Site icon Conservation news

Chocolate giant funds high resolution carbon map to protect forests

Satellite image of oil palm, agriculture, and forest in East Kalimantan, Indonesia. Image credit: Maxar Technologies.

Satellite image of oil palm, agriculture, and forest in East Kalimantan, Indonesia. Image credit: Maxar Technologies.

  • A new carbon map based on high resolution satellite imagery that will help companies avoid deforestation in their supplies chains is expected to be published by the end of 2021.
  • The map builds on the High Carbon Stock (HCS) approach, a methodology that differentiates between six categories of vegetation cover, from native forest areas that conservationists say should be protected to degraded lands low in carbon and biodiversity that may be appropriate for conversion to other uses.
  • The map was developed by the EcoVision Lab at ETH Zurich and financed by Barry Callebaut, the world’s largest chocolate maker.
  • The initial release of the map covers Indonesia, the Philippines, and Malaysia.

A project by the EcoVision Lab at ETH Zurich expects to have every quarter of the tropics mapped and classified by carbon stock by the end of the year, providing a complete picture of which lands store the most carbon and allowing companies to track deforestation in their supply chains. The map is being constructed by passing publicly available satellite imagery through a deep machine learning algorithm in order to inventory the world’s landscapes at a resolution of 10 by 10 meters.

Developed by EcoVision Lab and financed by Barry Callebaut, the world’s largest chocolate maker, the project builds on the High Carbon Stock (HCS) approach, a methodology that differentiates between six categories of green cover, from native forest areas that conservationists say should be protected to degraded lands low in carbon and biodiversity that may be appropriate for conversion to other uses. Backed by WWF and Greenpeace, the HCS approach has become a widely recognized standard and was adopted by the Roundtable on Sustainable Palm Oil (RSPO) in 2018.

The initial release of the EcoVision Lab project is a complete map of Indonesia, the Philippines, and Malaysia, with each square decameter (10×10 meters) of those countries classified according to the High Carbon Stock methodology. They expect to release a tropics-spanning map by the end of 2021. After that, the next step will be to determine the interval at which new maps are produced in order to be able to see changes in landscapes over time.

Indicative High Carbon Stock (HCS) map of Indonesia, the Philippines, and Malaysia, courtesy of the EcoVision Lab at ETH Zurich.
Indicative High Carbon Stock (HCS) map of Indonesia, the Philippines, and Malaysia, courtesy of the EcoVision Lab at ETH Zurich.

Although it is technically complete, it will need to be complemented with information pertaining to indigenous land rights and local community initiatives in order to fully represent HCS categories. The map will have to be adjusted to the vegetation patterns on each continent, but since it is agnostic to raw materials or supply chains, it could potentially be used by any company to monitor its sourcing—or anyone seeking to hold companies accountable. At a resolution of 10 by 10 meters, the map is appropriate for tracking tree cover changes even at the smallholder level. A one-hectare farm would be represented by a 100-pixel array.

EcoVision Lab was founded in 2017 by Dr. Jan Dirk Wegner, a specialist in geodesy, the field that deals with the geometry of the planet, and geoinformatics, which applies information science to geography and earth sciences. ETH Zurich, the university that hosts the lab, is a mecca for technically gifted students and researchers. Its alumni, which include Albert Einstein, have racked up more than a dozen Nobel Prizes, in addition to Fields Medals, a Pritzker Prize, and a Turing Award.

At the lab, Dr. Wegner mentors a growing cadre of PhD students and postdoctoral researchers that are pushing the envelopes of engineering, robotics, and artificial intelligence. There are no ecologists on the team. “We have a lot of colleagues who are active in that area, of course,” Wegner said in an interview with Reset. “There are many good researchers, but they often don’t have as much technical know-how, especially when it comes to the field of machine learning.”

That’s where EcoVision comes in. Operating at the intersection of machine learning, computer vision, and remote sensing, the team at EcoVision invents original, data-​driven methods to automate the analysis of environmental data at very large scales, solving ecological problems that biologists cannot.

“EcoVision are hardcore machine learning guys, improving their code day in and day out, and they have the computers for that,” said Oliver von Hagen, Barry Callebaut’s global director of ingredient sustainability. The mapping project runs on ETH Zurich’s supercomputer, equipped with a potent cluster of graphics cards that can crunch massive sets of satellite images.

Vegetation Stratification. Module 4 Forest and vegetation stratification, HCS Approach Toolkit v2.0 May 2017. © 2017 High Carbon Stock Approach Steering Group.
Vegetation Stratification. Module 4 Forest and vegetation stratification, HCS Approach Toolkit v2.0 May 2017. © 2017 High Carbon Stock Approach Steering Group.

Barry Callebaut conceived of the project after running into difficulties with the implementation of the High Carbon Stock approach. “We told our suppliers, this is what we expect from you, to protect these areas, and develop only areas indicated for that purpose by HCS,” von Hagen said. “Sure enough, they got back to us and said, that’s fair, but we don’t have the capacity to develop these maps at a large scale.”

“HCS takes biomass, vegetation height, and land use as the three main components to develop categories,” von Hagen told Mongabay. “For instance, biomass data is very difficult to find on a large scale. You typically need guys running around on the ground measuring tree diameter at breast height.”

The European Space Agency’s Sentinel-2 satellite provides the world with high-resolution images of every part of the Earth’s surface every three to five days. But to interpret those photographs in terms of forest type and ergo carbon storage, EcoVision’s algorithm is trained with data from NASA’s GEDI lidar (a space-based laser scanner), run from the International Space Station. Although emitted at sparser intervals, lidar gives rich information on the volume of the vegetation. The machine intelligence thus learns from the 3D lidar data how to infer forest height and structure from the optical Sentinel-2 images.

Copernicus Sentinel-2 image of North Kalimantan, Borneo, Indonesia, 2021, processed by the European Space Agency.
Copernicus Sentinel-2 image of North Kalimantan, Borneo, Indonesia, 2021, processed by the European Space Agency.

By driving down the cost of using the High Carbon Stock approach, reducing the need for in-house technical expertise, and providing a reliable and neutral standard, the machine learning approach being invented at EcoVision will allow HCS, proposed since 2010, to be implemented at scale.

“You see a mushrooming of companies that offer these kinds of services. They aim to monitor deforestation in your supply chains, but they each use different methods,” von Hagen said. “We wanted to make sure that there is a public tool based on a credible and transparent methodology with no business model behind it.”

“It’s a step towards democratizing this HCS approach,” said Nico Lang, a fourth-year PhD student who is running the project at EcoVision Lab. “So far it’s been very expensive to be certified by HCS. Smallholders cannot pay a company to produce a HCS map for them. But if there is a map available for free and with very high resolution, it’s a step towards democratizing that.”

Forest and vegetation stratification, HCS Approach Toolkit v2.0 May 2017. © 2017 High Carbon Stock Approach Steering Group.
Forest and vegetation stratification, HCS Approach Toolkit v2.0 May 2017. © 2017 High Carbon Stock Approach Steering Group.

Despite these laudable intentions, unless there is massive buy-in the project could wind up being just one more project in a cluttered field, adding ambiguity to the already complex task of monitoring supply chains for deforestation.

“The initiative is right, but it adds to the proliferation,” Samuel Mawutor told Mongabay. Mawutor is a senior advisor for Mighty Earth’s campaigns in Africa, where Barry Callebaut sources most of its cocoa. “What we’ve always asked companies is, you need to understand your supply chain better so you can manage risks. So going solo is a good effort from Barry Callebaut, but it doesn’t go with the spirit of collective effort.”

If the project is to become the global reference it promises to be, factors such as the cachet of ETH Zurich and the absence of a profit motive will likely prove crucial.

“Deforestation information needs to be public, not privately held,” von Hagen said. “It’s important that future funding come from a neutral place as well.”

Cacao pod in Sulawesi, Indonesia. Photo credit: Rhett A. Butler
Cacao pod in Sulawesi, Indonesia. Photo credit: Rhett A. Butler

For Barry Callebaut, the project developed at EcoVision represents their contribution to the global information commons. Having financed the research for four years, they are actively searching for a long-term home for the map and its generative technology, a neutral and respected institution that will take on the costs of running the algorithm on fresh satellite imagery at the established intervals and make the archive of HCS maps available to the public free of charge.

“We’ve always thought of our chocolate as a vehicle for creating a movement towards sustainability,” von Hagen told Mongabay. “For that, we need our suppliers to help us, we need our customers to support us. So it was logical to invest in something that our suppliers, our customers, and even our competitors can use. There’s no gain if we stop deforestation in our supply chain but our competitors do not.”

Exit mobile version