How Nabat Is Rebuilding the World's Ecosystems With AI

Something is dying, and we keep not quite looking at it. Not in the abstract. Not in some projected model of a world thirty years from now. Right now, today, in the time between reading one paragraph and the next, an ecosystem somewhere on this planet has shifted past a threshold it will not cross back. A coastline has lost another metre of mangrove buffer. A forest understory has thinned past the point where it can shelter the species that shelter everything else.

A wetland that once filtered water for tens of thousands of people is becoming scrubland, and almost no one has measured it, recorded it, or told anyone with the power to act.

This is the condition Nabat was built inside of. Not as a response to a trend report or a funding opportunity, but as a direct answer to a specific, observable failure: the people and organisations responsible for protecting the natural world are routinely asked to act without knowing what they are acting on. The data is incomplete. The coverage is patchy. The tools that exist are either too slow, too disruptive, or too expensive to deploy at the scale the problem actually demands.

Nabat intends to fix that. Not with ideology, and not with the breathless confidence that has made so much environmental technology feel more like marketing than science. With data, with hardware, with artificial intelligence, and with something rarer than any of those things in this sector: the willingness to say when the right answer is to do nothing at all.

The Diagnosis Comes First

The central problem Nabat identified is not that people lack the will to restore ecosystems. It is that they lack the information to do it correctly. Organisations tasked with conservation frequently work with incomplete data: snapshots of ecosystems taken at long intervals, with limited geographic coverage, and using instruments that disturb the very environments they are attempting to protect.

"Many organisations are asked to act on ecosystems without having a complete, up-to-date understanding of where those ecosystems stand," says Mehdi Ajana, Nabat's Head of Strategy. "How healthy they are. How fast they are changing. What interventions are most appropriate?"

Take mangroves as the sharpest illustration. Successful mangrove restoration is not simply a matter of planting seedlings in shallow water. It depends on an intricate web of interacting conditions: soil composition, tidal rhythm, salinity levels, the presence or absence of particular species, hydrological flow. Get those factors wrong, and a restoration project fails silently. The seedlings die, the investment evaporates, and the ecosystem continues its decline while the records show "planting activity completed."

Nabat's answer is to establish understanding before action. Every deployment begins with a baseline assessment: a comprehensive reading of the ecosystem's current condition, its dynamics, and its trajectory. Only after that analysis is complete does the question of intervention arise. And in some cases, the answer is to do nothing. To protect rather than plant. To let natural recovery proceed without disturbance.

"When restoration succeeds," Ajana says, "the result is not just planting activity. It is a functioning ecosystem, supporting coastlines, marine life, carbon storage, and local resilience."

What the Machines See

Nabat's technology is built around a fundamental insight: ecosystems change continuously, but our ability to observe that change has historically been intermittent and coarse. The team has developed AI models capable of processing large volumes of environmental data, drawing on satellite imagery, aerial surveys and sensor readings, and translating that data into geospatial formats that ecologists can interpret and act upon.

The system can detect patterns that would be invisible to a researcher standing in the field: subtle shifts in vegetation health across thousands of hectares; the early signs of hydrological disruption; areas where natural recovery is already under way and intervention would be counterproductive; areas where the window for successful restoration is narrowing and delay carries a measurable ecological cost.

But Ajana is careful about what he claims for the technology. When asked where AI actually makes decisions today, his answer is deliberate. "AI supports scale and clarity," he says. "Final decisions remain with scientific teams, often in collaboration with local agencies and nearby communities."

This is not modesty for its own sake. It reflects a genuine philosophy about where technology adds value and where human judgement remains irreplaceable. An AI model can flag that restoration conditions appear viable in a given area. It cannot weigh the social dynamics of the community living adjacent to that area, or the institutional relationships that will determine whether a project has long-term stewardship. Those judgements belong to people.

The Hardest Constraint

Ask most technology companies about the hardest operational constraints they face, and you will hear about power systems, connectivity, hardware maintenance, regulatory approvals, extreme weather. These are real problems. Nabat faces them. But when Ajana is asked this question, he gives an answer that stops the conversation.

"Time," he says. "In many ecosystems, the window for recovery is narrowing. Delays increase costs and reduce options, even when intentions are strong."

There is a particular weight to that framing. Nabat is not describing time as an inconvenience or a project management challenge. They are describing a biological clock: the reality that some degradation processes, once past a certain threshold, become self-reinforcing and impossible to reverse. A mangrove system that loses its hydrological connectivity does not simply stagnate. It collapses inward, and the collapse accelerates.

"We see this not as a constraint imposed by others," Ajana says, "but as a responsibility we choose to carry. It shapes our pace, our focus, and our commitment to long-term engagement."

Scale Without Uniformity

One of the most common failure modes in environmental technology is the assumption that a solution proved in one context can be replicated directly into another. A drone seeding protocol developed for a semi-arid savanna does not transfer automatically to a tropical coastal wetland. An AI model trained on one continent's forest imagery may misread the signals it encounters on another. And the institutional landscape varies enormously across regions: the regulators, the community organisations, the land tenure arrangements. None of that can be engineered around.

Nabat has built for this reality from the outset. "Scaling does not mean repeating the same solution everywhere," Ajana explains. "We design systems that can be adapted to different ecological contexts, while remaining accessible to the agencies and communities we work with. Each biome, regulatory framework, and local context requires adjustment."

Nabat's hardware, its AI models and its field protocols are built with modularity at their core: not locked to a single use case, but designed to be reconfigured as both the ecology and the institutional environment demand. Every engagement is also structured with continuity in mind. The goal is not to deliver a project and withdraw, but to build local capacity, shared ownership, and the trust required for restoration work to persist beyond any single contract or funding cycle.

"Lasting impact depends on shared ownership, trust, and the ability for others to operate and carry these systems forward over time," Ajana says.

Humility as Architecture

There is a phrase that appears, in some form, in almost everything Ajana says about Nabat's approach: "we design with humility and learning built in." It is easy to treat this as a rhetorical gesture, the sort of thing technology companies say to signal that they are thoughtful. In Nabat's case, it appears to describe something structural.

Their AI models are explicitly designed to be updated as ecological science evolves. Their deployment methodology includes monitoring not as an afterthought but as a core function: the feedback loop that determines whether an intervention is working and how it should be adjusted. The company explicitly reserves the right to tell a client that intervention is not appropriate, that protection is the better path, or that natural recovery should be allowed to proceed without interference.

In a sector frequently driven by the optics of activity, the planted seedling, the aerial photograph, the carbon credit certificate, this willingness to counsel restraint is notable. Ecosystems are living systems, and scientific understanding of how they function continues to evolve. A company that locks its models and its protocols is, in a real sense, refusing to learn.

Nabat's wager is that the opposite approach, building in the capacity to be wrong and to correct course, will produce more durable outcomes than any amount of confident action.

The forests are still thinning. The coastlines are still retreating. Nabat is not claiming to have solved this. What they are doing is carefully, at pace, building the infrastructure for decisions that are better informed, more ecologically sound, and more honestly measured than those that came before. In a field defined by the distance between ambition and outcome, that may be the most important thing a company can do.