Tackling deforestation and forest degradation with AI and remote sensing

(Image by Mario from Pixabay)

The challenge of deforestation is a big and serious one. The world has lost a third of its forests over the last 10,000 years - an area twice the size of the US - with more than half of them disappearing over the last century. 

But another rather less well-known issue is forest degradation. This is where the canopy thins due to a reduction in tree density in a given area. Unlike deforestation, it does not come about due to changes in land use, for example, to agriculture or mining. It is instead the result of less permanent events, such as wildfires, shifting agriculture or logging for timber.

One country that is no stranger to either problem is Australia. Barney Swan, Co-Founder of nature regeneration charity ClimateForce and son of polar explorer Robert Swan, explains:

The past 200 years have resulted in the degradation or permanent conversion of half of the country’s forests, impacting millions of hectares of endemic ecosystems. This has placed numerous Australian ecosystems at risk of collapse by 2060, threatening biodiversity and the provisions it supports, including clean air, healthy soil, water, and climate regulation. 

The main drivers include demand for resources, population growth, poorly managed livestock, and hard-to-govern environmental policies. Tropical rainforests cover only 6% of the earth’s surface but are home to an estimated 50% of all terrestrial plant and animal species, highlighting why preserving locations like the Wet Tropics World Heritage Management area [in Queensland, where the charity is based] is critical.

Creating an environmental restoration blueprint

ClimateForce, which was set up in 2018, is currently working on developing a proof-of-concept methodology and testing ground for tropical regeneration techniques in part of the Daintree Rainforest World Heritage Site. Another key aim is to support the economic resilience of the local indigenous communities that are its custodians. 

The project started in 2021 to remove invasive plant species, restore biodiversity in what was a 527-acre razed section of the Rainforest and optimize its management. Key activities have included cleaning up waste and planting 12,000 trees of 75 different species. As Swan says:

Given the cultural significance of this forest for First Nation’s people plus the staggering complexity of species within this space, we’re working in the perfect location to build out a blueprint and develop exportable models for land management…The vision encompasses offering frameworks, methodologies and systems that will support farmers, First Nations people, and rural communities.

Doing so is important, Swan explains, because:

There’s a need for more robust business models to accelerate conservation, promote future green economies, and diversify monoculture land use.

Moreover, he says, technology is “pivotal” to the organization achieving its goals:

It enables us to better track biodiversity outcomes as we test various regeneration techniques, innovate new reforestation methods, and conduct advanced impact reporting. We’re using interactive GIS [geographic information system] mapping, tracking the health of each tree, and listening to the forest to capture vital data that supports the flourishing biodiversity market. Technology also supports us in modelling variables within low-cost reforestation models and builds transparency in our reporting.

To transform the data it gathers into actionable insights, ClimateForce is using NTT Data’s AI-based Smart Management Platform and Analytics technology. Activities include assessing the value of various organic reforestation techniques and predicting and modelling regenerative pathways for different landscapes.

Once the modelling process is complete, the aim, Swan says, is:

To replicate our environmental restoration blueprint in other tropical rainforests globally, leveraging data and local partnerships to create scalable, commercial reforestation that encourages big industries like mining, cattle ranching and big energy to transition, instead of demonizing them in terms of sustainability. Our goal is change and becoming one of many impetuses to move industries towards sustainable practices.

Tackling forest degradation in Vietnam

Another organization that is trying to address the challenge of forest degradation is international conservation charity, Fauna & Flora. It has operated in the Mu Cang Chai Mountains at the south-eastern tip of the Himalayas for the last 20 years. The area is home to the last remaining western black crested gibbons in Vietnam.

A key issue in this remote spot is that local communities tend to be subsistence farmers with limited access to resources. As a result, they clear shrubs and small trees under the forest canopy to grow cardamon, which is an important cash crop for those living at high altitudes in the country. Emma Scott, the charity’s Senior Technical Specialist for Agriculture, explains:

Cardamon likes shade but not too much, so the threat is around forest thinning rather than deforestation. Also, you process it by drying so local people cut down trees for fuel, which creates a dual threat. These communities are living in poverty and have limited access to alternatives, so this is a way for them to make money and feed their families. But it’s causing a threat to the forest at the same time.

To gain a better understanding of the extent of this threat, Flora & Fauna got together with AIforgoodAsia. An international NGO, it brings together third-party organizations, experts and governments together to conduct research and implement AI systems that help solve community problems in the region. 

The aim in this instance was to track changes in the forest and measure the extent of habitat degradation. As Scott points out:

We knew cardamon cultivation was happening, but in a mountainous landscape that’s very difficult to access, it was hugely difficult to know how expansive it was. The forest still retains a high level of cover at the top so it’s hard to see what’s happening using standard satellite imagery. The issue is you can’t tackle the problem and offer possible alternatives unless you know where it is and what communities are involved. So, we’d been struggling to know what to do next.

Using algorithms to detect what the human eye can’t

The partnership with AIforgoodAsia involved developing a proof-of-concept (PoC) computer vision-based system based on Crayon’s remote sensing and machine learning technology to see if it was possible to detect cardamon cultivation in satellite imagery. The key objectives were to evaluate both the feasibility and cost-effectiveness of using such technology. Harriet Branson, Technical Specialist for Remote Sensing and GIS at Flora & Fauna, explains:

It’s very difficult for the human eye to identify cardamon growing beneath the canopy, so training the algorithm to do it is a big goal. But the data needs to be of good quality and there’ll have to be a lot of trials before it’s likely to come into practice.

When undertaking phase one testing, it soon became clear that cardamon production was not visible in low or medium-resolution satellite images from open data access providers, such as NASA’s Landsat. But the same was not true of the high-resolution commercial imagery purchased from EarthImages. When combined with ground reference data, it proved effective in helping to train the Crayon systems’ algorithms to spot cardamon cultivation. 

As a result, Flora & Fauna is currently applying for funding to support phase two of its initiative. This will involve collecting further training samples both to improve accuracy and cover a wider area of the forest. The ultimate aim here, Scott says, is:

To scale up and get an idea of what’s happening in the wider landscape, so we can make informed decisions on how to work most effectively with local communities based on real data.

My take

Both the ClimateForce and Flora & Fauna initiatives signal a growing use of advanced technologies in conservation activity. As Branson points out:

Technology, such as machine learning and AI, may be fairly new in conservation terms, but it’s starting to become quite big, particularly in mapping habitats and forest loss. So, identifying tricky landscapes like we’ve done is a novel next step, and it’s only likely to get bigger.