Using aerial photos, a neural network calculates forest height.

 

To monitor the natural environment, infrastructure, and timber supply, researchers built a neural model to determine the height of trees.

 

According to this research, a neural network could scale up trees from satellite pictures.

Many people demand precise estimates of forest tree size, from environmental scientists to civil engineers to wood industry workers. This data is crucial for determining how much carbon dioxide the trees absorb from the atmosphere, if they pose a threat to power lines, and how much timber is available for logging.

Because drone technology is unsuccessful in big and difficult-to-reach areas, these estimates are now based on satellite images and several cameras spanning various bands of infrared light. However, multispectral data is both scarce and costly to get.

Researchers from the Moscow-based research institute Skoltech may well have discovered the ideal solution, since they were able to train a neural model to reliably and cost-effectively measure tree height.

The approach published in IEEE Access, unlike previous methods, does not require drone footage or images outside the visual range. Instead, standard optical satellite imagery is used by the neural networks.

"The ability to analyse spatial data and texture characteristics is the single most important factor that makes our neural network successful," stated Skoltech PhD student Svetlana Illarionova.

"We added new features in the form of ArcticDEM, a freely available high-resolution model, in addition to the optical images." It's a 2-meter-resolution simulation of the Earth's naked topographic surface spanning the boreal areas."

To build the high-quality model, the researchers took use of the connection between tree crown shape and height.

The canopy height forecasts are rated based on how closely they match lidar observations performed on location in that region with drones, and the training data originates from Arkhangelsk in northern Russia. The researchers claim that their approach can be used in any location with similar vegetation.