The authors of the project were scientists from the Swedish University of Agricultural Sciences (SLU), the Mid Sweden University, the Inland Norway University of Applied Sciences in Koppang, as well as from the Mammal Research Institute of the Polish Academy of Sciences in Białowieża. The researchers described their work on the model and possible applications in the latest issue of Communications Earth & Environment, in the paper 'The conservation value of forests can be predicted at the scale of 1 hectare'.

Sweden is one of the most forested countries in Europe, and one with very intensively used forests. Since forestry and the wood industry are important parts of the economy, Swedish authorities, services and forest researchers have been collecting a lot of data for years - from the ground, planes and satellites. According to one of the authors of the publication, Professor Grzegorz Mikusiński from SLU, data have been collected systematically for over 100 years, and from 2023 - as part of the Swedish National Forest Inventory.

This resource has been put to use once again: scientists have developed a computer model that makes it possible to precisely identify forest areas with the greatest ecological value.

'Our goal was to create a tool that will enable automatic scanning of forest landscapes on a large scale, with high spatial resolution and with high thematic accuracy. Using this tool, we want to find forest stands, areas and landscapes with the highest natural value (the so-called High Conservation Value Forests - HCVFs). The distinguishing features of such forests include long history of forest continuity, complex structure, the presence of dead wood and relatively little human impact,’ one of the authors of the analyses, Dr. Jakub Bubnicki from the Mammal Research Institute of the Polish Academy of Sciences in Białowieża, tells PAP - Science in Poland.

The task is only seemingly simple, Dr. Bubnicki adds. Why? Because the natural value of forests is determined by a number of variables: the structure of the forest stand, the historical impact of humans on the forest stands (including forest management), topography, the structure of the surrounding landscape, and the animal and plant species composition. This is a complex issue that is not easily captured in a simple equation or model that would lead researchers to say, 'This is what we were looking for'.

To facilitate the search for valuable forest areas, scientists have developed a special computer model that can identify High Conservation Value Forests. The model gained this ability thanks to 'training' on well-known data for Sweden, i.e. 'mapped' and field-verified HCVFs. 'We also used open data describing the structure of forest stands and landscape, available throughout Sweden,’ says 

Dr. Bubnicki. This includes: data from aerial scanning of local forests regarding the structure of trees and their height, detailed maps of land cover and global Forest Cover Change data sets, showing the dynamics of forest stand loss after 2000.

The researcher checked their model several times, comparing its results with data from completely different sources, e.g. those provided by the largest forest owner in Sweden, Sveaskog, data from national forest monitoring, as well as expert assessment with representatives of Swedish regions, forestry companies, other researchers and the Swedish Environmental Protection Agency.

The model not only identifies valuable forests, but also determines their level of naturalness. 'Our goal was not only to determine whether such forests exist in a given place or not - but also to assess the level of their naturalness on a continuous scale, e.g. from 0 to 100. Using the results of our model, we can say, in a broader perspective, to what extent a given forest landscape is natural, and to what extent it has been transformed, for example, by forest management,’ says Dr. Bubnicki. 'When planning protective measures, the goal is not to select random areas that have no connection with each other, have too small areas and do not fulfil their ecological role, but in total they will add up to the required 10 or 30 percent. We also need to take into account the ecological connectivity between these areas at larger spatial scales.’

The model and map have been published and are available digitally online, so they can be used for further research or to identify naturally valuable forests that require urgent protection. 'They will also be useful for planning the restoration of degraded ecosystems. Imagine that we have two protected areas and around them - degraded forests, but with different degrees of degradation. Thanks to our results, we can identify the least degraded ones - and those that, in combination with already protected, valuable forests, will best create a functional ecological network,’ says Bubnicki.

The authors of the analyses draw attention to the Swedish research context. 'This is a country where forest management is carried out with great intensity, even on a global scale,' Bubnicki continues. 'Most of Sweden's area is covered by forest. But, apart from mountain areas or national parks, most of the tree stands are largely transformed. We are dealing with very heavily transformed forests with very valuable islands.’

Professor Mikusiński adds that the vastness of Sweden's forest areas and their dynamics create a situation in which it would be very difficult and extremely expensive to constantly check the natural value of each piece of forest in the field. 'That is why our model is so popular among various institutions and organizations,’ he says. 

Can a model trained on Swedish data be used to protect forests in other parts of Europe or the world? Dr. Bubnicki says that the methodology is universal, but the model itself needs to be trained using data from other regions it is to cover.

Although similar analyses and identification of High Conservation Value Forests have been carried out before on a global scale, the models used for this purpose were 'trained' on data from various places around the world. And forests in different places around the world are very different. The use of such models produces spectacular maps, but it does not work for local assessment and does not translate into activities carried out on a relatively small scale of provinces or regions.

What would have to happen so we can use a similar computer model in Poland? Bubnicki says that the model requires training on good data describing the landscape and tree stands - and Poland has such data.

'It would be a challenge to compile a training database for this model. It is possible, but it requires a team of experts: specialists in forest ecology, biodiversity, machine learning and foresters. When it comes to the technological aspect - there are no restrictions that would prevent us from performing such an analysis for Poland,' he says.

New analyses may be useful for spatial planning and protection of the most valuable forest parts, as well as for plans to restore the natural value of already degraded forests. This is important in the context of the assumptions of the EU Biodiversity Strategy, according to which at least 30% EU's land and sea areas in each country will be protected by 2030, including 10 percent under strict protection.

Interview by Anna Ślązak, PAP - Science in Poland

zan/ agt/

tr. RL