An unknown source of reactor radionuclides in the Baltic Sea revealed by multi-isotope fingerprints

We present an application of multi-isotopic fingerprints (i.e., 236U/238U, 233U/236U, 236U/129I and 129I/127I) for the discovery of previously unrecognized sources of anthropogenic radioactivity. Our data indicate a source of reactor 236U in the Baltic Sea in addition to inputs from the two European reprocessing plants and global fallout. This additional reactor 236U may come from unreported discharges from Swedish nuclear research facilities as supported by high 236U levels in sediment nearby Studsvik, or from accidental leakages of spent nuclear fuel disposed on the Baltic seafloor, either reported or unreported. Such leakages would indicate problems with the radiological safety of seafloor disposal, and may be accompanied by releases of other radionuclides. The results demonstrate the high sensitivity of multi-isotopic tracer systems, especially the 233U/236U signature, to distinguish environmental emissions of unrevealed radioactive releases for nuclear safeguards, emergency preparedness and environmental tracer studies.

(t ½ = 2.34 × 10 7 years) is an isotope of uranium that is produced by thermal neutron capture of 235 U via (n, γ)-reactions and through 238 U (n, 3n) 236 U reactions with fast neutrons. Even though a small amount of 236 U (~35 kg) occurs naturally in the Earth's crust, 236 U is (by mass) the largest secondary product created in nuclear reactors, estimated to be~10 6 kg 1 . 236 U is therefore a sensitive tracer of deliberate or accidental leakage from the nuclear fuel/waste cycle [2][3][4][5] . The known sources of reactor 236 U, i.e., deliberate releases from the two European reprocessing plants at La Hague, France (LH), and Sellafield, UK (SF) since 1950s, can be traced throughout the North Atlantic and the Arctic water currents 6 . Emissions from other known sources of reactor 236 U, e.g., the Springfield nuclear facility and the Fukushima accident, are negligible 5,7 .
A significant amount of 236 U (estimated at >1000 kg) was also delivered to the Earth's surface environments from the global fallout of atmospheric nuclear weapons testing in the 1950s and 1960s 8 . This ubiquitous fallout signature can make identification of sources of reactor 236 U challenging because of methodological difficulties in distinguishing the source of 236 U 9 . In addition, the 236 U/ 238 U ratio does not provide source information because of the prevalence of 238 U in nature.
Reactor 236 U can be differentiated from fallout 236 U because these sources have different and characteristic 233 U/ 236 U ratios due to different nuclear production mechanisms. 233 U was mostly produced during nuclear weapons testing by fast neutrons via 235 U (n, 3n) 233 U reactions or directly by 233 U-fueled devices, whereas almost no 233 U is produced in thermal nuclear power reactors or reprocessing plants 10 . Recently 233 U measurements at environmental levels have become possible with advanced accelerator mass spectrometry 10 .
The representative 233 U/ 236 U atomic ratio of global fallout from atmospheric nuclear weapons testing was suggested to be (1.40 ± 0.15) × 10 −2 9 . This is several orders of magnitude higher than the 233 U/ 236 U atomic ratio in nuclear reactors, e.g., 1 × 10 −7 -1 × 10 −6 in LH discharges 11 , which agrees well with reactor model calculations 12 . In the Irish Sea, an average 233 U/ 236 U atomic ratio of (0.12 ± 0.01) × 10 −2 has been measured 9 , reflecting a dominant reactor signal released from SF. The use of the 233 U/ 236 U atomic ratio helps to deconvolve the origin of 236 U based on the characteristic 233 U/ 236 U fingerprint from different source terms. In addition, the combination of 236 U with other radionuclides, e.g., 129 I, can be useful to trace the transport of 236 U from specific source points, e.g., releases from LH and SF [13][14][15][16] .
The Baltic Sea is a highly polluted sea, with anthropogenic radionuclides demanding specific attention because of the risk to ecosystem and humans from radioactivity in the environment. It receives radionuclides from global fallout, discharges from the two European reprocessing plants, releases from the Chernobyl accident, and from any other local sources. In this study, we use a novel combination of three anthropogenic radionuclides-233 U, 236 U, and 129 I-to identify a previously unknown local source of radionuclide pollution to the Baltic Sea.

Results and discussion
Study area and sampling. The Baltic Sea is a landlocked intracontinental sea in Northern Europe with about 80 million inhabitants in the surrounding states and constitutes one of the largest brackish water environments on Earth 17 . The water exchange of this large brackish estuarine-like water mass with the Kattegat and the North Sea takes place through the narrow and shallow Danish Straits (Fig. 1). The driving force for the water circulation is freshwater surplus from river runoff, estimated at 473 km 3 per year, together with "recycled" North Sea inflowing water as Baltic outflow that sum to a total water exchange rate of 753 km 3 per year 18 . A mean residence time for the 21,721 km 3 Baltic water volume 19 was estimated to be 29 years, which is equivalent to a "half-life" for the water volume of 20 years 18 .
In the investigation presented here, water and sediment samples were collected from the Baltic Sea and related water masses including the western Danish coast, from 2011 to 2016 (Supplementary Tables 1 and 2). The majority of water samples are from the surface (0-5 m depth), with a few samples from deep water, and one lake water from the Lake Mälaren, which receives downstream discharges from a nuclear fuel fabrication facility (Westinghouse) in Sweden and finally drains into the Baltic Sea. In addition to the Baltic Sea water, we analyzed sediment samples to assess the accumulation trend of the isotopes in the Baltic Sea. A more detailed description of the study area and samples can be found in the "Methods" section.
To facilitate the presentation of results and related discussion, we grouped the sampling locations into five geographical regions ( Fig. 1) Tables 1 and 2) vary within (5-52) × 10 −9 , with the higher ratios in the central and northern parts of the Baltic Sea and lower ratios in the western parts (Danish Straits, Kattegat, Skagerrak, and Danish west coast). The highest value reported here is sixfold greater than the average value found in the North Sea in 2010 ((7.6 ± 3.7) × 10 −9 ) 20 .
The spatial patterns (Fig. 2) suggest a general decline of 236 U with distance from higher values in the North Sea which is expected to be dominated by discharges from LH and SF. However, high 236 U concentrations ((6-9) × 10 7 atom/l) are observed in the surface water of the Bothnian Sea and Borthnian Bay, which are comparable to values ((3-10) × 10 7 atom/l) in the central North Sea 20 . Compared to the Kattegat-Skagerrak region, the average 236 U/ 238 U atomic ratio in the middle and north Baltic region increases by a factor of 3, from (10 ± 3) × 10 −9 to (32 ± 7) × 10 −9 . This pattern of increasing in 236 U/ 238 U ratio highlights an additional, likely local, source of 236 U in the Baltic Sea 7 . 233 U/ 236 U atomic ratios obtained here are in the range of (0.14-0.87) × 10 −2 , with the lowest 233 U/ 236 U atomic ratios in the western parts of the Baltic, including the Danish coast, and the highest ratios in the central Baltic Sea. As the typical 233 U/ 236 U ratio for global fallout is (1.4 ± 0.1) × 10 −29 , the high 233 U/ 236 U in the central Baltic Sea could indicate either strong influence of global fallout or addition from a local source.
Distribution of 129 I concentration, 129 I/ 127 I and 236 U/ 129 I atomic ratios. The measured 129 I concentrations ((3-232) × 10 9 atom/l) and 129 I/ 127 I atomic ratios ((101-1286) × 10 −9 ) in the seawater collected in this work show comparable values and distribution trends as observed in an earlier investigation 21  Aldahan et al. 22 reported that the average concentration of 129 I in the rivers around the Baltic Sea was 3.9 × 10 8 atom/l, which suggested some minor contribution of 129 I from riverine water to the Baltic Sea. The 129 I concentrations obtained in this work show a larger gradient (two orders of magnitude) compared to the 236 U concentrations (15-fold) along the Baltic Sea. 236 U/ 129 I ratios are within the range of (5-133) × 10 −4 and indicate a reversed geographical distribution compared to 129 I concentration and 129 I/ 127 I atomic ratio (Fig. 2).
Potential sources of uranium and iodine in the Baltic Sea. Five different sources of uranium and iodine in the Baltic Sea are: (1) Natural ocean water, with salinity of 35‰, which contains 60 μg/l 127 I, 3 μg/l 238 U, but negligible 129 I, 236 U, and 233 U.
(2) Natural freshwater with salinity <1‰, negligible 129 I, 236 9,20 . As they are distributed in the Baltic and mix with waters from various rivers, ratios can be altered by addition from local sources of 236 U and 233 U (and minor 238 U in river waters). Removal of uranium from Baltic water will not alter the ratios. The increase in 236 U/ 238 U observed within the Baltic Sea points clearly to a local source of this anthropogenic radionuclide. 236 U source identification via binary mixing. The concentration of 238 U (Fig. 3A) demonstrates a strong positive correlation (R 2 = 0.91) with salinity. The intercept corresponds to the average riverine input with a 238 U concentration of 0.33 ± 0.05 μg/l, which falls in the range (0.2-0.7 μg/l) of 238 U for some rivers in the Baltic Sea region 28 . We will use the typical value 0.4 μg/l in the following calculations. There is more scatter in the 238 U concentration for low salinities, which might be attributed to differences in regional riverine input. 129 I also shows a general positive linear correlation with salinity demonstrated by two mixing lines for the western (KGR-DS, R 2 = 0.89) and interior (SBR-MBR-NBR, R 2 = 0.97) region (Fig. 3B). The scatter at the high salinity end can be attributed to the mixing of 129 I enriched North Sea coast water with 129 I depleted North Atlantic water in the Kattegat-Skagerrak region. The 238 U and 129 I trends with salinity suggest that their concentrations in the Baltic Sea are mainly controlled by the saline water input from the North Sea via Kattegat-Skagerrak, mixing with fresh waters in the basin.
Both the 236 U/ 238 U and 236 U/ 129 I atomic ratios increase with the decreasing salinity as waters mix in the interior of the Baltic Sea. The 236 U/ 238 U ratio increases by a factor of 3, while the 236 U/ 129 I ratio increases greater than an order of magnitude from an average of (8 ± 2) × 10 −4 in the Kattegat-Skagerrak region, corresponding to reprocessing derived 236 U and 129 I, to 1 × 10 −2 in the central Baltic Sea. Both ratios indicate addition of 236 U from a local source. If the source does not contain any 129 I, the tenfold increase in 236 U/ 129 I suggests that ca. 90% of 236 U in the central Baltic Sea is from local sources. If the source does contain 129 I, the portion of 236 U derived locally must be still larger.
To understand the source terms of 236 U in the Baltic Sea, a binary mixing model is applied with two respective end members representing 236 U input from the North Sea and freshwater input via river runoff. Parameters for the first end member representing the North Sea water entering from the west Baltic Sea are well defined by previous studies (Supplementary Table 3) 20,29 . The deviation of the observed 236 U/ 238 U atomic ratio in the binary mixing (line L1, Fig. 4A) of the North Sea water and an assumed freshwater end member containing no 236 U (neither 233 U) from the best-fit model L reflects additional 236 U sources besides North Sea water. The spatial distribution of deviations in the 236 U/ 238 U atomic ratio enable determination of the location of the additional 236 U source ( Supplementary Fig. 2). The distribution pattern is compatible with the introduction of additional riverine 236 U input from the north Baltic region, which has most river runoff.
Nevertheless, it is challenging to define the 236 U/ 238 U ratio of the riverine input to the Baltic because a component of global fallout may still be present in runoff from the land surface. The 236 U/ 238 U and 236 U/ 129 I ratios cannot be used to determine the extent to which the excess 236 U is from global fallout or an  additional, previously undiscovered, source that has directly released 236 U to the Baltic Sea.
Application of 233 U/ 236 U atomic ratio for 236 U source identification. If we assume that the excess 236 U originates only from global fallout, the 236 U/ 238 U atomic ratio of the riverine input in the best-fit binary mixing is (6.79 ± 0.75) × 10 −8 (line L, Fig. 4A). However, there is a clear deviation of the observation from the model for 233 U/ 236 U atomic ratios (Fig. 5A). A subgroup of samples from the Kattegat-Skagerrak reveal a relatively stable 233 U/ 236 U atomic ratio of 0.20 × 10 −2 (blue dash-dotted line in Fig. 5) independent of 236 U/ 238 U and salinity. This behavior can be explained by assuming an end member of North Sea water with 233 U/ 236 U atomic ratio = 0.20 × 10 −2 (a mixed signal of global fallout plus nuclear reprocessing) and salinity 35‰, which is mixed with natural uranium or water with neither 236 U nor 233 U. This feature shows the notable impact of nuclear reprocessing from SF and LH in the region.
On the other hand, a cluster of samples with the majority from the south, middle and north Baltic Sea region, representative for a large part of the Baltic surface water and with median salinity (6.92 ± 0.29)‰, show a typical 233 U/ 236 U atomic ratio of (0.53 ± 0.03) × 10 −2 (the green dash-dotted line in Fig. 5). This cluster lies significantly below the binary mixing model L, indicating an additional local 236 U sources besides the global fallout, which is characterized by low 233 U/ 236 U atomic ratio. A low 233 U/ 236 U atomic ratio is typical for releases from nuclear reactors, thereby we assume such a reactor-related source of 236 U with negligible 233 U in the following.
About two-thirds of the anthropogenic uranium observed in the middle and north Baltic Sea region seems to originate from this additional local source (Eq. (1)), indicating a strong contribution of 236 U without 233 U, i.e., from a thermal nuclear reactor 236 U.
where R s , R f , and R r represent, respectively, the 233 U/ 236 U atomic ratio of the Baltic seawater, global fallout, and nuclear reactor; N 233, f and N 233,r refer to the atomic number of 233 U from global fallout and nuclear reactor, respectively; N 236, f and N 236, r refer to the atomic number of 236 U from global fallout and nuclear reactor, respectively. Therefore, With R s = (0.53 ± 0.03) × 10 −2 , R f = 1.4 × 10 −2 , and R r = 0.12 × 10 −2 (the Irish Sea ratio), we calculate the 236 U contribution from our assumed reactor source to be 2.1 ± 0.2 times that of global fallout.
To locate this additional reactor 236 U source, we apply another binary mixing line L2 (Fig. 4A) of the North Sea water with riverine water, the latter carrying global fallout that accounts for 1/(1 + 2.1) of the average 236 U concentration of our samples in the Baltic Sea. Thus, the freshwater end member is characterized by salinity = 0, 238 U = 0.4 μg/l, 236 U = (3.56 ± 0.39) × 10 7 atom/l, which is calculated to match the 233 U/ 238 U atomic ratio ((1.70 ± 0.18) × 10 −10 ) for the cluster of samples from SBR, MBR, and NBR at the media salinity of (6.92 ± 0.29)‰ ( Supplementary  Fig. 2). The resultant 236 U/ 238 U atomic ratio of the freshwater end member is (3.52 ± 0.39) × 10 −8 . The excesses of the 236 U/ 238 U atomic ratio from the mixing curve L2 and their spatial distribution are shown in Fig. 6. The data indicate that the extra reactor 236 U source input is not from places where salinity is particularly low or where there are rivers, but in the middle and north basins of the Baltic Sea which is probably linked to direct releases of 236 U into these locations.
Properties of the unknown 236 U source. To narrow down the possible sources of the excess 236 U, 236 U inventories and fluxes need to be estimated. It should be noted this calculation is a firstorder approximation based only on our data on surface waters from a multi-year survey. A precise interpretation will require more data, and to account for many different effects such as vertical distribution of 236 U in the Baltic water columns, interannual variation in distribution pattern and on the scavenging of uranium into the sediment (especially in the anoxic regions).
The existence of an additional source of anthropogenic 236 U in the Baltic Sea is indicated by the difference between the models L2 and L (Fig. 6). The amount of 236 U required to explain this difference can be calculated by the following approximation and with the uncertainty estimated according to Müller 30 : where X 236 is the excess mass of 236 U in the Baltic Sea, S is the salinity, 238 U S ð Þ ½ is the 238 U concentration corresponding to S taken from Fig. 3A. S is the average salinity of the Baltic Sea. S was taken as 7.36‰ based on the reported mean salinity of the Baltic Sea during 1902-1998 31 . Our data from SBR, MBR, and NBR, which comprise the bulk of Baltic Sea water, show an average of salinity of 7.06‰ and a median of 6.92%, comparable to the reported value. 238 U S ð Þ ½ is 0.9 μg/l. V BS is the volume of the Baltic Sea (21,721 km 3 ) 19 and ðL À L2Þð SÞ = (1.02 ± 0.24) × 10 −8 is the difference of the model curves at the average salinity. The approximation in the formula is possible because in the models L and L2, the 236 U concentration is a linear function of S. Therefore, 200 ± 47 g of 236 U is from the additional reactor source.
This calculation is a snapshot in time based on the uranium isotope ratios and salinity. While uranium concentrations in water may be altered in the partly anoxic Baltic Sea by precipitation of inorganic U(IV) or binding to organics in the sediment, the uranium isotopic ratios will only change by mixing of difference sources. Total salinity is slightly affected by precipitation (rain and snow) and evaporation (net balance 63 km 3 per year) 32 , which may, be neglected at the present level of precision. Large intrusions of the North Sea water can change salinity patterns and introduce anthropogenic uranium from the North Sea. These intrusions add up to 5.2 × 10 9 metric tons of salt 33 , which is about 3% of the salt inventory of the Baltic 32 . The spatial pattern may not be constant throughout a multi-year survey, nevertheless, a minor change in the calculation is expected as we use only the average salinity for our estimate.
Taking into account that the ratio between the additional source and global fallout is N 236,r /N 236,f = 2.1, it suggests that 95 ± 22 g of 236 U is related to global fallout introduced into the Baltic Sea directly or via riverine input. It is estimated that a total inventory of 1000 kg of anthropogenic 236 U was distributed via global fallout mainly on the Northern Hemisphere 7 . Considering the surface area of the Baltic Sea of 3.77 × 10 5 km 2 (without the catchment area) in comparison to the Northern Hemisphere (i.e., 2.55 × 10 8 km 2 ), the total 236 U deposition from direct global fallout is estimated as 1.5 kg. However, if considering the 29-year mean residence time (equivalent to 20-year half-life) of Baltic seawater, then most of the deposited 1.5 kg 236 U was transported out of the region after 60 years (i.e., three half-lives), leaving behind~0.19 kg. In addition, some 236 U fraction from global fallout might be removed from the water body and incorporated into the Baltic sediment 34 . Therefore, the above estimation of 95 ± 22 g remaining 236 U in the Baltic seawater from global fallout seems plausible, considering the uneven distribution of global fallout.
If we include the Baltic catchment area (1.64 × 10 6 km²) in the calculation, the input of global fallout 236 U in the Baltic region can be up to 8 kg (1.5 kg in seawater + 6.4 kg in catchment area). However, only a small fraction of the particle associated 236 U deposited on land can be leached and transported to the Baltic Sea through river runoff. If we assume this fraction accounts for 10% of the 6.4 kg of 236 U deposited in the catchment, the total amount of global fallout 236 U in the Baltic Sea might be about 0.64 + 1.5 = 2.14 kg.
Emissions from the Chernobyl accident may contribute additional 236 U to the Baltic Sea, but it is difficult to quantify. Nuclear dumping and/or nuclear installations around the Baltic countries are also possible source candidates. As marked in Fig. 1, there are many nuclear installations in surrounding Baltic countries, but there is very limited documentation with poor, unreleased or missing data about the 233 U and 236 U release records from these installations (Supplementary Table 4) 11 . Data for 236 U are available from Westinghouse during 1998-2017, with a total reported release of 1.06 × 10 6 Bq of 236 U, equal to 0.44 g. In addition, we measured one lake water sample collected in Lake Mälaren (Supplementary  Table 2), which receives waste discharges from the Westinghouse facility and finally drains into the Baltic Sea. The results show that the 236 U/ 238 U ratios is at the level of 2 × 10 −8 , which is comparable with the seawater samples collected in the central Baltic Sea. The lake water shows a 233 U/ 236 U atomic ratio of (0.18 ± 0.05) × 10 −2 , a signature of reactor material.
The amount of 236 U released from the Westinghouse installation (0.44 g) is negligible compared to the above estimated 280 g of the unknown reactor source in the Baltic Sea. For the Lake Mälaren, the 238 U concentration was measured to be 1.5 ± 0.1 μg/l in this work, together with a flux of 166 m 3 /s, it means an input of 0.1 g per year of 236 U, which is negligible also.
Another candidate for the additional source may be reactor fuel, dumped into the Baltic. The atomic ratio of 236 U/ 238 U can be as high as 1 × 10 −2 in conventional nuclear reactors, which would require only 27 kg of dumped/dissolved fuel (a commercial nuclear reactor contains~100,000 kg of fuel). 235 U enrichment in reactor fuel is 3% for light-water reactors, up to 10% for thermal gas-cool reactors and up to 20% for fast reactors 35 . The concentration will be even higher in the core of a nuclear reactor for marine applications, where enriched or highly enriched 235 U is used; Russian submarine reactors were reported to contain 50-200 kg of 235 U 36 . The former Soviet Union (USSR) was accused of dumping radioactive waste in the Baltic Sea, but it is not possible to assess the dumped amount 37,38 .
The geographical distribution of 236 U/ 238 U atomic ratio in surface seawater of central Baltic Sea shows high values nearby the Swedish coast close to Stockholm, which is within~100 km of a nuclear research company Studsvik AB, Nyköping that has been in operation since 1950s. It was reported that during 1959 and 1961, 64 tons of radioactive waste with total radioactivity of 14.8 GBq were dumped into the coastal area nearby Studsvik 39 . The aquatic discharges of radionuclides (except 3 H) from Stusvik into the Baltic Sea in 1999-2010 were reported to be 0.45 TBq with the majority consisting of 90 Sr, 137 Cs, 60 Co, and 134 Cs 40 . Our measurements on some sediment samples from the Studsvik area show very high 236 U content ((2.02 ± 0.12) × 10 13 atom/kg), which is three orders of magnitude higher than sediment collected from the North Baltic Sea region (Supplementary Table 2). The 233 U/ 236 U atomic ratio ((0.36 ± 0.05) × 10 −2 ) for the Studsvik sediment clearly indicates a higher contribution of reactor input compared to the other five sediments collected in the Baltic Sea with 233 U/ 236 U ratios between 0.59 × 10 −2 to 0.83 × 10 −2 .
Even though the release of 236 U from Studsvik is not well documented due to its low specific radioactivity, it is not surprising that waste discharges from Studsvik contain 236 U. The high 236 U levels in the sediment samples measured most likely originate from scavenging of waterborne 236 U from liquid waste discharges by particles into the sediment. Waste dumping/ discharges in the Studsvik area are our most plausible candidate for the excess 236 U in the Baltic Sea.
Outlooks for future study. The radiological risk associated with 233 U, 236 U, and 129 I observed in this work is negligible due to their low specific activities and radiotoxicities. However, the observed unknown 236 U reactor source may be an indication of leakage from a previously unrecognized (or unreported) additional radioactive source in the Baltic Sea, e.g., disposed nuclear waste in the seabed. Such source could potentially contain 137 Cs and many other radionuclides imposing high radiological risks. Recent studies of the distribution of 137 Cs inventories in the Baltic Sea indicated that 137 Cs deposited in surface sediments is not permanently buried, but may be re-suspended and redeposited by currents, bioturbation, or anthropogenic activities 41 . This leads us to suggest that radioactive release from such a source although currently low, might become more significant in the future with climate and environmental changes (e.g., sea level, temperature, and pH) in the Baltic Sea. It will be important to further understand the sources of anthropogenic radioisotopes in the Baltic regions, so that prediction and monitoring can prevent any associated radiological risk in the future. Further observation and forensic work will be needed to tighten the constraints in the binary mixing models, provide clear source terms and radiation risk assessment.

Methods
Detailed description of the study area and sampling. The Baltic Sea features three major basins, the Bothnian Bay, the Bothnian Sea, and the Baltic Proper. The two northerly basins (Bothnian Bay and Bothnian Sea) are characterized by low salinity water mass (1-3‰ and 3-7‰, respectively) and weak vertical salinity stratification, although strong thermoclines usually develop during the summer 42 . The Bothnian Sea represents a large reservoir of brackish water mass that can be divided into two layers blocked by a weak halocline around a depth of 60 m. The long-term circulation of the Bothnian Sea water is dominated by an estuary circulation, where the bottom dense waters can be traced as surface water in the Baltic Proper 43 . The Baltic Proper is the largest basin in the Baltic Sea, permanently stratified in the central part with a strong halocline around a depth of 75 m separating the surface water (salinity 7-8‰) from the deep water (salinity 9-20‰) and a long-term cyclonic current circulation pattern 44 . Water exchange in the Baltic Proper happens through renewing of the deep water during extreme inflow events from the Kattegat. The water mass circulation is further associated with outflow of surface water to the Kattegat and inflow of fresher surface waters from the Bothnian Sea, the Gulf of Finland and the Gulf of Riga (Fig. 1).
Water Standards and reagents. Uranium standard solution (1.000 g/l in 2 M HNO 3 ) purchased from NIST (Gaithersburg, MD) was used after dilution as a standard for the ICP-MS measurement to quantify 238 U in seawater. All reagents used in the experiment were of analytical reagent grade and prepared using ultra-pure water (18 MΩ cm). UTEVA resin (100-150 μm particle size) was purchased from Triskem International, Bruz, France and packed in 2-ml Econo-Columns (0.7 cm i.d. Analytical methods for determination of 238 U, 236 U, 233 U, 127 I, and 129 I. The concentration of 238 U and 127 I in seawater was measured by ICP-MS (X Series II , Thermo Fisher Scientific, Waltham, MA) after 10-50 times dilution with 0.5 M HNO 3 and 0.1 M NH 3 ·H 2 O, respectively. The ICP-MS instrument was equipped with an Xt-skimmer core and a concentric nebulizer under hot plasma conditions. The typical operational conditions of the instrument have been given elsewhere 46 . Indium (as InCl 3 ) as an internal standard and 0.5 M HNO 3 as a washing solution between consecutive assays were applied for 238 U, and caesium (as CsCl) as an internal standard and 0.1 M NH 3 ·H 2 O as a washing solution were used for 127 I.
The radiochemical method for 233 U and 236 U separation from seawater was applied according to Qiao et al. 47 . Each seawater sample (0.8-10 l) were filtrated with filter paper (Munktell 00 K, particle retention 5-6 μm) to remove large particles and then acidified to pH 2 with concentrated HNO 3 . Purified FeCl 3 solution (0.05 g/ml of Fe) was added to a final Fe concentration of 0.1 g/l. The sample was vigorously stirred with air bubbling for 5-10 min in order to decompose carbonate complexes. In total, 10% NH 3 ·H 2 O was slowly added to adjust the pH to 8-9 for the co-precipitation of U with Fe(OH) 3 . The precipitate was allowed to settle for 0.5-1 h in order to decant most of the supernatant. The sample slurry was centrifuged at 3000 × g for 5 min and the supernatant was discarded. The final residue was dissolved with 15 ml of 3 M HNO 3 and the solution was loaded onto a 2-ml UTEVA column which was preconditioned with 20 ml of 3 M HNO 3 . The UTEVA column was rinsed with 40 ml of 3 M HNO 3 , followed by 20 ml of 6 M HCl. Uranium absorbed on the column was eluted with 10 ml of 0.025 M HCl. The flow rate for the chromatographic separation was controlled manually to 1.0-1.5 ml/min.
A 100-μl aliquot of U eluate from the column separation was taken for measurement of 238 U by ICP-MS to evaluate the chemical yields by comparison with ICP-MS analysis on diluted seawater samples. The 238 U content measured in the eluate was also used for blank subtraction to calibrate the actual 236 U/ 238 U and 233 U/ 238 U atomic ratios 47 . The remaining fraction was used to prepare target for the AMS measurement of 236 U/ 238 U and 233 U/ 236 U. For sediments, 5-10 g of each dried sample was ashed overnight at 450°C in a muffle oven and leached with 100 ml of aqua regia on a hotplate for 30 min at 150°C and then 2 h at 200°C. A 100-μl aliquot leachate was taken for direct measurement of 238 U by ICP-MS, which was used to calculate the 238 U concentration in the sediment sample. The remaining leachate was processed following the same procedure (i.e., Fe(OH) 3 co-precipitation and UTEVA column separation) as for seawater samples.
The AMS measurement was carried out at the 3-MV tandem accelerator facility Vienna Environmental Research Accelerator (VERA) at the University of Vienna, Austria 9,10,48 . To summarize, U, which is extracted as UO − from a cesium sputter ion source, has to pass a first mass separation stage before it is injected into a tandem accelerator. For the analysis of actinides, the accelerator is operated at a terminal voltage of 1.65 MV and a rather high helium pressure in the terminal stripper is used to suppress molecular background 49 . The relatively high stripper gas pressure causes losses of a significant fraction of U 3+ ions to angular scattering and change exchange outside of the stripper assembly. This gives an effective stripping yield of around 21% for the charge state 3+ 50 , which is selected by the subsequent 90°analysing magnet. The combination of the analysing magnet with a Wien filter, an electrostatic analyzer, and a second 90°magnet, efficiently suppresses isotopic background on the high-energy side. Possible isotopic background is mainly caused by 235 U and 232 Th that are injected into the accelerator as 235 U 16 O 1 H and 232 Th 16 O 1 H, respectively. At the end of the AMS set-up, a Bragg-type ionization chamber is installed in order to detect and identify the remaining ions.
238 UH 3+ which escapes destruction in the stripping process gives a background to mass 239, 3+ lower than 238 UH 3+ / 238 U 3+ = 10 −14 . A similar suppression is expected for 235 UH 3+ / 235 U 3+ , which suggests an instrumental background for 236 U below 235 UH 3+ / 238 U 3+ = 10 −16 , which is negligible compared to the background of real 236 U extracted from the ion source. The mass 239, 3+ background is monitored for every sputter sample. The situation is different for 233 U 3+ , where the potentially interfering molecular isobar is 232 ThH 3+ . In fact, an even higher intensity of these molecules was found from a similar ion source 51 . As thorium is a different chemical element, the behavior of hydride during stripping cannot be predicted from uranium ions. However, because thorium is only a trace element in our sputter samples, much less suppression than for 235 UH 3+ would be sufficient to render 232 ThH 3+ insignificant as background for 233 U 3+ . For quality control, 232 Th 3+ is monitored for all sputter samples, which is extracted as 232 ThO − . Though substantial rate above 100 kHz (too high for quantification by our detector) were observed in some cases, no correlation with the mass 233, 3+ count rates were found. This suggests that 232 ThH 3+ is also sufficiently supressed by the high stripper gas pressure.
A detection efficiency of 2 × 10 −4 for environmental samples and a detection limit for 236 U/U below 10 −14 has been reported for the VERA set-up 10 . Because of the small relative mass difference (ca. 1%), fractionation effects between 233 U and 236 U are negligible, therefore a detection efficiency comparable to 236 U is assumed for 233 U. In samples with low 236 U content, e.g., procedure blanks, the uncertainty of 236 U/ 238 U atomic ratio measured by AMS is mainly attributed to the counting statistics, while for environmental samples the precision usually is limited by the reproducibility of multiple measurements which is taken into account in the overall uncertainty of 1-5% as well. Due to the low count rates of environmental 233 U, the uncertainty of the 233 U/ 238 U atomic ratio is dominated by counting statistics of 233 U. As the 238 U content in the sample determined by ICP-MS was used for blank correction of the atomic ratios measured by AMS, the overall uncertainty of the blank corrected values presented in Supplementary Table 1 is therefore a combination of the corresponding AMS and ICP-MS uncertainties calculated by error propagation.
For the determination of 129 I in seawater, 100 ml of sample was transferred into separation funnels. In total, 2.0 mg of 127 I carrier (prepared using iodine crystal purchased from Woodward company, USA, with a 129 I/ 127 I ratio of 2 × 10 −14 ), 500 Bq of 125 I − tracer, and 0.5 ml of 0.5 M Na 2 S 2 O 5 solution were added to the funnel, and then the pH of the solution was adjusted to 1-2 using 3 M HNO 3 to convert all iodine species to iodide. With addition of 20-50 ml chloroform (CHCl 3 ) and 2-5 ml 1.0 M NaNO 2 , iodide was oxidized to I 2 and extracted to CHCl 3 phase by shaking. The extraction procedure was repeated three times to extract all iodine. The CHCl 3 phases were combined to a new funnel, 20 ml H 2 O and 0.3-0.5 ml 0.05 M Na 2 SO 3 solution was added to the funnel to reduce I 2 in chloroform phase to iodide and back-extracted iodine into water phase. This extraction and back extraction processes were repeated once for further purification.
The separated iodine (in iodide form) in a small volume (5-7 ml) was transferred to a centrifuge tube, 1.0 ml of 0.5 M AgNO 3 solution and 1 ml of 3.0 M HNO 3 were added to form AgI precipitate. The AgI precipitate was separated using centrifugation at 2300 × g for 3-5 min, and washed in sequence using 10 ml 3 M HNO 3 and two times of 10 ml deionized water to remove possibly formed Ag 2 SO 3 and Ag 2 SO 4 which are soluble in acidic solution. The precipitate was transferred to a 1.5 ml centrifuge tube. 125 I in the precipitate was measured using a NaI gamma detector for calculating the chemical yield of iodine. The prepared AgI precipitate in small tube was dried at 70°C and weighed. The dried precipitate was ground to fine powder and mixed with five times by mass of niobium powder (325 mesh, Alfa Aesar, Ward Hill, MA), which was finally pressed into a copper holder using a pneumatic press. 129 I/ 127 I atomic ratios in the prepared targets were measured by the 5 MV AMS system at the Tandem Laboratory, Uppsala University. The standard used in the measurement was prepared by dilution of 129 I standard (NIST-SRM-4949c) and mixed with 127 I carrier to a ratio of 129 I/ 127 I of 1.0 × 10 −11 . All samples, blanks, and standards were measured for six cycles and 5 min per sample in each cycle. It should be noted that only the samples collected in 2015 by research vessel Argos were analyzed for 129 I. Other samples were not feasible for 129 I analysis, since the samples have been acidified before receiving, resulting in loss of iodine due to its high volatility in acidic conditions.

Data availability
The data that support the findings of this study are available on request from the corresponding author upon reasonable request.