Looking for shrubs in an alvar: Investigating classification of orthophotos as a way of mapping shrub species Juniperus communis and Dasiphora fruticose on Stora Alvaret, Öland

Abstract: Loss of biodiversity is an immediate threat to the planet and if no actions are taken it is expected to accelerate. The highest species richness on smaller scale is found in semi-natural temperate grasslands, with one of the documented records from Stora Alvaret on Öland, Sweden. The area is protected under the Natura 2000-network and Länstyrelsen Kalmar are responsible for monitoring while local farmers execute most of the management with support. One of the main threats to the biodiversity on Stora Alvaret is encroachment by shrubs when the grazing pressure is decreased. The study investigates the possibility to use classification of high resolution orthophotos as an aid in monitoring if and where shrub encroachment is happening. The studies main aim is to develop a method to successfully classify the species of interest Juniperus communis and Dasiphora fruticose with the software ArcGIS pro. The result shows that it is possible to successfully classify the species of interest if following the proposed method. The method consists of a first classification on 5 m resolution with removed agricultural fields, to extract the forest class which then is removed from the 1 m resolution. The 1 m resolution without forest is then used as a base for the final classification. The final result had a kappa value of 0,92 which translates to an almost perfect classification. As the study is mainly focused on the shrub species it is important to also note the producer’s and user’s accuracy for their classes. Juniperus communis had a producer’s accuracy of 0,86 and user’s accuracy of 0,38 while the corresponding values for Dasiphora fruticose were 0,92 and 0,86. The weakness in the classification thus lies with other classes being included in the juniper pixels. The removal of forest also introduces uncertainty and the producer’s and user’s accuracy of that class should be considered when interpreting and using the results. Further research should focus on improving the user’s accuracy for juniper and investigating the use of change detection between classified results from different years.