Welcome to the 2026 GIS in Action Annual Conference hosted by the Oregon & SW Washington Chapter of the Geospatial Professional Network & Cascadia ASPRS.
Geohazard monitoring in the real world is dominated by messy data, weather interference, occlusions, sensor drift, dropouts, and site-to-site variability. Cascade Geomatics is building a robust alerting pipeline that turns imperfect sensing into reliable, georeferenced risk intelligence for rockfall, landslide, mudslide, and avalanche hazards. We begin by constructing a large, local database of labeled geohazard events and near-miss conditions, aligned in space and time with multi-modal sensor streams. We then train machine-learning models to sift signal from noise and detect hazard-specific precursors, deformation rates, fracture evolution, moisture/loading indicators, and thermal/seasonal dynamics, while producing calibrated confidence scores. In parallel, we maintain a physics-informed neural network (PINN) for each hazard class, using sensor-derived inputs as boundary/forcing terms to model near-real-time dynamics and constrain predictions when data quality degrades. The system continuously compares current conditions to historical analogs, fusing outputs from the historical neural models and the PINNs to generate GIS-ready risk layers: evolving hazard polygons, runout/impact overlays, and asset exposure scores, each with uncertainty and confidence. This talk focuses on the practical path from “bad data” to actionable alerts, its limitations, and how reliability, explainability, and geospatial delivery are engineered into the system end-to-end.
