Skip to main content
v2026.11,986 entries · CC-BY 4.0

Lidar point cloud: returns, classes and data quality

Interpret lidar point cloud records through return order, surface classes, quality flags and the metadata needed for reproducible research.

Written and maintained by CASRAI Editorial Board

Published

On this page

A lidar point cloud is a collection of three-dimensional points derived from laser measurements of reflecting surfaces. Each point has spatial coordinates and may carry return, intensity and classification attributes. For research, the crucial distinction is between where a return was measured, what surface it represents and whether quality flags permit its use.

Short answer

The cloud is the point-level evidence behind many terrain and vegetation products. It can retain observations from different heights at similar horizontal locations, whereas a derived elevation grid summarizes a selected surface. A cloud containing roofs, branches and ground therefore needs interpretation before it becomes a bare-earth model.

Three fields answer different questions: return number describes a reflection’s position in a pulse’s return sequence; classification identifies an assigned surface category; flags describe additional characteristics or processing exclusions. The data table below combines these distinctions with their source documentation. Keeping them separate makes a research methods description more informative than simply reporting that lidar data were used.

Definition and scope

Lidar stands for light detection and ranging. NOAA describes an airborne system combining measured ranges with position, orientation, scan angles and calibration to locate reflecting targets. A point cloud is the resulting collection of locations. It belongs to the wider family of remote sensing methods, but its spatial records should be distinguished from an aerial photograph and from a raster elevation product.

The term describes a data representation as well as its measurement origin. ASPRS defines LAS as an open format for exchanging point clouds from lidar or other sources. A LAS extension alone consequently does not prove that a file originated from a laser scanner. The source description matters when comparing a measured laser cloud with points reconstructed from images.

The representation also has a file-level context. Point coordinates need their coordinate reference system, units and encoding to become interpretable locations. LAS records store integer X, Y and Z values with scale factors and offsets in the header. Those stored integers are not automatically the final map coordinates, and a large integer does not by itself indicate an unusually distant or high target.

How lidar point clouds work in practice

From a laser pulse to several returns

NEON describes ranging through the travel time of emitted light and its reflection. A pulse encountering vegetation may generate several recorded returns: some light reflects from foliage or branches while some continues through gaps. A solid surface can instead yield a single return. Esri’s introductory tutorial illustrates both cases.

The return fields preserve that sequence. In ASPRS point record formats 6–10, the return number identifies a point’s place among the returns from its pulse, and the number-of-returns field gives the total. A first return is therefore an ordering label. It is not a vegetation classification. Likewise, the last recorded reflection does not, merely by being last, establish a ground surface.

Coordinates, attributes and classification

ASPRS specifies reconstruction of each coordinate by multiplying the stored integer by its corresponding scale and adding the corresponding offset. This is an encoding operation. It does not establish the measurement’s accuracy. A methods record should distinguish the numerical precision retained in storage from evidence about how accurately the survey locates real surfaces.

Classification adds a separate interpretation. Standard categories in formats 6–10 include ground, vegetation, buildings and water. NEON explains that classifications are assigned during additional processing and vary among products. A file may contain usable spatial measurements without a completed surface classification; Esri’s example initially displays unassigned points.

Flags add another layer without replacing the class. ASPRS defines withheld points as excluded from processing. The overlap flag identifies points within overlapping swaths or takes while preserving their classification. A point can therefore carry a ground class and an additional flag simultaneously. Treating every label as one mutually exclusive surface category loses this distinction.

Discrete returns and full waveforms

NEON distinguishes discrete systems, which record individual return peaks as points, from full-waveform systems, which record the distribution of returned energy. These are different forms of measurement information. A downloadable point collection should not be assumed to include the original waveform simply because the acquisition system could record one.

ASPRS point record formats specify different attribute packages. Format 6 provides the core record; format 7 adds red, green and blue channels; format 8 also includes near-infrared information. Other formats support waveform-related data. Thus software compatibility depends on the point record format as well as the LAS version. A colored display does not establish that all of these fields are present.

Topographic and bathymetric lidar

NOAA distinguishes topographic lidar, typically using near-infrared light for land mapping, from bathymetric lidar, using green light to measure underwater terrain. These names concern sensing applications. They should not be confused with file formats, surface classes or the choice to display points by elevation.

Point clouds and elevation models

A digital elevation model is a derived surface representation. NOAA lists elevation, canopy and building models among products generated from point clouds. The product name and its surface definition belong in a research description alongside the source cloud. Selecting ground observations and generating a raster are analytical choices, not consequences of changing a filename.

USGS makes the distinction concrete in its 2025 revision A delivery specification: classified point data and a bare-earth raster are separate deliverables. Its terrain raster uses ground-classified points that are not withheld. That is a programme-specific requirement, useful as an explicit example of how surface labels and exclusion flags jointly determine a derivative.

Where lidar point clouds matter for research labs

Environmental research teams use point clouds to study vertical vegetation structure, while NOAA describes coastal applications including shoreline mapping and inundation modeling. The relevant research question determines which observations and derivative products matter. A canopy analysis and a bare-earth analysis can start from the same acquisition yet use different parts of the evidence.

For a laboratory or research group receiving a dataset, the practical issue is documenting the handoff between acquisition and analysis. The following evidence checklist is an editorial synthesis of ASPRS’s format definitions and the USGS mapping-report requirements. It identifies documentation to preserve rather than prescribing a universal processing procedure.

  • Coordinate interpretation: preserve the reference system and units with the cloud. LAS 1.5 requires a WKT coordinate reference description in its variable or extended variable length records.
  • Observation selection: record the return fields, classes and flags actually used. State whether exclusion flags were honored rather than leaving a filter described only as ground points.
  • Processing provenance: retain the software and relevant processing descriptions. USGS mapping reports separately document point creation, calibration, adjustment, classification and raster production.
  • Quality evidence: preserve the assessment behind a quality claim. USGS reports include density methodology and geometric assessments; a dense-looking visualization alone supplies neither.

This is a role for descriptive and technical metadata, not merely a note about a download location. An archive such as OpenTopography helps with discovery and access; the individual dataset’s documentation remains necessary to interpret its records.

Format standards and delivery requirements

ASPRS’s published specification is LAS 1.5 R00, whose introduction records approval in August 2025. It removes support for legacy point record formats 0–5 and requires WKT reference-system encoding. Those facts describe that version; they should not be retroactively imposed on every older LAS file encountered in an archive.

USGS’s 2025 revision A deliverables page separately specifies LAS 1.4 with point record formats 6–10 and compressed LAZ 1.4 delivery. The apparent version mismatch illustrates two different authorities: a format standard defines a representation, while a programme names its required deliverable. Record both when claiming conformance. The existence of a newer format does not establish that an older programme delivery is defective.

Interpreting a point record: an illustrative example

This illustrative composite is invented to explain fields, not a measurement from a real survey. Suppose a record reports return number 2, a total of 3 returns, ground classification 2 and a withheld flag. It is the second recorded return from its pulse, has been assigned the ground category and is flagged against processing. The class does not cancel the exclusion flag.

A second illustrative record might be return 3 of 3 with an unclassified label. It is last in its recorded sequence, but the record does not establish that the target was ground. Calling it a ground observation would add an interpretation not supplied by those fields. These examples follow the ASPRS field definitions; they demonstrate why return-based selection and class-based selection must be described separately.

The same reasoning applies when reviewing a methods sentence. A statement that the analysis used last returns specifies an ordering filter. A statement that it used ground-classified, non-withheld points specifies a surface interpretation plus an exclusion rule. Those descriptions answer different questions even when some selected points coincide. Neither sentence alone supplies an accuracy assessment.

Coordinate encoding in a worked example

For a separate illustrative composite, suppose the stored Z integer is 12,345, the Z scale is 0.01 and the Z offset is 100. Applying the ASPRS coordinate equation gives 12,345 × 0.01 + 100 = 223.45. The result is expressed in the coordinate system’s applicable units. It is not 12,345 units above the reference surface.

This calculation decodes a record; it neither validates a datum nor converts between reference systems. It also provides no evidence that the observed target was ground. A numerical coordinate, a surface class and an accuracy claim remain separate pieces of information even when a viewer displays them together in one scene.

Common misunderstandings

Point density is the same as pulse density or accuracy

These quantities describe different properties. USGS defines nominal pulse density using first or last returns as surrogates for pulses; counting every return can answer a different question. Accuracy concerns agreement with a reference, while relative accuracy concerns internal agreement. Report the quantity actually assessed instead of treating point count as an accuracy certificate.

Intensity is an ordinary photograph or a calibrated material property

ASPRS describes intensity as optional and system-specific; an unavailable intensity value is stored as zero. A zero therefore need not prove a physically dark target. USGS also cautions that discrete-return intensity is affected by instrument settings and viewing geometry. A grayscale intensity image should not automatically be treated as a calibrated reflectance map.

A class label guarantees that every point is correct

A class field records an assigned interpretation. It is not independent validation of that interpretation. NEON’s description of classification as additional processing, together with ASPRS’s separate exclusion flags, explains why a research report needs both its selection rules and its quality evidence. A colored cloud may be an effective inspection view, but color alone does not validate the underlying classes.

LAS and LAZ describe different landscapes

NEON identifies LAZ as a compressed representation of LAS data. Compression is a storage distinction, rather than a new ground, vegetation or water category. Record the format and version for interoperability, and use the acquisition and processing metadata to establish what landscape evidence the file contains.

Sources

Data: Lidar point cloud fields and research interpretation

An original crosswalk compiled from the ASPRS LAS 1.5 field definitions and USGS Lidar Base Specification 2025 revision A. It separates coordinate encoding, return order, surface classification and flags from interpretation of intensity and bare-earth derivatives.

Lidar point cloud fields and research interpretation
Record or product featureDocumented meaningResearch interpretationSource attribution
Coordinate encodingInteger X, Y, Z with scale factors and offsetsReconstruct coordinates using the header; storage precision is not an accuracy test.ASPRS LAS format definition
Return sequenceReturn number and number of returns for the pulseA position in a return sequence does not assign a ground or vegetation class.ASPRS LAS format definition
Surface classification2 = ground; 3–5 = vegetation; 6 = building; 9 = waterThese are assigned classes in point record formats 6–10.ASPRS LAS format definition
Withheld flagA flagged point should not be included in processingA ground class does not override the withheld flag.ASPRS LAS format definition
Overlap flagPoint lies in overlapping swaths or takesSeparate flag preserves the point classification; flagging is not mandatory under LAS alone.ASPRS LAS format definition
IntensityOptional, system-specific return magnitude; zero if unavailableZero is not proof of a dark target; discrete intensity is affected by gain and incidence.ASPRS LAS format definition; USGS glossary
Bare-earth derivativeUSGS 2025 rev. A uses ground-classified points not flagged withheldThe raster is a separate deliverable from the classified point cloud.USGS 2025 rev. A deliverables

Sources

Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · free to try

Ask about Lidar point cloud: returns, classes and data quality

2 free questions a day, no account.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →