Quantitative assessment of 2D-3D thematic convergence in UAV-based terrain inventory

While geometric accuracy, expressed as Root Mean Square Error (RMSE), serves as the primary benchmark in Unmanned Aerial Vehicle (UAV) photogrammetry, thematic consistency between planar (2D) and volumetric (3D) deliverables remains insufficiently quantified. This study investigates the "accuracy paradox" questioning the assumption that high georeferencing precision inherently ensures thematic reliability. Research was conducted over a 22.6-hectare rural site in southern Poland. Using a UAV photogrammetric model (3 cm Ground Sampling Distance (GSD), 0.136 m RMSE), independent manual land-use delineations were performed on a high-resolution orthomosaic and a dense point cloud. To evaluate discrepancies, the Convergence Index (CI) was introduced, quantifying relative area differences between representations. Results indicate a total thematic discrepancy of 2.4% (0.5419 ha). However, errors were non-uniformly distributed: infrastructural classes showed near-perfect agreement (CI < 1%), whereas forested areas exhibited CI values exceeding 6%. 2D projections tended to overestimate forested areas, creating a false horizontal expansion of tree canopies. Furthermore, a strong negative correlation (Pearson r = –0.95, p < 0.001) was found between image overlap redundancy and thematic divergence. The study concludes that object morphology and 3D reconstruction completeness exert a more significant impact on inventory reliability than nominal georeferencing accuracy. For high-precision cadastral and environmental surveying, integrating 3D point clouds may be necessary, particularly in vegetation-dominated environments, to mitigate systematic distortions inherent in 2D orthomosaic projections.