Tritium (3H) is a radioactive isotope of hydrogen (Alvarez and Cornog, 1939) with a half-life of 12.32 years (4500±8 d; Lucas and Unterweger, 2000). Natural tritium is formed mainly by spallation reactions of protons and neutrons of primary and secondary cosmic radiation with atmospheric nuclei, mainly by the interaction of fast neutrons with atmospheric nitrogen (Lal and Peters, 1967). Tritium emission by thermonuclear tests between the 1950s and 1980 enormously exceeded the natural production (Araguas-Araguas et al., 1996; Palcsu et al., 2018). Since that time, tritium emission to the atmosphere from anthropogenic sources (e.g., nuclear industry, medical applications, luminizing industry) has corresponded to ∼10 % of the natural production and influences 3H content in precipitation mainly at local to regional scales (Araguas-Araguas et al., 1996). Starting from the 1980s, technogenic tritium has become the prevailing anthropogenic atmospheric tritium input signal over bomb tritium in Central Europe (Hebert, 1990).

Tritium is introduced into the hydrological cycle following oxidation to tritiated water (3H1HO). Tritium is an excellent tracer for determining time scales for the mixing and flow of waters and is ideal for studying processes that occur on a time scale of fewer than 100 years (Kendall and McDonnell, 2012). It proved to be a powerful tool in various applications in hydrological research (Jasechko, 2019), such as estimating mean residence time for surface water and groundwater (Michel, 1992; Stewart and Morgenstern, 2016; Zuber et al., 2001), dating cave drip waters (Kluge et al., 2010), understanding water circulation and mixing in geothermal (Ansari et al., 2017; Chatterjee et al., 2019) or permafrost settings (Gibson et al., 2016), and many other fields (Eyrolle et al., 2018).

A prerequisite for such applications is either a measured or modeled reference of precipitation tritium activity (Stewart and Morgenstern, 2016). Long-term measurements for precipitation tritium activity are rare worldwide, and even the longest time series are usually intermitted by gaps. In the absence of on-site measurements, either remote monitoring data have to be used as references (Huang and Pang, 2010; Thatcher et al., 1961), or estimations are required. There are several methods to reconstruct precipitation tritium time series for geographical locations (Li et al., 2019). The prediction of the first global model for tritium distribution in precipitation from 1960 to 1986 (Doney et al., 1992) was improved and provided a higher-accuracy estimate for precipitation 3H variations (Zhang et al., 2011) extending up to 2005 called the “modified global model of tritium in precipitation” (MGMTP). Unfortunately, the key parameters of the MGMTP are only available as isoline maps (Zhang et al., 2011), from which the model's coefficients can be extracted with high uncertainty in a manual way, which leads them to be ambiguous. In addition, the quality of the estimated precipitation tritium activity values by the MGMTP become quite poor after 1990 (Zhang et al., 2011); for instance, in the studied region it produced uninterpretable, negative values (Sect. 4). The most recent global model for precipitation tritium activity covering the period 1955–2010 (Jasechko and Taylor, 2015) used inverse distance weighting for interpolation, and its output is available in gridded format. However, this is based only on precipitation 3H activity concentration records of the stations of the Global Network of Isotopes in Precipitation (Rozanski et al., 1991), and it does not represent the most recent decade.

Although global models are available, due to the differences in tritium activity around the globe it is beneficial to define local precipitation 3H input curves (Stewart and Morgenstern, 2016). In the northern part of the Balkan region, for instance, it was shown that 3H content in precipitation deviated from the Vienna record considerably after 1980 (Miljević et al., 1992). This record is popularly used as a remote reference station in hydrological modeling and calculations in the Adriatic–Pannonian region. The quality of such curves is vital for the reliability of a hydrological model's output when employed as an input signal or data in hydrological modeling and calculations (Koeniger et al., 2008; Miljević et al., 1992). Indeed, it has recently been found that the (in)accuracy of the precipitation tritium time series used is the key uncertainty factor for groundwater recharge estimations (Li et al., 2019).

Measurements of precipitation tritium activity in the Adriatic–Pannonian region began in Vienna Hohe Warte in 1961, which is the longest continuously operating station in Central Europe and in the world (IAEA, 2019). Additional stations started operation in the past ∼50 years with frequent interruption in data collection (Araguas-Araguas et al., 1996; Krajcar Bronić et al., 1998; Rozanski et al., 1991; Vreča et al., 2008). The demand for long-term precipitation 3H reference time series in various hydrological and hydrogeological applications across the Adriatic–Pannonian region called forth the use of remote stations (e.g Gessert et al., 2019; Kanduč et al., 2014, 2012) and/or motivated the derivation of case-specific “composite” tritium reference curves (e.g., Kern et al., 2009; Krajcar Bronić et al., 1992; Ozyurt et al., 2014; Szucs et al., 2015). Derivation of these ad hoc “composite” 3H reference curves usually applied different imputation methods to the “gappy” time series and/or employed different interpolation techniques. Differences in the absolute values due to methodological differences might seem marginal if the peak concentration of the mid-1960s is used as a time marker; however if data are used as input in tritium mass balance models in the postbomb period, differences become highly important (Li et al., 2019).

The aim of this study is to create a spatially continuous gridded database for tritium (isoscape) in precipitation across the Adriatic–Pannonian realm for the decades around the turn of the 21st century with a special focus on post-2010, which is not covered by the existing global models.

The obtained regional gridded amount-weighted annual mean precipitation 3H activity time series for the Adriatic–Pannonian region (AP3H_v1 database) capture the well-known decrease in precipitation 3H activity (e.g., Fig. 2).

Besides this most striking long-term temporal pattern prevailing in the whole region seen from the isoscapes, subregional differences in precipitation 3H activity within the region are also reconstructed. Although no significant relationship was documented between latitude and/or continentality, still-increasing precipitation 3H activity was observable inland, with the lowest values documented along the Slovenian and northern Croatian coast in all years (e.g., 1982, 2010, 2014; Fig. 2). This pattern can be related to the generally observed lower activity at maritime coastal stations due to the higher contribution of primary marine evaporation practically free from 3H (Eastoe et al., 2012; Rozanski et al., 1991; Tadros et al., 2014; Vreča et al., 2006) and higher contribution of recycled modern meteoric water over the continent. For instance, moisture originating from continental Europe and the Atlantic Ocean was found to be distinct regarding tritium concentrations (8.8 and ∼0 TU, respectively; Juhlke et al., 2020).

Two of the longest records from Slovenia (Ljubljana) and Hungary (Budapest) illustrate the performance of the estimations and their potential in mitigating lack of data. Budapest and Ljubljana 3H records – complete years were used both in the variograms of the “anchor years” and in the interpolation – were compared to the interpolated product's time series of the nearest grid cell (Fig. 3). In the complete years, when the measured values were used in interpolation, there is an expected perfect match between the measured and modeled values. It becomes clear that the estimated records are more than capable of filling the gaps of the measured time series when there were no measurements (e.g., Ljubljana: 1985 and 1996; Fig. 3b) and usually provide a higher mean annual precipitation 3H in the case of incomplete years. The magnitude of differences between the measured and modeled data in the incomplete years varied between ∼16.5 and 1 TU for Budapest and Ljubljana, with a general tendency of obtaining higher differences with a higher ratio of precipitation not represented by tritium measurements.

The distinctive interannual fluctuation of amount-weighted annual mean 3H activity at Budapest and Ljubljana (Fig. 3) also indicates that the AP3H database produced differing subregional variability over the modeled time. In the whole investigated period (1976–2017) the coefficient of variation in the annual differences between the precipitation 3H values of the complete years in Budapest and Ljubljana was 62 %, calling for the need for spatially representative estimates, such as the ones presented from the AP3H database. In the period when both stations were active at the same time (1981–2003, excluding the incomplete years) the precipitation 3H activity was higher in Budapest than in Ljubljana by 5.5 TU on average, and this difference is quite well reflected by the AP3H data, which were 4.57 TU higher at Budapest on average. Besides these overall differences, distinct interannual variability can be observed at the Ljubljana and Budapest stations. For example, the maxima of the modeled precipitation 3H activity occurred in 1976 at the Ljubljana station and in 1978 at the Budapest station, and a minor peak seen in 1988 at Budapest (Fig. 3a) was without a counterpart in the Ljubljana records (Fig. 3b).

The estimated effective range shows a decrease in the spatial autocorrelation of tritium activity concentration of precipitation from the 1970s to the 2010s: ∼440 km in the 1970s, ∼425 km in the 1980s, and ∼360 and 235 km in the 2000s and 2010s, respectively (Fig. 4). This period (1970–2010) was characterized by the removal of bomb tritium from the atmosphere (Araguas-Araguas et al., 1996; Palcsu et al., 2018). The overwhelming activity of bomb-produced 3H was several orders of magnitude higher than the natural background (Rozanski et al., 1991) and largely masked the smaller-scale natural variability. During recent decades the tritium activity in precipitation has declined globally and regionally, approaching the natural prebomb levels, indicating that bomb tritium is now barely present in modern precipitation. Since in the Adriatic–Pannonian region 3H activity in precipitation approached natural levels by the 2000s (Krajcar Bronić et al., 2020; Palcsu et al., 2018; Vreča et al., 2008), it can be expected that the ∼200–300 km range obtained for the 2010s reflects the range of similarity of natural 3H variability in the study area (southeastern Europe and eastern Central Europe).

The performance of the AP3H database replicating measured precipitation 3H activity was tested via out-of-sample verification. The actual amount-weighted annual mean precipitation 3H activity time series at a Hungarian (Nick, Fig. 5a) and a Slovenian (Zgornja Radovna, Fig. 5b) station were compared to the AP3H data of the grid cell closest to the specific stations. The annual mean precipitation 3H activity estimates from the AP3H database fit the amount-weighted annual mean precipitation 3H activity data nicely for both independent records even if a small portion (<8 %) was not represented with measured 3H in the experimental data (Fig. 5a, b). The agreement between the independent mean annual precipitation 3H record and the AP3H estimates is superb for Zgornja Radovna, where the difference between modeled and actual values ranged from 1 % to 6 % (mean: 3.2 %). The difference between modeled and actual values in complete years for Nick ranged from 0 % to 31 % (mean: 8.3 %). The relatively larger mean error for the Nick record is due to an overestimation of the model values in 1994 (Fig. 5a); however, the model data fit the rest of the measured data as well as is seen for Zgornja Radovna. The pattern observed for the incomplete years strengthens previous impressions that (i) modeled mean annual precipitation 3H is higher than the mean calculated from just a few monthly measurements, and (ii) there is a tendency of obtaining higher differences with a higher ratio of precipitation not represented by tritium measurements. The comparison between the modeled and actual values for the stations' recorded precipitation 3H for multiple but exclusively incomplete years further strengthens these impressions.

Station records from Nick and Zgornja Radovna were not used in the interpolation to derive the AP3H database, yet these perfectly match the actual values when 3H activity concentration was measured for the total volume of annual precipitation. This excellent agreement supports the assumption that the AP3H database provides similarly accurate estimates as at other stations for the incomplete years as well. If this is the case, then differences between the AP3H data and the mean precipitation 3H activity estimates accompanied by a certain portion of precipitation amount lacking 3H activity can represent a weighting bias due to the fragmented observation record. Comparing the differences between (i) the estimated mean precipitation 3H activity values from these fragmented records and the AP3H estimates with (ii) the percentage of precipitation amount lacking 3H activity data in the incomplete years for Zgornja Radovna, Nick, Jósvafő, and Malinska revealed the relationship between these (Fig. 5e). If 3H activity was lacking for less than 15 % of the annual precipitation total, the difference between the calculated mean precipitation 3H activity values and the AP3H estimates is practically negligible (Fig. 5e). The difference between modeled and calculated mean precipitation 3H considerably increased (>6 TU) if the proportion of the precipitation lacking 3H activity was between 15 % and 30 % of the annual precipitation total (Fig. 5e). According to the GNIP protocol the required completeness is 70 % of the precipitation total for calculating amount-weighted mean isotopic values for a certain period (IAEA, 1992). The above results (Fig. 5e) confirm that a proportion of missing data of 30 % is indeed a critical limit because, if exceeded, each calculated mean precipitation 3H activity value of these scanty annual datasets showed great deviation from the modeled values. However, occasionally a precipitation deficit over 15 % yet under 30 % causes a noticeable difference (Fig. 5e), suggesting that a proportion of missing data of less than 30 % may be advisable for the calculation of representative amount-weighted isotopic means. In addition, these reinforce the decision made in the present study to only include values for which more than 85 % of fallen precipitation was analyzed for 3H.

Although at Nick and Zgornja Radovna – as seen above – the AP3H database provides a realistic estimation (Fig. 5a, b), one may ask why the calculated annual mean precipitation 3H activity values are systematically higher at Siófok (mean difference in the complete years: 1.52 TU) than the AP3H data (Fig. 5d). It could be explained by the closeness of the largest shallow lake in Central Europe, Lake Balaton (Hatvani et al., 2020). The mean residence time in the largest basin of the lake, Siófok basin, was estimated to be between 2 and 6 years in the 1990s (Istvánovics et al., 2002), which is presumably in the same range in the 2010s as well. Keeping in mind the gradual decrease in 3H in meteoric waters (in the region; e.g., Fig. 4), the local evaporation from this “aged” reservoir can provide an isotopically detectable contribution to the atmospheric moisture, which can be transferred to the falling rain droplets via molecular exchange mechanisms (Bolin, 1959), resulting in higher tritium activity values at the Siófok station than the ones from the AP3H database (Fig. 5d). This is a simple example illustrating the potential of this new regional annual mean precipitation 3H activity database (AP3H_v1) in hydrological applications.

The presented AP3H_v1 database of tritium activity was compared with the spatially corresponding output of both currently available global precipitation tritium isoscapes: the modified global model of tritium in precipitation (MGMTP; Zhang et al., 2011) and the global inverse distance weighted model (GIDW) at Budapest (Fig. 3a) and Ljubljana (Fig. 3b). Between 1975 and 1980 the estimates from the AP3H database and the MGMTP are very similar and resemble the actual weighted annual mean precipitation 3H at Budapest. However, only at Ljubljana is the MGMTP capable of steadily reproducing the actual measurements until the late 1990s. Afterwards, it indicates solely negative values, which are uninterpretable, just as most of the MGMTP-predicted values at Budapest after 1980. Meanwhile, the AP3H database provided much more accurate and reliable results (Fig. 3) as discussed above. Note here that the weak estimation of the MGMTP can be attributed to the difficulties in reading the precipitation tritium activity values from the only available output (isoline map) of the model and the undocumented factors of the model in given years.

The GIDW (Jasechko and Taylor, 2015) model was capable of reproducing the measured precipitation tritium values much more accurately at all locations than the MGMTP (Fig. 3). Nevertheless, the GIDW model produced a striking overestimation at the beginning of the modeled period, for example in 1977, when the measured values at Budapest were overestimated by >18 TU (Fig. 3a). In contrast, the GIDW model underestimated the actual values from from 1981 to 1992 except for 1990 (Fig. 3a).

The skills of the AP3H database and the GIDW model could only be compared at the Nick station because the record of Zgornja Radovna starts in the terminal year of the GIDW model's coverage. In the first complete year of the Nick record the GIDW estimation is closer to the actual measurements compared to the AP3H estimations; however, from 1992 to 1994 the GIDW estimations show much higher (mean error: 4.6 TU) annual mean 3H activity compared to the values calculated from the monthly measurements. Meanwhile, the difference from the AP3H estimations is smaller (mean error: 2.2 TU). Interannual variations of the estimated precipitation 3H activity from the GIDW and the AP3H database are in agreement at Jósvafő, and the models are largely in agreement regarding the absolute values and the decreasing trend, too.

However, the MGMTP had practically no success in estimating the actual measured values. For instance, in both 2000 and 2001 the MGMTP produced negative – thus meaningless – values at Malinska, when direct measurements were available, while the present database provided more reliable estimates (Fig. 5c).

Taken all together, the AP3H_v1 database gave better regional estimates than either of the global models. This may be because (i) the AP3H database is based on more local station data for the study area compared to the global models relying only on the GNIP records (Table 1), and (ii) the improved performance of the database obviously benefitted from incorporating the spatial correlation structure of precipitation tritium activity into the presented geostatistical prediction.

The final product (AP3H_v1.nc), the spatially continuous annual (1976–2017) 1km×1 km grids of amount-weighted annual mean precipitation tritium activity for the Adriatic–Pannonian region, is provided in a netCDF-4 (Network Common Data Form) format available at PANGAEA (10.1594/PANGAEA.896938; Kern et al., 2019) and compiled using the EPSG 3857 projection. For a solely visual inspection of the annual grids Panoply (NASA, 2020) is recommended. An R script written to be able to browse the dataset and convert the projection to EPGS 4326 is provided in the Supplement. This publication describes the AP3H_v1 database. As the database is updated with corrected or extended new data the new versions will be indexed with incremental integers.

Continuous long-term records of tritium in precipitation are scarcely available worldwide; thus estimation or modeling is necessary to exploit its potential in hydrological research. In order to decrease the uncertainty of tritium activity in the hydrological models, the application of regional 3H models has to be increased since these are more capable of producing accurate estimations than global ones (Stewart and Morgenstern, 2016).

Instead of using remote station data or ad hoc composite curves, site-specific time series retrieved from the presented AP3H_v1 database of amount-weighted annual mean precipitation 3H isoscapes should be used. These isoscapes (Kern et al., 2019) can serve as a reference dataset for studies on infiltration dynamics, water transport through various compartments of the hydrological cycle, mixing processes, and run-off modeling, for example, to estimate mean residence time in surface waters and groundwater (Kanduč et al., 2014; Ozyurt et al., 2014; Szucs et al., 2015). As a specific type of hydrogeological application, the regional model of 3H time series will serve as a benchmark in estimating the mean infiltration age of drip water (Kluge et al., 2010), which can provide an additional tool for ongoing cave monitoring studies from the region (e.g Czuppon et al., 2018, 2013; Fehér et al., 2016; Surić et al., 2010) in a spatiotemporally accurate way.

The higher precipitation 3H activity observed at a lakeshore station (Fig. 5c) likely reflects moisture recycling from the aged lake surface water via evaporation to local precipitation. The observed deviation highlights the potential of the database to reveal subregional anomalous local sources in the hydrological cycle. As a special case the post-2010 isoscapes can serve as benchmarks for background tritium activity for the region, helping to determine local increases in technogenic tritium from these backgrounds.

The AP3H database was able to provide better estimates than either of the currently available global models for the study area. Its values seem more capable of reproducing the actual annual mean precipitation 3H activity than the fragmented 3H data, which represent less than 85 % of the total annual precipitation. Prior to 1975 we encourage the use of the GIDW model's estimations (Jasechko and Taylor, 2015) as a reference for studies dealing with precipitation tritium activity. The AP3H and the GIDW model data should be spliced together in 1975 and can be used together in the need of a semicentennial precipitation tritium activity dataset.