Towards scalable, GCP-free UAV photogrammetry using PPP-RTK: repeatability tests at a Baltic Sea cliff site (Wustrow, Germany)

A major limitation of using UAV photogrammetry for coastal erosion and geohazard assessment is the effort associated with GCP-based georeferencing. To optimize this, we evaluate the German national PPP-RTK (precise point positioning real-time kinematic) service GEPOS®, provided by the Federal Agency for Cartography and Geodesy (BKG). The focus is on its ability to enable repeatable, GCP-free, and more time-efficient UAV-derived 3D products, suitable for operational upscaling.

In total, we conducted 42 UAV surveys with three platforms (two DJI Mavic 3 Multispectral; RGB imagery only, and one DJI Mavic 3 Enterprise) in 10/2025, covering both nadir-only and multi-view oblique acquisition geometries at an actively eroding cliff section near Wustrow (Baltic Sea coast). Here a former East German military bunker collapsed in February 2024, thus providing a benchmark. GEPOS® corrections were generated by the BKG using three dedicated access points located within the area of interest and at distances of approximately 350 m and 1000 m. This setup enabled an initial sensitivity assessment of repeatability as a function of the correction-access configuration.

In the absence of independent geodetic reference, we focused on repeatability (relative precision). To this end, we compared point clouds and DSMs across near-contemporaneous surveys to avoid impact of surface change. We applied a two-stage evaluation strategy: (i) we quantified end-to-end repeatability from unaligned model-to-model differences, reflecting combined georeferencing and photogrammetric effects; and (ii) removed rigid offsets by masked ICP-based co-registration on a stable concrete reference surface, transferring the derived rigid-body transform to the entire dataset before reassessing residual differences. We assessed distances using M3C2 and robust summary statistics such as Normalized Median Absolute Deviation (NMAD) and 95^th and 99^th percentiles of absolute M3C2 distances after excluding change-prone areas.

Preliminary results indicate that oblique acquisition achieves centimeter-level repeatability without co-registration and improves to around the centimeter scale after co-registration. In contrast, nadir-only surveys show substantially larger inter-model discrepancies prior to co-registration, consistent with predominantly rigid offsets, but converge to low-centimeter residuals after alignment on stable surfaces. The final analysis will quantify repeatability across platforms, acquisition geometries, and correction-access configurations, and evaluate the implications for scaling UAV-based coastal monitoring while minimizing field effort.