To restore a forest, know when to step back

A forest is not simply a collection of trees, any more than a city is just a collection of buildings.

It is soil, water, fungi, animals, genetic variation, and chance. It is also countless interactions, unfolding over decades or centuries. It is biodiversity. Planting trees can begin the process of restoration and is increasingly promoted as a climate solution. But humans cannot manufacture a forest.

In a new commentary in “Trends in Ecology & Evolution,” Pedro Brancalion of the University of São Paulo and Marcelo Medeiros, a Climate Action Fellow at Harvard’s Salata Institute, argue that restoration projects too often borrow the logic of construction, as if a forest were just another infrastructure development. That approach may produce tree cover. A self-sustaining forest, however, is a complex system that takes shape on its own over time.

“By failing to ‘see the forest for the trees,’ tree planting promotes a universal, visually compelling, and marketable solution,” they write. Yet where native forests have recovered extensively, the process has often been driven not by deliberate intervention but by natural succession after farmland was abandoned, “with forest communities emerging through self-assembly and following trajectories that are only partly predictable.”

Forest restoration, the authors argue, must embrace that complexity. It cannot be treated like commercial forestry, which makes landscapes simpler and more uniform, or agriculture, where a farmer can optimize conditions for a single crop. Forest restoration must account for hundreds of species, shifting relationships and outcomes that may remain uncertain for decades.

To optimize restoration efforts, the authors’ offer an “integrated framework” where humans can use AI to address the need, without depending on AI alone: People establish goals, contribute knowledge, and decide when to intervene. Artificial intelligence helps interpret large amounts of ecological data. “Natural intelligence” – succession, adaptation, and self-organization – does much of the actual forest building.

AI can help, to a point. A machine-learning model might analyze soil, hydrology, climate, and land-use history to identify where planting is needed and where a forest may recover on its own. But ecological data are patchy. Models may favor familiar species and overlook poorly studied ones. They may reinforce the assumption that planting is always better than natural regeneration. They may even encourage tree planting in places that are not naturally forests.

“Given that most plant, animal, and microbial species remain understudied ─ or have not even been formally described ─ AI-driven restoration could inadvertently promote biotic homogenization by favoring the limited subset of species for which data are abundant,” they write.

Thus technology’s role is not to design the forest, Brancalion and Medeiros contend, but to use human judgment and technology to give nature room to rebuild itself, what they call a “paradigm of informed humility.”