RTK, GCP and Checkpoint Accuracy Validation in Agisoft Metashape

Begin with the decision the final product must support.

• Identify the deliverable being validated: camera solution, point cloud, orthomosaic, DEM, mesh, measured feature or another product.
• State whether the requirement concerns absolute position, relative geometry, repeatability or change detection.
• Separate horizontal and vertical requirements where the project specification does so.
• Record the required units, reporting statistic, confidence convention and maximum allowable error where applicable.
• Identify the contractual, survey, scientific or internal standard governing acceptance.
• Define who is qualified and authorized to approve the result.
• Avoid deriving the acceptance threshold after seeing the processing result.

Required record: a written accuracy and acceptance specification linked to a defined deliverable.

A consistent numerical result can still be wrong if the reference definition is inconsistent.

• Confirm the coordinate reference system used by camera positions, GCPs and checkpoints.
• Verify EPSG or custom CRS definitions rather than relying only on a project name.
• Determine whether heights are ellipsoidal, orthometric or referenced to a local datum.
• Identify the geoid model or vertical transformation required for the project area.
• Check units, axis order, decimal separators and coordinate-column mapping during import.
• Confirm antenna, sensor or lever-arm offsets when they are relevant to the coordinate source.
• Test at least one known point after transformation and before interpreting residuals.

Required record: source, project and output CRS definitions, including vertical datum and transformation resources.

Do not treat every coordinate tagged as RTK as equally reliable.

• Record the aircraft, camera, GNSS receiver and correction workflow used.
• Distinguish RTK, PPK, autonomous GNSS and coordinates modified by another application.
• Inspect solution status, timestamps, base or network information and any quality fields available.
• Confirm the datum, height convention and coordinate epoch where relevant.
• Verify whether coordinates refer to the antenna, camera centre or another physical point.
• Review whether per-camera accuracy estimates are available and meaningful.
• Look for systematic offsets by flight, altitude, direction, battery or acquisition session.
• Retain the original metadata and any processing log that produced corrected coordinates.

Required record: a camera-reference audit stating coordinate source, correction status, datum, offsets and adopted uncertainty.

Spatial coverage and geometric strength matter more than a point count in isolation.

• Distribute control points across the site rather than concentrating them near the centre.
• Represent boundaries, corners, elevation range and areas where accuracy matters operationally.
• Reserve checkpoints before optimization and keep their role documented.
• Avoid locating every checkpoint close to a GCP or in only one portion of the project.
• Use stable, unambiguous targets that can be identified at the image resolution available.
• Consider terrain, vegetation, occlusion, reflective surfaces and access constraints.
• For corridor, façade, vertical or mixed aerial-terrestrial projects, design the network for the actual capture geometry.
• Document points that were excluded and the technical reason for exclusion.

Required record: a map and point register showing control/check roles, distribution, survey source and point condition.

A precise survey coordinate cannot correct an imprecise image observation.

• Confirm each marker identifier matches the correct surveyed point.
• Inspect target visibility, sharpness, contrast and obstruction in every relevant image.
• Review projections at sufficient zoom and correct misplaced observations.
• Use several geometrically useful images rather than relying on repeated views from a similar direction.
• Check for motion blur, rolling-shutter effects, shadows or target deformation.
• Remove uncertain projections rather than forcing a measurement where the target is not identifiable.
• Document manual changes and maintain consistent operator rules where several people mark points.
• Reinspect markers showing unusual residual magnitude or direction after optimization.

Required record: a reviewed projection set with traceable corrections and notes for ambiguous or excluded observations.

Accuracy settings express trust in the observations; they are not a target result.

• Base camera and marker accuracy values on survey method, equipment records and coordinate provenance.
• Use separate horizontal and vertical values when the measurement process supports different uncertainty.
• Avoid unrealistically small values that force the solution to fit reference observations beyond their credible precision.
• Avoid broad default assumptions when project-specific evidence is available.
• Apply individual accuracies where certain observations differ materially from the rest.
• Record every change to reference accuracy settings and why it was made.
• Evaluate sensitivity when uncertainty in the adopted weighting could materially affect the result.
• Do not alter weights solely to make the reported residual appear smaller.

Required record: an observation-accuracy table supported by survey evidence and a change log.

Preserve a baseline and understand what each adjustment changes.

• Save or duplicate a baseline project before reference and calibration changes.
• Review camera alignment, tie-point distribution and weakly connected areas before interpreting reference error.
• Confirm calibration groups and sensor models reflect the actual equipment and acquisition sessions.
• Enable only the intended control observations; verify checkpoints remain excluded from adjustment.
• Apply one documented group of changes at a time and compare the result with the previous stage.
• Inspect camera, marker and reprojection residual patterns rather than only aggregate values.
• Investigate systematic direction, flight-line grouping or elevation trends.
• Avoid excessive parameter freedom that is unsupported by image geometry or calibration evidence.
• Record software version, relevant settings and the operator responsible for approval.

Required record: a staged optimization log with baseline, changes, observations and approved final state.

Interpret the distribution of errors, not only one summary value.

• Confirm checkpoints were disabled as control throughout the final adjustment.
• Review horizontal, vertical and three-dimensional residual components as appropriate.
• Calculate or report the statistic required by the project specification.
• Inspect mean error or directional bias that may be hidden by an aggregate magnitude.
• Identify maximum residuals and investigate outliers without automatically deleting them.
• Compare interior, boundary, high, low and operationally critical areas.
• Assess whether the number and spatial distribution of checkpoints support the scope of the conclusion.
• Distinguish a failed checkpoint measurement from evidence of a local model problem.
• Repeat the analysis after any change that affects alignment, calibration, reference or geometry.

Required record: checkpoint residuals, summary statistics, distribution map, outlier decisions and comparison with acceptance criteria.

State exactly what has—and has not—been demonstrated.

• Identify the validated deliverable, project area and coordinate reference system.
• List the control and checkpoint observations used, including survey source and stated accuracy.
• Report horizontal and vertical results separately when relevant.
• Include the statistic, units, point count and any confidence convention used.
• Provide maps or tables sufficient to reveal the spatial distribution of residuals.
• Describe excluded points, outliers, manual edits and deviations from the planned protocol.
• Record the Metashape version and processing state to which the result applies.
• Disclose weak coverage, limited elevation range, small validation samples or other constraints.
• Avoid extending a local validation conclusion beyond the area and products actually tested.

Required record: a signed or approved validation statement with results, scope, evidence and limitations.