Prepare LAS/LAZ for a traceable project
Start with a vendor-preprocessed, georeferenced LAS/LAZ point cloud. Supply CRS, units, vertical datum and relevant acquisition/trajectory quality information. Raw laser packets and proprietary sensor files need their own manufacturer processing or an explicitly integrated conversion workflow.
The reviewed Suite pipeline includes ground classification, terrain profiles, DSM/DTM generation and EPT point-cloud viewing. PDAL and Entwine workers must be available and configured. Choose a terrain profile as a starting point, then inspect its behavior on the project sample rather than treating a preset as an accuracy certificate.
Separate reduction, filtering and classification
Point reduction can make a dataset easier to process or review, while filtering removes points according to the configured criteria. The photogrammetry processing-node options include pc-filter and pc-sample where supported by that node. Confirm option availability and settings before submitting a large job. These controls are distinct from assigning semantic classes.
Inspect retained edges, thin structures, vegetation and the terrain surface after changing density or filtering. Keep the original dataset and record processing settings. A smaller point count alone does not prove a better ground model or a more accurate volume.
Use optional AI with explicit model requirements
The reviewed source includes optional ONNX imagery-model assistance: road, building and forest signals can be combined with geometric rules. This path requires a configured runtime, suitable model and usable RGB imagery. It is separate from universal point-cloud denoising. Assess the resulting ground, building, vegetation and road grouping against the actual scene.
Review uncertain boundaries and classes before using derived geometry. Urban building-footprint extraction can produce a GeoPackage plus summary metadata, but touching roofs, terraces and canopy boundaries need attention. Authoritative footprint datasets remain useful references when available.
Define the base surface for volume calculations
Saved measurements support cut, fill, net and total volume relative to the chosen base surface. Net equals cut minus fill; total equals cut plus fill. Available base methods include triangulated, plane, average, custom, highest and lowest surfaces. State the AOI, method, units, elevation datum and date, then compare the boundary and surface with control information before commercial use.
Automated stockpile detection is a separate path from a reviewed saved measurement. Its source notes that real-mine validation remains outstanding. Do not treat an automatic stockpile result as a certified quantity without a project-specific validation procedure.
Deliver terrain evidence with context
Potential deliverables include DSM/DTM layers, classified point-cloud review, slope/hillshade and bench/ramp/face layers. Saved area, distance and height measurements can complement volume analysis. Keep the source survey, processing configuration and reviewed measurement geometry connected to the export.
Absolute elevations depend on correct datum/geoid handling and available grids. A photogrammetric cloud cannot reveal ground hidden by dense canopy simply because it is classified as terrain. Agree the required checks, accepted errors and final export package with the receiving team before scaling the workflow.
Manufacturer and data references
Dataset questions
Is point-cloud reduction the same as AI classification?
No. Filtering and sampling change which points are retained. Semantic classification assigns categories to points or uses supporting model inference. Each needs its own configuration and quality review.
Can an automated stockpile result be used as a certified quantity?
The reviewed automated stockpile code does not establish real-mine validation. Use an agreed reference surface, control and independent project checks before accepting commercial quantities.