Friday 31 January 2025
Forests are crucial for our planet’s health, but accurately measuring their height is a challenge. Traditional methods like LiDAR (Light Detection and Ranging) devices require expensive equipment and can’t be used on a global scale. Synthetic Aperture Radar (SAR) technology offers a scalable solution, but processing the data is complex.
Researchers have developed a new approach that uses deep learning to estimate forest canopy height from SAR data. They’ve created a model that takes into account the radar’s ability to penetrate through the tree canopy and measure the strength of the signals reflected back.
The team tested their method using data from an aerial survey of a forest in Germany. They compared the performance of different numbers of input SAR images, finding that increasing the number of images improved accuracy. The best-performing model used seven images and achieved a test mean absolute error (MAE) of just 4.17 meters.
The researchers also explored the impact of polarisation channels on the results. They found that using vertical transmit and receive (VV) polarisation channel produced the best results, with an MAE of 3.33 meters. This is likely because VV is more sensitive to vertical structures like tree trunks.
To further improve accuracy, the team applied a height filter to remove areas under 5 meters in height, which are prone to errors due to ground-level scattering. This reduced the MAE by 27%.
The study’s findings have significant implications for monitoring forest health and carbon sequestration. By using SAR data and deep learning, scientists can create accurate maps of forest canopy height on a global scale. This will help them track changes in forest biomass over time, enabling more effective conservation efforts.
In the future, the researchers plan to integrate additional geometric parameters from the SAR data to improve their model’s performance. They also hope to use hyperspectral data to gain even more insights into the forest ecosystem.
The development of this new approach is an important step towards creating a global monitoring system for forests. By combining cutting-edge technology with machine learning, scientists can better understand and protect our planet’s vital ecosystems.
Cite this article: “Accurate Forest Canopy Height Estimation Using Synthetic Aperture Radar and Deep Learning”, The Science Archive, 2025.
Forests, Sar, Deep Learning, Forest Canopy Height, Lidar, Radar, Machine Learning, Carbon Sequestration, Conservation, Ecosystems
