Abstract
Natural disaster prevention requires the maintenance of the proper function of river flow during droughts and heavy rains caused by abnormal weather as well as the minimization of damage via rapid and accurate predictions. This study analyzed hydraulic characteristics by utilizing D-LiDAR (Drone LiDAR) topographic data and D-HSI (Drone Hyperspectral Image) data to quickly and accurately collect and analyze related data. The HEC-RAS model was used to compare and analyze the results obtained via the D-LiDAR topographic data of this study, the roughness coefficient of the river master plan, and the results of applying parameters by adding NDVI. As a result, in the case of a 100-year frequency flood discharge, the average velocity decreased by an average of 2.23 m/s when applying the resistance coefficient by area, based on NDVI. Moreover, the flood level increased by an average of 0.30 m, and the cross-sectional area of the flow capacity increased by an average of 20.17 m2. These results confirm that vegetation has a significant influence on the flow characteristics of natural streams. Hence, the data construction and analysis method used in this study can provide useful information regarding the rapid measurement and analysis of river flow characteristics during floods as well as the development of prediction techniques and warning-related systems for drought and flood prevention.
Introduction
Accurate flood prediction is a cornerstone of disaster prevention, yet many hydraulic models rely on "River Master Plans" that use static, generalized values for surface roughness (Manning’s n). In reality, natural streams are complex environments where vegetation acts as a significant physical barrier to flow. Mapping this vegetation manually is labor-intensive, and satellite imagery often lacks the resolution needed for small-to-medium streams. This study demonstrates a high-resolution approach to flood modeling by fusing drone-based LiDAR (D-LiDAR) and Hyperspectral Imaging (D-HSI). By using the Normalized Difference Vegetation Index (NDVI) to accurately map riverbank vegetation, the researchers address the gap where models often fail to account for the "drag" created by real-world plant life.
Methods and Data Acquisition
The researchers focused on a 370-meter reach of the Seongyeon Stream in South Korea. Their workflow integrated two cutting-edge drone technologies with standard hydraulic simulation:
- Topographic Mapping (D-LiDAR): A DJI M300 RTK equipped with a Greenvalley LiDAR sensor was used to create a precise Digital Terrain Model (DTM). This provided the "skeleton" of the river's geometry.
- Vegetation Mapping (D-HSI): A DJI M600 carrying a Headwall Hyperspectral Camera captured 273 spectral bands. From this data, the team calculated the Normalized Difference Vegetation Index (NDVI) to identify exactly where vegetation was growing and how dense it was.
The study primarily utilized original data collected on-site, including drone-based LiDAR point clouds for DTM generation and 273-band hyperspectral data for NDVI mapping. The Seongyeon River Master Plan (2009) provided the baseline boundary conditions and standard roughness values for comparison.
Hydraulic Simulation and Results
The authors used HEC-RAS (v6.5) to simulate a 100-year frequency flood, comparing a baseline model (uniform roughness n=0.030) with an NDVI-informed model (n=0.045 for vegetated areas).
The comparison revealed that vegetation acts as a "hydraulic brake," significantly altering flow behavior:
- Velocity Reduction: The average flow velocity decreased by an average of 2.23 m/s when vegetation was included.
- Flood Level Increase: The predicted flood level (stage) rose by an average of 0.30 m (with a maximum increase of 0.44 m at certain stations).
- Flow Area Expansion: The cross-sectional area of the flow increased by an average of 20.17 m² to accommodate the slower-moving water.
Discussion and Conclusion
The most critical insight is that vegetation is a major energy sink. Traditional models that ignore vegetation might predict high flow velocities, which can lead to incorrect assumptions about bridge scour and bank erosion, while simultaneously under-preparing for the actual height of the floodwaters. Sensor fusion is key: LiDAR provides the geometry, while Hyperspectral provides the surface characteristics.
While the study proves the efficacy of the D-LiDAR/D-HSI fusion, future work should address seasonality—as NDVI values change throughout the year—and species-specific roughness. The data construction and analysis method used in this study provide a framework for rapid measurement and analysis of river flow characteristics, essential for developing robust drought and flood prevention systems.
Summary updated