Monitoring deforestation across millions of hectares of Amazonian jungle has always been an onerous ask. When illegal logging first became an issue, the authorities relied on word-of-mouth and reports from indigenous communities, before buzzing the jungle in small planes to find the deforested areas.
In the 1980s the first satellite images became available, with resolution steadily improving over the decades. Still, these images only showed deforestation after it had occurred. Now, breakthroughs in artificial intelligence (AI) mean it’s possible to predict areas most at risk of illegal deforestation, allowing the authorities to take preventative action.
Carlos Souza, a senior researcher at Imazon, the NGO behind AI platform PrevisIA, says AI is central to developing a proactive rather than reactive approach to deforestation. “When we detect deforestation with satellite data, the only thing we can do is to send the information to the authorities, and they can enforce the law, but the forest is gone,” he explains. “(But) if you can show them the areas that are likely to be deforested, then you can prevent it from happening… it’s a big shift in the process of combatting deforestation.”
Vivian Ribeiro, senior data scientist at the Stockholm Environment Institute (SEI), agrees with Souza that AI could help turn the tide. Riberio leads the spatial intelligence team at Trase, a partnership between SEI and the NGO Global Canopy. Trase is a data-driven, supply chain mapping initiative, which is supporting a new Supply Chain Data Partnership focused on using AI to plot the facilities, such as silos, refineries and mills, that are often linked to deforestation.
As part of a trial in Brazil, images and satellite maps of existing facilities are turned into data sets, and fed into the model, which has been trained to look for similar patterns of development using machine learning. With this information, says Riberio, it’s now possible to process a huge amount of data and pinpoint these facilities with much greater accuracy.
Central to this is the importance of monitoring roads, which Souza describes as “arteries of destruction” for a forest. Traditionally a laborious job, it took Imazon two years to manually trace road lines over the top of digital images in order to plot all the routes across the Brazilian Amazon alone.
Now algorithms can be trained to detect objects like roads on satellite images. “If we keep track of the roads, we could potentially anticipate what will be the future deforested areas, because 95% of deforestation is carried out within 5km of a road,” he says.
Once the roads have been detected, models are run based on spatial regression using a range of variables, from historical deforestation levels, to the GDP and population densities of a municipality, and its topography. This, he adds, “allows them to narrow down the areas with a high probability of deforestation”.
So far the system has proved highly accurate, with 75% of the areas pinpointed falling within 4km of the areas where deforestation was shown to be taking place.
The effectiveness of this approach relies on strong partnerships, Souza continues, and Imazon works with the public prosecutors in four Brazilian states. But the system is also proving attractive to banks, investors and trading companies, he says, which can “use the information to make better decisions, both from an economic and an environmental point of view”.
Meanwhile, the Rainforest Alliance certification scheme has developed proprietary forest layer remote sensing technology to help counter deforestation in high-risk countries where the commodities it certifies are grown. It uses AI to provide companies with an advanced assessment of deforestation risk at the farm level, explains Laybelin Dijkers, senior GIS manager at Rainforest Alliance.
Open-source data doesn’t capture the nuances of the forest, she explains, and the alliance’s system offers a sharper distinction between plantations, such as rubber and cocoa, and true forested areas. The high accuracy of the methodology, when combined with land-cover changes that are available through open data sources, helps pinpoint areas that are at greatest deforestation risk, and is now an important part of their certification process.
But it’s not just data-crunching algorithms and satellite images that are helping in the fight against deforestation.
Polygon mapping is another way of tracing a farm’s boundaries, and helping to ensure that companies know exactly where the commodities they are buying have been produced.
Ofi (formerly known as Olam food ingredients) is one of several agrifood businesses using GPS positioning, in this case across its coffee and cocoa supply chain. The system maps the landscape of a farm as well as tracing its perimeter using GPS polygons. This can ensure that the volume of coffee or cocoa purchased doesn’t exceed the farm’s capacity, and that land hasn’t been illegally cleared to grow more crops.
So far the system has proved highly accurate, with 75% of the areas pinpointed falling within 4km of the areas where deforestation was shown to be taking place.
Polygon mapping is another way of tracing a farm’s boundaries, and helping to ensure that companies know exactly where the commodities they are buying have been produced.