Accuracy assessment of point clouds measured by UAV imagery in war-disaster-affected areas

Unmanned aerial vehicles (UAVs) are increasingly used for post-conflict and disaster mapping because they enable rapid acquisition of high-resolution imagery and dense 3D point clouds in hazardous or inaccessible areas. This study evaluates the accuracy of UAV-derived point clouds for Kafranbel, Syria, a 143-hectare war-damaged urban area where rubble, vegetation, and damaged structures complicate ground control placement and photogrammetric reconstruction. Four UAV missions were conducted using a DJI Mavic 2 Pro: three nadir flights at 45 m, 50 m, and 65 m altitude with approximately 80% front and side overlap, and one oblique flight at 38 m with a 70° camera angle. The missions captured 400–1000 images each. Fifty-four independent checkpoints were surveyed using a Topcon 105N total station with <2 cm accuracy and used as external reference data for validation. Images were processed in Agisoft Metashape using a structure-from-motion workflow to generate dense point clouds, digital elevation models, and orthophotos. Accuracy was evaluated by computing RMSE in X, Y, Z, and RMSE3D. The 45 m nadir flight achieved the highest accuracy, with an RMSE3D of about 0.13 m, compared with about 0.35 m at 50 m and 0.45 m at 65 m. Planimetric errors in the nadir models were generally 5–10 cm. The 38 m oblique mission improved modelling of vertical structures such as façades and walls, although horizontal precision on flat surfaces decreased to about 14 cm. Overall, a flight altitude near 50 m provided the best compromise between accuracy, coverage, and efficiency. The results recommend combining nadir and oblique imagery with a well-distributed checkpoint network for reliable 3D documentation of war-damaged urban areas.