A neural network has learned to identify tree species from satellites


A detailed land cover map showing the forest in the state of Chiapas in southern Mexico. The map was created with optical data from Copernicus Sentinel-2 from April 14, 2016. The picture is not part of the study discussed.

Much of what we know about forest management today comes from aerial photographs. Whether it’s drones, helicopters or satellites, a bird’s eye view of forests is crucial to understanding how our forests are doing – especially in remote areas that are difficult to monitor on the ground.

Satellite images, in particular, offer an inexpensive and effective tool for surveillance. The problem with satellite data, however, is that the resolution is often quite low and what you are seeing can be difficult to see.

but a new study Using neural networks to differentiate between satellite images can help with this.

Hierarchical model structure / Svetlana Illarionova et al., IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.

“Commercial forest tax providers and their end users, including wood procurers and processors, as well as the forestry industry can use the new technology for the quantitative and qualitative assessment of the wood resources in leased land. In addition, our solution enables a quick assessment of underdeveloped forest areas in terms of investment attractiveness, ”explains Svetlana Illarionova, first author of the paper and PhD student at Skoltech.

Illarionova and her colleagues at the Skoltech Center for Computational and Data-Intensive Science and Engineering (CDISE) and the Skoltech Space Center used a neural network to automate the identification of dominant tree species in high and medium resolution images.

Class award of the study area. Photo credit: Illarionova et al.

After the training, the neural networks were able to identify the dominant tree species in the test area of ​​Leningrad Oblast, Russia. The data was confirmed by ground-based observations in 2018. A hierarchical classification model and additional data such as vegetation height helped to further improve the quality of the forecasts while improving the stability of the algorithm to facilitate its practical application.

The study focused on identifying the dominant species. Of course, among the forests of different composition, there will be forests in which the distribution between two or even more species is roughly the same, but the composition of these mixed forests was not the subject of the study.

“It is noteworthy that the“ dominant species ”in forestry does not exactly match the biological term“ species ”and is primarily associated with the wood class and quality,” the researchers write in the paper.

Overall, the algorithm appeared to be able to identify the dominant species, although the researchers find that better training markup can improve the outcome, which they plan in future research

“In future research, however, we will cover mixed forest cases that fall entirely within the hierarchical segmentation scheme. The other goal is to add further forest inventory features that can also be estimated from the satellite images, ”the study concludes.


Leave A Reply