AI helps to map thinning requirements in Swedish forests

Monitor natural values, get the trees to grow better or create a forest for recreation? Thinning is a forest management measure that can fulfil several purposes, and regardless of which one, digital decision support can improve both planning and implementation. Currently, Mistra Digital Forest is further developing an AI-trained model that helps forestry operators to map thinning requirements.  

Thinning increases the value of the trees and is a way of creating a forest that is varied, healthy and is a place where people want to spend their time. At the same time, thinning of the Swedish forest needs to increase, and these needs differ greatly between forest owners. In a Vinnova-funded project at Skogforsk, researchers have found ways of mapping the thinning required in large forest areas, using machine learning, which is an area within AI. By combining field data and satellite images from the EU's Copernicus Earth Observation programme, it is possible to train models that assess the thinning requirements in large stands. This work is now continuing within Mistra Digital Forest.  

- We have a basic model that has good potential. Now we are going to take important steps on the road to operational management, and develop a model that meets the different requirements that exist, both within and between groups such as forest owners and forestry entrepreneurs. Hopefully, the models will encourage more people to thin their forests and make thinning more efficient, says Liviu Ene, researcher at Skogforsk.  

So far, the model has only been trained on data from forests in central Sweden. The next step is to test the model in new areas, and train it on data from forests in other parts of the country. This is important in order to arrive at a model that can handle the great variation in Swedish forests, with all the deciduous trees in the south and the rich spruce forests in the north. The researchers will also be testing the use of more high-resolution satellite data; the satellite data used so far is sensitive to weather conditions, which affects its reliability. In this case, the researchers see a potential solution in combining data from different satellites. 

Many possible fields of application  

Södra, Sveaskog and Mellanskog are participating in the project and are positive about the possibility of estimating the need for thinning "at a distance". One item of feedback from the project participants which is now being taken further within the project, is the development of models that can be used to evaluate how long the thinning work will take. For example, in addition to facilitating planning, this would be a good basis for decision-making in the business dealings between forest owners and forestry entrepreneurs. Precisely how the results should be made available is a focal question within Mistra Digital Forest. Another possible area of use is the analysis of different approaches to thinning work, and the pinpointing of potential improvements. 

- Now we have the opportunity to develop decision support that helps us to prioritise correctly on several levels. Firstly, we can target resources more intelligently at the planning stage of forest management, and avoid going out to make an inventory only to find that there is no need for thinning. In addition, the thinning teams can have a better basis for organising their work and for concentrating their efforts on site, says Marcus Abrahamsson Månsson, technical specialist at Sveaskog.  

Whilst the researchers are validating and further developing the model, the former project partner Field Sweden has already started developing a commercial service in the form of a nationwide decision support system. 

- We are seeing a great deal of interest from the industry, but the requirements vary, and they will determine the direction of our work in the future. Ultimately, we are keen to develop solutions that are useful and applicable to industry and to forest owners, says Liviu Ene.