Workshop “Methods and applications of GeoML”

Abstract

The workshop explores the application of modern computational methods—including machine learning, large language models, and Earth observation—to complex challenges in geospatial analysis and environmental risk management. The presented studies demonstrate how heterogeneous and unstructured data, from text and informal addresses to multispectral satellite imagery, can be transformed into robust, spatially explicit geodatabases for environmental monitoring, predictive modeling, and risk assessment. Particular attention is given to climate-driven hazards and environmental processes, including Arctic coastal thermoerosion, glacial lake outburst floods, and seasonal transport infrastructure. Methodological contributions include hybrid neural-symbolic architectures, attention-based semantic segmentation, spatial cross-validation, and conformal prediction for uncertainty quantification. Together, these approaches advance scalable, transparent, and uncertainty-aware GeoML solutions for climate adaptation, disaster risk reduction, and sustainable resource management.