Oxygen and carbon isoscapes for the Baltic Sea: Testing their applicability in fish migration studies

Abstract Conventional tags applied to individuals have been used to investigate animal movement, but these methods require tagged individuals be recaptured. Maps of regional isotopic variability known as “isoscapes” offer potential for various applications in migration research without tagging wherein isotope values of tissues are compared to environmental isotope values. In this study, we present the spatial variability in oxygen (δ18OH2O) and dissolved inorganic carbon (δ13 CDIC) isotope values of Baltic Sea water. We also provide an example of how these isoscapes can reveal locations of individual animal via spatial probability surface maps, using the high‐resolution salmon otolith isotope data from salmon during their sea‐feeding phase in the Baltic Sea. A clear latitudinal and vertical gradient was found for both δ18OH2O and δ13 CDIC values. The difference between summer and winter in the Baltic Sea δ18OH2O values was only slight, whereas δ13 CDIC values exhibited substantial seasonal variability related to algal productivity. Salmon otolith δ18Ooto and δ13Coto values showed clear differences between feeding areas and seasons. Our example demonstrates that dual isotope approach offers great potential for estimating probable fish habitats once issues in model parameterization have been resolved.


| INTRODUCTION
Several marking approaches have been employed to address questions in migration ecology. Until recently, conventional extrinsic markers (i.e., tags) applied to individuals have been used to investigate their movement (Lucas & Baras, 2000), but these methods require tagged individuals be recaptured to acquire spatial information. During recent years some investigations have been conducted to study long-term movements of individual adult Atlantic salmon (Salmo salar L.; Figure 1) in the sea using tags that record environmental characteristics along the migration routes (e.g., Chittenden, Ådlandsvik, Pedersen, Righton, & Rikardsen, 2013). However, due to the present size of the tags, the studied fish have to be large and, therefore, do not necessarily represent the majority of the population.
Intrinsic biochemical markers such as stable isotopes can provide an alternative approach to track individual movements over large geographical distances such as between continents (Hobson & Norris, 2008). All animals are isotopically marked by the environment they live in and by their diet. The assignment of an individual to a certain area works by estimating probabilities of occurrence for animal individuals by comparing values obtained from tissue samples to isotopic landscapes (i.e., isoscapes; Wunder, 2010). A great advantage in using biochemical markers is that they can be linked to those individuals that actually survived the migration to their breeding habitats, and therefore, better represent the population. As fish otoliths are almost completely mineralized from the carbonate of the environmental water (Kim, O′Neil, Hillaire-Marcel, & Mucci, 2007;Patterson, Smith, & Lohmann, 1993;Solomon et al., 2006), analysis and comparison of otolith and water stable isotopes can reveal the locations where the otolith of an individual fish is formed. However, if the chosen tissues/ materials are dissimilar, spatial assignment (matching of tissue and source isotope values) needs fractionation equations between the chosen tissue and source isotope values due to the different fractionation of the element isotopes via environmental and physiological factors.
Unfortunately, the availability of isoscape data for studies at a more local spatial scale appears to be sparse or the distance of the survey stations of global isoscape data may be too large for adequate local isotopic discrimination (see Bowen & Revenaugh, 2003). Therefore, additional isotopic data are needed to increase the resolution of global isoscapes to enable more precise reconstruction of animal locations and movement.
The aims of this study were (1) to provide horizontal and vertical isotopic gridded data sets (i.e., isoscapes) of oxygen (δ 18 O H 2 O ) and dissolved inorganic carbon (δ 13 C DIC ) for the water of the Baltic Sea. (2) As an example we demonstrate the potential of these isoscapes using two Atlantic salmon individuals from the River Simojoki. Combined with the isotope data from the salmon and spatial probability surface maps, we show probable locations of individual fish in various time points during their sea-feeding migration phase in the Baltic Sea. We also demonstrate how parameterization of the models influences on location estimates.

| Sampling and isotope analyses of water
Baltic Sea water samples were collected during three different cruises by the R/V Aranda of the Finnish Environment Institute (SYKE). To evaluate a possible seasonal impact on sea water isotope values, we collected two summer sets and one winter set of samples. The dates of the cruises were (1)   Rosette sampler records conductivity, from which salinity was automatically calculated. From every sampling occasion of water for isotope analysis purpose, salinity was recorded. Sample water collected for δ 13 C DIC analysis, were injected into 12-ml borosilicate Exetainer vials (cat. no 438B; Labco Ltd., High Wycombe, UK) prepared in the laboratory, where 0.2 ml of 85% orthophosphoric acid (H 3 PO 4 ) was added into each vial, which was then sealed with a cap (containing a rubber septum) and flushed and filled with a helium atmosphere. In the field 2-4 ml of sea water from each station and sample depth was injected through the rubber septum into the vial. Sample water for δ 18 O H 2 O analysis were collected in 20-ml glass scintillation vials and filled full ensuring no air bubbles. All samples were stored in an unlit  software version 4.5.5 (Schlitzer, 2002(Schlitzer, , 2011.

| Otolith sampling, micromilling and stable isotope analysis
In order to test applicability of Baltic Sea isoscapes in fish migration studies, we analyzed otoliths from two example female Atlantic salmon (hereafter Baltic salmon or salmon) originally caught, as they were returning to the River Simojoki ( Figure 2) to spawn in 2008, by the Natural Resources Institute Finland (LUKE) as part of the program to monitor yolk-sac fry mortality (M74 syndrome: e.g., Keinänen et al., 2012; original LUKE code numbers SS5447 and SS5461, hereafter FISH 1 and 2, respectively). Both salmon had spent two years feeding in the sea. The origin (wild or hatchery-reared) and age of the salmon was determined from the scale nucleus and scale growth pattern (Hiilivirta, Ikonen, & Lappalainen, 1998). FISH 1 was of wild origin (5,300 g, 80 cm) and FISH 2 was of hatchery-reared origin (6,300 g, 81 cm). Both sagittal otoliths were removed from the head of the salmon, cleaned in deionized water to remove any remaining organic tissue, and dried overnight at 60°C.
Both otoliths were sampled for δ 13 C and δ 18 O analysis using the custom-built three-dimensional computer-controlled micromilling system in the Saskatchewan Isotope Laboratory at the University of Saskatchewan following the procedure of Wurster, Patterson, and Cheatham (1999  using package RNetCDF (Michna, 2012). Interpolated maps were produced using Data Interpolating Variational Analysis (DIVA) gridding software (Troupin et al., 2012) included in ODV (DIVA parameters: scale lengths chosen automatically; signal-to-noise ratio = 40; quality limit = 3.0; excluding outliers). Vertical profiles were also created using DIVA gridding (scale lengths chosen depending on used data; signal-to-noise ratio = 40; quality limit 3.0; excluding outliers).

| Creating isotopic (isoscapes) and temperature maps for the Baltic Sea
The coastlines in the maps are based on the Global Self-consistent Hierarchical High-resolution Shorelines database v 2.1 (Wessel & Smith, 1996).
As the spatial coverage of collected Salmon summer temperatures were fixed at 11.5°C, calculated from preferred true 10 m staying depths of salmon in Baltic salmon data storage tag study by Westerberg, Sturlaugsson, Ikonen, and Karlsson (1999). For winter temperatures (2007)(2008), we used all available temperature profiles collected from HELCOM database (Andersson, 2014) covering the whole Baltic Sea (mean temperatures from depths 5-15 m) and created an interpolated surface for the preferred 10 m staying depth. Interpolated metabolically derived bicarbonate δ 13 C diet values were derived from δ 13 C values of salmon dietary species around the Baltic Sea (Appendix S3).  Patterson et al., 1993). Moreover, the δ 13 C oto value is a mix of bicarbonate δ 13 C DIC from ambient water and metabolically derived bicarbonate δ 13 C diet (Solomon et al., 2006;Wurster & Patterson, 2003).

| Creation of otolith-related isoscapes from
We therefore corrected δ 18 O H 2 O and δ 13 C DIC values using literaturederived fractionation equations, after which the corrected values are considered as "otolith isoscape" of both elements (Appendix S1). This permits a direct comparison of otolith and Baltic Sea water isotopes, thereby providing a probabilistic spatial assignment of salmon during their sea-feeding phase. Oxygen isotope values of otoliths reflect those of the ambient water (Campana, 1999;Farrell & Campana, 1996;Thorrold, Jones, & Campana, 1997), with a temperature-dependent fractionation (e.g., Patterson et al., 1993). Following common practice, we used the linear temperature-dependent fractionation (e.g., Patterson et al., 1993).
where T is temperature (10 3 /K), where K is ambient water tempera-  An otolith-related δ 13 C oto isoscape was calculated as follows (e.g., Solomon et al., 2006;Wurster & Patterson, 2003) to allow comparison between δ 13 C DIC and δ 13 C oto : where δ 13 C diet is a mean δ 13 C value of salmon primary prey species in

| Salmon assignment using otolith and water isotope values
To estimate the locations of salmon individuals in their 1SW, 2SS and second 2SW from δ 18 O oto and δ 13 C oto values, we calculated probability density surfaces for each salmon by using a deterministic grid covering the Baltic Sea following the approach presented in Wunder (2010) (Pebesma, 2004). The probability density surface calculation approach assumes water-temperature-dependent The resulting probability surfaces do not represent twodimensional probability surfaces (total probability in each maps is not scaled to be 1) and the minimum and maximum probability values are not the same in all maps as is common in this kind of approach (see e.g., Wunder, 2010). Instead, these surfaces visualize where the probability of obtaining the measured value of the otolith (or the probability of presence of particular individual at a certain time) is the greatest, and where it is (much) lower. In those maps where both oxygen and carbon isotopes have been taken into account, probabilities were calculated by multiplying isotope-wise probabilities, assuming independence, which may not be strictly true for this kind of phenomenon. However, our data did not allow for full estimation of covariance between true data values due to partially differing measurement or observation locations. Nevertheless, the covariance of interpolated surfaces was about 10-fold smaller than the variances of individual isotopes. Therefore, we expect that any possible error in the results, due to dependence in isotope values, is small.
To study, how sensitive our approach is to measurement errors related to the stable isotopes, we conducted a simple sensitivity anal- The difference was smaller in the winter (Figure 3d). An increasing trend was observed in δ 13 C DIC values from the Gulf of Finland to the Baltic Proper in both seasons, from around −1‰ to around 2‰ in summer and from around −2‰ to around 0‰ in winter (Figure 3c,d).

Vertical interpolation of the transects showed that δ 18 O H 2 O values in
the Central Baltic Proper were higher below ~50 m from around −6.5‰ to around −5.5‰ at the bottom and as high as −4‰ in the Southern Baltic Proper (Figure 4a,b), whereas a clear decreasing trend in summer   decreasing it causes the most probable area to shift slightly toward east. Increasing δ 13 C value seems to make the assignment more concentrated at one location and decreasing it causes the result to be less determined.

| Isotope values of the Baltic Sea
We Distance from nucleus (%) isotopic maps from a depth of 10 m and also via vertical cross-sections.
Differences between sea areas in both isotope values were observed.
Horizontal δ 13 C DIC values clearly differed between summer and win-  (Schmidt, Bigg, & Rohling, 1999) and has been recorded from the Southern Baltic Sea (Frohlich, Grabczak, & Rozanski, 1988;Punning, Vaikmae, & Maekvi, 1991). Temperature-related autumn overturn of the water column breaks down the summer stratification and mixes the water above the halocline (down to ~50 m), but the deeper water column remains rather stable maintaining more marine characteristics . However, in our study, δ 18 O H 2 O values remained stable between summer and winter, and only occasional strong salt water pulses (e.g., Dickson, 1973;Matthäus & Lass, 1995)   High probability Low probability  (Fry, 2006) was reflected in the distinct differences between summer and winter in the surface water values of δ 13 C DIC . This mechanism also offers a possibility to reveal animal movements (MacKenzie et al., 2011;Trueman, MacKenzie, & Palmer, 2012

| Isotope values of otoliths and the location estimates of salmon during their sea-feeding phase
Both otoliths' isotope values showed clear and similar variations from the nucleus to the otolith edge for the two example salmon individuals in this study. Our results are consistent with measurements for salmon in the Atlantic environment (Hanson et al., 2010(Hanson et al., , 2013, the Pacific (Zazzo, Smith, Patterson, & Dufour, 2006) and even in early salmon-like fish 172 million years ago in the proto-Atlantic during the Jurassic Period (Patterson, 1999); the similar isotope values in the otolith nuclei and the outermost edges clearly show the hatching and spawning times of the individual salmon in the river. Values between the nucleus and the edge indicate the salmon life cycle from a parr in the river, through the following juvenile post-smolt phase to the final marine-feeding phase. Low δ 18 O and δ 13 C values in the nucleus and shortly after indicate the juvenile riverine phase of salmon (e.g., Zazzo et al., 2006). Rapid increases in both δ 18 O and δ 13 C isotope values indicated the beginning of the salmon post-smolt migration phase continuing with the first and second sea winter feeding phases in the Baltic Sea. Temperature-related fractionation is clearly observed be- Temperature-fractionation relationships (Godiksen et al., 2010;Hanson et al., 2013;Patterson et al., 1993;Storm-Suke et al., 2007) between  (Aro, 1989;Jutila, 2008;Kallio-Nyberg et al. 2011;Salminen, Kuikka, & Erkamo, 1994;Torniainen et al., 2014) have indicated that majority of River Simojoki salmon migrate to the Baltic Proper area for feeding.
However, smaller but appreciable proportion feed in the Bothnian Sea.
Based on those findings, we see the Model 1 provides most sensible results.
The Baltic Sea temperatures are highly variable between seasons and markedly higher during summer than in the Atlantic. This could be seen in the otolith δ 18 O value profiles via the clear decrease in the values between winters (e.g., Hanson et al., 2010). Summer temperatures used in our assignment method are based on Westerberg et al.
(1999) study conducted using data storage tags. Therefore, the salmon preferred summer temperature should be realistic. In winter, temperatures are average values of measured temperatures from the assumed depth range occupied by salmon. The exact temperature range occupied by wild salmon in the winter in the Baltic Sea is unknown, but it has been shown that the interannual differences in winter (February/ March) sea surface temperature (SST) are small, average temperature being ~0.5°C and ranging roughly between 1.5 to −0.5°C (Siegel, Gerth, & Tschersich, 2008) In addition, temperature does not change much until unsuitable hypoxic halocline for salmon is reached in winter .
Carbon isotope values have been shown to offer potential for salmon assignment in the Atlantic (MacKenzie et al., 2011). Therefore, we tested whether δ 13 C oto and δ 13 C DIC values could differentiate especially between the Gulf of Finland and the Bothnian Sea, which is difficult from oxygen isotopes alone. Substantial part (~20%) of the otolith δ 13 C is derived from diet (Solomon et al., 2006) and knowledge about the spatial variability of salmon diet δ 13 C in the Baltic is still insufficient. Therefore, possibly false assignments are at least partly due to missing δ 13 C values for prey species covering the whole Baltic Sea area. In general, our assignment models have uncertainty due to lack of present knowledge of the exact parameters for Baltic salmon and therefore there is urgent need for species-specific experimental studies to better understand temperature-fractionation relationships and ambient environment of Baltic salmon. Such information could be obtained by conducting controlled experiments of temperature-fractionation relationships between water and salmon otoliths. Regardless of these small shortcomings, our examples illustrate the great potential of the isoscape method used in this and other studies (Correia, Barros, & Sial, 2011;Dufour, Höök, Patterson, & Rutherford, 2008;Hanson et al., 2013) for revealing fish movements during their sea-feeding phase which are difficult to study using other methods.
In conclusion, we emphasize that δ 18 O oto and δ 13 C oto values offer great possibility to show habitats and estimate migration pathways of salmon and other fish species alongside other methods (Chittenden et al., 2013;Healey et al., 2000;Lucas & Baras, 2000;Peterson, Morgan, Fisher, & Casillas, 2010), but uncertainties in the set parameters need to be better resolved. In addition, we recommend that all the data (e.g., prey species, water isotope values and water temperature) used in the assignment model and the temperature-fractionation corrections should cover the whole study area to achieve the most reliable area assignments.

ACKNOWLEDGMENTS
We thank Sandra Timsic, Dinka Besic and Bruce Eglington in the Saskatchewan Isotope Laboratory for valuable help with otolith micromilling and statistical guidance. We are grateful for access to the laboratory facilities of R/V Aranda (SYKE) and for the cooperation of its crew and staff in sample collection during the cruises. This research was supported by awards from the Maj and Tor Nessling Foundation (#2010150, #2011105, #2012506, #2013040 to JT) and the Academy of Finland (#134139 to MK). Anonymous reviewers provided useful comments on an earlier draft of this manuscript.

CONFLICT OF INTEREST
None declared.

DATA ACCESSIBILITY
We provide the measured δ 18 O H 2 O and δ 13 C DIC values from the Baltic Sea and salmon otoliths for any scientist to use in the ISOBANK repository https://github.com/BrianHayden/IsoBank. R codes can be found in Dryad Digital Repository.