Repeat lidar and photogrammetry reveal how geohazards reshape terrain, but raster differencing may confuse noise with real change. In this hands-on workshop, emergency managers and analysts build robust, uncertainty-aware change detection workflows in GRASS using Hurricane Helene data.
Geomorphic change detection is a foundational tool for identifying and mitigating landscape hazards. Repeat LiDAR and structure-from-motion (SfM) photogrammetry are now routine across federal, state, and local programs, so analysts increasingly hold sequential topographic observations that span the timescale of extreme events. The hard part is turning those observations into defensible measurements of change. Errors in vertical accuracy, horizontal co-registration, interpolation, and resolution mismatch all propagate into topographic derivatives, and a naive DEM of Difference can present measurement noise as though it were a real geomorphic signal. In hazard work, where the result may inform where people rebuild or how a watershed is managed, that distinction matters.
During this workshop we will build a robust, reproducible change detection workflow in GRASS, with every participant running the analysis on their own laptop. We start from the most common approach, the DEM of Difference, and use it to make its limitations concrete: how vertical uncertainty sets a minimum level of detection, why spatial autocorrelation of error complicates simple thresholds, and why co-registration is often the single largest source of apparent change. From there we work through the methods GRASS provides for managing and propagating uncertainty across a sequence of assets, including spatially variable error models, masking strategies, and the temporal framework for organizing and querying multi-date observations.
The session is anchored by two case studies from recent disaster responses. The first is post-hurricane terrain change in the southern Appalachians following Hurricane Helene, where extreme flooding and landslides reshaped channels and hillslopes across the Blue Ridge Escarpment. The second is post-fire landscape response, where loss of vegetation and soil structure drives erosion, debris flow initiation, and altered hydrologic behavior. Participants run each workflow end to end, from raw inputs through co-registration, differencing, uncertainty thresholding, and interpretation, using real LiDAR and photogrammetry rather than toy data.
By the end, participants will have computed a DEM of Difference and identified where and why it misleads; estimated a minimum level of detection and applied spatially variable, rather than uniform, uncertainty thresholds; diagnosed and corrected co-registration error as a distinct source of apparent change; organized and queried multi-date observations with the GRASS temporal framework; and run both the hurricane and fire workflows end to end.