New unique mapping of deciduous forest with potential conservation values
Sweden holds an important part of the EU’s natural heritage in its forests. This comes in the form of valuable coniferous natural forests, but also in the form of older deciduous forests with high conservation values. Older deciduous forests often have a unique flora, funga and fauna associated with them. For Sweden to be able to meet national and international environmental goals and the EU’s laws and regulations, all forests with conservation values must be preserved and extensive areas must be restored. Now Forest Monitor, run by the organization Protect the Forest, has developed entirely new and unique data on deciduous forests with potential conservation values in southern Sweden, excluding Gotland.
– An important step in the work of protecting and restoring vital forest ecosystems is to locate known and potential conservation values. This makes inventory efforts much more efficient and contributes to a unique overview of the values in forest landscapes, says Roger Adolfsson, spokesperson for the organization Protect the Forest.
The mapping tool forestmonitor.eu presents data on known and potential conservation values for the entire country, linked to older forests and so-called continuity forests, forests that have not previously been clear-cut. To find potential older forests and continuity forests, Forest Monitor has used, among other things, remote analyses of nationwide historical aerial photographs and satellite images, and then classified these with the help of inventory data and algorithms that use findings of species of conservation interest.
There are challenges when using classical forms of remote analysis of aerial and satellite images because these do not always, in a satisfactory way, capture deciduous forests with potential conservation values in southern Sweden. This is partly because some images were taken when deciduous trees had shed their leaves, which can cause the analyses to fail.
– Therefore, at Forest Monitor we have used machine learning, which uses many forms of data to produce a map layer with predictions of where the most valuable deciduous forests may be found in southern Sweden, says Jon Andersson, project manager and data developer at Forest Monitor, Protect the Forest.
The map layer of deciduous forests with potential conservation values was created using Random Forest, a highly effective machine learning algorithm. The algorithm was trained on satellite images, vegetation maps, terrain data, and data from tree and forest surveys in deciduous forest areas with known conservation values.
– We optimized a Random Forest model and managed, with about 90% accuracy, to predict the potential location of deciduous forests with conservation values, says Hainner Aparicio, remote sensing expert and analyst at Forest Monitor, Protect the Forest.
There is no scientific consensus on what constitutes “good” accuracy in forest classification studies that use machine learning. Similar studies in this area vary from acceptable and valuable results above 70% to excellent accuracy that typically exceeds 90–95%. High accuracy depends on factors such as the complexity of classifying different categories and the quality of the base data and the type of data used.
Forest Monitor wants to underline that it is important for all who use the new deciduous forest layer to be aware that this is a prediction, or forecast, of where in the landscape there could be deciduous forests with potential conservation values.
– The degree of accuracy determines how valuable this type of prediction is for conservation work, environmental monitoring, and when it comes to streamlining the prioritization of what should be inventoried in the field, says Viktor Säfve, project manager and communicator, Forest Monitor, Protect the Forest.
Forecasts generated using machine learning and that have high accuracy generally provide valuable information. But even if they are valuable and useful, all forecasts can sometimes be incorrect.
– This first version of Forest Monitor’s deciduous forest data has high accuracy. Even if it is usually correct, we have also seen examples where accumulations of standing dead coniferous trees have in some cases been detected as leafless large deciduous trees, says Jon Andersson, project manager and data developer at Forest Monitor, Protect the Forest.
The new map layer of deciduous forest with potential conservation values in southern Sweden is now published and can be used on ForestMonitor.eu This map layer and other layers on the ForestMonitor.eu mapping tool are free to use on the map service.
– Get out into the forest now and explore all the potential conservation values! We hope that Forest Monitor, and especially the new complementary deciduous forest layer, will help authorities, forest owners, and all other stakeholders to streamline inventory efforts, forest protection and restoration, and thereby help Sweden achieve its environmental goals and preserve biodiversity, says Viktor Säfve, project manager and communicator, Forest Monitor, Protect the Forest.
Fact box on the map layer Deciduous Forests with Potential Conservation Values 1.0
Area and method:
This layer contains deciduous forests and deciduous-dominated mixed forests that may potentially have conservation values in Svealand and Götaland except Värmland, Dalarna and Gotland. The analysis was carried out using machine learning (Random forest classifier), where several open datasets were used to train/test a model to identify and recognize deciduous forests with conservation values or deciduous-dominated forests with conservation values. The data do not differentiate between different species of deciduous trees or the degree of admixture of e.g. spruce and/or pine. The Swedish national land cover database (NMD) has not been used to create the data, but the consistency between the two data layers is high. The data will eventually also include Gotland County.
Read more about the methodology here
Area (deciduous forest with potential conservation values in the counties included in the analysis):
According to our results, the analysis includes just over 289,000 hectares of forest land with deciduous forest with potential conservation values in the studied area. Of these, only 18.5% (53,519 hectares) were within formally protected areas.
Total area including formally protected area: 289,000 hectares
Total area lacking formal protection: 235,481 hectares
Direct link to the map service
Forest Monitor’s website: www.forestmonitor.eu
Report on the new deciduous forest map layer
Contact: skogsmonitor@skyddaskogen.se
Watch short film about the new map layer:
