Scientists at ETH Zurich’s EcoVision Lab have developed a deep learning framework to map tree canopy height globally in high resolution, using publicly available optical satellite imagery as input. The Global Canopy Height Map is the first map of its kind. It could become a fundamental tool for monitoring carbon emissions that contribute to climate change and for planning sustainable regional development.
To acquire the data needed for the first global map of canopy height, EcoVision Lab researchers relied on two sources: NASA’s Global Ecosystem Dynamics Investigation (GEDI) and the Copernicus Sentinel-2 satellites operated by NASA. European Space Agency. GEDI, which has the highest resolution and densest sampling of any lidar ever placed in orbit, makes laser ranging observations of nearly all of Earth’s tropical and temperate forests and provides high-resolution measurements of Earth’s 3D structure. . It provides sparse but well-distributed data on canopy height around the world. Optical satellite images from the Sentinel-2 satellites capture every location on Earth every five days at a resolution of 10 × 10 meters/pixel. The Sentinel-2 satellites offer globally dense observations, although they cannot allow the measurement of vertical structures.
By combining GEDI data with Sentinel-2 data, the researchers created a deep learning probabilistic model to retrieve canopy height from Sentinel-2 images anywhere on Earth and quantify the uncertainty in these estimates with data from GEDI.
The researchers prepared a set of convolutional neural networks to map tree height globally. The neural networks were shown millions of sample images from two Sentinel-2 satellites.
“Since we don’t know what patterns the computer needs to look for to estimate height, we let it learn the best image filters on its own,” said researcher Nico Lang, who developed the networks.
The neural network algorithm accesses the appropriate answer to the height of the tree from the spatial laser measurements taken by GEDI. “The GEDI mission delivers sparse globally distributed data on vegetation height between latitudes 51° north and south, so the computer sees many different types of vegetation in the training process,” Lang said.
Researchers at ETH Zurich have developed a world map that uses machine learning to derive vegetation heights from high-resolution satellite images. Courtesy of EcoVision Lab.
With input from optical satellite images and lidar measurements, the algorithm can acquire the filters for textural and spectral patterns on its own. By sliding a 3 × 3 pixel filter mask over the satellite image, the algorithm learns about the image’s brightness patterns. “The trick here is that we stack the image filters,” said Professor Konrad Schindler. “This gives the algorithm contextual information, since each pixel, from the previous convolution layer, already includes information about its neighbors.”
Five neural networks independent of each other were trained, and each returned its own estimate of the height of the tree. “If all the models agree, then the answer is clear based on the training data,” Lang said. “If the models arrive at different answers, it means that there is a greater uncertainty in the estimate.”
The models also incorporate uncertainties that may arise from the input data; for example, when a satellite image is blurry, the uncertainty is greater than when atmospheric conditions are clear.
Once trained, the neural networks only require image data, meaning the map can be updated annually with Sentinal-2 satellite imagery. At the same time, the longer the GEDI mission collects data, the denser the reference data for the Global Canopy Height Map.
Once a neural network has been trained, it can automatically estimate vegetation height from the more than 250,000 images (about 160 TB of data) needed for the global map. According to the researchers, calculating the global vegetation height map would take a single powerful computer three years. “Fortunately, we have access to the ETH Zurich high-performance computing cluster, so we didn’t have to wait three years for the map to be calculated,” Lang said.
The Global Canopy Height Map can provide information on carbon emissions, as tree height is a key indicator of biomass and the amount of carbon stored. “About 95% of biomass in forests is made up of wood, not leaves. Therefore, biomass is strongly correlated with height,” Schindler said. According to the map, only 5% of the global land mass is covered by trees taller than 30m, and only 34% of these tall canopies are within protected areas.
The ETH Zurich model will enable consistent, uncertainty-informed global mapping and support ongoing monitoring to detect change and inform decision-making. It can serve ongoing efforts in forest conservation, and has the potential to foster advances in climate, carbon and biodiversity modelling.
The global map and underlying source code and models are made publicly available to support conservation efforts. For more information, visit N. Lang et al., “A High-Resolution Canopy Height Model of the Earth” (www.doi.org/10.48550/arXiv.2204.08322).