Underwater UXO detection using magnetometry on hovering AUVs

The EU‐funded project BASTA (Boost Applied munition detection through Smart data inTegration and AI workflows, http://www.basta-munition.eu) aimed at improving underwater unexploded ordnance (UXO) detection approaches and advancing data acquisition techniques. One aspect of the project was performing autonomous underwater vehicle (AUV)‐based magnetic measurements. In this paper, we present the first results of integrating three submersible fluxgate magnetometers to a Girona 500 AUV in the context of underwater UXO detection. The hovering capabilities of these AUVs allow them to maintain a fixed position or to precisely navigate at very low velocities and altitudes. The magnetic sensors are rigidly attached to the nose of the AUV at a lateral distance of 2 m and are arranged in the shape of a vertical triangle, thereby allowing for the calculation of three spatial magnetic gradients. A series of surveys was performed when visiting several munitions dumpsites in the German Baltic Sea. Furthermore, we successfully conducted a test survey with surrogate objects of known magnetic moments in a naval port basin in Kiel, Germany. With a noise floor of approximately 2 nT, the system is capable of reliably detecting munitions similar in size to 81 mm shells from altitudes of 1 m above the seafloor. For ground‐truthing purposes and for a concluding confirmation or rejection of a UXO suspicion, the AUV is equipped with a high‐resolution camera system. This newly developed system aims at improving the industry standard's technical potentials of autonomously discriminating between hazardous UXO and anthropogenic debris or rocks and therefore reducing the number of target points before underwater UXO clearance campaigns.


| Background on munitions in the sea
More than 75 years after World War II, the remnants of past wars and military activity still pose a threat to life and work of people in the coastal areas of Europe and especially Germany today. As a result of warfare, postwar dumping as well as military test and training activities, huge amounts of unexploded ordnance (UXO) and discarded munition material (DMM) are located at the seafloor of the North Sea and the Baltic Sea. According to estimations, around 1.6 million tons of conventional and 5000 tons of chemical munition are expected to remain in German waters alone. They are mainly distributed over but not restricted to 71 known munitionscontaminated sites (Böttcher et al., 2011). Besides the threat to the marine ecosystems (Maser & Strehse, 2020), this legacy represents a direct threat to offshore construction work (dredging [Shum et al., 2021], pipeline and cable laying, windfarm constructions), maritime traffic, the fishing industry (Helsinki Commission, 2018), and beach visitors (Böttcher et al., 2011). With the growing development of offshore energy production in Europe, the topic of marine UXO and DMM became increasingly prominent over the last decade. Preparing for offshore construction and development requires the execution of a technical survey with the aim of detecting, subsequently identifying and ultimately removing legacy munitions. However, underwater UXO detection surveys are not restricted to marine environments; lacustrine, riverine, or estuarine environments are affected as well.
Being exposed to saltwater, the munitions' metal shells are gradually corroding and over time the contained explosive chemicals leak out into the marine environments. The fact that corrosion of the munitions' casings gradually reduces their detectability and accelerates the contamination of marine ecosystems, stresses the urgency of finding solutions for this issue. Recent toxicological studies suggest that these chemicals are currently unlikely to cause acute toxicity to marine organisms directly, but in the long run they may enter the marine food web and potentially affect human health through seafood consumption (Maser & Strehse, 2021).

| Background on maritime UXO surveys
Marine UXO surveys are typically performed by specialized companies using a standard repertoire of geophysical and hydroacoustic approaches. State-of-the-art UXO detection methods in marine environments are ship-towed magnetometer arrays, side-scan sonar (SSS), multibeam echosounder (MBES), and subbottom profiler (SBP; Frey, 2020;Wehner & Frey, 2022). SSS and MBES are capable of producing high-resolution images of the seabed in which surficial objects are detectable. Magnetic measurements can detect exposed and buried ferrous objects in depths of several meters depending on survey and sensor configurations as well as the sought-after objects.
Beyond that, buried ferrous objects and nonferrous munitions made from aluminum or austenitic steel are detectable by active electromagnetic (EM) systems, which measure electrical conductivities. While towed magnetic sensors are widely used for mapping large areas, EM systems are usually applied only for target point investigations. In contrast, SBP is another acoustic method which uses low-frequency soundings that penetrate the seabed and allow the detection of buried objects due to the sonar reflection from those objects.
After a UXO survey of an area of interest is completed, a list of potentially hazardous targets that may be UXO is produced. These targets are then investigated in detail to classify them as UXO or non-UXO. The necessary confirmation and characterization steps rely on visual inspections either through specially trained divers (Schultz, 2016) or the deployment of Remotely Operated Vehicles (ROVs) operated by qualified personnel. However, the vast majority of target points under investigation eventually turn out to be harmless debris, scrap, or natural objects, such as boulders. Guldin (2021) states that during UXO clearance campaigns in which their company has been involved, only 6% of the investigated targets is identified as UXO.
The ability to efficiently discriminate between hazardous munitions objects and the large amount of other anthropogenic metal waste is therefore an important factor in making offshore undertakings economically more viable. Lowering the expenses of munitions clearance by reducing the number of target points could in turn decrease energy costs for consumers and reduce costs of public infrastructure projects for the taxpayer.

| Project background
This paper is an outcome of the EU-funded project BASTA (Boost Applied munition detection through Smart data inTegration and AI workflows, www.basta-munition.eu). The project aimed at standardizing munitions detection approaches and at advancing data acquisition techniques through intelligent autonomous underwater vehicle (AUV)-based magnetic measurements. The overarching projects goal was to allow for an instant generation of a target list while surveying an area of interest. To do so, data acquisition, data quality assessment, and data interpretation were improved and object classification algorithms based on artificial intelligence (AI) were developed. While this greater goal is not addressed in this paper, the results presented here can contribute to its achievement in the future.

| AUV-based magnetic measurements
Here, we present the first results of integrating a set of three submersible 3-axis fluxgate magnetometers into a Girona 500 AUV (Ribas et al., 2012). According to our knowledge, our paper is the first publication presenting magnetic measurements on AUVs with hovering capability using a solid and fixed sensor construction. The hovering capability of the Girona 500 AUVs prevents the need to continuously sustain horizontal motion at a constant velocity to SEIDEL ET AL. | 849 maintain a fixed altitude above the seafloor, that is, a requirement of most torpedo-shaped AUVs (e.g., the REMUS AUVs made by the Woods Hole Oceanographic Institution). Instead, low operational velocities between 0.1 and 1 m/s and flight altitudes of 1 m or less above the seafloor allow for a precise underwater navigation and target localization. For ground-truthing purposes and for a concluding confirmation or rejection of suspected UXO targets, the AUV is equipped with a high-resolution camera system. We developed a system that autonomously investigates potential targets that were located by methods for large area surveys (like, MBES, SSS, or towed magnetometer arrays) beforehand and thus overcome some limitations of diver-deployed or ROV-based individual underwater UXO target inspections.
An AUV in operation mode is expected to be a significant source of EM noise. Hence, integrating sensitive instruments like fluxgate magnetometers into an AUV, which contains electric propulsion systems and other devices producing EM noise, gives rise to the challenge: How to reduce the AUV-inherent and motion-induced magnetic noise without overly affecting the maneuvering capability and automatic positioning control of the vehicle?
The first feasibility studies of integrating magnetometers into AUVs and related noise mitigation approaches appeared around the millennium change (Allen et al., 2001;Wynn & Bono, 2002). Later, Armstrong et al. (2009) conducted magnetic field measurements deploying a fluxgate magnetometer right underneath a small, torpedo-shaped AUV to magnetically investigate surface vessels with a fleet of AUVs equipped with magnetometers. The authors state that when traveling in a straight line, the measured signals have standard deviations of 21 nT. Further developments of the same project, mainly calibration methods and an increase in the distance between the sensors and the propulsion systems, reduced the noise floor to less than 10 nT (Walker et al., 2011). Pei et al. (2010) installed a set of fluxgate magnetometers at the nose of a torpedo-shaped AUV to form a magnetic gradiometer and conducted field tests. In their paper, the authors discussed a noise compensation method, a sensor calibration approach, and a UXO object parameter estimation by inversion but did not publish any sensitivity thresholds of their system. Tilley et al. (2012) investigated the performance of an Overhauser magnetometer towed behind a REMUS 100 AUV.
According to the authors, the system is characterized by a vehicleinduced sensitivity error of less than 1 nT with a sensor-to-vehicle distance of 30 ft (~9 m). Gallimore et al. (2020) presented their work about integrating scalar magnetometers into the nose of a REMUS 100 AUV. The work includes the development of different digital filtering approaches to improve signal quality and to allow for automatic real-time target classifications on-board the AUV. Moreover, the authors discuss methodologies to correct magnetic data for interference and errors induced by the AUV to improve overall sensor performance. All aforementioned publications have in common that the utilized platforms are torpedo-shaped AUVs that cannot maintain a stationary position.
To mount fluxgate magnetometers onto the Girona 500 AUVs with hovering capabilities, we developed an aluminum construction to which the magnetic sensors are rigidly attached at a distance of 2 m to the nose of the AUV. Thereby, the maximum distance to the rear thrusters, which are considered to be the main source of platform noise, is approximately 3 m. The three magnetometers are arranged in the shape of a vertical triangle perpendicular to the direction of travel. This allows for the calculation of all three spatial magnetic gradients while the vehicle is moving straight ahead at a constant velocity. More than 100 surveys were performed visiting known munitions dumpsites at Kiel Bay, Eckernförde Bay, and Lübeck Bay (all in the German Baltic Sea). After data acquisition, the recorded data were processed in several steps, like, positioning correction and lowpass filtering. With a noise floor of approximately 2 nT, the system is capable of reliably detecting munitions similar in size to 81 mm shells at a distance of 1 m. While facilitating precise object localization using a fixed and solid construction, this result is a significant improvement in terms of the noise floor, when compared to the aforementioned publications that deal with the integration of magnetometers into AUVs. Under decent environmental conditions, that is, low current velocities and thus low thruster activity, even smaller objects can be detected with this system. We do not claim that our system outperforms other systems in the field of UXO detection and classification. Instead, the system described here is a new addition to the portfolio of those who are dealing with detection and classification of UXO.
The main achievements of our research are: • The successful integration of submersible magnetometers into a hovering AUV.
• The lateral offset of 2 m between the AUV and the magnetic sensors yields a noise floor of approximately 2 nT during operation.
• The fixed and solid aluminum construction allows for a precise object localization and stable sensor control during operation.
• Arranging the three magnetometers on the shape of a vertical triangle allows for the calculation of all three spatial magnetic gradients.
• Test measurements of surrogate objects in a controlled environment indicate that the system is capable of detecting UXOs with magnetic moments of 0.1 Am 2 .

| METHODOLOGY
AUVs are untethered and self-propelled submarines that emerged during the second half of the last century as an alternative platform to traditional research vessels that suffer from shortcomings related to their size, mobility, and surface restriction. For the management of UXO, unmanned systems are considered a preferred option in an effort to reduce human risk (SERDP & ESTCP, 2007). AUVs usually operate completely underwater, beyond the control of and with limited communication with any piloting personnel while conducting preprogrammed missions. They navigate autonomously and therefore provide an efficient method to conduct fast and precise surveys.
While ROVs are usually wire-connected to a vessel and therefore come with an "infinite" power supply, AUVs have to cope with a limited amount of energy and consequentially with payload restrictions. Additional challenges for AUVs to deal with are the danger of collision or entanglement with underwater obstacles (submerged ropes, fishing gear, etc.) as well as difficulties regarding navigation in areas with strong water currents. On the other hand, AUVs are usually relatively small, easily transportable, and independent from the vessel which can perform other tasks during AUV missions. In comparison to vessel-towed systems, AUV-based surveys can provide a much higher spatial precision (e.g., for the navigation on narrow survey lines) and resolution. This is a consequence of lower flight altitudes and of the fact that towed systems usually depend on the navigation precision of the vessel, from which they are tens or  . Combining these gradients, the 3D analytic A (see Equation 1) signal can be calculated, a derived value that exhibits maxima over magnetization contrasts and that determines the outlines of magnetic sources (Roest et al., 1992). (1) Comparing single spatial components of the fluxgate magnetometers requires a precise and stable relative alignment of all three sensors which is challenging from an engineering point of view. Thus, only total magnetic intensity (TMI) values are considered at the current stage of the project. During the development of the system and data acquisition for the presented results, operational altitudes above the seabed and velocities of the new system were usually between 1 and 1.5 m and between 0.2 and 0.5 m/s, respectively. In most cases, the line spacing during magnetic AUV surveys was 1 m. for these survey parameters.
The system is designed for operation only in areas where the bathymetry is sufficiently known. As the DVL, which is measuring the altitude above the seafloor, is installed at the back of the AUV, the lateral distance between the DVL and the magnetometers is >3 m. Operating in unknown territories would F I G U R E 2 GEOMAR's Girona 500 AUV "Luise" with three magnetometers (S1 = top, S2 = bottom-port, and S3 = bottom-starboard) and the camera system incl. light-emitting diodes.
The horizontal distance between the tip of the AUV and the magnetic sensors is approximately 2 m. The data acquisition unit (DAU) is positioned near the center of the AUV. The entire magnetometer construction incl. the sensors (but without the DAU) has a total weight of approximately 5 kg. AUV, autonomous underwater vehicle; DVL, Doppler Velocity Log.
T A B L E 2 Typical values of parameters during the UXO surveys using the system.

| Noise
We distinguish between AUV-inherent platform noise and motion- The rotation rates of the thrusters are therefore volatile and unpredictable. All thrusters were active to a medium degree for the time of the measurements, that is, the AUV was not idle. The vertical magnetic gradient, calculated from the lowpass-filtered TMI values of sensor 1 and the average of sensors 2 and 3, is displayed in Figure 3f.

| AUV-inherent noise
The gradient values during the noise measurements are below ±1 nT/m and the standard deviation is~0.3 nT/m. The negative peak of~−1 nT/m during the first seconds is related to noisy data of sensor 1 in that period.
Power spectrum density (PSD) plots of the magnetic noise and corresponding thruster data are displayed in Figure 4. The data were collected during a noise investigation maneuver with a sampling rate of 200 Hz when the AUV was moving at 0.5 m/s in the water column. The activity of the heave and yaw thrusters was around 300-400 RPM, denoting frequencies of 5-7 Hz (marked with "A").
For the surge thrusters, the RPM values were around 700, related to frequencies between 10 and 12 Hz in the PSD plot (marked with "B").
The peaks C (~26 Hz, central sensor) and D (~45 Hz, lower sensor) in the PSD plot cannot be explained by thruster activity and are probably related to supplementary on-board devices.

| Motion-induced noise
Motion-induced noise is generated by changes in the vehicle attitude, that is, the yaw ( Figure 5). This approach caused significantly higher thruster activities and therefore higher noise levels.
Motion-induced noise is lowest when the AUV is traveling at a constant velocity in a constant direction or when it is maintaining a fixed position. This is particularly true, when the AUV is oriented in parallel to the direction of the ambient water current. When planning AUV missions that consist mostly of parallel tracks, like, typical lawn-mowing patterns, we usually set the main mission direction to be parallel to the expected water currents. This reduces unnecessary thruster activity, which would arise as a response to transverse water currents to straighten the AUV track lines.

| Synthetic magnetic dipoles
To assess the detection capabilities of the new AUVmagnetometer construction under controlled conditions, we placed 10 surrogate objects containing magnetic dipole moments of known strengths between 0.1 Am 2 (DP-10) and 30 Am 2 (DP-01) in a naval harbor in Kiel. The benefits of this test site were low water currents and the existing infrastructure. Disadvantages are an increased density of sources of magnetic interference like submerged metallic waste, as must be expected in a harbor, and nearby anchored or passing ships.
In the first step, the designated testing area was scanned by an  Table 3. With the dipoles put in place, the main AUV mission was launched covering an area of approximately 80 × 24 m 2 .
The resulting analytic signal of this mission is displayed in Figure 6b (blue) on top of the reference measurement (gray). We placed the dipoles in a magnetically relatively undisturbed corridor. This was fortunate, as manually laying the magnetic dipoles precisely on their seabed destinations, while navigating a small boat at a minimum velocity, was challenging. Acquiring meaningful images with the AUV on-board camera at an altitude of 1.8 m was impossible due to poor underwater visibility conditions in the basin. Hence, no photographic evidence of the synthetic objects could be gained. As seen from the comparison of the reference and the main survey data, the weakest dipole DP-10 with a magnetic moment of 0.1 Am 2 was not reliably detected, DP-09 (0.2 Am 2 ) was detected with an acceptable confidence level, all stronger dipoles were detected convincingly. Moreover, dipole DP-10 was unintentionally dropped outside the area of the reference measurement. The annotated locations of dipoles DP-01 to DP-09 are manually picked from the analytic signal of the main mission. The location of DP-10 was determined from anomalies in the raw magnetic data that resembled the expected magnetic patterns. However, there are strong indications that dipole DP-10 was successfully detected by the magnetometers as well. Figure 7 shows a section of the raw and filtered TMI anomaly data of all three F I G U R E 5 Magnetic field intensity, AUV heading in degrees, and velocity during a circular calibration mission. AUV, autonomous underwater vehicle; TMI, total magnetic intensity.
sensors along with synthetic data of a vertical (upward-directed) magnetic dipole of 0.1 Am 2 moment strength for flight altitudes of (a) 2.1 m, (b) 1.9 m, and (c) 1.6 m. The raw TMI data were downsampled to 20 Hz.
Although the designated flight altitude of the lower sensors 2 and 3 was 1.8 m, the synthetic data for an altitude of 1.6 m (blue) fit the recorded data of sensor 3 best (Figure 7c). This can be explained by the approximately 3 m horizontal offset between the magnetometers and the DVL, which is measuring the vehicle altitude. In 3 m distance to the DVL location, where the sensors are located, the bathymetry can easily vary by 20 cm. Other possible explanations are that the dipoles are elevated~10 cm by design or that the dipole landed tilted on the seafloor.
Furthermore, it is unclear whether the sensors passed the dipole precisely above or whether a lateral offset of a few decimeters (considering the 1 m line spacing) occurred. The altitudes of the synthetic data in Figure 7a (green) and 7b (orange) are 1.9 and 2.1 m, resembling realistic distances.
The synthetic data fit the measured data to a decent degree in all three cases. Yet, we cannot prove whether these anomaly signals are related to DP-10, but they are very similar to signals of an upward-directed vertical magnetic dipole of moment strength 0.1 Am 2 . An additional indication is that these signals are recorded at a location where DP-10 was expected to be located (see Figure 6). The synthetic dipole moments were

| Field data
During the BASTA project, several single-day and multiday cruises were conducted in areas of the German Baltic Sea including several munitions dumpsites. The majority of those field campaigns took place at the Kolberger Heide, a marine munitions dumpsite in Kiel Bay with water depths ranging from 8 to 20 m. It was established as a munitions dumpsite after WWII by order of the British military forces.
According to historic records, an estimated 30,000 tons of munitions, including torpedoes, sea mines, ground mines, depth charges and smaller objects like grenades and artillery munitions can be found here (Kampmeier et al., 2020).  where the presence of a UXO object is suspected. If the investigation yields a confirmation of the UXO suspicion, clearance commences.
The magnetometry on hovering AUVs described in this paper can be added to the toolbox for these phases of the EOD workflow.
While theoretically possible, the technical survey of larger areas with the system described here is not recommended due to a lack of efficiency. Even though, survey speed depends on the required alongtrack data point spacing, moving with the maximum 2 kn of the AUV is substantially slower than the 10 kn, which are considered possible for towed systems (McDonald, 2003). In addition, a towed array can have  Another area of application is harbor areas, which are difficult to navigate for large vessels. Here, the AUV may even be more efficient than small boats with fixed-frame magnetometer systems. Acquired data may be superior, due to the AUVs ability to compensate for roll, pitch, and yaw, which is more challenging to achieve with a surface vessel.
During technical surveys of large areas, it would be useful to have a number of the AUVs on standby at the water surface. Once one of the above scenarios arises, the AUV could be dispatched to the concerned location and perform a survey pattern. During the use of towed systems, gaps may be introduced, where adjacent survey lines were not sufficiently close to each other to detect a target object. These gaps could be filled instantaneously by the AUV system, which may be more efficient than having to close it with the vessel. Finally, the system can be used to aid the generation of the target list, by acquiring additional and more detailed information on the target point locations, while the surveyor continues performing the campaign on the larger area.
The system may also assist during the investigation of target points, which is commonly done by another organization than the survey of the area of interest, due to the fact that target point investigation requires knowledge of the handling of explosive materials (Frey, 2020). This organization may arrive months after the original survey was executed and the target object may have been moved as a consequence of natural mechanisms or anthropogenic influences, such as bottom trawling (Böttcher et al., 2011). The hovering AUVs can seamlessly move from target point to redetect them. Simultaneously, the EOD specialists on the vessel can focus on the clearance of those targets, where the suspected object turns out to be UXO. After clearance or in situ detonation has been performed, the AUV can return to the site to perform an as-left survey as proof of the operation or to ensure that the point is free of further anomalies that may be buried underneath the original object.

| CONCLUSIONS
In an effort to autonomously perform underwater UXO target investigation operations, we installed three fluxgate magnetometers in the shape of a vertical triangle on a Girona 500 AUV with hovering capabilities using a solid construction. The AUV-based magnetic noise was primarily reduced by increasing the distance between the AUV and the magnetometers to 2 m, which does affect the maneuverability and positioning control of the vehicle but to an acceptable degree. This system allows for measuring all three spatial magnetic gradients with noise floors <2 nT when traveling at a constant speed and direction. Supplementary photographic images may assist with the object classification. A survey under controlled conditions with surrogate objects showed that by using the system it is possible to detect even small UXO objects. Being able to detect objects with magnetic dipole moments around 0.1 Am 2 at altitudes of 1.8 m potentially allows for the detection of projectiles as small as, for example, 60 mm grenades (Billings et al., 2002). Even smaller objects become detectable at lower altitudes. Under favorable environmental conditions, objects with the size of hand grenades become detectable with this system (Stanley & Clark, 2003).
Drawbacks of the system are its inability to operate in unknown or uneven territories because of possible ground collisions or obstacle entanglement. Furthermore, operating in areas with water currents of more than 2 kn is impossible because the maximum velocity of the Girona 500 AUVs is 1 m/s. This issue could be resolved by attaching the magnetometers to an AUV with a higher operational velocity. However, this would certainly introduce new and unforeseeable challenges depending on the replacement AUV.
We point out that the navigational precision of the Girona 500 AUV with a 2-m long sensor arm is certainly affected by strong water currents. Turnaround maneuvers at the end of each survey line are more time-and energy-consuming than turnarounds without the sensor arm. Over the course of the BASTA project, more than 100 AUV missions using magnetometry were conducted in different munitions dumpsites (e.g., Kolberger Heide, Lübeck Bay) covering a variety of different water current velocities. Even with the 2 m long sensor arm, the intended line spacing was usually maintained to a satisfying degree, which could be verified through photomosaics of the surveyed sites and by using INS data to plot the AUV's positions.
The lower the ambient water currents are, the better is the overall data quality. Moreover, water currents do have an impact on total mission endurance because of higher battery power consumptions.

| FUTURE WORK
To gain higher levels of certainty of the system's performance and to further prove its potentials and limits, the project consortium aims at running advanced tests in an underwater UXO test site with magnetic reference targets in 2023. To allow the system to be also operated in unknown territories, an obstacle avoidance system for the Girona SEIDEL ET AL. | 859 500 AUVs using a 45°downward looking multibeam device is currently being developed at GEOMAR.
We are currently working on the development of a backseat driver algorithm that independently creates new waypoints or entire AUV mission tracks based on in situ magnetic measurements. During a 2-weeks cruise in October 2022, the first tests were conducted indicating the huge potential of this approach. In the future, the system is supposed to be able to autonomously detect, locate, investigate, and assess magnetic anomalies and eventually generate an object report for each individually investigated target.
Beyond its application within UXO detection and classification surveys, the system can potentially act as a cable tracking tool or be utilized during archeological underwater surveys.

DATA AVAILABILITY STATEMENT
Selected data available upon request that does not contain coordinates due to security reasons (may contain positional information of potential munition objects).