AI-driven hazard monitoring in Albania: combining CLIP, image segmentation, and Web GIS for floods, fires, and deforestation
AI-driven hazard monitoring in Albania: combining CLIP, image segmentation, and Web GIS for floods, fires, and deforestation

Summary/Abstract: Albania is increasingly exposed to natural and anthropogenic hazards, including recurrent floods in the Shkodra basin, forest fires in mountainous protected areas, and ongoing deforestation linked to land-use change. Effective monitoring of these processes is challenged by fragmented datasets, delayed reporting, and the limited integration of advanced analytical methods into operational geoinformation systems. This article proposes a hybrid AI-GIS pipeline that combines semantic image classification with quantitative geospatial analysis to support hazard detection and management.The approach integrates the Contrastive Language–Image Pretraining (CLIP) model for zero-shot classification of hazard-related imagery with pixel-level segmentation and spectral indices derived from remote sensing data. CLIP enables the automatic labeling of images and tiles according to natural language prompts such as “flooded farmland”, “burned forest” or “deforested hillside” providing semantic context without the need for retraining. Segmentation methods and indices, including NDWI for floods, NDVI for vegetation loss, and dNBR for burn severity, are then applied to quantify the spatial extent of affected areas. The resulting outputs are structured in a PostGIS database, where hazard layers and attributes are stored and linked to spatial queries. A Web GIS environment built with Leaflet provides interactive visualization, dashboards, and temporal comparisons for end users.Three scenarios are presented: flood extent mapping in Shkodra, wildfire impact in Lurë National Park, and deforestation monitoring in Tropoja. Results demonstrate that the integration of semantic classification and quantitative extraction enhances both the interpretability and accuracy of hazard assessments. The framework highlights the potential of combining AI and GIS technologies to create scalable, reproducible, and policy-relevant observatories for environmental risk monitoring in Albania.

• Details
• Contents