Methodology & Formula Reference

AccuracySuite computes positional accuracy metrics from user-supplied checkpoint coordinates. It is a numbers-only tool — it does not process imagery, point clouds, or 3D meshes. It operates exclusively on coordinates and residuals, computing statistics aligned to ASPRS Edition 2.

ASPRS Positional Accuracy Standards for Digital Geospatial Data, Edition 2 (published 2023; Version 2 update 2024) defines how to report positional accuracy for geospatial datasets. It specifies:

• Accuracy classes based on RMSE values
• The exact language for "Tested to meet" and "Produced to meet" statements
• Checkpoint count requirements (≥30 for "Tested")
• CE90/LE90 confidence conversions
• Non-vegetated and vegetated vertical accuracy (NVA/VVA)

Root Mean Square Error for each axis:

RMSEx = √( Σ(x_model - x_truth)² / n )
RMSEy = √( Σ(y_model - y_truth)² / n )
RMSEz = √( Σ(z_model - z_truth)² / n )

Horizontal radial RMSE:
RMSEr = √( RMSEx² + RMSEy² )

Where n = number of independent checkpoints, and residuals are (model - truth) at each checkpoint.

CE90  = 1.5175 × RMSEr     (Circular Error at 90% confidence)
LE90  = 1.6449 × RMSEz     (Linear Error at 90% confidence)
NVA   = 1.9600 × RMSEz     (Non-Vegetated Vertical Accuracy at 95%)

CE90: 90% of horizontal positions are within this radius of their true positions. LE90: 90% of vertical positions are within this distance. NVA: 95% confidence for non-vegetated terrain.

These conversion factors assume residuals approximate a normal distribution (circular for horizontal, Gaussian for vertical).

The ASPRS accuracy class is determined by the computed RMSE value. The class is the nearest standard class value ≥ the computed RMSE. Standard accuracy classes include:

ASPRS Edition 2 distinguishes two levels of reporting:

"Independent" means the checkpoints were NOT used to constrain the photogrammetric model (i.e., they are not GCPs). "Higher accuracy" means the checkpoint survey accuracy exceeds the target accuracy class.

AccuracySuite identifies potential blunders as checkpoints whose radial residual exceeds 3× the median radial residual of the dataset. These may indicate:

• Misidentified checkpoint location
• Point ID mismatch between truth and model datasets
• CRS or datum mismatch at specific points
• Genuine positional error (processing artifact)

Blunders are flagged but included by default. The user must explicitly exclude them.

Correct CRS and unit selection is critical. Common pitfalls:

• US Survey Feet vs International Feet: Differ by 2 ppm (1 US ft = 1200/3937 m; 1 int'l ft = 0.3048 m exactly). This is enough to affect accuracy class assignment.
• Ellipsoidal vs Orthometric heights: Mixing these produces meaningless vertical residuals. NAVD88 requires a geoid model (e.g., GEOID18) to convert from ellipsoidal.
• Geographic vs Projected CRS: Lat/lon coordinates cannot be directly differenced for RMSE without reprojection.

Every report stamps two version identifiers:

• ASPRS Edition: The standard edition used (currently Edition 2, 2024).
• Formula Bundle Version: The specific implementation version of our classification module (currently v1.0).

This ensures any report can be reproduced and audited, and survives future standard revisions.

• Results are limited by the accuracy of the supplied checkpoint coordinates.
• CE90/LE90 assume residuals approximate a normal distribution.
• The tool does not assess whether checkpoints are truly independent of the model.
• No assessment of spatial distribution quality (ASPRS recommends well-distributed checkpoints).
• Not a substitute for professional surveyor judgment on network design.

• ASPRS Positional Accuracy Standards for Digital Geospatial Data, Edition 2 (2023; Version 2, 2024). American Society for Photogrammetry and Remote Sensing.
• ASPRS Guidelines for Reporting Positional Accuracy (supplementary materials).
• NGS GEOID18 — National Geodetic Survey geoid model for NAVD88 orthometric heights.