1 Introduction

Two major wars took place in Europe in the twentieth century. Unexploded ordnances, which have been fired, dropped, or launched but failed to detonate have been found in shallow waters surrounding the European continent. It has been estimated that 1.6 million tons of ammunition from the two World Wars (WW1 and WW2) remain in the Baltic and North Sea. These dangerous devices may explode due to an accidental disturbance or even reaction with salt water, causing environmental damage. Several spontaneous detonations have been observed and recorded at local seismic networks (Albright, 2012). To avoid accidental explosions, ordnances found close to human activity are removed, moved into designated areas, and destroyed in a controlled way (Robinson et al., 2020). An attempt is made to reduce the volume of explosive material in a process of deflagration; however, explosion of ordnances may never be ruled out. Authorities issue warnings to residents to avoid entering areas endangered by the controlled explosion. Therefore, the time and geographical location of these events are known publicly, and they provide ground-truth events, useful in the evaluation of the detection and location processing performed at the International Data Centre (IDC) of the Comprehensive Nuclear-Test-Ban Treaty (CTBT) Organization (CTBTO). The main purpose of this paper is to document these in-water explosions as they provide examples of low-magnitude events of interest to the CTBTO. It also allows to complement the few yield-magnitude relationships for in-water explosions available in the literature (Gitterman & Shapira, 2001; Heyburn et al., 2018) on the lower side of the magnitude scale. Analysis of these events contributes to assessing both the location accuracy of detected events and magnitude estimates calculated by the IDC algorithms with known charge size of explosive materials.

The International Monitoring System (IMS) is a vital part of the CTBT verification regime. It is a global network of four technology stations, split between radionuclide and waveform technologies. The waveform technologies stations, namely seismic, hydroacoustic and infrasound have been designed to detect signals from nuclear test explosions and locate seismic events triggered by these explosions. Seismic stations are primarily used to detect underground nuclear test explosions; however, they also record other signals originating from both natural phenomena (e.g. earthquakes) and man-made (e.g. chemical explosions) (CTBT Treaty, 1996; CTBTO Verification Regime, 2018). All events which have been detected by at least three seismic stations are included in a so-called Late Event Bulletin (LEB) and a subset of these LEB events is published in the Reviewed Event Bulletin (REB), which is one of the main products of the IDC.

The IMS hydroacoustic network, which is composed of the smallest number of stations, is designed to detect underwater explosions. Six underwater hydrophone stations are particularly sensitive and detect even small explosions with yields of a few kilograms. Events shown in this analysis took place in shallow waters with no clear underwater path to any of the IMS hydroacoustic stations, which prevented hydroacoustic network from detecting these underwater events. However, they were detected by regional IMS seismic networks as seismic and hydroacoustic networks complement each other. Detection threshold of the IMS seismic stations is much higher than the hydroacoustic threshold. It is fortunate that seismic stations provide coverage in areas not covered by the hydroacoustic network and that in-water explosions couple better to the surrounding media than underground explosions, triggering events of larger magnitudes for the same yield.

Several controlled underwater detonations, powerful enough to be recorded by the IMS network, were examined during this study. Position and charge size of WW2 ordnances were documented in the press (Sielski, 2020; Madej, 2020) as they were of interest to residents of affected areas. This paper includes results of analysis of events linked to controlled underwater explosions of WW2 ordnances mostly in 2020. Examined events were triggered by devices of different charge sizes at several locations. Other previously detected WW2 ordnance underwater explosions (Larsen et al., 2013; Lenhardt et al., 2013) could be compared to events from 2020. Recordings of chosen detonations demonstrate how accurate these relatively small explosions can be located by the IMS network. The charge size of the explosive material is gauged against the event magnitude in this special case of underwater explosions. A few papers have been published in the literature empirically deriving the relationship between charge size and magnitude and these relationships are compared to the classic underground yield-magnitude relationship (Murphy, 1981). This paper complements the literature of observations on the lower range of magnitudes.

2 Bulletins Available at the IDC

The IMS seismic network has been designed to detect and locate underground seismic sources. When completed it will consist of 50 primary and 120 auxiliary seismic stations. As of January 2022, 44 primary and 109 auxiliary stations are part of the network. The primary network, consisting mostly of arrays of seismometers, is continuously collecting data, and transferring them in almost real time to the IDC via Global Communication Infrastructure (GCI). Once received, waveforms from each station are processed to extract signals from the noise background. Sets of detections from different stations are combined to form events with the help of the Global Association algorithm (GA) (Hanson et al., 2001; Le Bras et al., 1994), which produces automatic bulletins, called standard event lists. The first list of events, Standard Event List 1 (SEL1) is ready one hour after the real time. Events listed in SEL1 are used to request data from the auxiliary network to improve event location. The next automatic bulletin, Standard Event List 2 (SEL2) is available after four hours. The last and most complete automatic bulletin, Standard Event List 3 (SEL3) is available after six hours. This automatic product is reviewed by human analysts. All events considered as valid are improved and included in the LEB. LEB also contains events which were missed during automatic processing and have been manually curated by waveform analysts. Events in the LEB, which pass an event definition criterion consisting of a weighted sum of defining attributes (number of defining time, azimuth and slowness attributes) are promoted to the REB. The REB is a product distributed to the member states of the CTBT, LEB is available internally but also shared for research purposes via the virtual Data Exploitation Centre (vDEC) platform (Vaidya et al., 2009; vDEC, 2021). All automatic products are also distributed to the member states of the CTBT.

There are ongoing efforts to improve network automatic processing in the IDC. Currently GA is complemented with NET-VISA (Network Processing Vertically Integrated Seismic Analysis, Arora et al., 2013). This algorithm uses a Bayesian approach to seismic network processing. An automatic bulletin, VSEL3 (NET-VISA SEL3), is produced by NET-VISA and available at the same time as the SEL3 bulletin and uses the same set of input detections. NET-VISA has been shown to perform better than GA when measuring the overlap with the REB (Bondár et al, 2019; Le Bras et al., 2017, 2021). This means it misses fewer events and produces a completer and more accurate automatic bulletin. Since January 2018, it is used by IDC analysts in complement to GA. After they finish reviewing the SEL3 bulletin, analysts are presented with the VSEL3 events that were not obtained by GA and review them as well. This step is important, especially in the case of events with a small number of associated stations.

3 IDC Bulletin Events Related to Controlled WW2 Ordnance Explosions

In 2020 several published news reports described efforts to reduce the danger of WW2 ordnances. Three areas were identified where WW2 ordnances were destroyed in 2020: Gulf of Gdansk, Polish-German border (both in Poland), and vicinity of Brest (France). Some additional examples were found, namely from offshore Denmark from 2012, as well as an example of explosion in the Danube River, near Vienna, Austria, in 2012 (Lenhardt et al., 2013).

These five areas were examined to check whether the REB contained any additional past events. Next, press articles were searched to identify whether these newly found events could relate to WW2 controlled explosions of ordnances, which would confirm their origin. This was successful in case of the Gulf of Gdansk.

An attempt was made to estimate the accuracy of automatic event location. In case of REB events identified as related to controlled underwater explosion, SEL3 was searched for a corresponding automatic solution. Both automatic and reviewed locations were compared. Based on the location given in the news, which quoted the Polish Maritime Authority, the IDC reviewed solution was compared to the reported position of the explosive source. If the news report provided the size of explosive materials, it was possible to investigate the relationship between the reported size and the IDC calculated magnitude.

Analysis of all IDC events was done without prejudice to the nature of the source. Free depth solution was preferable if depth uncertainty was not too large. All events chosen for this study were located at depth of 0 km.

Location accuracy and list of stations which detected analysed events are shown in Table 2. IMS network is sparse so it is expected that location could be improved if observations from other stations contributed to the solution. For some events it was possible to find more accurate locations in other bulletins (i.e. ISC bulletin). In this case IDC location was compared to the more accurate one.

4 Events Related to Controlled WW2 Ordnance Explosions Listed in the REB Bulletin

Nine REB events were found around the Gulf of Gdansk between 2017 and 2020. For three events it was possible to find the explosion place given by authorities, the Maritime Office in Gdynia (Poland), and reported in the news. Events with known explosion location are listed in Table 1 (1–3), (Figs. 1, 2).

Table 1 Events related to controlled WW2 ordnance explosions
Full size table
Fig. 1
figure 1

IMS seismic stations which detected events related to controlled WW2 ordnance explosions described in this study. Location of explosion sites are marked with red (A, B – Gulf of Gdansk, C- German-Polish border, D – Offshore France, E – Offshore Denmark, F – Danube river Austria). Blue circles mark primary seismic arrays, green circles mark auxiliary seismic arrays, an auxiliary seismic three component station is marked with a green triangle. Infrasound station I26DE, which contributed to location of event 8 in Table 1 is collocated with seismic array GERES

Full size image
Fig. 2
figure 2

Location of REB events in the Baltic Sea area; IDC location marked in blue, reported controlled explosion marked in orange. Ellipse indicates 90% confidence area. Position of explosion site in Fig. 1 marked with red letter, event number as listed in Tables 1, 2

Full size image

Events which had detections associated to SEL3 events are indicated as Yes in the “SEL3” Column. If the event had associations to VSEL3 events, it is indicated as Yes in the “VSEL3” column. All automatic events were poorly located due to detections falsely associated to the automatic event. The REB events were systematically located in the area of controlled explosions and the related 90% confidence ellipses included the ground-truth explosion location, which was found in the news reports. Location accuracy can be seen in Table 2. All events were located at less than 20 km from the place given by the authorities.

Table 2 IDC location of WW2 ordnance explosions
Full size table

Events in the area of Gulf of Gdansk were recorded at four seismic stations located in Scandinavia, HFS (Sweden), FINES (Finland), NOA and ARCES (both in Norway) (the naming convention used is in agreement with the International Registry of Seismograph Stations). The reported size of explosive materials was one Ton, and the ML magnitude calculated at the IDC based on the IMS network measurement ranged between 2.3 and 2.4. ML is a local magnitude calculated in IDC from the incoherent beam filtered in the frequency band 2–4 Hz.

It is noticeable that events in the Gulf of Gdansk were well recorded as far as 20 degrees to the north (i.e. ARCES) but not at stations to the south. For instance, GERES (Germany) is a high-performance array located at a similar distance from the Gulf of Gdansk as the closest IMS station in Scandinavia (670 km from HFS and 770 km from GERES). Intuitively the explosions in the Gulf of Gdansk should have been detected at GERES, which was not the case. This phenomenon was already observed during the GSETT-2 experiment (Schweitzer, 1995). The reason for the lack of signals at GERES could be explained by the difference in regional propagation conditions. Europe’s continental basement can be divided into two distinct regions: in the north and east a stable Precambrian craton known as the East European Platform (EEP), and in the south and west a mobile belt (Tesauro et al., 2008). These two regions are divided by the NW–SE-trending Trans-European Suture Zone (TESZ) also known as the Tornquist-Teisseyre Zone (TTZ). The fault line starts at the North Sea and crosses the continent until the Dobrogea region of the Black Sea. The Moho depth between the TTZ in Central Poland is intermediate between that in the EEP to the east (45 km) and the Palaeozoic Platform to the south-west (30–35 km). Schweitzer et al. proposed a model of low velocity zone below the TTZ which could explain blockage of seismic energy along the TTZ. They also suggested that this very heterogenous area will cause scattering of P wave energy and reflecting it back to the mantle. This fault is on the seismic propagation path between the source and GERES.

In October 2020, one of the largest WW2 ordnances was destroyed in the area of Baltic Sea shore near the German-Polish border, with a charge size of 5 Tons. Sappers planned to defuse the mine by deflagrating explosive material. It was considered that deflagration may turn into explosion which took place during this process. Related REB event was located in the canal which links the Baltic Sea with Szczecin Lagoon. All other REB events are located further north in the Baltic Sea at least 70 km away from the WW2 ordnance event.

This event was of a higher magnitude than events in the Gulf of Gdansk. In this case, a SEL3 solution was not formed, but it was possible to combine observations from seven stations (VRAC, GERES, HFS, NOA, FINES, EKA and ARCES) to include this event in the REB. Figure 3 shows generated signals recorded by seismic arrays GERES, HFS, NOA, FINES, EKA, and ARCES. Unlike explosions in the Gulf of Gdansk this event was recorded by both GERES and the Scandinavian stations, together with VRAC (Czech Republic) and EKA (United Kingdom). A reason for that distinct detection pattern may be that the explosion took place at a location with different propagation conditions. However, the main reason is more likely related to the larger size of the explosion. It is worth noticing that unlike for other stations, Pn was not automatically detected at GERES with Pg being the first arrival. This could pose a difficulty to the automatic system to correctly locate the event, which could be one of the reasons why this event was not formed by GA. The IDC local magnitude was estimated at ML = 2.9. It is worth mentioning that the event at the Polish-German border was listed in the VSEL3, which helped the IDC analysts to include it in the REB, although the initial location was far from the explosive source location.

Fig. 3
figure 3

Waveforms recorded at IMS array stations which detected the event at the Polish-German border. Signals filtered with band-pass filter for the best signal to noise ratio. Pg is the first arrival detected at GERES by the automatic system which could be one of the reasons why this event was not formed by GA

Full size image

Locations of all REB events described in this section has been shown in Fig. 2.

5 Events Related to Controlled WW2 Ordnance Explosions Detected by the IMS Stations But Not Listed in the REB Bulletin

On 15 September 2020 a controlled underwater explosion took place around Brest in France. This event was detected at two IMS stations EKA and ESDC (Spain). All other LEB events recorded in this area were built from associations of signals propagating in the atmosphere, that were recorded at infrasound stations. They are presumably related to above the ground anthropogenic activity, possibly from military supersonic aircrafts.

It is possible that other events linked to controlled explosions off-shore France were recorded by IMS seismic stations as well. They might however not have been included in the IDC bulletins since the number of detecting seismic stations was below the required threshold of 3. The reported charge of the WW2 ordnance was 850 kg, and the IDC local magnitude of this event 5 in Table 1 is ML = 3.0.

Event close to Brest was detected by only two stations, location was based on Pn arrival parameters. From eight investigated examples, location calculated at the IDC was at the largest distance from the ground truth, and so was the largest size of the 90% confidence ellipse (Fig. 4 D). Improved event location was found in the ISC bulletin. Additional local and regional network enabled to locate this event at the distance of 6 km from the source, whereas it was over 50 km in case of the IDC location.

Fig. 4
figure 4

IDC of events related to controlled WW2 ordnance explosions detected by the IMS stations but not listed in the REB; IDC location marked in blue, reported controlled explosion marked in orange. Ellipse indicates 90% confidence area. Position of explosion site in Fig. 1 marked with red letter, event number as listed in Tables 1, 2

Full size image

Several events related to WW2 ordnances explosions have been recorded in Danish waters. Larsen et al. (2013) presented a list of examples of large explosions in 2012 with the ground truth information provided by the Danish Navy. Publication contains a list of six events. One of them was found in the REB (event on 8 September 2012). The IDC location of this REB event was at a large distance (63.5 km) from the ground truth given by the authorities. The paper also mentioned large discrepancy between what is calculated and the ground truth location despite excellent recordings of this event.

Two other events from the paper (listed in Table 1 as event 6 and 7) were also detected by IMS stations. Both explosions were of relatively large yield (788 kg). Events were detected at HFS, NOA, EKA and FINES but were not present in SEL3 or REB. VSEL3 was not being produced in 2012. Number of detecting primary stations was below the necessary threshold for an event to be included in the REB.

Two events located at the IDC were up to 24 km away from the position given by the Danish Navy (Fig. 4 E). Location deviation mentioned by Larsen et al. (2013) was smaller than 10 km, but events were located by a denser network than the IMS.

During WW2 250 kg bombs were dropped on Vienna, Austria, causing extensive damage. It is estimated that over 21.000 unexploded devices have been recovered and neutralized since 1945. One of the WW2 ordnances was destroyed in the river Danube close to Vienna on 25 August 2012. Seismic waves were felt in many districts of the city. Two agencies, Zentralanstalt für Meteorologie und Geophysik (ZAMG) in Austria and the Institute of Physics of the Earth (IPE) in the Czech Republic recorded and located a related event that was saved in their national event catalogues. Both agencies could locate the source with a precision of 1 km from the ground truth. IMS auxiliary station VRAC also belongs to the IPE network and contributed to their location.

There was no relevant event in the IDC bulletins. However, upon closer examination, recordings were found of this event on the primary seismic array GERES as well as infrasound signal of characteristics and parameters consistent with the source at a collocated infrasound station I26DE.

The IDC applies Progressive Multi Channel Correlation (PMCC) (Mialle et al., 2019) algorithm to detect infrasound signal and estimate related wave parameter such as frequency content, direction of arrival or apparent trace velocity. Signal at I26DE was automatically picked up by IDC automatic processing. Figure 5 displays signal recorded by I26DE and the related information obtained by PMCC (the results have been reprocessed offline to enhance signal attribute extraction).

Fig. 5
figure 5

Infrasound detection at I26DE (Germany) related to the Danube River event. The top two panels display the DTK-PMCC (specialized software for station processing of array data based on PMCC method) results in back azimuth (108) and trace velocity (0.35 km/s) and the bottom panel exhibits the beamformed waveform for the displayed detection (frequency band 0.9–7 Hz). Note that detections not related to the event have been masked

Full size image

IDC location based on readings from GERES and I26DE was 7 km from the reported explosion location (Fig. 4 F).

6 Relating Charge Size to Seismic Magnitude for In-Water Explosions

Event magnitude is a measure of energy generated by a seismic source. The first seismic magnitude (ML) was introduced by Richter in the 1930s (Lay, 1995). His aim was to characterize earthquakes in the region of California which were measured at the same type of seismometer—Wood-Anderson torsion instrument. Magnitude was calculated as a comparison to the reference event, chosen as displacement of 1 µm at an epicentral distance of 100 km in a logarithmic scale. In the IDC signal amplitude for ML calculation is measured on an incoherent beam of vertical component. Amplitude maximum is picked in one second widow within four seconds from the arrival time, signal is filtered between 2 and 4 Hz. ISC computes magnitudes based on measurements provided by agencies which contributed their data (Di Giacomo, 2016).

ML is used for events recorded at local and regional distances. Event examples analysed in this paper were recorded at such distances, so ML was used to describe their size. The limitations of distance range make ML not suitable for global characterization of earthquake size. At teleseismic distances body wave magnitude (mb) is used. It is an amplitude over period measurement in the logarithmic scale with corrections for distance and depth. In the IDC amplitude and period measurements for mb calculations are made on vertical component coherent beam filtered between 0.8 and 4.5 Hz within 5.5 s from the arrival time.

Only a fraction of energy released during an in-water explosion couples into the ground and propagates as converted seismic waves. The efficiency of the coupling from in-water acoustic to in-ground seismic waves depends on the mean water depth, variations in water depth, seabed properties in the region close to the explosive source, as well as the frequency of underwater signal near the source. Many of these parameters are unknown and may affect the empirically determined charge size and magnitude relation of the explosive source. This causes uncertainty in relating yield from an in-water explosion to ML derived from seismic signals especially when comparing different explosions at various locations.

It is remarkable that comparatively small yield explosions are detectable when they are triggered in the water. The magnitudes recorded are approximately 1 order of magnitude higher than expected for explosions of the same yield when conducted underground (e.g. Murphy, 1981).

Figure 6 shows the magnitude-yield relationship resulting from this analysis for the WW2 ordnances listed in Table 1. Other examples of in-water explosions for which seismic magnitudes exist are also included in the graph and the values of their estimated yield and magnitude are listed in Table 3. The source for the CHASE explosions, shown in yellow on the graph, is from Talandier and Okal (2004), the source for the 2016 explosions offshore Florida, shown in purple on the graph, is Heyburn et al. (2018), the source for the 1999 Dead Sea explosions, shown in orange on the graph, is Gitterman and Shapira (2001), and the 2021 offshore Florida explosions use the IDC estimates for the magnitudes. The values for the events studied in this paper are shown in red on the graph and identified through a label. They are on the lower end of the magnitude and yield scales. Also shown on the graph are three lines: (1) the standard Murphy (1981) line; (2) the line derived by Gitterman and Shapira (2001) for the 1999 Dead Sea experiments; and (3) a line derived from the standard Murphy relationship by adding one unit of magnitude to its constant while keeping the same slope. It is labelled ‘modified Murphy’ on the graph. As expected, all in-water explosions fall well below the standard Murphy (1981) relationship which was derived from in-ground explosions. We propose this new magnitude-yield relationship for in-water explosions. It has the advantage of simplicity, being based on the widely used Murphy relationship. It uses the same slope and is just a static adjustment of one magnitude unit on all yield ranges. While there are outliers that do not fall close to the line in the log–log graph, it does account better than the Gitterman (2001) relationship, which has a different slope and is derived empirically for the very specific environment of the Dead Sea, with its high water density due to the salt content.

Fig. 6
figure 6

Relationship between magnitude (the type of magnitude used is given in Tables 1 and 2) and yield for underwater explosions with values published in the literature and all WW2 ordnances studied in this paper. It is clear that all points fall well below the Murphy (1981) curve. The curve labelled ‘modified Murphy’ is one unit of magnitude above the Murphy (1981) curve for the same yield and is a reasonable fit. Also shown is the Gitterman (2001) curve derived from experiments in the Dead Sea in 1999. The salinity and hence density of the water in the Dead Sea is much higher than in open oceans and may explain that the curve is higher still, meaning a higher magnitude for an explosion of the same yield than the ‘modified Murphy’

Full size image
Table 3 Other underwater explosions
Full size table

The in-water explosion events shown on Fig. 6 are from many diverse locations, with differing water depths and ocean or river bottom environments. Just as the geology and rheology of different test sites influence the magnitude measured at seismic stations, the specific environment of shallow water explosions will have a bearing on the magnitude measured at the stations. It is likely that it strongly depends on the coupling of the acoustic waves to seismic waves, the depth of the water column, the depth of emplacement of the explosion, and the nature and structure of the water bottom. As expected, the larger the yield, the larger the ML magnitude will be and that the slope of this relationship in the logarithmic scale does not seem to differ very much from the classical Murphy relationship. It may be worth simulating many types of environments, water depths, and emplacement depths to assert just how much these influence the magnitude measured for the simulated events. Perhaps the most remarkable feature of the empirical relationship on Fig. 6 is that it clearly indicates that in-water explosions result in magnitudes about one unit higher than an in-ground explosion with the same charge.

7 Conclusions

Events related to controlled underwater explosion of WW2 ordnances can be detected and located by the IMS network. This study concentrated on an area which could not possibly be covered by the hydroacoustic IMS network, therefore only seismic stations could detect and locate investigated underwater events. In the dataset, events related to ordnances of the order of 1 T to 5 T could be included in the REB. In all cases, the reported source location was within the 90% confidence ellipse associated to the IDC location. The distance between the IDC location and the source position was below 20 km.

The combination of stations, which detected the events described in this study depended, among other factors, on the crustal structure along the propagation path.

In-depth analysis of the studied SEL3 events led to improvements in event location accuracy. SEL3 included only events located in the Gulf of Gdansk while others were missed. SEL3 events were at a large distance from reviewed solutions, namely between 150 and 650 km. Consequently, there were far from the ground truth location. Reviewed and ground truth locations were separated by up to 20 km as stated above.

In this data set events related to ordnances of the order lower than 1 T were not included in the IDC bulletins, however, it was possible to locate them with the IMS network. The largest difference between the IDC location and ground truth was of 50 km.

Using events from this study we could extend the magnitude-yield relationship obtained from various published values in the literature to the lower magnitude range. The conclusion is that an in-water explosion will result in seismic waves with amplitudes equivalent to the amplitudes of seismic waves from an in-ground explosion with 17.2 times the yield in kT.