Assessing long-term landscape dynamics based on automated production of land category layers from Danish late 19th century topographic maps

Abstract Context Assessing long-term land use and land cover (LULC) change is critical for understanding of landscapes dynamics. Historical topographic maps contain valuable, spatially explicit information about past LULC. Yet, for analyses of landscape dynamics, it is necessary to "unlock" this information through recognition and extraction via map processing. Objectives Our main goal was to test automated extraction of machine-readable LULC categories from historical topographic maps. Furthermore, we explore landscape dynamics and discuss potential biases associated with category changes between historical and contemporary LULC data. Methods For two study areas in northern and central Jutland, Denmark we apply object-based image analysis (OBIA), vector GIS, colour segmentation and machine learning processes to produce machine readable LULC layers from topographic maps from the late 19th century. By comparison with contemporary maps, we investigate landscape dynamics over 140 years. Results An accuracy assessment applied to the extracted LULC categories indicated an overall obtained accuracy beyond 90 %. A comparison with a contemporary map revealed landscape dynamics, which are characterised by a decrease in heath, wetland and dune sand due to cultivation and afforestation. Dune sand was also characterised by a change to heath and dry grassland. Conclusions We conclude that automated production of machine-readable LULC categories from historical maps offers a less time consuming and more resource efficient alternative to manual vectorisation. Our results also underline that an understanding of mapped LULC categories in both historical and contemporary maps is critical to the interpretation of landscape dynamics.