Supercomputing Study Breakthrough Tree Mapping, Carbon Research – ScienceDaily

Using powerful supercomputers and machine learning algorithms, the team mapped the canopy diameter – the width of a tree when viewed from above – of more than 1.8 billion trees over an area of ​​more than 500,000 square miles, or 1,300,000 square kilometers. The team mapped how the diameter, cover and density of the tree canopy varied depending on rainfall and land use.

Mapping non-forest trees at this level of detail would take months or years using traditional analytical methods, compared to a few weeks for this study. The use of high-resolution imagery and powerful artificial intelligence represents a technological breakthrough for mapping and measuring these trees. This study is expected to be the first in a series of works aimed at mapping not only non-forest trees over a wide area, but also How much carbon they store – important information in understanding the Earth’s carbon cycle and how it changes over time.

Measure carbon in trees

Carbon is one of the main building blocks for all life on earth, and this element circulates through the carbon cycle between land, atmosphere and oceans. Some natural processes and human activities release carbon into the atmosphere, while other processes pull it out of the atmosphere and store it on land or in the ocean. Trees and other green plants are carbon “sinks,” which means that they use carbon to grow and store it in their trunks, branches, leaves and roots outside of the atmosphere. Human activities such as burning trees and fossil fuels or cutting down forest areas release carbon into the atmosphere as carbon dioxide, and rising concentrations of atmospheric carbon dioxide are a major contributor to climate change.

Conservation experts working to curb climate change and other environmental threats have targeted deforestation for years. However, those efforts don’t always include trees that grow outside of the forests, said Compton Tucker, senior biosphere scientist in NASA Goddard’s division of earth sciences. Not only could these trees be significant carbon sinks, but they could also contribute to the ecosystems and economies of nearby human, animal and plant populations. However, many current methods of studying the carbon content of trees involve only forests, not trees that grow individually or in small groups.

Tucker and his NASA colleagues, along with an international team, used DigitalGlobe commercial satellite imagery that was high-resolution enough to identify individual trees and measure their canopy size. The images come from the commercial satellites QuickBird-2, GeoEye-1, WorldView-2 and WorldView-3. The team focused on the arid areas – areas that receive less rainfall than what evaporates from plants each year – including the arid southern side of the Sahara, which extends through the semi-arid Sahel and into the humid subtropics of West Africa. By examining a wide variety of landscapes, from a few trees to nearly forested conditions, the team trained its computational algorithms to detect trees in different types of terrain, from deserts in the north to tree savannas in the south.

Learning on the job

The team ran a powerful computational algorithm known as a fully folded neural network (“deep learning”) on the University of Illinois’ Blue Waters, one of the fastest supercomputers in the world. The team trained the model by manually marking nearly 90,000 individual trees in different terrains and then “learning” which shapes and shadows indicated the presence of trees.

It took more than a year to encode the training data, said Martin Brandt, assistant professor of geography at the University of Copenhagen and lead author of the study. Brandt tagged all 89,899 trees himself and helped oversee the training and operation of the model. Ankit Kariryaa from the University of Bremen led the development of deep learning computer processing.

“In one kilometer of terrain, let’s say it’s a desert, there are often no trees, but the program wants to find a tree,” said Brandt. “It will find a stone and think that it is a tree. Further south it will find houses that look like trees. It sounds simple, you’d think – there is a tree, why shouldn’t the model know it’s a tree? But the challenges come with this level of detail. The more detailed it is, the more challenges there are. ”

Accurate tree counts in this area provide vital information for researchers, policy makers, and conservationists. In addition, measuring the variation in tree size and density depending on rainfall – with wetter and more populous regions supporting more and larger trees – provides vital data for conservation efforts.

“There are important ecological processes, not just inside the forests, but outside them as well,” said Jesse Meyer, programmer at NASA Goddard who oversaw Blue Waters processing. “For conservation, restoration, climate change, and other purposes, data like this is very important to establish a baseline. In a year or two or ten years the study could be repeated with new data and compared with today’s data to determine whether or not efforts to revitalize and reduce deforestation are effective. This has practical implications. ”

After the program measured the program’s accuracy by comparing it with manually coded data and field data from the region, it ran the program over the entire study area. The neural network identified more than 1.8 billion trees – surprising numbers for a region that is often thought to have little support for vegetation, Meyer and Tucker said.

“Future work in the series will build on the counting of trees, expand the areas studied, and find ways to calculate their carbon content,” Tucker said. NASA missions such as the Global Ecosystem Dynamics Investigation Mission (GEDI) and ICESat-2 or the Ice, Cloud and Land Elevation Satellite-2 are already collecting data that is used to measure the height and biomass of forests. The combination of these data sources with the power of artificial intelligence could open up new research opportunities in the future.

“Our goal is to see how much carbon is in isolated trees in the vast arid and semi-arid parts of the world,” Tucker said. “Then we need to understand the mechanism that drives carbon storage in arid and semi-arid areas. Perhaps this information can be used to store more carbon in vegetation by removing more carbon dioxide from the atmosphere. ”

“From the perspective of the carbon cycle, these arid areas are not well mapped in terms of the density of trees and carbon,” Brandt said. “It’s a white area on cards. These dry areas are basically hidden. This is because normal satellites just don’t see the trees – they see a forest, but when the tree is isolated they can’t see it. Now we are on our way to fill those blank spaces on the cards. And that’s pretty exciting. ”