10 lessons for better UAV mapping using GPS/GNSS techniques

Direct image georeferencing with GNSS reduces importance and dependence of GCPs

Integrating RTK and PPK GNSS receivers into UAV platforms can reduce the number of required ground control points (GCPs). At the same time it will make surveying in hardly-accessible or inaccessible locations possible.

But: if you are surveyor, you will almost always want to have at least 4 GCPs.

GCPs are important for long-baseline corrections

When using corrections from distant reference stations, GCPs are necessary to achieve surveying georeferencing precision, particularly in vertical direction. Without GCPs, model errors can exceed 20–50 cm depending on method and terrain.

Short-baseline PPK results in high-precision

Base-drone baseline distance should be short. Using a local GNSS base station near the measurement area will result in higher-precision of georeferences photos.

Long-baselines are uncertain

Precision in GNSS-corrected UAV mapping decreases as the baseline distance and elevation difference between base station and rover increase. Long-baseline reference stations introduce errors, especially in vertical (Z) direction.

GNSS sampling interval is important

Reference station data should be logged at a 1-second interval. Less frequent logging (above 15 seconds) is not recommended.

Position correction method impacts project workflow

RTK offers real-time correction during flight, PPK relies on post-processing of logged GNSS data. Both methods can achieve the same precision level, but differ in labor, equipment, and operational requirements.

GNSS receiver specifications matter

Centimeter-level precision requires multi-frequency receivers handling at least L1 and L2 signals.

Photogrammetric model precision is multifactorial

Image quality, camera calibration, flight plan, SfM photogrammetry algorithms, and ground surface texture all contribute to model quality. Standardization across datasets is essential for precision assessment.

Mean RMS error statistic for a model is not enough

Reporting a single mean RMS error value across all mapped surface types can mask significant variability. Surface-specific error analysis should be performed to transparently communicate project outcomes.

Error varies by surface type (ground, forest, construction)

Photogrammetric and GNSS auncertainty depends on the surface being mapped: solid textures like roads and bare ground yield the lowest errors, while trees and shadowed areas introduce larger elevation and alignment errors.