Remote sensing

Improved tree species classification using high-resolution data from multispectral instruments

The development is progressing rapidly, and new technology in airborne laser scanning enables analysis at the level of individual trees. A recent study shows that high-resolution data from instruments scanning with two frequencies improves tree species classification and can be used to estimate stem volume. This research may impact the technology selection for future national scans of Swedish forests.

The study within Mistra Digital Forest opens the door to a more detailed image of Swedish forests by developing models for classifying tree species and estimating individual tree stem volume. The study utilized airborne laser data with a point density of approximately 50 points per square meter. This can be compared to the ongoing national laser scanning, which has a density of 1-2 points per square meter.

The high point density allows for extracting information about individual trees. However, to classify tree species, airborne laser data from a multispectral instrument scanning in near-infrared and green wavelengths were also utilized. Currently, single-wavelength technology is the standard and is used in the ongoing national scanning of Swedish forests.

Kartor

The area in Siljansfors with classified tree segments. Image: Christoffer Axelsson. Orthophoto: Lantmäteriet.

– When comparing data from the multispectral scanning with laser data from single-wavelength instruments, it was found that the additional green wavelength was crucial for distinguishing deciduous trees. Multispectral laser scanning contributes to greater accuracy in tree species classification," says Christoffer Axelsson, postdoctoral researcher at the Department of Forest Remote Sensing at SLU and the study's performer.

Christoffer Axelsson, SLU
Christoffer Axelsson, SLU. Photo: Emma Sandström.
"Multispectral instruments have great potential in the long run."

The study is based on airborne laser data within a test area in Dalarna, and field data from the same area was used as a reference. The research is largely funded by Stora Enso, which owns forests in the test area.

– We are constantly working to find new cost-effective methods that promote the development of more adapted forestry. This is important in order to meet increased biodiversity objectives, and I believe that multispectral instruments for classifying individual trees have great potential in the long term, says Erik Willén, Manager, Precision Forestry Stora Enso.

Erik Willén, Stora Enso
Erik Willén, Stora Enso
Fotograf: Johan Olsson
Scaling up method produces good results

To classify tree species and estimate stem volume, the study uses so-called metrics. Metrics are different measurements, such as height and width, that are calculated from the tree's point cloud and act as a kind of identifier. By matching the metrics/characteristics of an unknown tree with a twin tree in the sample area, whose stem characteristics etc. are known, a detailed picture of the forest is built up.

– This upscaling method produces good results, and it demonstrates the information that it is possible to obtain when scanning with high point density and multiple wavelengths. The results could contribute to the choice of technology in the next round of the national scan, concludes Christoffer.

Christoffer Axelsson is demonstrating the method for upscaling, (in Swedish).