Comparative Analysis of UAV-Derived DEMs in Urban Terrain Mapping

Thesis Abstract

Urban terrain modeling relies on accurate digital elevation models (DEMs) generated from unmanned aerial vehicle (UAV) data to support planning, flood risk assessment, and infrastructure management in rapidly developing cities. Despite extensive UAV-based DEM production, there is limited systematic comparison of DEM accuracy, noise characteristics, and vertical fidelity across urban contexts that differ in building density, materials, and occlusion patterns. This study aims to evaluate and compare UAV-derived DEMs produced from multi-sensor data fusion in diverse urban environments to establish best practices for accuracy, reliability, and applicability in urban geomatics. The specific objectives are (1) to quantify vertical accuracy and noise characteristics of DEMs derived from RGB, multispectral, and LiDAR-enabled UAV surveys; (2) to assess the influence of flight parameters (altitude, overlap, and Ground Sampling Distance) and processing workflows (aerial triangulation, dense matching, and LiDAR-informed interpolation) on DEM quality; (3) to compare DEMs against a high-precision reference dataset obtained from terrestrial LiDAR scans and conventional total station measurements; (4) to examine the performance of error models under different urban morphologies using statistical tests and regression analysis; and (5) to formulate actionable guidelines for selecting UAV-derived DEM generation strategies in dense, medium-density, and mixed-use urban areas. The methodological framework combines a cross-sectional, multi-site design with a mixed-methods analysis, anchored in the theory of sensor fusion and error propagation. The population comprises urban sites in a major metropolitan region characterized by heterogeneous built forms. A stratified sampling approach selects three districts representing high-rise central areas, mid-rise residential zones, and low-rise peri-urban belts, with 15 to 20 UAV flight campaigns per site over a six-month period. Data collection integrates (i) UAV-acquired imagery (RGB and multispectral) at 60–90 m AGL with 80% forward and side overlap; (ii) LiDAR-derived point clouds collected with a terrestrial mobile cart system and fixed-wing UAV LiDAR where feasible; and (iii) ground truth measurements from static GNSS survey points and terrestrial laser scanning (TLS) benchmarks totaling 120 control points per site. DEM generation employs three workflows (a) multi-view stereo dense matching for RGB-derived DEMs, (b) structure-from-motion with scale refinement using ground control points, and (c) LiDAR-assisted interpolation to produce fused DEMs. Validity and reliability are evaluated through RMSE, MAE, and standard deviation of elevation residuals relative to TLS benchmarks; statistical methods include ANOVA to test differences among urban morphologies, multiple regression to model the influence of flight and processing variables on vertical accuracy, and Bland-Altman analyses to assess agreement with reference data. Hypothesis testing investigates whether LiDAR-assisted DEMs significantly outperform pure photogrammetric DEMs across all sites and whether higher GSD improves accuracy in high-occlusion zones. Anticipated findings indicate that LiDAR-informed fusion DEMs exhibit lower RMSE (expected range 0.10–0.25 m) compared with RGB-only DEMs (0.25–0.60 m) in dense urban cores, with marginal gains in low-density areas. The study also expects that closer flight heights (60 m AGL) and higher image overlap reduce vertical bias but increase processing time. The contribution to knowledge includes a systematic, cross-urban-context assessment of UAV-derived DEMs, a quantified understanding of error propagation under urban occlusion, and a robust, context-sensitive set of recommendations for practitioners on sensor selection, flight planning, and processing workflows. The research will inform urban planning and disaster risk management by providing validated guidance on the accuracy and reliability of UAV-based DEMs for flood modeling, line-of-sight analysis, and surface-based infrastructure inventory. The main conclusion anticipates that an LiDAR-informed fusion approach offers the most reliable DEMs for dense urban terrain, with photogrammetric methods augmented by low-altitude flights serving as viable alternatives when LiDAR data are unavailable. Recommended guidelines emphasize site-specific workflow selection, rigorous ground control deployment, and transparent reporting of accuracy metrics to support policy-making and standardized urban geoinformatics practices.