AI Sees the Forest and the Trees

A beetle no larger than a grain of rice is ravaging European forests, infesting and killing trees faster than they can be culled to slow the insects’ spread. It turns out the best way to spot the pests, and stop them, may be from space.

For years, Swedish forestry cooperative Södra has deployed hundreds of foresters to walk the widely spaced spruce on properties it helps manage, monitoring the trees’ bark for drilling holes that are a telltale sign of infestation. But it can take days to assess a single 100-acre estate by foot, and Södra oversees more than five million acres. Last year, the beetles damaged five million cubic meters of lumber, about a quarter of the season’s potential yield, says Johan Thor, an applied physicist and head of data science at the cooperative.

So in early 2019, Södra began working with the Dutch technology company Overstory to find the beetles from above—way above. By matching high-resolution satellite imagery with geographic readings of sick trees as recorded by the company’s harvesters, and integrating other satellite-derived data such as land-surface temperature, they were able to train a model to quickly and accurately locate infested areas. The complexity of the data—with a profusion of tree species and canopy heights—was “a sweet spot for machine learning,” says Overstory’s chief executive officer, Indra den Bakker.

“The preliminary results are really quite astonishing,” says Mr. Thor.

Click here to keep reading…