AbstractThe grasslands which are an important part of an ecosystem are endangered due to abandonment of traditional forms of land use. Preservation of biodiversity of semi‐natural grassland communities is of great importance, therefore it is important to identify threats and prepare a sustainable plan for their protection The goal of this study was to develop an approach to predict threatened grasslands hotspots, basing on multi‐factor machine learning analyses. The Gorce Mountains in the Polish Carpathians were chosen as a study area, as it is region partially protected by National Park, surrounded by villages with different socio‐economic conditions (agriculture or touristic oriented). The grasslands were identified and classified on Sentinel‐2 multispectral imagery. As lack of regular mowing of grasslands was found as main factor promoting forest succession. Therefore, in order to find endangered grasslands areas, support vector machines' algorithm was used to classify them in mowed and unmowed categories. Then the preservation potential of grasslands was modeled with the random forest method, based on the grasslands' mowing classification, digital terrain model, land‐use and population statistical data, and the historical forest extent. We found the grasslands above 750 m above sea level to be the most endangered. Also, the tourism activity and ongoing changes in employment structure from agriculture to services has had a negative influence on the grasslands preservation potential. On the other hand, the Gorce National Park's active grassland conservation by mowing and grazing was shown as a positive element in keeping with the high biodiversity of the area.
Read full abstract