rutenberg. 

TIK is considered dynamic and specific to a certain place or context. It is often encoded in rituals, stories and cultural practices. Yet Western research and data protocols often treat this knowledge as information that can be harvested selectively for scientific purposes, a practice that may deny Indigenous communities of sovereignty and control over their own data.

“There is a very long history of engaging traditional and Indigenous knowledge in an extractive approach, often stripping TIK of its cultural context,” Rutenberg says. 

The paper cites the potential operational delineations (PODs) framework as the most participatory technical approach to integrated fire management. First developed by the United States Department of Agriculture (USDA) Forest Service, PODs operate through inclusive workshops involving tribal members, land managers and fire crews to create shared frameworks for predicting fire behaviour and containment strategies. 

Reform or transform? 

Two broad governance approaches exist to integrate TIK with AI technologies. Reform-oriented approaches aim to influence existing AI development processes within universities, corporations and government agencies. Transformational approaches, by contrast, advocate for Indigenous-led AI initiatives that operate independently of mainstream institutions, according to the study. 

In the paper, Rutenberg recommends inclusive Indigenous governance from project inception; formal agreements establishing community ownership and authority over data and AI systems; and sustained capacity building that creates community technical expertise rather than dependence on external partners. 

These steps would involve a hybrid approach based on CARE principles (collective benefit, authority to control, responsibility and ethics) or on the First Nations Information Governance Centre’s OCAP® principles (ownership, control, access and possession), which assert that First Nations have collective ownership and authority over data concerning their communities. 

The way forward 

Consequently, there is an urgent need to develop legal and institutional frameworks that genuinely support Indigenous interests in data sovereignty and free, prior and informed consent (FPIC) in AI contexts. AI systems that operate according to Indigenous values – such as reciprocity, relationship and responsibility – also must be developed to replace those that are based on inappropriate assumptions. 

Furthermore, structural changes are required in research funding, academic publishing and technology development processes to remedy the under-representation of Indigenous voices in the literature. These changes should aim for genuine power sharing and Indigenous leadership, Rutenberg writes. 

Meaningful integration may lie in developing participatory, community-led AI governance models that treat Indigenous knowledge systems as co-equal sources of innovation, according to the paper. This would enable technological development that is not only efficient but culturally grounded, ethically responsive and socially inclusive. 

“If the engagement is done well, perhaps TIK and emerging AI systems can be mutually beneficial,” Rutenberg says.  

AI systems need better contextualization – or place-based understanding – for improved outputs, which TIK can usually offer. Meanwhile, AI systems could benefit TIK through the archiving and translation of such knowledge, if a community is interested in participating in such measures, he adds. 

*The research for this occasional paper was conducted under the Abundant Intelligences programme at Concordia University, supported by the Social Sciences and Humanities Research Council of Canada (SSHRC).