A LiDAR survey is only as useful as the data delivered at the end of the workflow. A dense point cloud may look impressive, but it cannot support design, quantities, asset decisions or measured drawings until it has been correctly referenced, cleaned, classified and checked. Knowing how to process LiDAR survey data is therefore as important as selecting the scanner, drone or GNSS equipment used in the field.
The right process depends on the capture method and required output. A mobile mapping scan of a road corridor, a terrestrial scan of an industrial structure and a drone LiDAR survey of a woodland site each create different processing challenges. The common objective is the same: produce an accurate, traceable dataset that is suitable for its intended use.
Start with the required deliverable
Processing should begin before files are imported. Define what the client, designer or asset team needs to receive. This might be a classified LAZ point cloud, a ground model, contours, a CAD drawing, a mesh model, stockpile volumes or a set of measured sections.
This decision controls the processing tolerance. A planning-level terrain model does not need the same level of feature extraction as a detailed as-built survey. Likewise, a visually convincing mesh is not automatically appropriate for dimensional verification. Establish the required coordinate reference system, vertical datum, accuracy specification, survey boundary and exclusion areas at the outset.
For UK projects, this commonly means confirming whether deliverables are required in OSGB36 / British National Grid and the relevant orthometric height datum. Confusion between ellipsoidal heights captured by GNSS and project levels can introduce significant vertical errors. Apply the correct geoid model and document the transformation used.
Import and protect the source data
Retain the original raw data in a controlled project folder before any processing begins. This includes LiDAR files, trajectory data, GNSS observations, IMU records, base-station files, photographs, control coordinates and field notes. Raw data is the audit trail if a question arises later about coverage, accuracy or a processing decision.
Import the scan data into the chosen processing software in its native format where possible. LAS and LAZ are widely used for point-cloud exchange, while E57 is common for terrestrial laser scanning. Keep a clear naming convention for survey dates, scanner positions, flight lines and processing versions. On larger schemes, a small amount of file discipline prevents costly confusion when data is revisited months later.
At this stage, inspect basic survey completeness. Check that all planned scan positions or flight lines are present, that no files are corrupt, and that the point density and coverage broadly match the site conditions. It is better to identify a missed area before processing time is spent on a model that will need to be remade.
Georeference the survey correctly
Georeferencing places the point cloud in the project coordinate system. The method depends on the equipment and survey approach. Terrestrial scanners may use surveyed targets, cloud-to-cloud registration, or a combination of both. Mobile and UAV LiDAR systems usually rely on GNSS and IMU data, supported by RTK, PPK processing or ground control.
For GNSS-enabled workflows, review the correction source, base-station coordinates, observation quality and any reported fixed or float solutions. A fixed RTK status alone is not proof that the final point cloud meets specification. Obstructions, multipath, poor satellite geometry and an unsuitable control setup can still affect results.
Where targets or control points are used, compare the measured point-cloud coordinates against independently surveyed coordinates. Distribute control across the site rather than concentrating it in one convenient location. Control should represent the full working area, including changes in elevation where relevant.
Registration and strip alignment
Registration aligns individual scans or LiDAR strips into one coherent cloud. Software can achieve close-looking overlaps while still leaving local error that matters for survey work. Inspect hard edges, building corners, kerbs, rails and other defined features in overlap zones. These reveal misalignment more clearly than vegetation or uneven ground.
For UAV and mobile LiDAR, review strip alignment across parallel runs and at cross-lines where available. Systematic offsets may indicate trajectory issues, timing errors, IMU calibration problems or weak GNSS performance. Do not conceal these problems by applying broad adjustments without understanding their source.
Clean the point cloud without removing evidence
Raw LiDAR data contains more than the site. It can include passing vehicles, people, birds, rain artefacts, reflective surfaces, scanner noise and points beyond the useful survey area. Cleaning removes points that compromise the output while preserving data that may be needed for validation.
Use a combination of automated filters and manual review. Automated noise filters are efficient, but aggressive settings can remove legitimate narrow features such as overhead wires, fence posts, sign supports and kerb edges. This is particularly relevant for utilities, rail, highway and asset-inspection surveys, where small objects may be the reason for commissioning LiDAR in the first place.
Clip the cloud to the agreed site boundary only after confirming that no contextual data is needed outside it. A modest buffer can be useful for drainage assessment, earthworks design and modelling of site access routes.
Classify points for the intended analysis
Classification separates points into meaningful groups, such as ground, vegetation, buildings, roads, water, vehicles and infrastructure. It turns a large collection of measurements into a dataset that can be analysed and modelled efficiently.
Ground classification is often the critical stage. The algorithm identifies the likely terrain surface by assessing local elevation changes, slopes and point relationships. It performs well on open ground, but can struggle around retaining walls, steep banks, dense undergrowth, spoil heaps and built-up areas. Parameters should be suited to the terrain rather than applied as a default across every project.
Review the ground class closely at breaklines and abrupt level changes. If building points are incorrectly retained as ground, the resulting digital terrain model can create false high areas. If genuine ground is removed beneath low vegetation, the terrain model may be incomplete. In forestry, LiDAR penetration can provide valuable terrain coverage under canopy, but results depend on sensor specification, flying height, pulse density and the density of the vegetation itself.
Build the right surface model
A digital terrain model represents bare earth using ground-classified points. A digital surface model includes the visible upper surface, such as buildings, trees and vehicles. Neither is universally better; the appropriate model depends on the task.
Use a terrain model for cut and fill calculations, drainage studies, route design and contours. Use a surface model for canopy height, visibility work, roof analysis and broad site context. Grid resolution should reflect point spacing and required accuracy. An overly fine grid can imply detail that the source survey does not support, while an overly coarse grid can smooth out operationally significant features.
Check accuracy before creating final outputs
Quality assurance should be a formal stage, not a quick visual check at the end. Compare independent check points to the final cloud or derived surface, rather than only assessing the control that was used to constrain the survey. Record horizontal and vertical residuals, along with the method used to obtain them.
A practical QA review should confirm four areas:
- coordinate system, datum and units are correct;
- coverage meets the agreed survey extent;
- point-cloud registration and strip alignment are within tolerance; and
- independent check points support the stated accuracy.
Also examine the dataset at the scale at which it will be used. A cloud that appears clean when viewed across an entire site may show duplicated façades, noisy edges or gaps when a designer zooms into a kerb line or steel connection. Accuracy, precision and point density are related but different measures. High point density does not correct poor georeferencing.
Produce CAD-ready and analysis-ready deliverables
Once the cloud has passed QA, export only the formats and layers required for downstream use. A classified LAZ file is efficient for archiving and specialist analysis. CAD users may need extracted linework, breaklines, contours, sections or an agreed surface format. Asset teams may require a registered point cloud, panoramic imagery and clearly identified features for inspection planning.
Avoid supplying an unstructured, multi-gigabyte point cloud as the only deliverable unless that has been specifically requested. It transfers the processing burden to the client and may be impractical for their software or hardware. Clear layer naming, sensible clipping, a coordinate statement and a concise survey report make data easier to use and defend.
For repeat surveys, use the same control strategy, coordinate system, classification rules and reporting method where possible. Consistency makes change detection, stock monitoring and asset comparison far more reliable.
LiDAR Tech UK can support the full workflow, from selecting capture equipment and GNSS corrections through to processing and survey deliverables. The best result is not simply a detailed point cloud; it is dependable spatial information that your project team can use with confidence.