1. Introduction
The domain characterized by shallow waters, where the dynamic interplay between the sea and land is particularly pronounced, constitutes a critically significant region for a multitude of hydrographic, oceanographic, and topographic survey applications. This encompassing domain is instrumental in supporting various endeavors including coastal construction, marine safety assurance, resource assessment and development, fisheries and marine industries, marine transportation and shipping logistics, environmental preservation, and management, as well as research pertaining to the coastal zones of islands or peninsulas [
1,
2,
3,
4,
5,
6,
7].
Traditional methodologies for the measurement of bathymetry in shallow waters primarily rely on shipborne single-beam or multibeam echosounders (MBES) [
8], airborne Light Detection and Ranging (LiDAR) [
9], Synthetic Aperture Radar (SAR) [
10], SAR-based techniques to extract bathymetric features, and optical remote sensing [
11,
12], with each approach presenting a distinct set of advantages and limitations.
MUSs are increasingly being integrated into military operations, serving either as a supplementary addition to traditional vessel operations or as an autonomous solution for conducting environmental monitoring [
13]. This trend reflects the growing reliance on technologically advanced methodologies in naval strategies [
14,
15]. The military objectives of the North Atlantic Treaty Organization (NATO) extend to crisis response, peace support, humanitarian operations, and conventional warfighting, thereby fundamentally altering the demands of military environmental support [
16]. The dynamic nature of these operations, often characterized by short notice and deployment in highly variable, inadequately monitored, and potentially hostile physical environments, underscores the requirement for dynamic and responsive processes to provide operational environmental information.
The hydrographic surveys delineated in this study are not confined solely to operational applications within the military domain; rather, they find utility in civilian contexts, particularly in the realm of coastal mapping. Coastal mapping stands as a fundamental tool essential for supporting coastal engineering efforts [
17]. It constitutes a primary component in the ongoing monitoring and assessment of both shorelines and coastal regions. This practice enables the identification of areas particularly susceptible to alterations induced by erosion, thereby facilitating the efficient and cost-effective management of shore protection strategies. In instances of protected shores, coastal mapping facilitates expeditious verification of the functionality and efficacy of applied reinforcement, thereby enhancing engineering practices. Consequently, the efficacy of coastal management is contingent upon the consistent monitoring and mapping of coastal environments [
18].
The presented experiments form an integral component of the activities conducted within the framework of the 2023 edition of the REPMUS multinational exercise [
19]. The REPMUS exercise series is a multinational initiative wherein collaborative efforts among military entities, industry, academia, and other institutions are orchestrated to experiment with the deployment of MUSs for both military and civilian applications. The system integration of vehicles, sensors, and survey methodologies represents an innovative solution for conducting environmental surveys in a shallow water domain, particularly in the vicinity of a coastal site, with the specific objective of a fast environmental assessment. REA is a concept that is widely used in the military domain, to rapidly assess the area of operations before a military operation, or in civilian use for disaster research or any other monitoring activity of the environment [
20].
The uncrewed systems employed in this evaluation consist of surface and aerial uncrewed systems integrated with state-of-the-art commercial off-the-shelf and prototype sensors specifically engineered for shallow and medium water bathymetric measurements. The application of conventional crewed survey techniques is characterized by significant financial implications and limited time efficiency. This can be effectively complemented by the comprehensive utilization of MUSs, which can deliver precise results in shallow water environments, achieving this with a reduced personnel risk and a substantial decrease in time and cost.
In conjunction with the surveys conducted using surface and aerial uncrewed systems, an evaluation of SDB was undertaken. The escalating availability of satellite remote sensing data, particularly from satellites such as WorldView, Landsat, Airbus, and Sentinel, has contributed to the advancement of bathymetric assessments through remote sensing technology. SDB involves utilizing multispectral satellite images and mathematical models, often of linear, polynomial, exponential, or similar nature, derived from radiative transfer formulas. Subsequently, the model is applied to calculate the depth of the water body under consideration [
21].
The conceptualization of this study was based on addressing the common limitations and deficiencies characteristic to the domain of shallow water hydrographic surveying:
The main contributions of this study are to address the above-mentioned challenges using cutting-edge MUSs for rapid shallow water hydrographic surveys:
A novel integration of a medium-depth multibeam sonar with an Unmanned Surface Vehicle (USV) was successfully trialed in a shallow water environment.
The challenging surf zone area was rapidly surveyed with high accuracy using an innovative LiDAR survey Uncrewed Aerial System (UAS).
Advanced SDB techniques demonstrated the capability to produce high-resolution products, facilitating remote assessments in hydrographic surveying.
Recent studies have concentrated on the use of MUSs for hydrographic surveys [
26], cartographic and safety of navigation aspects [
27,
28,
29], and geomorphological surveys using UAVs [
30] of shallow water bathymetry, using commercial or innovative bathymetric sensors [
31], emphasizing the accuracy and precision of the data collected.
In contrast, this study concentrates on the rapid and accurate evaluation of the depths in any environmental context, including the surf zone, with the objective of enabling prompt task execution using innovative integrations of MUSs and bathymetric sensors, combined with recent remote sensing technology methods. The focus of the study is not predominantly on the cartographic rigor of the collected data. Overall, the paper aims to contribute to promoting the effectiveness of the use of MUSs in challenging shallow areas of surveying, thus improving the monitoring capability in areas of interest.
This article is organized in the following manner:
Section 2 provides a description of the bathymetric survey areas, followed by a description of the MUSs used, including technical specifications, survey parameter details, and data processing methodologies.
Section 3 is devoted to the presentation of the survey outcomes for each methodology, with an emphasis on the statistical characterization of the resultant bathymetric surfaces. Additionally, a discussion pertaining to comparative evaluations with a reference survey area and the corresponding nautical chart is expressed. Finally,
Section 4 delineates the summarizing conclusions and outlines prospective directions for future research.
3. Results and Discussion
This section is dedicated to the processing and analytical examination of the hydrographic data derived from each system. It involves a comparative evaluation of the datasets with the reference survey, as well as inter-comparisons among the datasets themselves.
3.1. Reference Bathymetric Survey
This analysis is not aiming to assess the compliance with hydrographic or cartographic standards used for charting purposes but rather intends to evaluate it under the scope of a REA operation. In this context, the rapid exchange of data and products’ dissemination holds greater significance than adherence to demanding standards typical of traditional bathymetric surveys.
The bathymetric model from the ref”renc’ survey (
Figure 10) was conducted following the International Hydrographic Organization (IHO) S-44 standard survey order for special order surveys [
58].
The node density of the reference survey was computed (
Figure 11a and
Table 6), highlighting higher values in the shallow part of the area and decreasing in density in the deeper part, with a mean value of 139.66 nodes. Moreover, the standard deviation, which represents one component of the bathymetric uncertainty in the final bathymetric model, was computed for the reference surface (
Figure 11b and
Table 6), with 0 m as the mean value, and 0.1 m as the maximum value, accentuating the high accuracy and precision of the bathymetric surface.
3.2. USV MB Survey Results
The raw multibeam data, sent from the DriX USV to the Datahub, via the Starlink connection, were imported into the Teledyne Caris HIPS and SIPS software for postprocessing. Differential GNSS corrections were integrated into the multibeam dataset to reduce the height of the tide and reference the depths to the local chart datum, followed by the generation of a preliminary CUBE model [
59,
60]. A meticulous visual inspection of the bathymetric model was conducted by the team of hydrographers to identify potential systematic errors or gross inaccuracies within the data. After the inspection of the surface and the removal of noisy data (and spikes) and data consistency, a detailed analysis of the final hypotheses generated by the CUBE algorithm was undertaken. This step was crucial to ascertain whether the noise reduction process performed by the CUBE algorithm inadvertently eliminated valid data points. Afterwards, the statistical parameters of the CUBE surface, including standard deviation, histogram details, and the correlation with depth data from the electronic nautical chart (ENC), were rigorously examined.
Simultaneously, a graphical representation of the surveyed area, indicated by the CUBE surface (
Figure 12), was generated, accompanied by the statistical metrics.
The values indicated in
Figure 13 and
Table 7 highlight the accuracy and precision specific to IHO special order surveys. Moreover, the multibeam dataset presents a high point density and a node standard deviation tending to 0 m.
3.3. UAV LiDAR Survey Results
The processing workflow for the LiDAR data commenced with the synchronization of raw laser data with GNSS positioning information, followed by the processing of waveforms and the generation of the laser point cloud data file (LAS file). After its creation, the LAS file underwent a quality analysis check, which included the integration of GNSS data and the identification of various errors. A data cleaning process was then applied to the LAS file, aimed at minimizing noise (spikes) and implementing statistical quality control, finishing in an auto-processing procedure that produces the final LAS file. The final LAS file was imported into the Teledyne Caris HIPS and SIPS software, where a geoid model (Geoid PT08) was applied for reducing the depth to a certain vertical datum (in this instance, the Portuguese Chart Datum), followed by a final review of data quality. The team of hydrographers then conducted a comprehensive crossline quality control, surface statistics analysis, and visual inspection prior to the production of the final deliverables (
Figure 14).
Furthermore, the distinctive pushbroom technology inherent to the system, in conjunction with the sophisticated water column processing algorithms it employs, distinguishes it as the single system currently capable of facilitating water column volume analysis and augmenting data density through the exploitation of this feature. The profound understanding of the water column afforded by the pushbroom technology enables difficult analyses of underwater contacts. This is achieved by fully leveraging the technological foundations of the system, particularly tailored for littoral mapping endeavors [
61,
62].
In recent years, enhancements in waveform processing have been implemented within the PILLS (RAMMS) system. These advancements encompass refined methodologies for water surface modeling, backscatter modeling, and signal attenuation, in addition to automated seabed detection. Moreover, the integration of machine learning techniques for the analysis of water parameters and seafloor surface characterization has been introduced, further augmenting the system’s capabilities.
The demarcation of the survey area into two distinct zones, one in shallow waters and the other in deeper waters, was implemented to facilitate a more streamlined processing workflow.
The LiDAR survey exposed an average node density of 65.60 (
Figure 15a and
Table 8) with a mean standard deviation of 0.2, higher in greater water depths, suggesting a diminishing accuracy of LiDAR beyond a water depth of 25 m, in this specific area.
An inconsistency step was observed between the two datasets’ calibration procedures (the shallow water surface and the deeper water surface due to boresight and offset angles between the inertial navigation system and the LiDAR sensor). The deeper surface is improved since more time was given for the surveyors to refine the boresight calibration. It was discovered that the system, still being in its first use with a high scan rate of 60 Hz for the 2 AIRTRAC lasers, had a 1 nanosecond offset in the electronics of the laser and this resulted in an inconsistency step of around 0.5 m between the surfaces (highlighted on the profile in
Figure 16) in the LiDAR data. However, with extra processing, these errors were acknowledged and eliminated after the experimentation.
3.4. SDB Results
After calibrating and validating the SDB processing models, the time series were merged to remove possible outliers and smooth the final composite model. A different number of statistics were computed for each pixel image to evaluate in detail the results of the final SDB model surfaces: minimum value, maximum value, mean value, median value, standard deviation value, variance value, and range (maximum–minimum) value. The mean depth values are represented in
Figure 17 (chart datum) with a pseudo band coloring to better match bathymetric symbology.
In the case of each of the five selected images that had undergone prior preprocessing, the process of calibrating and validating the SDB model was thoroughly executed. ML algorithms were employed to estimate bathymetry values for depths with available training data, while the Stumpf algorithm was utilized for points at greater depths than those covered in the training data. In each instance, the dataset of soundings specific to the Troia Peninsula region were employed. Within this framework, the models were systematically trained using 80% of the 300 ground truth sounding points accessible, while the remaining 20% were reserved to rigorously assess the model’s performance.
3.5. Dataset Comparison
Given that the systems employed in this study are at an emerging stage of development with respect to their integration with unmanned systems (as is the case with the multibeam echosounder) and considering the innovative nature of the LiDAR solution trialed in this exercise, it was not feasible to predetermine a comprehensive a priori uncertainty and error assessment. Consequently, a conventional hydrographic survey was executed using a crewed vessel to fulfill the IHO S-44 special order survey requirements, in the days preceding the unmanned survey operations, in the same area of interest. The reference hydrographic survey resulted in an independent, accurate, and precise bathymetric model, thereby functioning as the definitive reference standard for MUS surveys.
The primary methodology for quantifying errors in the datasets assessed in this study involves a comparative analysis with the results from the reference survey, focusing on the standard deviation and density values for each of the resulting MUS bathymetric models. Moreover, a histogram analysis between the MUS models and the reference survey and a comparison between depth contours between the MUS data and the local electronic nautical chart (ENC), complemented by cross-comparisons among all datasets to evaluate vertical and horizontal accuracy and precision, were conducted.
3.5.1. Horizontal Accuracy Comparison
In the context of hydrographic surveying, horizontal accuracy typically possesses a broader tolerance margin compared to vertical accuracy, in accordance with most IHO and industry standards. Nonetheless, from a military perspective, horizontal accuracy and precision are paramount due to the imperative need for precise object detection capabilities inherent in these systems.
Figure 18 illustrates the overlay of three datasets upon an identical geomorphological sand formation, showing a coherent correspondence. This alignment indicates the accurate horizontal resolution of the systems in question, underscoring their capability for precise spatial delineation.
The horizontal accuracy of SDB is naturally linked to the spatial resolution of the employed satellite sensor, resulting in bathymetric data uncertainties that typically correspond to the dimension of one pixel. In this study, Sentinel-2 imagery, which possesses a spatial resolution of 10 m, was utilized. Consequently, for a product with a 10 m resolution, the horizontal accuracy is anticipated to be in the same range. This limitation in spatial resolution is the primary reason SDB techniques have not yet been extensively adopted for underwater object detection applications.
3.5.2. Vertical Accuracy Comparison
In the context of dataset comparison, the designated approach involved comparing the UAV LiDAR and SDB datasets with the reference survey bathymetric model and creating several Difference surfaces to assess the accuracy and precision of the data. However, it is noteworthy that the coverage area of the USV MB dataset did not align with the coverage of the reference survey; therefore, cross-comparisons with the other available survey datasets were conducted.
The comparative analysis of the difference surface generated by the UAV LiDAR survey and the reference survey (
Figure 19) exposes discrepancies predominantly along the track of the trajectory of the LiDAR scan, where minor artefacts are observed, mainly due to the boresight calibration and the delay between the internal LiDAR systems (
Figure 20). These variations, however, remain within a maximum threshold of 0.2 m.
Concerning the differential analysis between the SDB and the reference survey (
Figure 21), it is observed that variances are discernible with a maximal deviation of 1.8 m in the shallow water part, below a depth of 12 m. In deeper waters, the disparity becomes considerably more pronounced, attributable to the fundamental constraints of the SDB in penetrating the ocean.
Given that the USV MBES survey did not encompass the area of the reference survey area, an analytical comparison was conducted with the UAV LiDAR survey, as seen in
Figure 22. It can be seen from this comparison that the differences between the two surveyed surfaces are constrained within a vertical discrepancy not exceeding 0.5 m.
Furthermore, upon examination of the histogram of differences (
Figure 23), the bathymetric measurements exhibit a robust level of reliability. The minimum, maximum, mean, and standard deviation values are presented in
Table 9, with the SDB dataset demonstrating higher values attributed to the deeper areas where satellite imagery is less effective, while most LiDAR survey data are below a 0.5 m difference from the reference survey data.
Furthermore, various profile differences were implemented to accentuate the distinctions between the reference bathymetric survey and the two SDB and LiDAR surveys (
Figure 24).
3.5.3. Bathymetric Contours Comparison with the ENC
To further validate the precision and accuracy of the bathymetric data, contour lines were generated for each surface, the reference surface (
Figure 25), MBES USV (
Figure 26), LiDAR UAV (
Figure 27), and SDB (
Figure 28), and subsequently superimposed and compared with the contour lines derived from the ENC.
The results of this comparative analysis revealed a high level of bathymetric accuracy within the domain of extremely shallow waters. However, as seen in
Figure 28, the contours generated by SDB do not exhibit the same level of spatial resolution as those produced by the LiDAR and MBES surveys, attributable to the inherent constraints on horizontal accuracy associated with satellite remote sensing technology.
The survey operations were characterized by a series of challenges, including the presence of turbid waters in the river inshore region, which blocked the penetration capabilities of the LiDAR system. Furthermore, the prevalence of calibration errors underscored the criticality of thorough calibration processes to optimize data quality. This includes the priority for boresight calibration of the LiDAR system and the execution of the patch test for the MBES survey, underscoring their crucial roles in ensuring the precision and accuracy of the collected data.
Furthermore, the processing methodologies associated with SDB present a considerable challenge, owing to their extensive nature and the requisite expertise of a subject matter specialist for their execution. Additionally, there persists a requirement for ground truth soundings to calibrate the model, thereby enhancing the accuracy of the resultant data. It is also noteworthy that the efficacy of SDB is profoundly influenced by the transparency of the water column and prevailing meteorological conditions. Moreover, SDB frequently encounters challenges in the precise quantification of uncertainties, a limitation that curtails its extensive scientific and operational application. Nevertheless, SDB retains the potential to greatly help in planning survey operations since it provides an approximate bathymetric picture of a location, without necessarily requiring the deployment of either assets or personnel.
Surveying in shallow water remains a particularly challenging activity within hydrographic studies. This paper delineates various methodologies aimed at addressing this issue, with the employment of uncrewed systems and remote sensing techniques, elucidating the respective advantages and disadvantages of each approach. Furthermore, the paper presents findings from pioneering systems, notably the integration of PILLS LiDAR with the Schiebel CAMCOPTER® S100, as well as the implementation of a state-of-the-art medium water multibeam system, the Kongsberg EM 712 USV, onboard the DriX USV. The investigation successfully demonstrated the ability to survey the surf zone both efficiently and in a safe manner.
It is imperative to highlight the outcomes produced by the advanced LiDAR system utilized in this study. This system demonstrated a level of accuracy and precision that is comparable to that of an MBES system, operating in challenging areas like the surf zone. Notably, it achieved survey coverage at a significantly accelerated rate compared to traditional sonar-based hydrographic survey methods.
Furthermore, the innovative integration of a medium water multibeam echosounder into an USV demonstrated a creditable degree of accuracy and precision. Consequently, this asset holds potential for future application in high-resolution hydrographic surveys within shallow water regions or offshore environments. However, it is imperative to note that the vehicle requires constant monitoring by a minimum of three personnel, and in emergencies, the presence of a manned asset in proximity is necessary for rescue purposes. The extensive endurance test granted successful results, with no instances of communication loss observed. As expected, the duration of the survey closely matched that of a manned asset survey, as the surface vehicle followed the same navigation and hydrographic survey planning and execution procedures.
From a military point of view, the uncrewed systems used for environmental monitoring present numerous operational benefits, including the augmentation of battlefield tactics, amplification of combat range, and the diminution of human life exposure to risk. They provide instantaneous data and increase situational awareness, thus supporting rapid decision-making processes and enhancing precision for other equipment employment. Moreover, MUSs limit civilian casualties and safeguard military personnel by mitigating combat zone exposure.
4. Conclusions
This study elucidates various methodologies employed in shallow water bathymetric surveys, employing diverse MUSs equipped with distinct payloads, when compared with a reference traditional bathymetric survey and cross-comparisons. In conjunction with the experimented MUS surveys, SDB was examined and evaluated using the Sentinel-2 constellation. As a disclaimer, the results do not represent the performance of each system or equipment, as some misconfigurations might have been overlooked and that is not the purpose of the research.
The SDB bathymetric model exemplified the capabilities of satellite remote sensing surveys in evaluating bathymetry within shallow oceanic regions, demonstrating accurate outcomes with inconsistencies not surpassing 1.8 m relative to the reference hydrographic survey. The penetration depth of the SDB was confined to 12 m, a constraint linked to the distinctive water transparency characteristics of the surveyed area. A suite of ML algorithms were incorporated into the processing pipeline, aiming to expedite the workflow and enhance the accuracy of the final products. The satellite constellation utilized in this process, the employment of low-orbit satellites with superior revisit frequencies, can facilitate the production of higher-resolution bathymetric outputs.
The LiDAR system described in this manuscript represents a revolutionary advancement, being utilized for the first time in this experimental context, and it is characterized by its advanced technical specifications and the remarkable efficacy of its products. The survey was strongly improved by the Schiebel CAMCOPTER® S-100 efficiency and maneuverability. The system was tested in the turbid waters of the Sado River; however, its efficacy was compromised due to the opacity of the water. The LiDAR UAV survey demonstrated rapid data acquisition, with the results indicating a vertical depth uncertainty not exceeding 0.3 m, in comparison to the reference hydrographic survey. However, certain discrepancies were observed along the survey lines, attributable to a 1 nanosecond delay between the two AITRAC lasers within the PILLS system. Additionally, the delay produced a discontinuity between the two processed bathymetric surfaces (shallow and deep water).
The USV MBES survey demonstrated a vertical depth uncertainty not surpassing 0.5 m when juxtaposed with the LiDAR survey (considering the vertical uncertainty of 0.3 m between the LiDAR and reference survey surfaces). However, it is noteworthy that this experimental assessment was constrained by the inability to survey depths shallower than 10 m and the absence of the SVP winch, necessary for a completely autonomous hydrographic survey.
The combination of mild winds, minimal rainfall, and calm seas during the survey period resulted in optimal conditions for the deployment of both UAV-based bathymetric lidar and USV-mounted multibeam echosounder systems. Additionally, these systems should be subjected to rigorous testing in more demanding conditions, such as high wave scenarios, rocky seabed, heavy precipitation, and strong wind environments, to ascertain the operational thresholds of each system and to examine the integrity of data under adverse conditions.
Future research endeavors should explore the development of SDB processing methodologies that are independent of supplementary ground truth soundings. Such advancements would facilitate a complete and entirely remote evaluation of underwater bathymetry across specified areas.
The concurrent deployment of a USV and a UAV (or a swarm version of MUSs) can potentially optimize operational efficiency and expand the scope of data acquisition and resolution by integrating complementary sensing equipment across the two platforms. This multiplatform approach may result in significant reductions in both time and financial resources by facilitating the simultaneous deployment of survey assets.
For the continuation of the study, it is imperative to conduct a comprehensive evaluation of the horizontal and vertical uncertainties and errors associated with each system integration. Subsequent research endeavors should focus on enhancing the detection capabilities of surface and seabed features, as well as the identification of contacts. Currently, the field lacks definitive standards or criteria pertaining to REA bathymetric surveys, necessitating further research to establish tailored bathymetric specifications within this domain.