Deforestation solutions, anyone? Machine learning can help!
AI can tell us where to rebuild the green lungs of our planet.
Forests cover 30% of the Earth’s land. After the oceans, they’re the world’s largest reserve of carbon. The entire life on Earth depends on forests for its survival. Not only do the forests support diverse ecosystems, reduce carbon dioxide in our air, prevent soil erosion, but also absorb other greenhouse gasses that human activities emit.
We need forests for what they do for our climate and beyond. Yet, both human-driven and natural processes increasingly endanger the life of trees.
- Between 1990 and 2016 we’ve lost 1.3 million square kilometers of forest globally.
- Each year we lose an area of forest the size of the UK because of deforestation. The trend continues despite global pledges to halt it.
- Agriculture remains the most significant driver of global deforestation, e.g. 13 million hectare per year in South America and Africa and Southeast Asia is converted from forest to agricultural land.
- Additionally, e.g. in Europe, the bulk of abandoned agricultural land (4.8 million ha gross) is likely to remain unused within 2015–2030 because of negligible re-cultivation.
Both prevention of forest loss and promotion of the restoration of forest help significantly with climate change mitigation. Organisations and institutions globally are trying to find long-term, systemic solutions to this multidimensional and complex challenge.
Armed with our trained machine learning neural network and satellite images, we looked under the magnifying glass for a single, but important factor in this battle:
How to identify areas with the potential to contribute to forest sustainability easily and cost-effectively?
What is land abandonment and why is it important?
The implications of how we manage agricultural land is critical to our ecosystem and biodiversity. Despite many thinking otherwise, agriculture today still remains the most significant driver of global deforestation. There is an urgent need to better link agriculture and forestry.
If we could map and manage areas where potential forest could grow spontaneously, it would greatly contribute to natural forest regeneration.
A phenomena called “land abandonment” is essential here. Why?
What is land abandonment?
It is a non-deliberate stop in using a given area for agriculture. Abandoned fields, meadows, and pastures are subject to spontaneous plant regeneration and succession. These are places where biodiversity can flourish, and new trees can grow.
The detection and management of unused land can help us maintain and enhance the values of forests for the benefit of present and future generations. Even though we wouldn’t think that agricultural land use is a modern day challenge, there is a growing need for research on the topic.
Around 0.9 billion hectares of land worldwide would be suitable for reforestation, but such areas need to be identified first. The challenge is how to identify spots where new forest can grow in a time-efficient and cost-effective way. Usually it is done in a form of field studies, which involve riding around the countryside, or poring over aerial photographs. It is both expensive and time-consuming. But what if we would use machine learning?
Let’s tap into the predictive power of AI…
With advances in machine learning we can now tap into the predictive power of AI to make better data-driven models of environmental processes.
The wide availability of multispectral satellite imagery through projects such as Landsat and Sentinel, combined with the introduction of deep learning in general and Convolutional Neural Networks (CNNs) in particular, has allowed for the rapid and effective analysis of multiple classes of problems related to land coverage.
One such problem is land abandonment detection. It is also an example of a nature-based solution to climate change. If we let nature do its job, the new trees could grow on abandoned areas human is not exploiting anymore.
That’s why Maciej Adamiak and Mikołaj Koziarkiewicz teamed up with the scientists from the Geographical Sciences Faculty from University of Łódź to develop a Machine Learning model that can do land abandonment detection automatically, based on relatively cheap satellite images.
Results are here, in a peer-reviewed paper in MDPI Land Journal.
What’s next?
Land abandonment detection, where we can track down the important vectors to forest regeneration, present huge climate mitigation opportunities. However the global patterns, drivers, and implications of land abandonment pose a challenge for scientists to map, predict, and manage this dynamic process.
The principal benefits of the proposed approach are both time-efficiency and cost-effectiveness. There are several processes of improvement to implement though. However, with training new neural network models for different data on the basis of existing ones, the model could prove advantageous when classifying additional pieces of land.
We hope this forms a bedrock that could help guide policy-making and planning processes that focus on forest sustainability further.
