Satellite Classification of Tree Species

Posted 01/6/16

The forest industry is an important part of the economy, and forests harbor amazing biodiversity and provide recreational opportunities for people. However, the economic, ecosystem service, and recreational value of forests depends on their tree composition. Is the forest deciduous or coniferous? Is it an open pine or a dark spruce forest? How widespread are different tree species? In the course of his PhD, Konrad Turlej is going to answer these question.

Konrad Turlej, who brought his great expertise of remote sensing to SILVIS, enthusiastically started his PhD project in 2015 focusing on the mapping of tree species in Poland and in Wisconsin. The idea of the project is to map Polish and Wisconsin forests with 20-30 m resolution imagery. Konrad’s goal is ambitious: he wants to map not only where forests are, but also tree species with this medium resolution satellite data.

The reason why this is complicated is that a single tree is less than 30 mand there can be 2-4 trees species in a single pixel. Furthermore, in a satellite image, many tree species look very similar during the peak of summer.  However, phenology varies greatly among tree species. Some trees have their leaves earlier, some ater. In the fall, some trees are losing leaves in September while others keep them till the first frost.

Forest Study Areas in Poland and Wisconsin
Forest Study Areas in Poland and Wisconsin

By analyzing satellite images for the entire growing season, one can analyzes so-called phenology curves. Over the course of a year, this curve looks different for different species because of phenological differences. This idea sounds quite promising but there is another challenge: Landsat, the source of 30 m imagery, provides only 1-2 images per months. In Poland, where leaves can fully come out in two weeks, this is not frequent enough to build a good phenological curve. This is why Konrad will combine Landsat data with imagery from other satellite sensors, including MODIS and Sentinel 2a.

Ultimately, the maps that Konrad is creating will be beneficial for several purposes. Describing and counting trees in the field takes a lot of time and money. Mapping maps of tree species from satellite images instead will save money and provide more timely information. With such maps, foresters can then estimate current situation on various species, amount of timber and its economic value, and eventually provide better management.

Story by Bragina, Eugenia