Impact of Environmental Backgrounds on Atmospheric Monitoring of Nuclear Explosions

Abstract

Radionuclide monitoring for nuclear explosions includes measuring radioactive aerosol and noble gas concentrations in the atmosphere. The International Monitoring System (IMS) of the Comprehensive Nuclear Test-Ban Treaty has made such measurements for decades, revealing much about how atmospheric radioactivity impacts the sensitivity of the network. For example, civilian emissions of radioiodine make a substantial regional impact, but a minor global impact, while civilian radioxenon emissions create major regional and complex global impacts. The impacts are strongly influenced by the minimum releases anticipated to be interesting. The original design of the IMS anticipated relatively large releases, and the current IMS network substantially meets or exceeds the sensitivity needed to detect those levels. Much lower signal levels can be motivated from historical tests. Using a release that corresponds roughly to a one-ton equivalent of fission in the atmosphere rather than the design level of one-kiloton equivalent, the network detection probabilities for ¹⁴⁰Ba and ¹³¹I are quite good (~ 75%) and for ¹³³Xe is still considerable (~ 45%). Using measured and simulated background concentrations, various possible desired signal levels, and an innovative anomaly threshold, maps of sensitivity and a station ranking are developed for IMS radionuclide stations. These provide a strong motivation for additional experimentation to learn about sources and the potential plusses of new technology.

1 Introduction

Several phenomena created by nuclear explosions can be used to remotely monitor for their occurrence (Maceira et al., 2017). The International Monitoring System (IMS) (CTBTO PrepCom, 2019) is composed of seismic, hydroacoustic, infrasound, and radionuclide (RN) networks that monitor the earth for events that would violate the Comprehensive Nuclear-Test-Ban Treaty (1996).

Industrial activities, such as medical isotope production facilities, nuclear research reactors, and nuclear power reactors, also release radioactive materials to the atmosphere. Quantifying the impact of these anthropogenic backgrounds of ¹⁴⁰Ba, ¹³¹I, and ¹³³Xe on IMS radionuclide stations is a major goal of this work.

The historical design-basis studies in working paper WP.224 (IMS Expert Group, 1995) considered 50-, 75-, and 100-station networks for ¹⁴⁰Ba and also a 100-station network for ¹³³Xe, with a caveat that there was insufficient time to evaluate 50- and 75-station ¹³³Xe networks. The CTBT calls for 80 RN stations, but one of the 80 was not assigned coordinates. Also, noble gas samplers are planned for 40 stations, although the possibility exists to install them at additional RN locations after the treaty goes into force. Therefore, the noble gas results provided here consider both 39- and 79-station networks, using the coordinates defined in the text of the Treaty.

The WP.224 aerosol detection design goal is > 90% probability of detecting a 1 kiloton TNT equivalent atmospheric nuclear test within 10 days (Schulze et al., 2000) using the isotope ¹⁴⁰Ba. While nuclear fission of actinides creates many radioactive isotopes, ¹⁴⁰Ba is considered a top signature because of its high fission yield, favorable decay scheme, and moderate half-life which facilitates detection at global scales.

1.1 Isotopes Released from Historical Nuclear Tests

Radioactive xenon is an important indicator used in detecting underground nuclear explosions, as noble gases are by far the most likely to leak through geologic or engineered containment. Dubasov (2010) recorded xenon detections for over 40% of underground nuclear explosive tests conducted at Semipalatinsk, for example. By comparison, many aerosol isotopes are generated in a nuclear explosion, but only a few have been observed at the surface from underground nuclear explosions.

Isotopes useful for monitoring should have half-lives long enough to allow the isotope to travel 1000 km or further downwind before it decays below detectable levels, but not so long that it does not register many decays in a measurement period. Likewise, the fission yield should be as high as possible to increase detection probabilities. As will be shown in Sec. 2.1, in the case of ¹³³Xe, the maximum inventory is reached about 3 days after the fission occurs if ¹³³I is held together with the xenon.

Both ¹³³Xe and ¹³⁵Xe were frequently observed leaking from U.S. underground nuclear explosive tests (Schoengold et al., 1996). Due to their relatively large yields and inert nature, ¹³⁵Xe is much less likely to be detected far from the release location because of its relatively short half-life. The number of times selected isotopes were observed leaking from U.S. underground explosive tests (Schoengold et al., 1996) are tabulated in Table 1 along with their half-life and fission yield. The IMS noble gas network frequently detects ¹³³Xe but rarely detects other xenon isotopes despite their release by nuclear power plants and medical isotope production facilities.

Table 1 Frequency of detection by isotope for the 824 underground nuclear explosive tests conducted in Nevada (Schoengold et al., 1996)

Next-generation noble gas systems have just been being deployed (Ringbom et al., 2017) in the IMS or are nearing the completion of pre-deployment tests (Chernov et al., 2021; Hayes et al., 2013; TBE, 2020; Topin et al., 2020). These systems will have better detection thresholds for all xenon isotopes and may result in more detections from background sources.

Molybdenum, barium, and lanthanum are considered ‘refractory,’ i.e., non-volatile, however, barium and lanthanum isotopes have short-lived xenon precursors, ¹³⁹Xe (T_1/2 = 39.7 s) and ¹⁴⁰Xe (T_1/2 = 13.6 s). Perhaps the ¹⁴⁰Ba and ¹⁴⁰La entries in Table 1 are observed with a greater frequency than ⁹⁹Mo because of very prompt leaks of ¹³⁹Xe and ¹⁴⁰Xe rather than direct emission of the refractory particles.

In Table 1, the frequency of detections at U.S. underground nuclear explosive tests are listed, but this frequency is without regard for the relative sensitivity of measurement systems employed. Systems in the IMS today have noble gas sensitivities in the 0.2–0.5 milli Becquerel per cubic meter (mBq/m³) range, while aerosol systems have sensitivities around 10 µBq/m³. This is because the aerosol minimum detectable concentration (MDC) formula depends (Miley et al., 2019) on the inverse square root of the sampled volume, which for an aerosol system is as much as a thousand times larger than a xenon system, but still using less power than that needed to remove tens of cubic centimeters of xenon from tens of cubic meters of air. Thus, it is possible that the reported ratio of xenon to aerosol detections in Schoengold et al. (1996) are skewed toward aerosol detections by a factor of 20 or more. In any case, for underground test leakage, the three iodine isotopes ¹³¹I, ¹³³I, and ¹³⁵I were detected far more often than the aerosol isotopes ⁹⁹Mo and ¹⁴⁰Ba that are favored for a direct release of fission products to the atmosphere.

1.2 Types of Background Radionuclide Signals

Like many measurement systems, IMS radionuclide measurement systems must contend with background signals. These can come in two varieties and are dealt with quite differently. First, there is natural radioactivity in Earth’s atmosphere. For aerosol systems, radon decay products such as ²¹²Pb, ²¹²Bi, and ²⁰⁸Tl collected on the sample provide a background of Compton scattered gamma ray signals across a wide spectral energy range, obscuring the gamma ray signals below 2615 keV. These background signals increase the MDC of ¹⁴⁰Ba, ¹³¹I, and all other aerosol isotopes of interest, because the gamma ray signals from the isotopes of interest must significantly exceed fluctuations in the background signals to be registered. Because these decay products originate from radon upwelling in continents, their signals are stronger at interior continental locations than at coastal locations, which are in turn stronger than at island locations. The daily ²¹²Pb concentrations at RN79 (Oahu, Hawaii, USA) and RN70 (Sacramento, California, USA) shown in Fig. 1 illustrate the large variation in ²¹²Pb background concentrations. As seen in Fig. 1, there are also significant seasonal variations at many locations. The ¹⁴⁰Ba MDC values for the network vary from 3.1 to 23.2 µBq/m³ as shown in Table 9 in the appendix and corresponding MDC fluctuations would occur in all other isotopes. Aerosol removal due to rain and radon variation due to barometric pressure changes add an additional element of variability to these signals. In xenon measurement systems, variations in radon can likewise impact the spectra results, but filtration, chemical separation, and energy spectrum analysis are employed to greatly minimize this effect.

Fig. 1

Daily ²¹²Pb concentrations for 9 years at RN79 (Oahu, Hawaii, USA) (left panel) and RN70 (Sacramento, California, USA) (right panel). The same vertical scale is used on both graphs to emphasize the large variation of values at different locations

A second type of background occurs when activities unrelated to nuclear explosions release the same isotopes of interest. Nuclear power plants (Kalinowski & Tatlisu, 2020; Kalinowski & Tuma, 2009) and medical isotope production facilities (Bowyer et al., 2013; Saey, 2009; Saey et al., 2010a, 2010b; Stocki et al., 2008; Wotawa et al., 2010) and nuclear research facilities (Hoffman & Berg, 2018) also release the same xenon isotopes to the environment as a nuclear explosion. Each of these sources of fission products has an associated leakage rate and have been studied in relation to IMS signals (Achim et al., 2016; Gueibe et al., 2017; Schoeppner & Plastino, 2014). Figure 2 shows the monitoring locations in the IMS, the locations of 181 active nuclear power production facilities, and the locations of the 11 largest fission-based producers of medical isotopes. The facile assumption that nearby IMS stations suffer the most from anthropogenic backgrounds is generally true but quantifying the impact on both nearby and distant stations is challenging and a major goal of this work.

Fig. 2

Map of IMS radionuclide monitoring locations, nuclear power reactors, and medical isotope production facilities

2 Methods and Data

To study the impact of background radioactivity on a network of RN sensors, the network design, sensor sensitivity, and the intended source term must be considered. In this study, we take as a given the 79 locations for radionuclide monitoring in the IMS and the sensitivity of currently deployed IMS systems. These will be considered versus a wide range of source strength intended to explore the entire range of network response. Rather than a simple formula for detection sensitivity, the authors also employ a frequency-based approach for some backgrounds, such that an anomaly or action threshold is achieved at the 95th percentile of frequently seen backgrounds. In the future, as the natural backgrounds and many anthropogenic backgrounds become sufficiently well-known and, on average, predictable, studies could be done to predict the performance of different networks, different sensors, and different detection criteria.

2.1 Source Terms for Network Detection Analyses

The original IMS design document, WP.224 (IMS Expert Group, 1995), identified the magnitude of the aerosol source term M = 2 × 10¹⁵ Bq of ¹⁴⁰Ba and presented the rationale that this activity corresponds to lofting 90% of the fission products from a 1 kiloton fission explosion in the atmosphere. As seen in Table 1, ¹³¹I is far more likely to be released from a nuclear explosion contained underground than other aerosol species. Despite the original thought of a network of aerosol samplers developed to detect radioactive releases from an atmospheric nuclear explosive test, this study will consider the utility of ¹³¹I leakage from underground nuclear explosive tests. Compared to ¹⁴⁰Ba, this isotope also has a high fission yield (3%), a favorable decay scheme for gamma ray spectroscopy, and a useful half-life of 8.24 days. There have been a number of ¹³¹I detections in the IMS, presumably from the production and use of medical isotopes, so it is conceivable that backgrounds could hamper the use of the isotope for detecting underground nuclear explosive tests. Other fission and activation products (e.g., legacy ¹³⁷Cs and cosmogenic ²⁴Na) are frequently detected by the IMS, but because of the importance of iodine releases from U.S. underground nuclear explosive tests, the authors will use ¹³¹I to explore the impact of aerosol backgrounds.

The xenon source term in WP.224 is differentiated between a ¹³³Xe source magnitude of 10¹⁵ Bq for ‘evasive atmospheric tests’, where rain eliminates aerosols, but 90% of instantaneous xenon isotopes are lofted, 10¹⁵ Bq of instantaneous release for underwater nuclear explosions, and 10¹⁴ Bq released over 12 h, or 10% of the xenon, for a 1 kiloton underground nuclear explosion. The release timing is important, because the radioactive precursor to xenon is iodine, which is less likely to escape from underwater or underground explosions. The amount of ¹³³Xe as a function of time past a nuclear explosion with a 1 kiloton yield for cumulative xenon and iodine isotopes and fractionated ¹³³Xe is shown in Fig. 3. Isotopic inventories were generated using a combination of the MCNP6 code (Goorley et al., 2012) for estimation of neutron fluxes and the ORIGEN2.2 code (Croff, 1980) to calculate the resulting irradiation material balances from those fluxes. From the fractionated ¹³³Xe curve (lowest curve in the plots), it is evident that leakage of ¹³³Xe from a nuclear explosion during the earliest hours would be strongly suppressed due to the lack of ¹³³I ingrowth during later containment. The average from the cumulative ¹³³Xe curve during the earliest 12 h, as in WP.224, is 6 × 10¹⁴ Bq before loss of containment. The WP.224 estimate implies a 17% leakage of ¹³³Xe over this time.

Fig. 3

Comparison of some ²³⁵U fission products from a one kiloton equivalent nuclear explosion: the initial independent quantity of ¹³³Xe, compared with the ingrowth and decay of ¹³³I cumulating into ¹³³Xe. Also, ¹³¹I is shown. Note that ^133mXe, not shown, also decays into ¹³³Xe. The log-time left panel illustrates the rapid growth of ¹³³Xe due to ¹³³I decay that is lost when the xenon is released (fractionated) at early time

To fully understand the range of responses of the 79-station network, it will be tested with a range of source strengths wide enough to determine where the network sensitivity begins, and where it maximizes. The source strengths used in this study run from 10⁹ to 10¹⁶ Bq for the three isotopes mentioned above, ¹⁴⁰Ba, ¹³¹I, and ¹³³Xe, such that this activity range corresponds to nuclear explosion yields in the atmosphere from 100 g to 1 kiloton.

2.2 Minimum Detectable Concentrations and the Influence of ²¹²Pb

The original IMS design document (IMS Expert Group, 1995) calls for an MDC of 10 µBq/m³ for ¹⁴⁰Ba, achieved in a 3-day sampling period including collection and measurement. The MDC for aerosols depends on several factors related to the decay of background sources and of the analyte of interest, including the efficiency of the detector and the branching fraction of the isotope in a region of interest. The MDC equation (Miley et al., 2019) includes the square root of the observed count of background signals that occur in the region of interest for an isotope. This can be thought of as related to the uncertainty or fluctuation of the background signals. Many such MDC formulations could be created, but many use the statistical approach similar to that described by Currie (1968), in which a 5% false positive and 5% false negative choice is made. In general, the MDCs for different isotopes can differ greatly (a couple of orders of magnitude or more), and their response to different background concentrations can be different.

The concentration of ²¹²Pb is usually thought to be the main driver of background signals in IMS aerosol systems and varies widely from day to day and location to location. The MDCs of many analytes are calculated for each sample, but most of these are strongly impacted by ²¹²Pb. The IMS has gradually expanded since 2000 to include 72 of the 80 planned IMS stations. Sample measurement results include the MDC of ¹⁴⁰Ba and ¹³¹I achieved in each measurement. A summary of the relevant historical MDC data for each station is provided in Appendix 1 for approximately 200,000 measured samples since the beginning of 2012, i.e., not including Fukushima-related measurements. The authors selected data from Reviewed Radionuclide Reports (RRR) that pass air flow-rate quality checks, acquisition time (counting) quality checks, and have a ²¹²Pb concentration of no more than 400,000 µBq/m³. Information on the ²¹²Pb concentrations are included here because high ²¹²Pb concentrations can reduce the sensitivity of a detector (Werzi, 2010).

Setting target MDC values is useful when determining the ability of a sampling network to monitor for specific isotopes. However, after the target level was set and sampling systems were developed, the collected data showed that no sampler design can achieve the target MDC for ¹⁴⁰Ba of 10 µBq/m³ in areas where the background levels of ²¹²Pb are extremely high. Thus, the definition of the target MDC level was revised to apply to the situation where no ²¹²Pb is present in the samples. A computational technique was then created to extract the performance the station would give with no ²¹²Pb. This approach uses groups of MDC values for an isotope (e.g., ¹³¹I or ¹⁴⁰Be, etc.) and depends on the assumption that background counts originate from ²¹²Pb. The group of sample MDCs are fitted with a constant value, representing the radioactivity intrinsic to the system, and a term linear in the ²¹²Pb concentration. The following equation uses the ²¹²Pb concentrations, Pb_i, and the MDC values, MDC_i, from a group of samples (indexed by i):

$${MDC}_{i}^{2}=a+b\cdot {Pb}_{i}$$

(1)

where a and b are fitting constants obtained from a linear regression model.

Radionuclide Aerosol Sampler/Analyzer (RASA) systems (Miley et al., 1998) are deployed at 20 IMS radionuclide sampling stations. For aerosol sampler/analyzer systems, the concentrations and MDC have units of µBq/m³. Using 3722 samples collected in 2019 at 11 RASA systems operated by the United States for the IMS, one obtains the following functional fits for ¹⁴⁰Ba and ¹³¹I given the ²¹²Pb concentration, Pb, in each sample:

$${MDC}_{Ba140}=\sqrt{72.32+0.02546\cdot Pb}$$

(1a)

$${MDC}_{I131}=\sqrt{8.86+0.03318\cdot Pb}$$

(1b)

When the ²¹²Pb concentration is zero, MDC_Ba140 = 8.50 and MDC_I131 = 2.98 µBq/m³ for these RASA systems.

The square of the sample MDC is plotted against the sample ²¹²Pb concentration in Fig. 4, showing that a linear fit is reasonable, and supports the assumption that ²¹²Pb and its decay products dominate background contributions to the MDC. But because the fitted line does not go through the origin, one can be certain that ²¹²Pb does not represent the entirety of the background signals.

Fig. 4

The square of measurement MDCs for ¹³¹I (left pane) and ¹⁴⁰Ba (right pane) for 3,722 samples collected at 11 different RASA samplers in 2019 for a range of ²¹²Pb concentrations. The functional fits to adjust the MDC for varying levels of ²¹²Pb concentrations are shown by the dotted grey lines

As shown in Fig. 1, the ²¹²Pb concentrations vary from day to day, and also show seasonal fluctuations, thus the ¹⁴⁰Ba and ¹³¹I MDCs also have daily and seasonal variations. An example of the range of MDCs is provided in Table 2 for 11 stations using RASA equipment. A station on Wake Island in the Pacific Ocean (RN77) has the smallest range. This makes sense, because the ²²⁰Rn that produces the ²¹²Pb mostly comes from atmospheric radon released from spontaneous fission of uranium and thorium in the surface rock. Of these 11 stations, the widest ²¹²Pb concentration range is found in RN74, which is located at Ashland, Kansas, near the center of North America.

Table 2 Sample MDC statistics for the ¹⁴⁰Ba MDC in µBq/m³ for 2012 through 2020 at 11 stations operated by the United States for the International Monitoring System

After adjustment for background ²¹²Pb, the diverse group of IMS aerosol systems currently meet the target 10 µBq/m³ MDC level for ¹⁴⁰Ba, although the lower ²¹²Pb background levels in some locations result in better (lower) MDCs. Data on the historical sample MDC values for ¹⁴⁰Ba and ¹³¹I is provided in Table 9 in the Appendix for all IMS stations.

Three types of noble gas samplers are currently deployed in the IMS. The SAUNA (Ringbom et al., 2003) has a 0.2 mBq/m³ MDC for ¹³³Xe using 12-h samples. The SPALAX (Fontaine et al., 2004) has a 0.15 mBq/m³ MDC for ¹³³Xe using 24-h samples. The ARIX (Dubasov et al., 2005) has a 0.5 mBq/m³ MDC for ¹³³Xe using 12-h samples. Information provided in Appendix 1 matches the sampler type with different sampling locations.

2.3 Creating an Anomaly Level Using Historical Detections, and Applying to ¹³¹I

Above, the limitation that uninteresting radioactivity imposes on monitoring for atmospheric signatures was discussed for aerosol systems. The second perplexing source of monitoring interference is the appearance of the specific signatures of interest arising from uninteresting processes. An example of this can be seen from 2019 ¹³³Xe data from one station in Fig. 13 in the Appendix. This system collects and measures two samples a day with an MDC of about 0.2 mBq/m³. The normalized integral of signals is also shown, such that by inspection it is clear that about 20%, or 150 reported signals are above the MDC of the system. If one assumes that there were no nuclear explosions in 2019, this distribution of background signals represents the noise above which a ¹³³Xe signal would have to rise to garner interest.

The authors chose the 95th percentile of the distribution to be an anomaly. In other words, a signal greater than 95% of the background might trigger additional study. There are other reasons to trigger such a study—if any other explosion-related signals are seen. This might include other isotopes of xenon, aerosols, or even vibrations in the Earth, oceans, or atmosphere. In this instance, however, only the detection of one isotope is considered. The choice of 95% is not completely arbitrary, as the statistical approach of Currie (1968) chooses this level of statistical background fluctuation to set the MDC.

At this point, an action threshold can be constructed with the maximum of either the MDC or the 95th percentile. For stations that observe the analyte often, the action threshold would be the 95th percentile, and for others, any signal exceeding the MDC would garner monitoring interest. While Fig. 13 in the Appendix represented ¹³³Xe at one station for one year, the situation is quite different for ¹³¹I. The frequency of historical ¹³¹I detections in the entire IMS in the years 2012 through 2020 are shown in Fig. 5 as a function of concentration. About half of the detections have concentrations above 3.52 µBq/m³ which is not too different from the MDC for that isotope. These 1234 detections occurred in 194,162 total samples.

Fig. 5

The frequency of ¹³¹I detections as a function of concentration. Data are global and start in 2012, which is long enough after the Fukushima event that related ¹³¹I will have decayed. The upper end of the horizontal axis (above 20 µBq/m³) is not to scale

Significant levels of ²¹²Pb are present in many of the samples with ¹³¹I detections and cause substantial variation in the daily ¹³¹I MDC. The concentration levels for ²¹²Pb in Table 3 are illustrated for the 9 samples with detected ¹³¹I collected at RN70 (Sacramento, California, USA) and the ¹³¹I MDC varies from 5 to 9 µBq/m³. Over half of the samples contain reviewer-confirmed ¹³¹I concentrations found below the calculated MDC for that sample. Only one sample is substantially above the MDC for the relevant day at the station.

Table 3 Influence of ²¹²Pb concentrations on the MDCs for ¹⁴⁰Ba and ¹³¹I for samples with detections of ¹³¹I at RN70 (Sacramento, California, USA)

There were one or more ¹³¹I detections at 45 IMS aerosol stations during 2012–2020. The average and 95th percentile of ¹³¹I concentrations at the IMS stations with 10 or more detections are shown in Table 4. Below this number, the 95th percentile becomes difficult to accurately estimate and is quite similar to the MDC. In Table 4, the 95th percentile ranges from about 13 times the MDC to about the same as the MDC. For half the entries in Table 4, the MDC is at most doubled by the ¹³¹I background. Thus, only 5 IMS aerosol stations have a particularly significant background effect for ¹³¹I. The number of ¹³¹I detections for all IMS stations is provided in Appendix 1.

Table 4 The number of ¹³¹I detections and the average and 95th percentile of the ¹³¹I concentrations of the detections for all IMS stations with 10 or more detections from January 2012 through February 2021

2.4 Atmospheric Transport Model

All atmospheric transport results in this paper used the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Stein et al., 2015), which incorporates both advective and diffusive processes. The transport runs were performed using the Linux version of HYSPLIT. Default model parameters (Draxler et al., 2020) were used. For example, vertical turbulence was simulated using the Clayson and Kantha (2008) model and horizontal turbulence is proportional to the vertical turbulence. The boundary layer stability was computed from heat and momentum fluxes read from the meteorological data files. The top of the atmospheric model domain was set to 10,000 m above ground level.

Wet and dry deposition mechanisms were deactivated for the xenon transport runs because xenon is a noble gas and its concentration in the air does not depend on rainout or deposition processes. Wet and dry deposition were included for both ¹⁴⁰Ba and ¹³¹I using the assumption the transport was in particulate form. Iodine can transport as inorganic, elemental, and particulate species, and can change forms during transport. In addition, the speciation depends on the local humidity (Fitzgerald, 1975; Winkler, 1973). Modeling the different iodine species is difficult, and while it is necessary when dealing with concentrations high enough to impact human health (Eslinger et al., 2014b), this work only used the particulate iodine form. This assumption may underestimate the detection probabilities for ¹³¹I.

The transport runs used to determine network performance used archived meteorological data for 2019 on a 1° spacing and 3-h time step (GDAS1, 2020). Each run modeled plume movement for 10 days and saved concentration data on a global 0.5° grid. One of the techniques developed to reduce the computational burden in a source-term analysis with far fewer samplers than possible release points is to use atmospheric transport runs done backwards in time (Hourdin & Talagrand, 2006; Hourdin et al., 2006; Rao, 2007; Seibert & Frank, 2004; Stohl et al., 2002). Although time-reversed runs don’t exactly match with forward-time runs (Eslinger & Schrom, 2019), these runs were performed in the reversed-time direction for computational convenience.

All releases from hypothetical nuclear explosions are assumed to be 3 h in length and the releases are transported through the atmosphere for 10 days after the release. This work models 8 releases per day for an entire year at a large number (258,839) of unequally spaced locations. The locations were selected on a global grid with constant 0.5° spacing in latitude and longitude. Thus, the detection statistics are based on 7.55 × 10⁸ possible releases.

2.5 Network Performance Model

As mentioned above, the detection design goal for the IMS radionuclide (aerosol) network performance was to achieve a 90% probability of detecting an atmospheric test of 1 kiloton TNT equivalent explosion within 10 days (IMS Expert Group, 1995; Werzi, 2009). In this work, a network detection metric is used to assess the probability of detection for a release at an unknown time and unknown location. As noted by Kalinowski (2001) and others, radionuclide samples and atmospheric transport modeling can be used for several purposes in addition to just detecting a release, such as locating the release point. Thus, we define three network metrics. The first metric is the probability a release is detected. The second and third metrics are the expected number of detecting samples and the expected number of detecting stations. In general, obtaining more samples with detections at more stations helps in determining the timing and location of the release event.

Consider the situation where there are N_s radioisotope samplers at different locations around the globe. In an abstract sense, each sampler has a probability of detecting a specific future release of a radioactive isotope. This detection probability depends on the location of the release, the magnitude of the release, the specific radioisotope, the sampling duration, the detection sensitivity of the equipment, and future atmospheric circulation patterns. One way to compare system performance for different network configurations is to define a detection metric for the entire global network. The same type of metric can be applied for smaller regions.

Suppose that the surface area of the globe is partitioned into a large number, N_r, of regions, each with surface area A_i, for i = 1,…, N_r. In general, the surface area of the regions may not be equal. Then, denote the probability, P(M,R_s), that a radioisotope release of a given magnitude, M, occurring in a specific region R_s at a series of times t_j, for j = 1, …, N_t is detected by one or more samplers. This can be expressed in the following mathematical form:

$$P\left(M,{R}_{s}\right)=\frac{1}{{N}_{t}}{\sum }_{j=1}^{{N}_{t}}I\left(One\; or\; more\; samplers\; detected \;release \;j\; from {R}_{s}\right)$$

(2)

where I(·) is the indicator function, which takes the value 1 when true and 0 when false. The network detection metric, D(M), defined for a release of magnitude M, given a network of N_k sampling locations, and a surface area partitioned into N_r regions, is:

$$D(M)=\frac{{\sum }_{i=1}^{{N}_{r}}{A}_{i}\times P\left(M,{R}_{i}\right)}{\sum_{i=1}^{{N}_{r}}{A}_{i}}$$

(3)

Below, we use the notation D(15)₃₉ to denote a release magnitude of 10¹⁵ Bq and a network size of 39 stations. If the surface areas of the regions are all of the same size, such as used by Schoeppner and Plastino (2014), then Eq. 3 simplifies to an average value:

$$D(M)=\frac{1}{{N}_{r}}{\sum }_{i=1}^{{N}_{r}}P\left(M,{R}_{i}\right)$$

(4)

The network metric satisfies $0\le D(M)\le 1$. A value of 0 indicates that a release of magnitude M will not be detected by any sampling station, no matter when or where the release occurs. A value of 1 indicates that a release of magnitude M will always be detected by at least one sampling station, no matter when or where the release occurs.

Determination of the release location from radionuclide samples becomes more important for small underground nuclear explosions than for large atmospheric nuclear explosive tests. Seismologists need a minimum of three sampling stations to pinpoint the epicenter of an earthquake. Similarly, samples at multiple locations helps in a source location analysis using airborne radionuclides. The detection metric, as formulated here, declares a detection of a release event if the concentration at one or more samples at one or more locations exceeds the detection threshold denoted by the MDC in response to that release.

We define, N(M), the expected number of detecting stations over all release events as:

$$N(M)=\frac{1}{{N}_{r}\sum_{i=1}^{{N}_{r}}{A}_{i}}{\sum }_{i=1}^{{N}_{r}}{A}_{i}\left(\sum_{j=1}^{k}{I}_{j}\left(M,{R}_{i}\right)\right)$$

(5)

where ${I}_{j}\left(M,{R}_{i}\right)$ is 1 if station j detects a release of magnitude M in the region R_i in one or more samples and takes a value of 0 otherwise.

We define, S(M), the expected number of detected samples over all release events as:

$$S(M)=\frac{1}{{N}_{r}\sum_{i=1}^{{N}_{r}}{A}_{i}}{\sum }_{i=1}^{{N}_{r}}{A}_{i}\left({C}_{i}(M)\right)$$

(6)

where ${C}_{i}\left(M\right)$ counts the number of samples, across all stations, that detects release i of magnitude M.

3 Results

Network performance results for ¹⁴⁰Ba are provided in Sect. 3.1, including a comparison with performance predictions made in 1995. Section 3.2 contains network performance results for ¹³¹I. Finally, network performance results for ¹³³Xe are provided in Sect. 3.3.

3.1 Network Detection Performance for ¹⁴⁰Ba

Historical estimates of the detection performance an aerosol network for ¹⁴⁰Ba for networks with 50, 75, and 100 stations for a range of transport times are shown in Fig. 6 for a design basis release of 10¹⁵ Bq of ¹⁴⁰Ba to the atmosphere and an MDC of 10 µBq/m³. Figure 6 is derived from Fig. 3-1 of WP.224 (IMS Expert Group, 1995). The detection probability, D(M), calculated using Eq. (3), is shown for a 39-station network (black circle) and a 79-station network (black triangle) event and historical data for detection limits. The current performance estimate for ¹⁴⁰Ba matches nicely with the historical estimates of the design-basis performance.

Fig. 6

Historical estimates of detection probabilities for different network designs, each assuming a release of 10¹⁵ Bq of ¹⁴⁰Ba. Our calculated performance of a 39-station network (black circle) and a 79-station network (black triangle) using 10 days of transport time past the release event and historical data for detection limits. The curves are derived from Fig. 3-1 of WP.224

The performance of 39- and 79-station networks for the detection of ¹⁴⁰Ba are shown in Table 5 for a large range of release magnitudes. The 39-station network uses the 39 stations with existing or planned noble gas systems (CTBTO PrepCom, 2020). The historical average ¹⁴⁰Ba MDC was used for 72 of the stations. The stations without operating systems were assigned the MDCs from other stations currently operating on the same continent. Details of the assignments of the MDCs are provided in Appendix 1. The performance statistics in Table 5 are averaged over a year for the entire globe. The same statistics can be evaluated for shorter time periods, such as a day, to give an idea of the temporal variation in detection capabilities. The ± values represent an approximate 95% uncertainty range on daily D(M), N(M), and S(M) values.

Table 5 Detection performance for ¹⁴⁰Ba for different levels of release for two network sizes

The network coverage varies in space as well as in time. The network detection probabilities for 39-station and 79-station networks are shown in Fig. 7, assuming a release of 10¹⁵ Bq of ¹⁴⁰Ba anywhere on the globe. Detection limits for each station were derived from historical measurements as described in Appendix 1. The average number of stations that would detect a release of 10¹⁵ Bq of ¹⁴⁰Ba anywhere on the globe for 39- and 79-station networks are provided in Fig. 8. This result was expected when the IMS system was designed (IMS Expert Group, 1995) but can now be evaluated through extensive transport simulations.

Fig. 7

Network detection probabilities for 39-station (upper panel) and 79-station (lower panel) networks assuming a release of 10¹⁵ Bq of ¹⁴⁰Ba anywhere on the globe. Detection limits for each station were derived from historical measurements. This activity level roughly corresponds to 100 tons of fission yield

Fig. 8

Average number of stations that would detect a release of 10¹⁵ Bq of ¹⁴⁰Ba anywhere on the globe for 39-station (upper panel) and 79-station (lower panel) networks. Detection limits for each station were derived from historical measurements

The overall network performance of the 79-station IMS RN network for a release of 10¹⁵ Bq of ¹⁴⁰Ba is remarkably close to the historical estimate. Even with the unexpected (in 1995) impact of ²¹²Pb on the ¹⁴⁰Ba detection limit, the average MDC across 79 stations is 9.92 µBq/m³, while the design basis was 10 µBq/m³. As shown in Figs. 7 and 8, coverage in the equatorial regions is poorer than at higher latitudes. This result was expected when the system was designed (IMS Expert Group, 1995) but can now be evaluated through extensive transport simulations. Increasing the number of noble gas stations from 39 to 79, as is allowed in the treaty after entry into force, improves D(M), but it nearly doubles N(M) and S(M). Having more detecting stations and samples is important for improving the accuracy of the estimated release location from the sampling data, as shown in Eslinger and Schrom (2016). The maps of the number of detecting stations calculated for this analysis are similar to those produced by (Werzi, 2009) for a few months.

3.2 Network Detection Performance for ¹³¹I

The performance of 39- and 79-station networks for the detection of ¹³¹I are shown in Table 6. The 39-station network uses the 39 stations with existing or planned noble gas systems (CTBTO PrepCom, 2020). The historical average ¹³¹I MDC was used for 72 of the stations. The stations without operating systems were assigned the MDCs from other stations currently operating on the same continent. Details of the assignments of the MDCs are provided in Appendix 1. The performance statistics in Table 6 are averaged over a year for the entire globe. The ± values represent an approximate 95% uncertainty range on daily D(M), N(M), and S(M) values.

Table 6 Detection performance for ¹³¹I for different levels of release for two network sizes

The network detection probabilities for a 79-station network are shown in Fig. 9, assuming a release of 10¹¹ Bq of ¹³¹I anywhere on the globe. Detection limits for each station were derived from historical measurements as described in Appendix 1. The decrease in the detection probability when the station detection limits are raised to the 95th percentile of historical detections at the station is also shown in Fig. 9.

Fig. 9

Detection probabilities for a 79-station network for a 10¹¹ Bq of ¹³¹I anywhere on the globe (upper panel). The decrease in the detection probability when the station detection limits are raised to an anomaly level equal to the 95th percentile of historical detections at that station is shown in the lower pane. For example, the coverage in a region with a bright red contour has an absolute decrease in D(M) of around 0.1, which could correspond to as much as a 40% decrease in the detections in that local region

The network-level results in Table 6 and the spatial coverage in Fig. 9 show that the coverage of the IMS radionuclide network is poor for a nominal release of 10¹¹ Bq of ¹³¹I, especially in the equatorial regions. As shown in Table 6, the percent decreases in detection probabilities due to background sources of ¹³¹I are nonnegligible, but they are more pronounced for a 39-station network than for a 79-station network. Also, next generation aerosol samplers currently under development (Miley et al., 2019) may have lower MDCs and shorter sampling periods. These changes would improve the detection capabilities, including providing more samples with concentrations above the detection limits.

Medical isotope production facilities that release some ¹³¹I to the atmosphere are located in Argentina (South America), in China, and in the Russian Federation. The fourth region with elevated background levels of ¹³¹I is in Panama, Central America. There is no known medical isotope production near this station, but exhalation of ¹³¹I by multiple patients treated with ¹³¹I for medical procedures (Gründel et al., 2008) near the sampler might release enough ¹³¹I to result in occasional detections (Miley et al., 2021).

3.3 Network Detection Performance for ¹³³Xe

As of the start of 2021, only 25 locations had xenon systems certified for operation in the IMS, and these fall into three system types (Dubasov et al., 2005; Fontaine et al., 2004; Ringbom et al., 2003) with different performance characteristics. Many of the certified systems have processed thousands of samples over several years and thus have a well-established background history, such as that in Fig. 13 of the Appendix. This history indicates that presence of the analyte of interest, ¹³³Xe, is the dominant monitoring challenge for many locations.

The network performance estimates for ¹³³Xe use modeled concentrations from nuclear power plants and medical isotope production facilities to determine the 95th percentile anomaly level. As shown in Appendix 1, the 95th percentile of modeled ¹³³Xe concentrations can be close to the measured values for some stations. The results are similar to that of Achim et al. (2016) and Schoeppner and Plastino (2014). In addition, Gueibe et al. (2017) compared modeled and measured concentrations in the IMS for four xenon isotopes. An action threshold is then created using the maximum of either the MDC for the system presumed to operate at that station or the 95^th percentile of the estimated background.

The performance of 39- and 79-station networks for the detection of ¹³³Xe are shown in Table 7. Published MDCs are used for currently deployed sampling equipment. Details of the assignments of the MDCs are provided in Appendix 1. The columns in Table 7 with a Δ in the header show the percent decrease in performance when the MDC is set to the maximum of the equipment level and the 95th percentile of the modeled background concentrations. The performance statistics in Table 7 are averaged over a year for the entire globe. The ± values represent an approximate 95% uncertainty range on daily D(M), N(M), and S(M) values. Increasing the MDC to an anomaly threshold based on the background levels causes significant reductions in detection performance, especially for lower release magnitudes. Adding additional sampling locations improves the overall performance but does not mitigate effects of background ¹³³Xe.

Table 7 Detection performance for ¹³³Xe for different levels of release for two network sizes

The estimated performance of a 100-station network given in Fig. 3-3 of WP.224 was about 0.2 for a release of 10¹⁵ Bq, assuming a detection limit of 1 mBq/m³. Our calculations of the detection probability for the smaller 39-station network for a release of 10¹⁵ Bq is 0.7, much better than the historical performance estimate. We calculated a D(M) of 0.9 for a 100-station network formed by adding another 21 stations to the existing radionuclide stations while assuming a 10¹⁵ Bq release of ¹³³Xe and an MDC of 1.0 mBq/m³ for every sampler.

The network detection probabilities for a 39-station network are shown in Fig. 10, assuming a release of 10¹⁴ Bq of ¹³³Xe anywhere on the globe. The 10¹⁴ Bq release level was chosen mostly because two of the tests conducted by the DPRK (Murphy et al., 2013; Ringbom et al., 2009, 2014) may have had releases on the order of 10¹⁴ Bq of ¹³³Xe.

Fig. 10

Detection probabilities for a 39-station network for a 10¹⁴ Bq release of ¹³³Xe anywhere on the globe (upper pane). Change in the detection probability when the station detection limits are raised to an anomaly level equal to the 95th percentile of modeled background (lower pane)

Detection limits for each station were derived from equipment specifications and modeled backgrounds from nuclear power plants and medical isotope production facilities as described in Appendix 1. The decrease in the detection probability when the station detection limits are raised to the 95th percentile of modeled backgrounds is also shown in Fig. 10. Using the same assumptions, the network detection probabilities for a 79-station network are shown in Fig. 11.

Fig. 11

Detection probabilities for a 79-station network for a 10¹⁴ Bq release of ¹³³Xe anywhere on the globe (upper pane). Change in the detection probability when the station detection limits are raised to an anomaly level equal to the 95th percentile of modeled background (lower pane)

The network-level results in Table 7 and the spatial coverage in Fig. 10 show that the coverage of the IMS radionuclide network is poor for a nominal release of 10¹⁴ Bq of ¹³³Xe, especially in the equatorial regions. Following entry into force of the treaty, the addition of more IMS RN station locations would improve the coverage. As shown in Table 7, the percent decreases in detection probabilities due to background sources of ¹³³Xe are significant over large regions of the globe.

New generation noble gas samplers under development (Haas et al., 2017; TBE, 2020; Topin et al., 2020) and deployed in the IMS (Ringbom et al., 2017) and have lower detection levels and shorter sample collection periods than current systems. These new systems will improve the detection capabilities, especially in the number of detected samples.

4 Discussion

The results in this paper are based on atmospheric transport simulations that used 10 days of transport after the release events. These transport runs were performed in the reversed-time direction for computational convenience. Detection probabilities would likely increase, especially for larger magnitude releases, if longer transport times were used.

Using the values in Appendix 1, the average MDC of ¹³¹I is 3.34 µBq/m³, which is lower (better) than the average MDC of ¹⁴⁰Ba at 9.92 µBq/m³. Thus, ¹³¹I has a somewhat higher probability of being detected at low release magnitudes than ¹⁴⁰Ba. The average MDC for ¹³³Xe is much poorer at 0.231 mBq/m³ (or 231 µBq/m³), thus it has a lower probability of detecting smaller magnitude releases. However, the release of ¹³³Xe from a small underground nuclear explosion may be larger than the releases of ¹³¹I or ¹⁴⁰Ba. In addition, atmospheric transport of ¹³³Xe is not affected by atmospheric loss processes, such as dry or wet deposition, that reduce the ¹³¹I and ¹⁴⁰Ba concentrations.

The network performance for different release magnitudes of ¹⁴⁰Ba, ¹³¹I, and ¹³³Xe is discussed in Sect. 4.1. The stations mode impacted by background are discussed in Sect. 4.2. In Sect. 4.3, a hypothetical release scenario is discussed that would combine data from the aerosol and noble gas networks. Finally, the implications of other background thresholds is discussed in Sect. 4.4.

4.1 Network Performance for Different Release Magnitudes of ¹⁴⁰Ba, ¹³¹I, and ¹³³Xe

A summary of network performance for different release magnitudes of ¹⁴⁰Ba, ¹³¹I, and ¹³³Xe is provided in Fig. 12 using current equipment detection characteristics that are influenced by background. Although results for iodine, barium, and xenon are shown on the same graph, the data in Table 1 suggest that their release magnitudes may be quite different in a real event.

Fig. 12

Summary of network performance for different magnitude releases of ¹⁴⁰Ba, ¹³¹I, and ¹³³Xe using current equipment characteristics and measured or estimated background anomaly levels. The two curves for ¹³³Xe are for 39-station and 79-station networks. The detection probabilities, D(M), are shown in the top panel. The number of stations detecting the release, N(M), are shown in the middle panel. The number of samples detecting the release, S(M), are shown in the bottom panel. All results are for 10 days of atmospheric transport following the release event

The results shown in Fig. 12 cover 7 orders of magnitude of releases, from 10⁶ to 10¹⁶ Bq. Using the rough scale that 1 kiloton of fission equals 10²³ fission atoms, applying fission yields and a half-life of around a week, this scale can be roughly translated to covering the nuclear explosion yield in the atmosphere from 100 g to 1 kiloton. The low end of the magnitude range is included to show that a sparse network of current sensitivity has little chance of detecting an extremely small release, say, 10¹⁰ Bq or 1 kg equivalent explosion, unless the release were to occur close to a measurement system and directly upwind of it. At 10¹³ Bq release, which corresponds roughly to one-ton equivalent of fission in the atmosphere, the detection probabilities for aerosols are quite good (~ 75%) and for xenon are still considerable (~ 45%).

Estimates of releases from the Fukushima nuclear power plants in 2011 for ¹³³Xe (Eslinger et al., 2014a) and ¹³¹I (Koo et al., 2014) of around 10¹⁹ Bq and 10¹⁷ Bq, respectively are larger than the upper end of the range in Fig. 12, perhaps corresponding to a megaton equivalent release of xenon and 10 kiloton equivalent release for iodine, and were detected all across the northern hemisphere (Biegalski et al., 2012). Of more interest, from a treaty monitoring perspective, is the ability to detect a test with a small yield, such as the tests conducted by the DPRK (Ringbom et al., 2009, 2014). The 2006 test yield was probably on the order of 1 kiloton (Murphy et al., 2013), while the yield of the 2013 test was a little larger in the 2.0–4.8 kiloton range (Murphy et al., 2013). Measured ¹³³Xe and ^131mXe for the 2013 test suggest a release for both tests on the order of 10¹⁴ Bq of ¹³³Xe. No aerosol samples detected ¹³¹I, so the release magnitude can be bounded, but not be estimated.

There are fewer industrial sources of ¹³¹I than ¹³³Xe, although occasionally there are accidental releases, such as in Hungary in 2011 (Tichý et al., 2017), not at production facilities. Thus, as seen in Fig. 9, the background interference is regional in scope, with stations in China and the Russian Federation being most impacted. The estimated background for ¹³³Xe is global in scope, as seen in Fig. 11.

In the data in Table 6 for ¹³¹I, and Table 7 for ¹³³Xe, the ∆D(M), ∆N(M), and ∆S(M) columns show that background interference degrades network performance more for small design basis magnitudes than for larger magnitudes. The tables also show that denser networks mitigate, to some extent, the degradation in detection performance from background emitters. However, network performance estimates need to rest on realistic release magnitudes. For example, analysis of data following the DPRK nuclear explosive test in 2013 yielded ¹³³Xe concentrations that would have been indistinguishable from background without the concurrent detection of ^131mXe (Ringbom et al., 2014).

4.2 Stations Most Impacted by Background

The data given in Tables 9 and 10 show the background levels of ¹³¹I and ¹³³Xe that impact each station. The data there can be used to rank the stations that are impacted the most by background. Table 4 shows the stations most impacted by background levels of ¹³¹I. Table 8 shows the 12 stations most impacted by the background levels of ¹³³Xe. The stations are ranked by the 95th percentile anomaly level, but the median concentrations are also given. The median value is at or below the MDC for 7 of the 12 stations.

Table 8 The 12 stations most impacted by the background concentrations of ¹³³Xe

The number of stations reporting and the number of detecting samples are also important metrics and were not considered in WP.224. Along with all the other assumptions made to compute results shown here, a tacit assumption is that all stations are in good working order, and do not suffer power outages or other issues when needed. Thus, it could be considered crucial to have more than one station detecting. The lack of a second detecting station does not weaken any detection results obtained, but the network is more robust in multiplicity. Similarly, the number of detecting samples add confidence in the result. But multiple detections in time and space can be used with atmospheric transport modeling to limit the size of the region in which the signal originated (Eslinger & Schrom, 2016; Eslinger et al., 2019).

4.3 Hypothetical Release Scenario Combining Aerosol and Noble Gas Networks

It is probably unwise to make an assumption about how radioactive material will be released from an underground nuclear test: the release pathway through the geologic containment, if any, may include a complex of wet or dry fractures, while the engineered containment may include filled tunnels and sealed doorways. These could combine in many ways to produce the various results found in Schoengold et al. (1996), including frequently no measured release at all. But it can be enlightening to create a hypothetical release scenario that elucidates how the xenon and aerosol networks could work together.

Let us hypothesize a 1 kiloton equivalent nuclear explosion with a release of xenon similar to that described in Ringbom et al. (2009). In this scenario, about 1% of the ¹³³Xe would be released, or 10¹⁴ Bq, after 3 days of decay chain ingrowth. The reader can choose from Fig. 3 some combination of ingrowth time and containment suppression that achieves a 10¹⁴ Bq release. Ely et al. (2021) estimate for U.S. underground nuclear explosive tests that an average ¹³¹I leakage would be about one part in 10⁵ or less, but without a timing estimate for when the release occurs. Using the approximation that 10²³ radioactive atoms are created in a 1 kiloton test, a release fraction of 10^–5 or slightly higher, the cumulative yield of ¹³¹I from Table 1, and a decay constant on the order of 10^–6 s⁻¹, one obtains an approximate 10¹¹ Bq release of ¹³¹I. In this scenario, there is only a factor of 1000 separating the activity of xenon and iodine.

If an explosion were to release 10¹⁴ Bq of ¹³³Xe and 10¹¹ Bq of ¹³¹I, then from Fig. 12, assuming 79-station networks for all samplers, the detection probability, D(14), is a very strong 0.76 for ¹³³Xe and D(11) is a poor, but not hopeless, 0.25 for ¹³¹I. The number of samples with detections, S(11), is 0.52 for ¹³¹I, but S(14) is a robust 6.4 for ¹³³Xe. By comparison, the DPRK nuclear explosive test in 2013 resulted in ¹³³Xe detections at 2 stations in a total of 5 ¹³³Xe detects (Ringbom et al., 2014), which adds some confidence that these modeling results are valid. Further, because D(11) for ¹³¹I is only 25%, the lack of IMS measurement results cannot rule out the hypothetical release scenario.

Because the 79 noble gas systems have a much better chance of detecting a leaking test than the 79 aerosol systems, it might be tempting to wonder if the aerosol network is needed. If, however, there were half as many xenon systems, say 39 vs 79 and thus, half as many detecting xenon samples, adding one or two iodine detections would materially increase the confidence in the detection. If the MDC for ¹³¹I were improved by an order of magnitude as suggested in Miley (2019), then for regions that do not have a serious ¹³¹I background issue, D(M) could improve to 0.54 and S(M) could improve to 2.0. It has not been mentioned to this point that the IMS also contains a network of 16 laboratories for the confirmatory remeasurement of aerosol samplers. Several of these are equipped with ultra-low background detectors, and it has been shown that these can obtain an order of magnitude sensitivity increase for ¹³¹I (Aalseth et al., 2009). This hypothetical scenario implies that sending the aerosol samples collected nearby the detecting xenon samples for ultra-low background measurement at IMS laboratories might yield additional evidence for or against the hypothesis that the source was a nuclear explosion.

While the scenario posited here is no more likely than others, it shows that the aerosol network plus its laboratories could substantially contribute to detecting small release underground nuclear test events now, and with station improvements, could make those contributions in near real-time. Another observation, without prejudice for the release mechanism, is that iodine, barium, and presumably other aerosols constitute a very sensitive detection means for small magnitude sources in the atmosphere that are orders of magnitude smaller than the detection measurement range of current xenon technology.

4.4 Other Possible Background Thresholds

If background estimates could be made sufficiently accurate, the 95th percentile could be adjusted temporally to lower the action threshold significantly. Approaches to accomplish this might include:

• Additional noble gas background measurement campaigns at IMS locations currently without a noble gas sampler. This would provide sampled data very useful for verifying and improving the modeled global concentrations. The change in network detection probabilities, ∆D(14), shown in Fig. 11 could be used to help select measurement locations with significant background concentrations.

• Besides IMS locations, the addition of local monitoring data, for example from local networks, safety systems, or stack monitors could greatly influence background calculations and potentially improve background estimates.

Another avenue of mitigation of the background radioactivity impact is to make the aerosol and xenon systems more supportive of each other. One order of magnitude additional iodine sensitivity in the aerosol network could allow ¹³¹I detections to occur in support of xenon detections, raising the confidence and location capability of the combined network substantially. There are six stations on both the most impacted aerosol and xenon lists in Tables 4 and 8. These stations and the regions around them, RN01, RN20, RN21, RN54, RN59, and RN61, might be prime candidates for a noble gas background measurement campaign to understand the sources and find ways to improve the action threshold. Such a field campaign could further help develop the concepts of fusion between aerosol and xenon networks.

Appendix 1: Additional Information about Detection Limits and Anomalous Background Levels

The operational data used to assign the detection limits for the four types of aerosol samplers in operation in the IMS are summarized in Table 9. Data to determine the average MDC values are obtained for 2 years (2019–2020) if the system regularly reported data. Two years is long enough to capture both daily and seasonal variations in ²¹²Pb backgrounds. Some stations were not in operation or collected a limited number of samples in 2019–2020, so the data period for them was extended to cover 2012–2020. Starting the data collection in 2012 means the ¹³¹I released from the Fukushima nuclear power plants following the 2011 earthquake would have decayed away (Biegalski et al., 2012). In a desire to only include data where the sampler was operating properly, the following screening criteria were implemented on reviewed radionuclide reports (RRR): (a) the sample had a spectrum category of 1, 2, 3, 4, or 5, (b) the air flow-rate quality flag was ‘PASS’, (c) the acquisition time was between 20 and 28 h, and (d) the ²¹²Pb concentration was less than 400,000 µBq/m³. For the 2012–2020 time period, approximately 200,000 samples met the screening criteria. The ¹³¹I concentration data in Table 9 are for samples taken between January 1, 2012 and February 15, 2021.

Table 9 Information on the detection limits and ¹³¹I detections used in the network performance analyses

Three types of noble gas samplers were deployed in the IMS at the time these calculations were performed. The SAUNA (Ringbom et al., 2003) has a 0.2 mBq/m³ MDC for ¹³³Xe using 12-h samples. The next-generation Swedish system (Ringbom et al., 2017), denoted by SAUNA III, uses a 6 h collection period and began operation at IMS station RN63 in Stockholm, Sweden in the middle of 2021. The SAUNA III system is not used in this analysis. The SPALAX (Fontaine et al., 2004) has a 0.15 mBq/m³ MDC for ¹³³Xe using 24-h samples. The ARIX (Dubasov et al., 2005) has a 0.5 mBq/m³ MDC for ¹³³Xe using 12-h samples. Fortunately, the noble gas concentrations do not depend on the ²¹²Pb background. For this analysis, all noble gas samplers were assumed to have a 12-h collection period. The stations with a SPALAX sampler were assumed to have a detection limit of 0.15 for all samples.

Currently, only 40 of the radionuclide sampling locations in the treaty have noble gas samplers. That number is reduced to 39 because RN35 does not have specified coordinates in the treaty. Thus, some assumptions are required to model a 79-station noble gas network. The entries in Table 10 are based on three assignment rules for stations where noble gas systems are not currently installed: 1) If the same country operates a noble gas sampler at another location in the IMS, the same equipment is used. These noble gas sampler assignments are denoted by using ‘()’, for example, (SPALAX), 2) All other assignments use a SAUNA system, denoted by [SAUNA]. The SAUNA and SPALAX have nearly the same MDC for ¹³³Xe and the 12 h collection cycle matches with the assumptions in the underlying ATM analysis. New generation noble gas samplers under development (Haas et al., 2017; Ringbom et al., 2017; Topin et al., 2020) have lower detection levels and shorter sample collection periods than current systems. Although they are expected to have better network performance that the currently deployed systems, this work uses only the existing deployed sampler technologies.

Determining the anomalous detection level as the 95th percentile of detections is complicated by the fact that sampling data do not exist for many of the noble gas stations used in this analysis. Thus, the 95th percentile is based on modeling nominal release quantities for medical isotope production facilities and operating nuclear power plant complexes. We provide summary results in Table 10 followed by additional data on the source terms and a brief discussion of the modeling approach. The average fraction of the modeled concentrations due to medical isotope production facilities over the 2 years is also provided in Table 10. In most cases, the detections are dominated by MIPF releases.

Table 10 Information on the sampler type assignments and the ¹³³Xe anomaly limits (mBq/m³) that will be compared with the sampler MDC

The average daily release values of ¹³³Xe (Bq) for the 12 medical isotope production facilities used in this study are provided in Table 11. The release rates were compiled from published sources over the last 10 years, with no attempt to modify the published values in response to changing global production levels of ^99mTc. However, the facility at Chalk River, Canada, has ceased operations and was not included in this study. In some cases, such as PINSTECH in Pakistan, the release rate was estimated using announced production rates of ^99mTc, combined with basic knowledge of the type of separations technology used in the facility.

Table 11 Daily release rates of ¹³³Xe (Bq) used for each medical isotope production facility

Daily release estimates were developed for the nuclear power plants in the online Power Reactor Information System (IAEA-PRIS, 2019) that were operating in 2019. These nuclear power plants are operating at 181 different locations. All release rates were set to 4.67 × 10⁹ Bq/d per reactor, which is derived from the combined continuous and batch releases (arithmetic average) in Table 4 of Kalinowski and Tuma (2009). Releases at each location accounted for the number of operating reactors. Many of the locations have only one reactor, but others have multiple reactors. For example, the Qinshan complex in China has 7 operating reactors.

The atmospheric transport runs in the Hysplit code (Stein et al., 2015) that propagate facility emissions of ¹³³Xe used archived meteorological data for 2015 and 2016 on a 0.5° spacing and 3-h time step (GDAS0P5, 2020). Each run modeled plume movement for 10 days after release and saved concentration data for each hour on a global 0.25° grid. The concentration data were then interpolated to individual sampler locations and aggregated to the sampler collection time periods.

A comparison of modeled versus measured ¹³³Xe values for 2019 at the IMS station RN38 in Takasaki, Japan, is provided in Fig. 13. The model predicts more values below about 0.25 mBq/m³ than are measured, but the cumulative frequency of predictions and measured values is quite close for values of 0.35 or higher. The sampled data are not censored below the detection limit of approximately 0.25 mBq/m³. The 95th percentiles agree quite nicely, providing evidence that the 95th percentile anomaly level based on the modeled concentrations is reasonable.

Fig. 13

Comparison of measured ¹³³Xe concentrations for 2019 at RN38 (Takasaki, Japan) with modeled concentrations based on nominal releases from nuclear power plants and medical isotope production facilities