Best Practices for Landmarking and Uncertainty Management in QuAnTeTrack • QuAnTeTrack

Introduction

This vignette complements the main QuAnTeTrack vignette by providing practical guidance for quantifying, controlling, and reporting uncertainty throughout the landmarking-to-analysis pipeline. The focus is on decisions that directly affect computed trajectories and movement parameters (e.g., turning angles, step lengths, sinuosity, velocity) and on transparent practices that make results reproducible and comparable.

While QuAnTeTrack implements analysis, simulation, and visualization tools, the automation of uncertainty quantification and propagation (e.g., automated repeat-digitization audits or error-aware resampling) is outside the scope of the current package. Nonetheless, this document outlines concrete procedures that users can follow before running downstream analyses. If community feedback indicates this would help new users, an automated uncertainty-quantification workflow can be considered for future versions.

Why Uncertainty Matters and Sources of Uncertainty

Objective, quantitative methods for trackway analysis do not imply false precision if the inherent uncertainty is explicitly measured, reported, and managed. Landmarking and geometric morphometrics have proven highly useful across biology and paleontology, but their reliability depends on transparent handling of error. Without careful consideration, uncertainty can accumulate from multiple sources and bias biological interpretations. Below, the main sources of uncertainty are outlined, together with examples of how they may affect results.

Anatomical definition of landmarks
If the anatomical basis for placing a landmark is vague or inconsistently applied, different samplers may digitize slightly different points. For example, defining a landmark at the “tip of the digit” may be ambiguous in a poorly preserved track where the digit outline is rounded. Such ambiguities inflate variance and reduce reproducibility.
Sampler/observer effects
Experience, training, and even fatigue of the sampler influence results. Intra-observer error (the same person digitizing multiple times) can lead to small but systematic differences, while inter-observer error (different people digitizing the same footprint) may reveal much larger discrepancies. Both affect the precision of derived metrics such as step length or pace angulation.
Preservation and sediment effects
Substrate conditions strongly influence footprint shape. Erosion can erase edges, infill can distort depth, and plastic deformation can shift toe positions. Landmarks digitized in these areas may not reflect true foot anatomy, and comparisons across trackways with different preservation states can be misleading if not carefully controlled.
Image acquisition and scaling
Digital recording methods—photographs, photogrammetry, laser scanning—introduce their own biases. Lens distortion, uneven resolution, or shadows can obscure features. If scale bars are tilted or not aligned with the footprint plane, all subsequent measurements (distances, velocities) are biased. Even small scaling errors propagate through the entire pipeline.
Georeferencing and orientation
When trackways are digitized from field images or slabs, orientation relative to north, slope, or bedding planes must be set. Misalignment alters calculated movement directions and turning angles. Inconsistent or incorrect reference systems can make comparisons across trackways invalid.
Interpolation of missing footprints
When parts of a trackway are reconstructed (e.g., a missing footprint inferred by stride length), uncertainty arises because the position is model-based. Overconfidence in interpolated points can bias trajectory smoothness or sinuosity measures.
Coordinate handling and unit conversion
Errors in coordinate systems (e.g., mixing cm and mm, or incorrect CRS transformations) propagate into all derived parameters. Even rounding during resampling may bias small-scale metrics such as step variability.

These uncertainties highlight the importance of integrating error evaluation into the workflow rather than treating digitized landmarks as exact data.

Best Practices for Dealing with Uncertainty in Landmarking

Anatomical definition of landmarks

Provide a precise, objective, replicable anatomical definition for each landmark.
Maintain written protocols and illustrated guides.
Run multiple rounds of digitization by the same sampler to evaluate internal consistency.

Sampler and experience

Involve multiple independent samplers with varied expertise.
Each sampler performs repeat digitization rounds to quantify inter-observer variability.
Use training/calibration sessions to harmonize criteria and reduce bias.

Preservation and sediment effects

Identify zones most affected by erosion, deformation, or infill.
Prioritize best-preserved anatomical regions when possible.
Where ambiguity exists, repeat digitization on alternative anatomical locations (e.g., heel vs. toe) to assess robustness.
Simulate uncertainty by randomly jittering landmarks within a realistic tolerance envelope (e.g., ±2–5 mm) and re-running analyses.

Image acquisition and scaling

Prefer high-resolution, calibrated methods (controlled photography, photogrammetry, laser scanning).
Ensure scale bars are in-plane with the surface and clearly visible.
Correct optical distortion; avoid oblique angles.
Regularly verify scale against known distances on the slab.

Georeferencing and orientation of trackways

Establish orientation relative to external references (north, bedding planes, slope).
Apply consistent coordinate systems across trackways to make turning angles comparable.
Where orientation is uncertain, digitize alternative alignments and test sensitivity of derived metrics.

Data management and documentation

Keep a digitization log (dates, observers, devices, versions, settings).
Save raw images/models, .TPS files, and scaling notes; record any interpolations done.

Recommended Workflow with QuAnTeTrack (for quantifying and controlling uncertainty)

Protocol definition
- Finalize anatomical criteria and imaging protocols (camera/scanner setup, scale placement, lighting).
Digitization rounds
- Perform repeated digitization of the same footprints by the same sampler (intra-observer error).
- Repeat with multiple independent samplers (inter-observer error).
Preservation assessment
- Identify best-preserved footprints for primary analyses.
- Where preservation is poor, create multiple plausible landmark sets (e.g., toe-focused vs. heel-focused) and compare outcomes.
Image and scaling verification
- Validate scaling against known in-field/slab distances.
- From photographs, test distortion by re-digitizing scale-bar points and comparing results.
Orientation and georeferencing
- Standardize all trackways to a common orientation framework.
- When orientation is uncertain, run parallel analyses using alternative plausible orientations and compare impacts on turning angles and trajectories.
Uncertainty simulation
- Apply random jitter to landmark coordinates within a plausible error range (e.g., ±2–5 mm); re-run parameter extraction and analyses.
- Compare standard deviations of outputs (sinuosity, turning angles, velocities) across replicates to quantify robustness.
Integration into analyses
- Use averaged parameters (means across repeated rounds) in the main analysis.
- Report variability (SD, range, or confidence bounds) alongside point estimates.
Transparent reporting
- Document who digitized what, when, and how, and provide uncertainty summaries in Methods/Appendices.
- When possible, share raw data (images or 3D meshes), .TPS, and protocol files.

How This Relates to the Package Documentation

The help page for tps_to_track() includes a detailed checklist of potential uncertainty sources and corresponding mitigation/quantification steps, aligned with the best practices above.

Throughout the vignette examples (simulation, similarity metrics, intersections, clustering), users are encouraged to repeat analyses over replicated or jittered datasets and to report dispersion of results.

Notes on Downstream Analyses

Direction and velocity tests: run on repeated/jittered digitizations; compare the spread of p-values/effect sizes.
Similarity metrics (DTW/Fréchet): test sensitivity to alignment choices (origin/centroid) and replicate datasets.
Intersection counts: evaluate whether conclusions (greater/fewer intersections than expected) are stable under scaling/orientation variants and jitter.
Clustering: when using cluster_track(), consider running the pipeline on replicate matrices (from repeated/jittered digitizations) and summarizing cluster stability (e.g., co-assignment frequencies).

Limitations and Future Directions

Automated, end-to-end uncertainty audits (e.g., built-in repeated-digitization managers, automated jitter-resampling with summary reports) are not part of the current release. If users and reviewers consider such tooling beneficial for non-specialists, a future version could include utilities to automate repeated-digitization aggregation, jitter-based sensitivity analyses, and standardized uncertainty reports.

References and Selected Reading

Zelditch, M. L., Swiderski, D. L., & Sheets, H. D. (2021). Geometric Morphometrics for Biologists: A Primer (3rd ed.). Academic Press.
Rohlf, F. J., & Slice, D. (1990). Extensions of the Procrustes method for the optimal superimposition of landmarks. Systematic Zoology, 39(1), 40–59.
Klingenberg, C. P., & McIntyre, G. S. (1998). Geometric morphometrics of developmental instability: analyzing patterns of fluctuating asymmetry with Procrustes methods. Evolution, 52(5), 1363–1375.
Bookstein, F. L. (1991). Morphometric Tools for Landmark Data. Cambridge University Press.
Dryden, I. L., & Mardia, K. V. (2016). Statistical Shape Analysis (2nd ed.). Wiley.
Slice, D. E. (2007). Geometric morphometrics. Annual Review of Anthropology, 36, 261–281.
Batschelet, E. (1981). Circular Statistics in Biology. Academic Press.
Benhamou, S. (2004). How to reliably estimate the tortuosity of an animal’s path: straightness, sinuosity, or fractal dimension? Journal of Theoretical Biology, 229(2), 209–220.
Alexander, R. M. (1976). Estimates of speeds of dinosaurs. Nature, 261(5556), 129–130.
Ruiz, J., & Torices, A. (2013). Humans running at stadiums and beaches and the accuracy of speed estimations from fossil trackways. Ichnos, 20(1), 31–35.

Additional references
- Fruciano, C. (2016). Measurement error in geometric morphometrics. Development Genes and Evolution, 226(3), 139–158. https://doi.org/10.1007/s00427-016-0537-4
- Robinson, C., & Terhune, C. E. (2017). Error in landmark-based geometric morphometrics: Testing the assumption of homology in fossils. American Journal of Physical Anthropology, 132(2), 301–310. https://doi.org/10.1002/ajpa.20616

Acknowledgments

These guidelines were shaped by feedback from reviewers and users seeking clear, field-ready practices for managing uncertainty from digitization to inference. They are intended to help users implement robust pipelines with transparent error reporting in QuAnTeTrack.