An AI Model Maps Alaska's Shrinking Tundra, Plant by Plant

From a satellite's usual vantage, the seam where Alaska's boreal forest gives way to open tundra looks like a smudge, a gradual fade from green to brown across hundreds of miles. Up close it is anything but smooth. It is a mosaic of knee-high shrubs, scattered spruce, sedge and bare ground, and that fine texture is exactly what matters as the Arctic warms and woody plants push north into terrain that was once too cold to hold them. Coarse maps miss it. A tundra patch turning to shrubland can be invisible at the resolution most large-scale vegetation products offer.

A study accepted July 22, 2026, in Environmental Research Letters tries to bring that texture into focus. A team led by Paul Montesano, drawing on researchers at NASA and Alaska-based ecological specialists, mapped vegetation height across roughly 40% of Alaska's boreal-tundra ecotone, the transition zone between forest and tundra, at 2-meter horizontal resolution and decimeter-scale vertical precision. That is fine enough to tell a low shrub from a tall one and a tall shrub from a young tree.

The engine behind the map is an AI foundation model, the same broad class of large, pre-trained systems now reshaping other corners of Earth observation. Rather than build a bespoke model from scratch, the researchers took one originally trained for global applications and fine-tuned it with airborne data and field observations specific to Arctic vegetation. It is a now-familiar recipe: adapt a general-purpose model to a narrow, data-poor problem by feeding it a targeted slice of local ground truth.

What the model produced is a detailed inventory of stature across the mapped zone. Low vegetation under about a third of a meter tall dominates, covering 63.4% (give or take 5.2) of the area, the sedges, mosses and dwarf shrubs that carpet the tundra floor. Low shrubs between roughly 30 centimeters and 1.5 meters account for about another 18%, and taller canopies (the trees and large shrubs that mark forest advance) make up a further 18% or so. Laid out at a 2-meter scale, that breakdown becomes a map of where the woody plants already stand and where the open ground still holds.

The point of such a baseline is what comes next. Shrubification (the spread of shrubs into tundra) and the northward creep of trees are among the clearest fingerprints of Arctic warming, and both are entangled with permafrost: taller vegetation traps snow, shades soil and alters how the frozen ground beneath thaws. A high-resolution snapshot of vegetation height gives researchers a precise starting line, so that when the same terrain is mapped again years from now, the change can be measured plant by plant rather than inferred from a blur.

Two key considerations help put these findings into context. First, this is a baseline, not a trend. The study maps what the vegetation looks like now in unusual detail; it does not, on its own, measure how much has changed or how fast, since that requires a second map down the road. Second, it covers 40% of the ecotone, a large but partial slice, and the figures describe that transition zone specifically, not the whole of Alaska. The method's accuracy also rides on the quality of the airborne and field data used to tune it and on how well a globally trained model generalizes to Arctic species it was rarely exposed to.

Still, as a demonstration, it points somewhere useful: fine-tuned foundation models paired with commercial imagery can resolve a warming landscape at a grain fieldwork alone could never cover, turning a smudge on the map into something detailed enough to watch change.