Finnish Geospatial Research Institute FGI has a team of scientists working on laser scanning data and detection algorithms for the Finnish forests. In Finland the bottleneck in forest research is rather in the efficiency of data analysis, than in the amount of data collected. AI, such as machine learning and deep learning, can provide tools to solve this problem. 

Forests are an important part of the Finnish economy, their health and growth has been followed closely for at least since the beginning of the forest industry. If artificial intelligence can be harnessed to analyse the combined data from different sources, such as from the forestry machine sensors, aerial and land-based surveys, this could create new possibilities to understand our forests better. Here are four examples of what we do at the FGI in order to increase forest knowledge. 

Accurate environmental analysis from combining modern sensors  

“The vulnerability of forested and water ecosystems to different distresses is escalating due to the warming climate” writes Research Professor Antero Kukko in his latest article in Laser Scanning Magazine. “Accelerating urbanization and use of natural resources have an impact on the environment and the livability of human settlements. The headlines in our daily news often report disasters arising from a chain-reaction beginning from a drought, or from wind- or snow-induced damages—these often lead to forest fires, plant diseases and pest infestations, such as spruce bark beetles.”  

Today, the national laser scanning surveys combined with data collected for example by sensors attached to forestry machines, drones and helicopters produce impressive amounts of accurate, up-to-date data on our forests. The sensors of the FGI vary from different types of spectrometers to in-house built advanced laser scanners and many more. Spectrometers can for example accurately detect changes in the forest canopy caused by forest pest infestation, even before a human eye can see any change in the forest health.

FGI has also built highly accurate laser scanner systems operating in multiple wavelengths. Such multispectral laser scanning can penetrate through vegetation, which allows analysis of undergrowth and terrain features under the forest canopy as well as any other geometric features on the surface, such as roads and buildings. We are currently running an international benchmarking study on tree species using multispectral laser scanning as primary data. As for AI solutions, both machine learning and deep learning approaches are tested by a large international consortium. Tree species information is crucial for understanding biodiversity stored in forests. For example, aspen is regarded as one of the most important species for biodiversity in some boreal forest regions.  

Individual tree identification in large scale powered by AI 

The UNITE competence center creates technologies that allow building an accurate digital replication of boreal forests at individual tree level, using airborne laser scanning surveys of National Land Survey of Finland as source. The technology is demonstrated in Metsäkanta forest database, which enables users to view individual trees sorted by length and tree species. The system is currently open to test users. The results have also contributed to automating field sampling of forests to be several dozen times faster than earlier. There are currently 2.4 billion trees in the Metsakanta, from each of which height, diameter, volume, biomass, species, crown base line, amount of log, pulp and energy wood, value and stored carbon are estimated. For field reference, 177 000 manually measured trees are applied.  

UNITE works towards multifunctional forestry, enabling forest sectors transition to meeting the demands of tomorrow and enabling climate smart, resource efficient forestry. Detailed data on the individual trees, species distribution and trunk size can be helpful in better recognition of areas for conservation purposes, for example by identifying key species for biodiversity, valuable biotopes or for example amounts of old, decomposing trees in certain areas. Tools based on this information could help the forestry machine operators in making an informed choice on where to operate. Ongoing IlmoStar project examines how the data collected by perception sensors, e.g. laser scanners installed on a harvester, could help in mitigating biodiversity loss and assisting harvester operators in decision-making for optimal carbon sequestration.

Combining hydrological knowledge to forest monitoring

In addition to species recognitions and general forest growth parameters, such as trunk size, the precise data allows the researcher’s to detect for example or canopy damage due to wind, snow or ice. Continuous terrestrial laser scanning (TLS) monitoring of forests can reveal even the smallest changes in the forest structure, such as the micro movements of trees during day and night. Research of Eetu Puttonen and collaborators has revealed that the trees go in a resting mode during the dark hours. Continuous monitoring of forests produces vast amounts of data, and detecting these small changes from days, weeks and months’ worth of data is a tedious job if certain automation did not exist. In one of our projects, LS-HYDRO, we combine the dense spatiotemporal TLS time series, high-resolution satellite data, and advanced biophysical ecohydrologic models to observe vegetation structure and hydrologic state and develop tools and means to measure and predict water fluxes in the forest. This combined with the laser scanning measurements carried out in wider hydrological settings like river environments may in future complete a large digital twin of a whole river basin, combining the existing hydrologic models, accurate forest data, tree-water interactions and a number of other functions, parameters and AI-based analysis tools to help decision-making, forest and water management. FGI is part of Freshwater Competence Centre that aims to build such a digital twin and recently started a large multi-million Digital Waters research project to complete the work. 

Helping berry pickers find berries using deep learning methods  

In Ferox project, we have already built a map that predicts the presence of berries at certain areas. The prediction is using certain environmental parameters, such as vegetation characteristics, forest type, closeness to lakes, and elevation of the area. The next step is to validate the model with in-situ data from photographs and drone images. Our scientists have compared the performance of various artificial intelligence algorithms for automatically finding individual berries, and estimating the berry yield, from images obtained in challenging Finnish forest and bog conditions.  

The technologies created have the potential to substantially improve the productivity of the wild berry picking industry, increase the sustainability of natural product extraction and improve the working conditions of berry pickers in the Nordic countries, and also globally. These maps can also help local people to pick local berries more effectively which could be socially important and beneficial. 

The long-term goal is to fully automate the mapping of berries, mushrooms, and other wild forest products in environments that are hard to traverse by foot and impossible to access by a vehicle. This can be done using a combination of drone mapping technologies, large-scale airborne laser scanning and highly accurate micro-scale weather predictions. Finding an optimal balance between the sustainable use of forest resources and the extraction of high-quality wild forest products in large quantities could be improved dramatically in the future with the introduction of these mapping technologies to the industrial sector and to the general public.