The height of trees provides a lot of information about the structure and condition of deciduous and tropical forests. Scientists have now succeeded in creating the first high-resolution global vegetation height map from satellite images with the help of an artificial neural network. The map could provide crucial clues for the fight against climate change and species extinction.
A United Nations (UN) initiative aims to restore as many damaged ecosystems as possible by 2030 and prevent further destruction. For such projects, however, the actors need precise data sets on the condition of the ecosystems, such as measurements and maps of the plant and tree populations. In order to cope with this, huge amounts of environmental data have to be viewed and interpreted – a task that is almost impossible for humans to perform, but which is well suited for artificial intelligence.
Training with satellite images
For this reason, Nico Lang from the Technical University of Zurich (ETH) and his colleagues have developed a machine learning algorithm that is able to automatically analyze large-scale environmental data. The algorithm is based on a neural network that can derive the height of trees from satellite images. To make this possible, Lang trained the computer with millions of images beforehand, showing it which satellite image represented which tree height. After this training phase, the neural network has connected itself in such a way that it can associate new satellite images with the correct tree height on its own – it has learned the assignment.
The images from the two Copernicus Sentinel-2 satellites of the European Space Agency ESA served as input for training the network. These satellites record every location on earth every five days with a resolution of 10×10 meters per pixel. These are the best quality images currently available to the public. The tree height information, which is also required for network training, was randomly determined in various regions of the world by the NASA Gedi mission. According to Lang, a decisive advantage is that the computer comes into contact with different types of vegetation during the training process – be it spruce forests in Northern Europe or kapok trees in South America.
Information on the condition and changes in forests
Thanks to this training, the researchers were able to use their algorithm to create a world map of tree heights based on data from more than 250,000 satellite images. The resolution of ten by ten meters is particularly noteworthy: users of the map can zoom in on any piece of forest on earth to read the height of the trees. According to Konrad Schindler from ETH Zurich, the possibilities arising from this information are far-reaching: “About 95 percent of the biomass in the forest is in the wood and not in the leaves. Therefore, biomass is strongly related to altitude.”
The tree height therefore also provides information about the amount of carbon that can potentially be stored – and thus about the importance of forests in the climate system. Coupled with the so-called high carbon stock approach, which classifies forests according to their carbon storage and biodiversity, the vegetation height map is an important basis for the preservation and strengthening of ecosystems. However, the data are only partially encouraging in this respect: According to the calculations of the research team, there are only five percent of the landmass with trees that are more than 30 meters high. Of these, only 34 percent are in protected areas.
How is the forest changing?
The map can therefore also be of interest to governments, administrations and associations of environmentalists: “Thanks to Sentinel-2, we could recalculate the vegetation heights every five days and thus have a monitoring system to observe the deforestation,” says Nico Lang. The damage caused by forest fires could also be better estimated with the updated tree height map. The map also offers an interesting insight into the structures of the different forests: “We have already been able to discover exciting patterns,” reports Schindler. “In the Rocky Mountains, for example, forestry is carried out in fixed squares and the rainforest also forms interesting structures that cannot be accidental”. Now it is possible for ecologists to interpret the recorded patterns and data globally.
Source: Swiss Federal Institute of Technology Zurich (ETH Zurich); Interactive map Global Canopy Height 2020