Terrain Classification
Definition
Terrain classification is the process of labeling landforms and surface morphologies from elevation and imagery into categories such as plains, hills, ridges, valleys, escarpments, dunes, and karst. In GIS this is achieved by extracting terrain derivatives—slope, aspect, curvature, topographic position index, roughness, openness—and applying rules or machine learning models that combine these measures at multiple scales. Classification can be categorical (assigning a single landform per cell) or hierarchical (broad classes subdivided by finer features). Because landforms are scale-dependent, robust workflows compute metrics with several window sizes and reconcile them, ensuring that a small gully inside a larger valley is not lost. Accurate terrain classification provides a common language for geomorphology, habitat modeling, hydrology, military mobility, and infrastructure siting.
Application
Planners use landform maps to steer growth away from unstable slopes and to site parks with scenic ridges. Ecologists target restoration in riparian bottoms and identify ridge habitats for wind-sensitive species. Hydrologists derive watershed divides and valley bottoms that feed flood models. Mining and engineering teams screen corridors by avoiding badlands and rock outcrops. Because the classes are interpretable, they help stakeholders discuss constraints without technical jargon.
FAQ
How do multi-scale metrics improve landform classification compared with a single window size?
Landforms express at different scales: a mesa may be flat at 30 m cells yet sit atop regional relief. Computing derivatives with several kernels and blending them detects both micro- and macro-features, avoiding mislabeling due to scale bias.
What role does curvature play relative to slope and aspect?
Profile and plan curvature capture concavity/convexity that distinguishes ridges from valleys even where slopes match. Curvature also highlights shoulders and footslopes that are critical for soils and hydrology.
How can classification be validated beyond visual inspection?
Use field points, geomorphologic maps, or expert-labeled samples; compute confusion matrices and kappa; and assess stability under DEM noise by perturbing inputs and reclassifying.
When should an object-based approach be favored over per-pixel labeling?
Object segmentation groups contiguous cells into landform units with shape and context, improving coherence in rugged terrain and aligning better with human interpretation.