Forests are crucial for sustaining Earth’s biodiversity and are key in absorbing atmospheric carbon dioxide, offering vital ecosystem services as defenses against climate and biodiversity crises. However, a significant knowledge gap exists due to most global forest data coming from low to medium resolution satellite imagery, which inadequately covers dynamic and dispersed forest systems such as agroforestry, drylands, and alpine forests.
Recognizing this deficiency, Meta and the World Resources Institute have launched a groundbreaking initiative featuring a global map of tree canopy height with a 1-meter resolution. This advancement in high-resolution forest mapping enables the detection of individual trees globally, greatly enhancing forest monitoring capabilities. The initiative supports open-source forest monitoring by making all canopy height data and the accompanying AI models publicly available, aiming to democratize access to this critical environmental data.
High-resolution forest mapping is not only a technological breakthrough but also a strategic element in Meta’s commitment to achieving net zero emissions by 2030. This commitment involves reducing corporate emissions and addressing any residual emissions through carbon removal strategies, including both natural and technological approaches. Forest-based carbon removal plays a pivotal role in these strategies, leveraging the enhanced monitoring, reporting, and verification capabilities provided by high-resolution forest mapping.
Such mapping is crucial for managing forest lands at scales necessary to combat climate change effectively. It aids in the rigorous monitoring and verification of forest-based carbon credits globally, a process underscored by the Paris Climate Agreement and various environmental stakeholders. The integration of AI with remote sensing not only bridges the data gap between reported and actual land use emissions but also facilitates comprehensive monitoring at multiple scales, from international to local.
Recent advancements in AI and foundational models have transformed remote sensing, particularly in forest conservation. The ability to map forests with high-resolution forest mapping and detect minute changes in canopy height enhances our ability to monitor deforestation and afforestation efforts accurately. For instance, while deforestation events, which typically involve the removal of larger trees, can be monitored with lower resolution imagery, the growth of young or sparse trees requires much finer resolution, achievable through the technologies developed by Meta and the World Resources Institute.
The released global dataset, which includes detailed canopy height maps covering the full global landmass, establishes a new benchmark for environmental monitoring. Analyzing satellite imagery from 2009 to 2020, this dataset provides a high-detail baseline that supports forest inventory accounting globally. More than one-third of Earth’s land surface, representing diverse forest types and conditions, has been mapped with this technology.
Moreover, the use of Self Supervised Learning (SSL) and the DiNOv2 model, trained on millions of unlabeled satellite images, exemplifies how AI can streamline the extraction of environmental data without extensive manual labeling. This model, capable of predicting canopy height with high accuracy, represents a significant leap forward in environmental monitoring and the application of AI in earth observation.
Overall, the capabilities of high-resolution forest mapping not only bolster forest conservation efforts but also enhance transparency and accountability in the carbon market and other restoration initiatives. As such, these technological advancements are indispensable tools for safeguarding the planet’s biodiversity and mitigating the effects of climate change.
www.sustainability.fb.com/blog/2024/04/22/using-artificial-intelligence-to-map-the-earths-forests