IPN students build AI to anticipate wildfires: citizen monitoring as a green shield

By Momentum IA · June 28, 2026.

Four students of Artificial Intelligence Engineering at the Higher School of Computing (ESCOM) of the IPN have just demonstrated something that should not be surprising, but remains refreshing: that well-oriented school projects can become tools with real impact. The team made up of Denys Monserrat Rodríguez Méndez, Mildred Valeria Lagunes Vázquez, Brisa María Lezama Tapia, and Aldo Díaz Martínez, with the guidance of researcher José Asunción Enríquez Zárate, developed a prototype web platform capable of identifying areas at high risk of forest fire before the fire begins, with an initial focus on El Tepozteco National Park, in Morelos, a reserve of more than 23,000 hectares.

The technical architecture deserves attention. The system does not rely on pretrained models taken from commercial black boxes: the team built its own models from scratch, including convolutional neural networks to analyze images of the terrain and an autoencoder to detect anomalies in the state of the vegetation. To this is added the verification of geographic metadata —to ensure that the uploaded photos actually correspond to the protected area— and the integration of real-time climate variables: temperature, humidity, and wind speed. The result is a risk classification in three levels (low, medium, high) projected onto an operational map. To train the models, they built three image banks with thousands of photographs from public platforms, digital resources, and material provided by local specialists.

What distinguishes this development from most existing systems is the approach: not to detect the fire once it is already burning, but to anticipate the conditions that make it likely. Accumulation of combustible material, dryness of the vegetation, presence of waste, human activity in sensitive areas —all factors identified in collaboration with the municipality of Tepoztlán and environmental experts— feed the model so that authorities can deploy cleanup brigades, sanitation work, or preventive surveillance before the emergency occurs.

There is also a dimension that tends to be overlooked in projects of this kind: the user architecture. The platform contemplates three profiles —administrators, government authorities, and citizens— which turns the park's residents and visitors into active monitoring agents. They can upload photographs of suspicious areas, expanding the surveillance coverage of a territory that, due to its size, cannot be exhaustively covered by any body of forest rangers. That is collective intelligence underpinned by computational models: a combination with more potential than it appears at first glance.

Our reading is this: projects like this clearly illustrate where the true value of teaching applied AI in public universities lies. Not in producing engineers who replicate what the big laboratories already do, but in training teams that tackle concrete problems in their environment with cutting-edge tools. Mexico loses tens of thousands of hectares of forest each year to fires, many of them preventable with earlier monitoring. The fact that the advisor describes this prototype as a 'minimum viable product with the potential to be implemented in other natural reserves of the country' is not academic rhetoric: it is the logical roadmap for a system that, scaled and maintained with adequate resources, could change the management of protected areas across the entire territory.

The question that remains open —and that no launch article ever answers— is that of continuity. School prototypes shine in their moment and then disappear for lack of funding, institutional will, or simply because the students graduate and move on with their lives. Whether this development reaches operational deployment in El Tepozteco, and beyond, will depend on whether there are institutional actors —Conanp, state governments, conservation organizations— willing to invest in taking it from the laboratory to the field. That gap between the functional prototype and sustained implementation is the historic bottleneck of public innovation in Mexico, and no AI model resolves it on its own.

In the long term, the potential of AI applied to ecosystem protection is one of the strongest arguments for technological optimism. Systems that monitor forests, predict droughts, optimize water distribution, or warn about soil degradation are not science fiction: they are being built today, in some cases by undergraduate students. The transition toward that collective capacity has costs and frictions —inequality of access, dependence on data, risks of overconfidence in the models— but the direction is unequivocal. What this IPN project does is not solve the problem of forest fires in Mexico; it is to demonstrate that we already have the tools to begin doing so.