Development of an Automated Riverbed Construction Using Airborne Bathymetric LiDAR Reflecting Depth-Dependent Characteristics in Large-Scale River Basins

Abstract

Airborne bathymetric LiDAR (ABL) has emerged as a highly effective technology for acquiring high-resolution riverbed information acrosswide riversystems. Thisstudy aimsto analyze the distribution characteristics of ABL point clouds with respect to varying water depths and evaluate the feasibility of applying existing geomorphological filtering algorithms to automate the construction of riverbed information over large-scale river networks. Unlike previous studies that focused on single locations or limited river sections, this study demonstrates the applicability and processing efficiency of the proposed approach across an entire basin featuring diverse depth conditions and complex channel morphologies. To achieve this, simultaneous airborne LiDAR surveys were conducted using near-infrared (NIR) and green lasersensors.The acquired point cloudswere integrated to generate isobaths, from which bathymetric maps and digital elevation models (DEMs) of the riverbed were constructed. The results confirm the potential of ABL forlarge-area, automated riverbed modeling and highlight its applicability in supporting river planning and management, as well as in establishing digital frameworks for flood risk assessment and disaster response.

1. Introduction

With the increasing frequency of extreme rainfall events driven by climate change, riverbed changes are having a direct impact on flood conveyance capacity and the structural stability of river infrastructure. Accordingly, the precise and regular construction of riverbed information is recognized as an essential component not only for river planning and management but also for disaster prevention. In particular, riverbed changes, caused by localized geomorphic processes such as sedimentation and scour, can occur within short periods of time. This has led to a growing demand for high-resolution spatial data to accurately detect and monitor such changes (Maddahi and Rahimpour, 2023). In South Korea, riverbed variation surveys are considered fundamental investigations for flood control and river design, and they are typically conducted in conjunction with the national River Master Plan on a 10-year cycle. However, for major national rivers such as the Han, Nakdong, and Geum Rivers-where sedimentation and scour are frequent-surveys are conducted biennially. Furthermore, areas with rapid riverbed changes may be surveyed annually, whereas regions with relatively stable beds may be surveyed every five years (Korean Law Information Center, 2024).

Despite these efforts, the current riverbed survey system still presents several limitations. First, the rigidity of the survey cycle and regional variability hinder timely responses in rapidly changing river sections. Second, coordination and data integration among related institutions are lacking, making it difficult to acquire the necessary datasets for specific areas. Many datasets are outdated and have limited practical value. These issues not only obstruct timely and accurate riverbed monitoring but also pose significant obstacles to transitioning toward a digital, data-driven river management framework. In this context, the need to establish a quantitative and time-series-based riverbed information system and automated processing framework for entire river networks is becoming increasingly urgent. Traditional riverbed surveys have mainly relied on direct methods such as echo sounders, depth poles, and cross-sectional surveys.

However, these approaches are inefficient in terms of manpower, time, and cost and are limited in their ability to generate continuous, high-resolution information over large river areas. To overcome these limitations, advanced remote sensing technologies-such as airborne LiDAR, drone surveys, and aerial photogrammetry-have been increasingly introduced into riverbed surveying. Among them, Airborne bathymetric LiDAR (ABL) has attracted attention for its ability to acquire terrain and underwater topographic data simultaneously. ABL employs green lasers (typically 515– 520 nm) capable of penetrating both the air and water interface, enabling the collection of high-density point clouds representing depth and bottom elevation. It is particularly advantageous in shallow water areas, inaccessible zones, or regions where traditional surveys are not feasible. Compared to acoustic echo sounding, ABL can cover larger areas and provide more detailed 3D terrain data. The latest ABL systems also integrate near- infrared (NIR) lasers with green lasers, enabling simultaneous measurement of terrestrial terrain, water surface, isobaths, and riverbed morphology.

Globally, ABL-based riverbed studies have been actively conducted, focusing on accuracy evaluation, standardization of quality control, and the development of automated data processing pipelines. One of the earliest and most notable studies applying ABL to river surveys was conducted by Hilldale and Raff (2007), who mapped approximately 220 km of the Yakima River in Washington State and the Trinity River in California using the SHOALS-3000 system. They compared the results with concurrent GNSS and sediment surveys, reporting an average depth root mean square error (RMSE) of approximately 0.15 m. However, they noted limitations in turbid water and heavily shadowed riverbank areas. The system was deemed most effective in depths shallower than three times the Secchi depth and under minimal surface reflectance conditions. Later, Kinzel et al. (2012) applied the USGS Experimental Advanced Airborne Research LiDAR (EAARL) system to three rivers (Colorado, Trinity, and Klamath) and found that in stable environments, RMSE was ±0.12 m, comparable to traditional echo sounding and in-situ surveys. However, under conditions of turbulence and increased suspended sediment, waveform separation between surface and bottom signals became problematic. Mandlburger et al. (2015) demonstrated the ecological monitoring potential of ABL by applying it to Austria’s Pielach River (mean width ~40 m). They achieved an average point density of 20 pts/m² and vertical accuracy under 10 cm across the surveyed area, detecting post-flood sediment movement of approximately 1,000 m³.

Awadallah et al. (2023) evaluated multiple systems, including CZMIL SuperNova and RIEGL VQ-880G (airborne) and VQ-840G (unmanned aerial vehicle [UAV]-based), over an 8 km section of the Lærdal River in Norway. Comparing results with MBES and terrestrial LiDAR, they reported RMSEs ranging from ±0.02 to 0.13 m depending on the sensor, while noting consistent signal loss in foamy, turbulent zones. They proposed a decision-making framework for selecting high-energy/low-density or low-energy/high-density sensors depending on the survey purpose and depth range. Frizzle et al. (2024) analyzed ABL survey data from 57 rivers worldwide to assess their suitability for use in hydrodynamic flood modeling. They found a mean RMSE of ±0.16 m and statistically identified turbidity, river width, and slope as primary determinants of signal attenuation thresholds. Furthermore, they developed empirical guidelines indicating that when depth is within 1.7× Secchi depth, river width exceeds 10 m, and vegetation cover is below 40%, over 90% of riverbed point acquisition is achievable, providing useful criteria for pre-survey feasibility assessments. Recently, drone-based bathymetric LiDAR systems, which offer superior deployment flexibility compared to aircraft platforms, have gained attention. Himmelsbach et al. (2025) conducted a comparative study over a 500 m mountainous reach of the Fischbach River in Austria, characterized by steep slopes and fast flow. Using both the RIEGL VQ-880G (airborne) and VQ-840GL (UAV), they found that drone data provided a finer representation (mean point density: 150 pts/m², RMSE: ±0.06 m), while aircraft-based systems were more efficient for broad-area DEM construction. Both systems, however, struggled with signal loss in foamy, turbulent zones, highlighting the necessity of hybrid deployment strategies and supplementary ground-based measurements.

In South Korea, the application of ABL to river surveys remains at a pilot stage. Most domestic studies have focused on evaluating the feasibility under Korean environmental conditions. Kim et al. (2021) surveyed three sites-Jeoncheon (Donghae City), Namhan River, and Gokgyocheon (Asan)-using the Leica Chiroptera 4X system. Flying at 400 m altitude and achieving 5 pts/m² point density, they analyzed observation success rates under different turbidity and depth conditions. In the shallow, clear waters of Jeoncheon and Gokgyocheon, the success rate exceeded 98%, whereas the deeper, more turbid Namhan River yielded a success rate of only 28%. RMSE values ranged from ±10 cm to ±13 cm. Recent domestic research (Lee et al., 2021; Kim et al., 2021) has focused on riverbed point classification from ABL data and proposed automated processing pipelines, but most have been limited to single sites or small-scale reaches.

In summary, despite initial validation studies, the application of ABL in Korea remains in its infancy. There is a clear lack of large-scale, basin-wide studies that demonstrate the practical utility and efficiency of ABL systems for constructing automated riverbed information. Therefore, this study aims to examine the distribution patterns of ABL point clouds with respect to water depth over a wide river area and, based on this, to evaluate the feasibility of constructing riverbed information by applying existing automation-based geomorphological filtering algorithms to seabed classification, which is the most critical step in DEM generation. Fig. 1 presents the overall research framework of this study.

Fig. 1. Overall research workflow.

2. Materials and Methods

2.1. Study Area and Data Acquisition

The study area comprises a river section flowing through the central region of Yeongcheon City, Gyeongsangbuk-do, South Korea, including the Geumho-cheon, Gohyeon-cheon, and Sinnyeong-cheon streams. The Sinnyeong-cheon and Gohyeon-cheon merge and flow into the main stem of the Geumho River. For this study, ABL data were acquired over the target area using the Chiroptera-5 system, the latest ABL sensor developed by Leica Geosystems, on April 27, 2025. The raw ABL data included both ground surface and riverbed return. Since the focus of the river terrain analysis is limited to in-channel features, extraneous points such as ground points, isolated water bodies near the riverbanks, and various noise elements outside the levee boundaries were excluded. These were manually filtered based on visual interpretation to extract only the data within the river levee zone. The processed dataset revealed that the total channel length of the study area is approximately 24.7 km, with ground elevations ranging from 86 m to 122 m. The spatial distribution of the raw ABL data and the extracted in-channel data are visualized in Fig. 2. The Chiroptera-5 sensor used in this study offers significant improvements over its predecessor, the Chiroptera 4X, particularly in terms of maximum penetration depth-enhanced from 2.7/K to 3.2/Kd-and meets the IHO Special Order accuracy standard. The system is recognized for its high-performance in-depth detection, measurement precision, and data density. Detailed sensor specifications are presented in Table 1, and an overview of the sensor system is shown in Fig. 3.

Fig. 2. Test sites: (a) raw ABL data and (b) ABL data within the river channel.

Table 1. Leica Chiroptera-5 sensor specifications

* Kd: the diffuse attenuation coefficient of light in water.

Fig. 3. Chiroptera-5 sensor system (Leica Geosystems, 2025).

2.2. ABL Data Overall Accuracy Verification

The Chiroptera-5 ABL system generates raw data that includes both riverbed and ground surface returns by utilizing green laser pulses reflected from the riverbed and NIR laser pulses reflected from the ground. To verify the accuracy of the raw dataset, a comparative analysis was conducted using ground control point (GCP) survey data. For this purpose, GNSS-based GCP measurements were carried out on July 3, 2025, within the study area, resulting in the acquisition of eight control points. The accuracy evaluation involved calculating the positional errors (X, Y, Z) at each GCP. The results showed a mean error of approximately 5 cm, with an RMSE of approximately 11 cm for all components. The detailed results of the accuracy assessment for the acquired ABL data are presented in Table 2. Fig. 4 visually presents the locations of the ground control point survey data.

Table 2. Overall accuracy assessment of ABL data

Fig. 4. Ground control point locations.

3. Experiment and Results

3.1. Depth-Based Distribution of ABL Point Clouds

The ABL system employs a green laser (515 nm) capable of penetrating the water surface and reaching the riverbed, enabling efficient measurement of underwater terrain based on the return signal. However, the quality and characteristics of these return signals vary depending on several environmental factors, including surface conditions, water depth, and turbidity. This study focused on analyzing environmental factors related to water depth, and it has the limitation of not considering other environmental factors, such as water surface conditions and turbidity, which affect the Kd value. In shallow water areas, the laser travel distance is short, causing return signals from the water surface, submerged objects, and riverbed to overlap, which complicates signal separation. In contrast, in deeper water, the laser path lengthens, and signal attenuation increases, potentially resulting in weaker return intensity and missing point cloud data from the riverbed. These depth-dependent variations significantly affect the classification accuracy of riverbed points and are a critical factor in determining the quality of the resulting DEM. In this study, water depth was estimated by calculating the vertical distance between water surface points, derived from ground returns, and riverbed points, extracted from bathymetric returns. Based on this analysis, the water depths in the study area were found to range from approximately 0.1 m to 2.6 m. To analyze point cloud characteristics more precisely, the study area was classified into three depth zones: Shallow depth zone (≤0.5 m), Moderate depth zone (0.5–1.0 m), and Deep depth zone (>1.0 m). For visualization and further analysis, the zones were color-coded: blue for shallow regions (8 segments), green for moderate-depth regions (10 segments), and red for deep regions (2 segments). The spatial segmentation and classification results are illustrated in Fig. 5.

Fig. 5. Depth-based zonal segmentation.

3.2. Classification of Riverbed Point Clouds

ABL data contains not only returns from the water surface and riverbed, but also includes mixed signals from underwater backscattering, reflections from submerged objects and suspended materials, multipath effects, and system-induced noise. Therefore, to extract accurate river terrain information from ABL data, a preprocessing step to classify the point cloud into water surface and riverbed points is essential. Fig. 6 illustrates the general workflow of ABL data processing in a schematic format.

Fig. 6. General workflow of airborne bathymetric LiDAR data processing (modified from Mandlburger et al., 2013).

In general, ABL system manufacturers provide point cloud classification functions through proprietary software platforms such as Leica LiDAR Survey Studio (LSS), Riegl RiHYDRO, and Teledyne Optech HydroFusion. However, the classification algorithms implemented in these software packages are proprietary, meaning that users are limited in their ability to understand or modify the underlying processes (Andersen et al., 2017; Saylam et al., 2018). As an alternative to overcome this limitation and enable automated extraction of riverbed points, ground filtering (GF) techniques have been utilized. GF methods were originally developed for terrestrial applications to extract the ground surface by removing non-ground objects such as buildings, trees, and vehicles.

However, since the riverbed is also a continuous surface located at the lowest elevation in a geomorphic structure, the GF approach is theoretically applicable to riverbed point classification as well. ABL point clouds typically include multiple return data, with water surface returns generally recorded first and riverbed returns last. Therefore, riverbed points are often located at the lowest vertical position, making them suitable for extraction using GF algorithms. In addition, most commercial ABL systems either do not provide waveform data or provide point clouds in unclassified LAS format (Class = 1). In such cases, filtering based solely on x, y, z coordinate information becomes necessary, and GF techniques that do not require waveform analysis offer a practical and accessible solution for point classification. In this study, we employed one of the GF algorithms, the cloth simulation filtering (CSF) method, to perform automatic riverbed point classification. Originally proposed by Zhang et al. (2016), the CSF approach conceptualizes a cloth-like surface that drapes over the inverted point cloud, simulating physical properties such as tension, gravity, and friction. Points that come into contact with the cloth are identified as ground-or in this case, riverbed -points.

This approach aligns well with the reflective structure of ABL data, where the riverbed is always lower than any surface or in-water objects, and the riverbed tends to form gradual or smooth curved surfaces, which are well-suited for the surface-fitting nature of CSF. The CSF algorithm is also highly practical due to its low number of parameters, fast computational speed, lack of training requirements, and implementation in open-source software such as CloudCompare. These advantages have led to successful applications of CSF in ABL-based studies on seabed point extraction, where it has demonstrated reliable performance (Zhao et al., 2020). However, previous applications of CSF have focused primarily on coastal or marine environments, and to date, no study has applied CSF to ABL data acquired in riverine environments. Furthermore, quantitative evaluations of CSF performance in relation to depth-dependent point cloud distribution -a key factor in river settings-remain insufficient in the existing literature.

3.2.1. Riverbed Point Cloud Classification Using the CSF Algorithm

In this study, the CSF algorithm, an automated classification method, was applied to classify riverbed point clouds over a large-scale river system. Considering the depth-dependent distribution characteristics of the ABL point clouds, the study area was segmented into shallow, moderate, and deep-water zones, and for each zone, point clouds were extracted and analyzed in grid units, each consisting of approximately 4,000 to 6,000 points. The CSF algorithm enables automatic classification of riverbed points from 3D ABL point clouds. As illustrated in Fig. 7, the method involves inverting the point cloud along the Z-axis and simulating a virtual cloth being dropped from above. The cloth conforms to the lowest surfaces it encounters-based on physical properties such as gravity and elasticity-and the points it touches are identified as representing the underlying terrain, which, in the case of bathymetric data, corresponds to the riverbed surface.

Fig. 7. Principle of the CSF algorithm (Zhang et al., 2016).

A virtual cloth is generated and dropped in the direction of gravity, with the positions of its particles adjusted iteratively by simulating spring forces between them. This process continues until no further deformation occurs. After the simulation stabilizes, the distances between the cloth surface and the point cloud are calculated. Points located within a short vertical distance from the cloth are classified as riverbed points, while those further away are considered non-riverbed points. The CSF algorithm uses three main parameters:

i) Scene type: steep slope is used for rugged mountains or rocky terrains. Relief is suited for moderately sloped coastal areas or rolling hills. Flat is recommended for generally smooth and flat terrains, allowing the cloth to fall gently and minimizing excessive distortion.

ii) Cloth resolution: this refers to the horizontal and vertical spacing between particles in the cloth mesh. A low resolution (smaller spacing) results in a denser cloth, better suited to capturing fine terrain features, but increases sensitivity to noise and slows down computation. A high resolution (larger spacing) improves processing speed but may oversmooth or distort fine-scale topographic variations.

iii) Classification threshold: this defines the vertical distance between the cloth surface and point cloud points used to classify them. If the threshold is set too large, there is a risk that submerged objects or water surface points may be misclassified as riverbed points. If it is too small, actual riverbed points may be excluded from classification.

However, the CSF tends to estimate the seabed more smoothly than it is in areas where the seabed topography is complex or where slope changes are abrupt, and it has the limitation of misidentifying protruding elements, such as suspended materials or underwater vegetation, as part of the seabed.

3.2.2. Depth-Based Parameter Optimization for CSF Algorithm Application

Among the three parameters of the CSF algorithm, the Scene Type was fixed as Flat across all water depth zones, taking into account the generally gentle slope characteristics of riverbeds. The remaining two parameters-grid size (Gs) and classification threshold (H)-were optimized separately for each depth category. For reference data during the optimization process, manually classified point clouds were used for each water depth zone. The classification accuracy corresponding to each parameter setting was evaluated using the following metrics: true positive rate (TPR), positive predictive value (PPV), overall accuracy (OA), and Kappa coefficient (K) (Fawcett, 2006). The Kappa coefficient is particularly important as it accounts for the likelihood of agreement occurring by chance. A Kappa value of 0.8 or higher is generally considered to indicate reliable classification accuracy (Landis and Koch, 1977). The following formulas were used for these calculations: TP = True Positive, TN = True Negative, FP = False Positive, FN = False Negative:

\(\begin{align}\text {TPR} = \frac{T P }{T P + F N}\end{align}\)       (1)

\(\begin{align}\text {PPV} = \frac{T P}{T P + F P}\end{align}\)       (2)

\(\begin{align}\text {OA} = \frac{T P + T N}{T P + F P + F N + T N}\end{align}\)       (3)

\(\begin{align}K = \frac{O A - P (e)}{1 + P(e)}\end{align}\)       (4)

where

\(\begin{align}{P(e)} = \frac{(T P + F N) (T P + F P) (F P + T N) (F N + T N)}{(T P + F P + F N + T N)}\end{align}\)

3.2.3. Accuracy and Consistency Analysis Results

Considering the characteristics of the river environment, the Gs parameter was set within a range of 0.25 m to 2.0 m, in 0.25 m intervals. The classification H was adjusted according to water depth: For shallow water zones, it ranged from 0.05 m to 0.4 m in 0.05 m intervals; for moderate and deep-water zones, it ranged from 0.05 m to 0.75 m in 0.1 m intervals. Experimental results showed that for moderate and deep-water zones, the highest classification accuracy was achieved when Gs = 0.5 m and H = 0.25 m, with Kappa coefficients of approximately 0.96 and 0.97, respectively. In the moderate depth zone, configurations with Gs = 0.25 m to 1.0 m and H = 0.15 m to 0.45 m consistently produced Kappa values above 0.9. Similarly, in the deep-water zone, all Gs values combined with H = 0.15 m to 0.45 m resulted in Kappa values exceeding 0.9. In contrast, in the shallow water zone (≤0.5 m), the highest Kappa value-approximately 0.80-was observed when Gs = 1.25 m and H = 0.1 m. However, for other threshold combinations, the classification accuracy was lower, and even the optimized thresholds did not yield satisfactory results.

Overall, in shallow waters, where the vertical spacing between the water surface, submerged objects, and riverbed points is minimal, a higher rate of misclassification was observed due to overlapping return signals. This resulted in reduced classification accuracy using the CSF algorithm. Therefore, in such zones, it is recommended to apply manual validation of ABL data or utilize GNSS-based ground surveys for riverbed data acquisition. On the other hand, in moderate to deep water zones (depth >0.5 m), the CSF algorithm demonstrated effective performance for automated classification of bathymetric points in ABL datasets. Fig. 8 and Table 3 present the optimized threshold results for each depth category. Fig. 9 illustrates the cross-sectional classification results obtained by applying the optimal parameters in depth. In the figure, blue points represent classified riverbed points, and red points indicate non-riverbed points.

Fig. 8. Optimization of CSF parameters: (a) Gs and (b) H.

Table 3. Classification accuracy by water depth

Fig. 9. Point cloud distribution and classification by water depth: (a) shallow depth, (b) moderate depth, and (c) deep depth.

3.3. Riverbed DEM Construction and Depth Distribution Mapping

Using the optimized threshold parameters of the CSF algorithm, riverbed points were automatically classified across the entire study area, which had been segmented by depth into two deep-water zones, ten moderate-water zones, and eight shallow-water zones. Based on these classified point clouds, a 0.5 m resolution grid-based DEM of the riverbed was constructed. The accuracy evaluation of the ABL riverbed data was subsequently performed using the riverbed DEM and GPS survey data acquired on August 15, 2025. After aligning the two datasets to the same horizontal axis, a vertical error analysis was conducted based on four river cross-sectional lines (a total of 33 GCPs). As a result, the vertical error RMSE was found to be approximately 9 cm. Although there is a possibility of riverbed changes due to the time gap of more than three months, cross-sectional GNSS surveys were carried out on stable riverbed structures (e.g., fishways, weirs) to minimize the impact of such changes. Fig. 10 visually presents the reference point locations of the acquired riverbed data, and Table 4 provides the detailed results of the data accuracy evaluation.

Fig. 10. Examples of evaluation reference points of ABL bathymetric data.

Table 4. Accuracy assessment of ABL-derived riverbed data

The depth distribution map was derived by extracting water surface points from simultaneously acquired NIR LiDAR data, generating a water surface DEM, and then differencing it with the riverbed DEM. The NIR sensor mounted on the ABL system operates at a wavelength of approximately 1,064 nm, which, due to its physical properties, cannot penetrate the water surface and is almost entirely reflected. As such, the NIR returns can be reliably interpreted as representing the water surface elevation, enabling the construction of a continuous water surface DEM across the river channel. Because the NIR sensor is synchronized with the green laser sensor, both water surface and riverbed measurements can be acquired at the same location. This allows the water surface and riverbed DEMs to be constructed with identical resolution and coordinate systems, enabling the precise calculation of water depth through DEM differencing. This method provides a high-resolution representation of the spatial and temporal distribution of the water surface. It is far more efficient than traditional methods such as echo sounding (ES) or cross-sectional surveying, especially for large-scale river terrain analysis.

Fig. 11 illustrates the visualization of the DEM and the resulting depth distribution map, where the depth map effectively represents the spatial variability of water depth within the river channel. In this study, the construction of the DEM and depth map enabled the quantitative and visual analysis of riverbed topography, allowing for detailed examination of depth variation and elevation patterns, and offering the potential for indirect monitoring of erosion and sedimentation. In particular, compared to conventional cross-sectional river surveying methods, the ABL-based 3D terrain produced in this study provides significant advantages in terms of spatial continuity, precision, and processing efficiency. This facilitates high-resolution analysis across the entire river reach, highlighting the clear potential of ABL technology for digital river management and the establishment of data-driven disaster response systems.

Fig. 11. Generation of riverbed DEM and bathymetric data: (a) riverbed DEM and (b) bathymetric data.

4. Discussion

This study demonstrated the applicability of ABL technology for a large-scale river section-approximately 24.7 km in length-located in Yeongcheon City, Gyeongsangbuk-do, South Korea. The depth-dependent variation in the distribution characteristics of ABL point clouds was clearly identified, leading to the segmentation of the study area into shallow (≤0.5 m), moderate (0.5–1.0 m), and deep (>1.0 m) water zones. For each zone, the parameters of the CSF algorithm were optimized. The results showed that the CSF algorithm performed effectively in the moderate and deep zones, yielding Kappa coefficients exceeding 0.96, whereas in the shallow zones, accuracy was somewhat lower due to the influence of surface reflections and submerged objects.

When compared with the study (Lee et al., 2021), which applied a ground filtering algorithm to separate riverbeds in domestic rivers, our method showed lower accuracy in shallow water but higher classification accuracy in mid- and deep-water areas. This outcome is considered to result from various factors, including the surveying sensors, riverbed topographic characteristics, and the properties of the applied algorithm. Using the classified riverbed point clouds, DEMs, and depth distribution maps, we generated DEMs, enabling quantitative visualization of the spatial characteristics of the riverbed topography. These outputs were found to be valuable as foundational datasets for various river management applications, including erosion and sedimentation monitoring and flood risk assessment.

For future research, technical enhancements are required to address the lower classification accuracy in shallow water zones (≤0.5 m). In particular, in environments where surface reflections, submerged objects, and riverbed returns overlap, signal separation becomes difficult, increasing the likelihood of misclassification. Therefore, it is considered necessary to conduct further research to complement surveys of very shallow river sections by developing methodologies that either integrate ABL waveform signal analysis results or combine them with UAV-based satellite-derived bathymetry (SDB) surveys. This study focused on analyzing environmental factors related to water depth, without considering other factors such as water surface conditions and turbidity. However, since the analysis results may vary when applied to rivers with different environmental conditions, future studies need to take into account other environmental factors (e.g., water surface conditions, turbidity) that influence ABL reflection signals in addition to water depth. The penetration depth of ABL is primarily governed by the Kd value; however, a standardized methodology for estimating Kd in real time using ABL has not yet been established. In addition, conventional turbidity measurement methods, such as Secchi depth, are difficult to apply over large-scale regions (e.g., the ~24 km section considered in this study).

Consequently, this study did not provide precise Kd values. Nonetheless, cross-sectional verification indicated that ABL signals penetrated stably to the maximum water depth of approximately 2.6 m within the study area. Accordingly, the development of methodologies for real-time measurement and integration of Kd with ABL surveys represents an important avenue for future research. In addition, to enable real-time acquisition and utilization of riverbed information over large-scale river networks, the development of an automated preprocessing and classification pipeline based on ABL data is essential. Such technologies could serve as critical infrastructure for real-time riverbed change detection and disaster response systems. Furthermore, the DEMs and depth maps developed in this study have the potential to be integrated with hydrological and hydraulic models, supporting practical applications such as flood simulations, flow capacity analysis, and riverbed erosion forecasting. Therefore, further research is needed to establish an integrated application framework for comprehensive river management and policy development.

Author Contributions

Conceptualization: An HW, Lee JB; Data curation: An HW, Lee JB; Methodology, Formal analysis: An HW; Validation: All authors; Project administration: Lee JB; Funding acquisition, Supervision: Lee JB; Writing–original draft, Writing–review & editing: All authors.

Conflicts of Interest

No potential conflict of interest relevant to this article was reported.

Funding

This research was funded by the Basic Science Research Program through the National Research Foundation of Korea (NRF), funded by the Ministry of Education, Grant No. 2021R1I1A3059263.

Data Availability Statement

The data presented in this study are available on request from the corresponding author.