Rural depopulation, agricultural land abandonment, and deforestation are massive concerns for Europe and elsewhere today and our planet's future. These interlinked phenomena can be analysed using land use and land cover (LULC) maps combined with dynamics of population geography, especially regarding urban sprawl. Modern LULC and spatially disaggregated population datasets go back to the 1980s and 1970s. Although we have earlier population data, these are not geomatched to locations in LULC maps. Earlier LULC maps are either not very reliable (extracted from historical maps) or limited in their geographical coverage (based on selected aerial photos or satellite imagery). These are severe limitations to developing longer and deeper perspectives and understanding the root causes of these detrimental changes in population geography and land use practices in large territories.
GeoAI_LULC_Seg funded by the European Research Council will develop an advanced, modular, and customizable geospatial artificial intelligence-based land use land cover segmentation process to accurately map LULC conditions for around 30,000 km² in a border region between Bulgaria and Turkey, including the cities Edirne, Istanbul, and Plovdiv, from historical aerial photographs and early reconnaissance satellite images (dating back to the 1950s and the 1970s respectively) by pairing them with geotagged historical population census data.
We are currently working on engineering solutions to orthomosaicking early aerial photographs from the 1950s and the latest panchromatic satellite imagery from the 1980s. We use aerial photographs from the General Directorate of Mapping, the Ministry of Defence, the Republic of Turkey and the Military Geographic Service, the Ministry of Defence of the Republic of Bulgaria. The satellite imagery we use is from the most sophisticated and the last film-based photo-reconnaissance satellite program, Hexagon (KH-9), online available at the Earth Resources Observation and Science (EROS) Center, the United States.
We aim to segment land use and land cover (LULC) patterns from these historical, high spatial resolution yet single band imagery and create commensurable LULC maps with current satellite-based data such as ESA WorldCover. Furthermore, these early panchromatic imageries can be used as ancillary data for gridded population exercises if successfully segmented.
Presentation
