Artificial Intelligence for Land Use and Land Cover Mapping in Austria

Land use and land cover (LULC) mapping is crucial for environmental management, urban planning, and sustainable development. Traditional methods for LULC mapping ofteninvolve manual categorization and timeconsuming field surveys. This study aims to evaluate the effectiveness of artificial intelligence for automation of LULC mapping in different geographical areas in Austria. We evaluate three convolutional neural network architectures and random forest classifiers trained on aerial images, digital elevation data, and reference labels. The AI-based approaches achieve an overall classification accuracy of over 90%, outperforming conventionalmethods by a substantial margin. The evaluation reveals thata UNet architecture with categorical focal loss provides the best results.