Research

Mapping tropical forest degradation with deep learning and Planet NICFI data

A deep learning model mapped forest degradation in Brazil's Mato Grosso at 8,443 km² per year, roughly double the area deforested, and outperformed all national and global monitoring products.

High-resolution maps show that rubber causes substantial deforestation

Rubber has caused at least 4 million hectares of forest loss in Southeast Asia since 1993, two to three times more than policy figures assume.

Improving above ground biomass estimates of Southern Africa dryland forests by combining Sentinel-1 SAR and Sentinel-2 multispectral imagery

Combining Sentinel-1 radar with Sentinel-2 optical data improved above-ground biomass estimates in Southern Africa's dryland forests, with random forest models using radar texture overcoming the saturation that limits optical-only approaches.

Mapping the presence and distribution of tree species in Canada's forested ecosystems

A machine learning model mapped 37 tree species across all 650 million hectares of Canada's forested ecosystems at 93.1 percent accuracy, revealing black spruce as the dominant species across 57 percent of the country's treed area.

Very high resolution canopy height maps from RGB imagery using self-supervised vision transformer and convolutional decoder trained on aerial lidar

A self-supervised vision transformer trained on aerial lidar produced sub-metre canopy height maps for California and Sao Paulo with an average error of 2.8 metres, a tenfold jump in resolution over prior Sentinel and GEDI based global maps.

Regional uniqueness of tree species composition and response to forest loss and climate change

A global map of 10,590 tree species shows historical forest loss has already restricted species ranges in every forest biome, with tropical dry and Mediterranean ecoregions facing the steepest projected losses.