Gou et al. (2026) Development of a stream DTM generation methodology using UAV-based SfM and LiDAR point cloud

Identification

• Journal: Scientific Reports
• Year: 2026
• Date: 2026-01-13
• Authors: Jaejun Gou, Hyeokjin Lee, Jinseok Park, 장성현, N Lee, Inhong Song
• DOI: 10.1038/s41598-026-35473-x

Research Groups

• Department of Rural Systems Engineering, Global Smart Farm Convergence Major, Seoul National University, Seoul 08826, South Korea
• Department of Electrical and Computer Engineering, Seoul National University, Seoul 08826, South Korea
• Research Institute of Agriculture and Life Sciences, Seoul National University, Seoul 08826, South Korea

Short Summary

This study developed a methodology for generating high-precision Digital Terrain Models (DTMs) in stream environments by integrating UAV-based Structure from Motion (SfM) and LiDAR point clouds with ground filtering, finding that the Simple Morphological Filter (SMRF) achieved the best overall accuracy (mean absolute error of 0.160 m, root mean square error of 0.214 m).

Objective

• To propose and validate a robust methodology for generating high-precision Digital Terrain Models (DTMs) in stream areas by integrating UAV-based Structure from Motion (SfM) and LiDAR point clouds.
• To implement an automated water area classification using the Normalized Difference Water Index (NDWI) to leverage the strengths of SfM for waterbed points and LiDAR for penetrating riparian vegetation.
• To evaluate and optimize the performance of various ground-filtering algorithms to identify the most effective strategy for accurate terrain modeling in complex riparian environments.

Study Configuration

• Spatial Scale: A 2.8 km long, approximately 250 m wide, 3.95 km² reach of the Bokha Stream in Icheon City, South Korea.
• Temporal Scale: Drone and Electronic Distance Measurement (EDM) surveys were conducted from February 26, 2024, to February 28, 2024.

Methodology and Data

• Models used:
  • Normalized Difference Water Index (NDWI) for water/non-water classification.
  • Ground filtering algorithms: Cloth Simulation Filter (CSF), Progressive Triangular Irregular Network (PTIN), Simple Morphological Filter (SMRF).
  • Triangulated Irregular Network (TIN) interpolation for DTM generation.
• Data sources:
  • UAV-based LiDAR point clouds (Zenmuse L1 on Matrice 300 drone).
  • UAV-based Structure from Motion (SfM) point clouds (Phantom 4 multispectral drone).
  • Ground Control Points (GCPs) measured with Real-Time Kinematic (RTK) GPS (Trimble GeoXR).
  • Ground-truth cross-sections measured with Electronic Distance Measurement (EDM) (Topcon ES-52) and GPS.

Main Results

• The NDWI-based water filtering achieved a Type 1 error of 1.2%, a Type 2 error of 3.0%, and a total error of 2.7%.
• The Simple Morphological Filter (SMRF) demonstrated the best overall accuracy for the integrated LiDAR-SfM DTM, achieving a mean absolute error (MAE) of 0.160 m and a root mean square error (RMSE) of 0.214 m across all areas.
• SMRF performed particularly well in vegetation areas (MAE = 0.160 m, RMSE = 0.207 m), effectively removing low shrubs and preserving riparian terrain details.
• The Cloth Simulation Filter (CSF) showed the highest accuracy in ground (MAE = 0.114 m, RMSE = 0.146 m) and water areas (MAE = 0.213 m, RMSE = 0.253 m) but was less effective in removing vegetation.
• SMRF exhibited a tendency to underestimate elevations in water areas due to sensitivity to outliers, while CSF was more robust to these error points.
• All filters struggled to accurately represent artificial structures like bridges, often retaining bridge decks as part of the terrain.

Contributions

• Proposed and validated a novel methodology for generating high-precision stream DTMs by integrating UAV-based SfM and LiDAR point clouds.
• Implemented an automated water area classification using NDWI, effectively combining the strengths of SfM for waterbed mapping and LiDAR for vegetation penetration.
• Conducted a comprehensive evaluation and parameter optimization of three widely used ground-filtering algorithms (CSF, PTIN, SMRF) in complex riparian environments.
• Identified SMRF as the most effective filter for the integrated approach, providing specific accuracy metrics and insights into its performance across different terrain types.
• Demonstrated the potential of the proposed method to enhance DTM accuracy for mesoscale hydrological and ecological modeling in small streams, particularly during periods of low vegetation vitality.

Funding

• National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. 2022R1F1A1075128)

Citation

@article{Gou2026Development,
  author = {Gou, Jaejun and Lee, Hyeokjin and Park, Jinseok and 장성현 and Lee, N and Song, Inhong},
  title = {Development of a stream DTM generation methodology using UAV-based SfM and LiDAR point cloud},
  journal = {Scientific Reports},
  year = {2026},
  doi = {10.1038/s41598-026-35473-x},
  url = {https://doi.org/10.1038/s41598-026-35473-x}
}