Funded by the Swedish National Space Agency until 2026, the project's objective is to develop a machine learning model for consistent monitoring of changes in species composition across southern Swedish forests using satellite data from Sentinel-1 and Sentinel-2.
Plants absorb some of the electromagnetic spectrum in the sun's rays for photosynthesis and reflect others. The method being developed aims to identify the differences in the reflected spectra between different tree species. The variation in the different spectra, which is linked to the moisture levels and chlorophyll content in the leaves and needles of the different species, will be observed during different seasons with the help of satellites. Using the spectral patterns, a machine learning model will analyze the distribution of the species each year. The model will be continuously trained on and tested against field validation data. The spectrum from seven different tree species in Götaland is included in the project, including pine, spruce and birch. Through the model, the aim is to produce annual maps of the composition of the forest stands.
Forest monitoring on a larger scale
"The idea here is to create maps that enhances forest monitoring on a larger scale. These maps could significantly benefit biodiversity monitoring and land management," states Hakim Abdi, a researcher at the Center for Environmental and Climate Science (CEC).
The essence of the project lies in the necessity for new specific measurements and methodologies. Existing tree maps are obtained every few years by SLU and Skogsstyrelsen. Other maps from the Environmental Protection Agency offer aggregated data on forests, categorizing different species into groups like broadleaf or coniferous.
"What's lacking is the breakdown of species proportions every year — such as the percentage of spruce, pine, or birch—which isn't regularly updated," explains Hakim Abdi, hopeful that his model will offer new tools through detailed data.
Aiding decision-making in forest management
"The data holds value for government authorities, land managers, and biodiversity researchers. It can be correlated with existing biodiversity surveys, such as birds, butterflies, and mammals. By linking species decline with forest composition, it becomes feasible to suggest management strategies to enhance biodiversity by, for example, favoring certain combinations of tree types or adjusting forest structures."
In cases of forest incidents like large fires, the detailed maps can provide information on the species composition of the affected areas, aiding decision-making in forest management. Moreover, these maps can help forecast future risks, such as the impact of drought or pests on forests, considering various species' resilience.
"Overall, we present a comprehensive approach that intertwines landscape composition and biodiversity data, enabling more informed decisions in conservation and management," concludes Hakim Abdi.
The product will be available for free download upon completion.
Anna Maria Erling