|
Listen to this story:
|
A new study by Meta and World Resources Institute (WRI) utilized artificial intelligence (AI) to create a global map of tree canopy height at a resolution of one meter. This allows for the detection of individual trees across the globe. In an effort to promote open-source forest monitoring, all data and AI models associated with canopy height are freely available for public use.
The canopy height map was derived from high-resolution satellite imagery analyzed by AI models. Both the data and the AI models are available for commercial use.
Meta set a goal of achieving net zero emissions by 2030 through a combination of reducing corporate emissions and utilizing carbon removal strategies. Forest-based carbon removal is a key component of Meta’s strategy, and the ability to accurately measure and monitor carbon sequestration is essential.
Nature-based carbon removal, including forest restoration, is critical for achieving the emissions reductions outlined in the Paris Agreement. Managed forests are a significant source of carbon credits, and improved monitoring of forest-based carbon credits is necessary to manage forests effectively for climate change mitigation. High-resolution, large-scale mapping with AI can improve data accuracy and enable monitoring across various scales.
Breakthroughs in AI and foundation models are transforming how we interact with the world. Satellite imagery for forest mapping has significantly improved in terms of scale, resolution, and refresh rate. While deforestation events can be monitored with lower-resolution imagery due to the large areas involved, monitoring afforestation and reforestation projects is more challenging as it requires individual tree-level data over large areas.
Meta and WRI believe that providing open access to AI can be a powerful tool for unlocking financial resources for climate change mitigation and adaptation efforts, while also increasing transparency. For the first time, the computational power and AI models required for global processing of high-resolution maps are available for public use. This data and model can be accessed on AWS, Google Earth Engine, and Github.
The continuously updated map provides a global baseline of tree canopy height, including individual trees and open-canopy forests. This data can be used for forest inventory and accounting. The analysis of satellite imagery from 2009 to 2020 revealed that more than one-third of the Earth’s landmass has a canopy height exceeding one meter, and 35 million square kilometers have a canopy height exceeding five meters. This data can be used as a reference point for supplementing field measurements of carbon in carbon credit monitoring. The publicly available model can also be used to detect changes in canopy height over time as new imagery becomes available.
To create the maps, a state-of-the-art model called DiNOv2, developed by Meta AI Research, was used. The model was trained on a massive dataset of satellite images encompassing over a trillion pixels. This allows for canopy height prediction with an average error of 2.8 meters, enabling the detection and measurement of individual trees. The model can also be applied to aerial and drone imagery.
Self-Supervised Learning (SSL) was employed to train the DiNOv2 model on unlabeled satellite imagery. This approach allows the model to extract general image features without the need for extensive manual labeling. The model can then be used to predict canopy height.
The released global earth foundational model has the potential for various applications beyond just canopy height estimation.
The canopy height map can be a valuable tool for estimating above ground biomass and establishing baselines for conservation and restoration projects. For instance, WRI is utilizing the data to connect high-resolution satellite data with field-based forest inventory data for the purpose of monitoring forest restoration efforts in Africa.
Related Article: Meta and Adapture Renewables Sign Agreements for 330 MW of Solar in Illinois and Arkansas
Forests are essential for the global carbon market and for preserving biodiversity in the face of climate change. Accurate data is paramount for monitoring forest conservation and restoration efforts. Regularly updated, high-resolution canopy height maps can significantly improve forest-based carbon credits and natural climate solutions. The advancements in satellite imagery, AI, and computational capabilities have enabled the creation of global, annual canopy height maps at the resolution of individual trees, which will be instrumental in preserving Earth’s forests. To maximize the potential of this work, the data and models are being made openly available.