STTR Phase I: Registration of Below-Canopy, Above-Canopy, and Satellite Sensor Streams for Forest Inventories

Gaia Ai, Inc, Somerville MA

Investigators

Abstract

The broader/commercial impact of this Small Business Technology Transfer (STTR) Phase I project is to increase the volume and improve the accuracy of data on the world’s forests. Presently, when collecting data on forests, surveyors must choose between slow, laborious methods, or quick but inaccurate ones. This project uses recent advances in sensors and machine learning to greatly improve data collection speed without sacrificing accuracy. The resulting rich datasets enable the construction of true “digital twins” of forests and open the door for higher fidelity modeling of forest growth trajectories. This information is useful both for timber firms seeking to maximize the potential of their assets and environmental groups projecting how changes today could impact a forest’s performance as a carbon-sink over the long term. The impacts on United States citizens are widespread. Here are two examples: improved efficiency in the timber industry brings down the cost and improves the quality of raw materials and turning forests into denser carbon sinks helps meet climate change goals. The availability of such broad and deep data on forests could also drive a boom in research and understanding about the more complex and nuanced relationships that drive forest health and productivity, launching entirely new sub-industries around forestry. The key technological innovations explored in this STTR Phase I project are in constructing the most high-fidelity forest model (digital twin) by combining disparate information sources, each with their own advantages and disadvantages. Light detecting and ranging (LiDAR) and camera sensors on backpacks provide high-quality inventory metrics nearly 1000 times faster than manual measurements, but still require someone in the forest to wear the backpack. Satellite imagery scales almost instantly to entire forests and also through time with historical data but is limited by the top-down nature of satellites and the resolution they offer, especially when historical and free data sources are considered. Drone-based imagery sits in-between, with advantages and disadvantages of both. In practice, combining information sources that measure in such different ways can be very difficult. In this project, the team explores how to express LiDAR-based metrics to best associate them with top-down imagery from satellites and drones. From these associations, one can then build powerful machine learning models and specialize them to individual forests. This ability may enable the company to provide forest inventories and forest management recommendations to timber companies at any scale: with satellite imagery only or with a combination of backpack-LiDAR and satellite for the highest accuracy over the entire forest. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.