Mapping multiple source effects on the strontium isotopic signatures of ecosystems from the circum‐Caribbean region

Abstract

A method for mapping strontium isotope ratio (⁸⁷Sr/⁸⁶Sr) variations in bedrock and water has been recently developed for use in the interpretation of ⁸⁷Sr/⁸⁶Sr datasets for provenance studies. The mapping process adopted the simplifying assumption that strontium (Sr) comes exclusively from weathering of the underlying bedrock. The scope of this bedrock‐only mapping method is thus limited to systems where the contributions of other sources of Sr are minimal. In this paper, we build on this ⁸⁷Sr/⁸⁶Sr mapping method by developing a mixing model of Sr fluxes from multiple sources to the bioavailable Sr pool. The new multiple source model includes: (1) quantitative calculations of Sr fluxes from bedrock weathering using an empirical rock weathering model; and (2) addition of sub‐models calculating the contribution of Sr fluxes from atmospheric aerosols based on outputs from global climate model simulations. We compared the performance of the new multiple source model and the bedrock‐only mapping method in predicting observed values from two datasets of bioavailable ⁸⁷Sr/⁸⁶Sr from the circum‐Caribbean region (Antilles and Mesoamerica). Although the bedrock‐only method performs relatively well in Mesoamerica (n = 99, MAE = 0.00011, RMSE = 0.00073), its prediction accuracy is lower for the Antillean dataset (n = 287, MAE = 0.0021, RMSE = 0.0027). In comparison, the new multiple source model, which accounts for the deposition of sea salt and mineral dust aerosols, performs comparably well in predicting the observed ⁸⁷Sr/⁸⁶Sr values in both datasets (MAE = 0.00040, RMSE = 0.00087 and MAE = 0.00014, RMSE = 0.0010). This study underscores the potential of using process‐oriented spatial modeling to improve the predictive power of Sr isoscapes over large spatial scales and to refine sampling strategies and bioavailable Sr dataset interpretations for provenance studies.

Introduction

Strontium isotope ratios (⁸⁷Sr/⁸⁶Sr) have been increasingly and successfully employed as provenance tracers of a wide variety of environmental processes (Graustein and Armstrong 1983, Capo et al. 1998, Kelly et al. 2005, Bentley 2006). The method relies on comparing the ⁸⁷Sr/⁸⁶Sr signature of a sample(s) of unknown origin to that of reference samples: generally local sample materials such as rock, soil, water, plant or animal. At larger spatial scales and/or for large datasets, where development of comprehensive reference collections can be prohibitive, the identification of provenance using ⁸⁷Sr/⁸⁶Sr data requires comparative analyses with reference ⁸⁷Sr/⁸⁶Sr maps (Price et al. 2002, Hodell et al. 2004, Bentley 2006, Evans et al. 2009).

Previous efforts have been made to map ⁸⁷Sr/⁸⁶Sr variations at large scales (Beard and Johnson 2000, Bataille and Bowen 2012) for use in the interpretation of ⁸⁷Sr/⁸⁶Sr for provenance studies. Bataille and Bowen (2012) developed a model for environmental Sr isotope variation based on the assumption that bedrock weathering represents the primary source of environmentally available Sr. Their GIS‐based model uses lithology‐specific model parameters and generalized ⁸⁷Rb decay equations to account for the combined effects of lithology and time on ⁸⁷Sr/⁸⁶Sr variations in bedrock. Although the predictive power of this method is still less than optimal, comparison with validation datasets shows promise for predicting geographic ⁸⁷Sr/⁸⁶Sr variation in different sample materials. Bataille and Bowen's model formulation only considers ⁸⁷Sr/⁸⁶Sr variations in bedrock and water and does not explicitly address other Sr sources that may contribute to ⁸⁷Sr/⁸⁶Sr variations in biologically available (bioavailable) Sr. Mapping bioavailable ⁸⁷Sr/⁸⁶Sr variations is fundamentally important for constraining local ⁸⁷Sr/⁸⁶Sr signatures for ecological (Capo et al. 1998) and archeological (Bentley 2006) studies of provenance.

In many locations, the bioavailable Sr pool can reflect the complex integration of multiple sources (bedrock and atmospheric sources) and sinks of Sr that interact with the ecosphere over different time scales. At the base of the ecosystem, plants uptake and incorporate Sr from the exchangeable Sr pool in soil, defined as the pool of Sr bound to organic matter and/or soil minerals and exchanging with plants and/or soil water (Capo et al. 1998, Stewart et al. 1998). This exchangeable Sr pool is operationally defined as the Sr leached from dry soil with reagents such as buffered ammonium chloride in methanol (Capo et al. 1998).

Stewart et al. (1998) proposed mathematical formulations to model bioavailable ⁸⁷Sr/⁸⁶Sr for local ecosystems, but this model is currently not applicable at regional scales because it requires a large number of site‐specific parameters as input. Consequently, most attempts to map regional ⁸⁷Sr/⁸⁶Sr variation in ecosystems have involved interpolating average ⁸⁷Sr/⁸⁶Sr measurements from local biomass (Price et al. 2002). However, this method can be data‐intensive and costly and in most cases does not explicitly consider the underlying spatial structure of factors which likely govern Sr isotope variation (e.g., bedrock distribution, climate, atmospheric deposition). This method is also hampered by the challenge of selecting appropriate samples, which is non‐trivial as different sample materials may integrate different spatial and temporal scales of ⁸⁷Sr/⁸⁶Sr variation.

Empirical mapping studies could greatly benefit from the parallel development of process‐oriented spatial models. These models can help refine the interpretation of bioavailable Sr datasets and produce cost‐effective reference isoscapes for provenance studies. These maps can be tested on compiled datasets of Sr isotope measurements from a range of materials. In this paper, we further develop the model of Bataille and Bowen (2012) by adding components that account for contributions from bedrock and non‐bedrock sources to a mixed bioavailable Sr pool. We couple Bataille and Bowen's model to an empirical, process‐oriented chemical weathering model to calculate estimates of soluble Sr fluxes from bedrock weathering to the bioavailable zone in soils. This model accounts for chemical weathering rate dependence on runoff and lithology (Jansen et al. 2010). We also include sub‐models representing the contribution of atmospheric sources (dust and sea salt aerosol) to the soluble Sr in soils. We use a simple box model to describe the mixing of these different sources within the soil, and test the new model against two datasets of bioavailable Sr isotopes from the circum‐Caribbean region.

The circum‐Caribbean region is an ideal area to validate models describing the interaction of multiple Sr sources to soil water because it receives: (1) spatially variable inputs of Sr from bedrock weathering due to its diverse geology; (2) large annual inputs of Sr from atmospheric aerosols through dry and wet deposition of Saharan mineral dust and sea salt from the surrounding ocean; and (3) more episodic inputs of Sr‐rich tephra from volcanic eruptions. In addition, large amounts of data have been gathered in this region, including studies assessing the comprehensive Sr budget of ecosystems by measuring ⁸⁷Sr/⁸⁶Sr in rainfall, soil, the exchangeable Sr pool, bedrock and plants (Bern et al. 2005, Pett‐Ridge et al. 2009b, Pozwa et al. 2002) and work producing several hundred ⁸⁷Sr/⁸⁶Sr measurements of local plants and animals for archeological provenance studies (Price et al. 2000, Hodell et al. 2004, Wright 2005, Price et al. 2006, Price et al. 2007, White et al. 2007, Price 2008, Price et al. 2010, Wright et al. 2010, Thornton 2011, Laffoon et al. 2012). The combination of these bioavailable Sr isotope datasets with the natural gradient of atmospheric deposition and varied lithologies of the circum‐Caribbean region offers an excellent opportunity to develop and validate our multi‐source mapping approach.

Material and Methods

Model derivation, calibration, and validation are described in the following sections. Additional details and documentation are available in the accompanying Appendix.

Model formulation

Previous research has contributed to our understanding of the factors controlling the mixing of multiple Sr sources in soils and ecosystems (Kennedy et al. 1998, Chadwick et al. 1999, Nakano et al. 2001, Pozwa et al. 2002, Bern et al. 2005, Chadwick et al. 2009, Pett‐Ridge et al. 2009a, Pett‐Ridge et al. 2009b). The contribution of atmospherically derived Sr to the soluble Sr in soil depends mostly on the magnitude of the bedrock weathering flux. When the flux of soluble Sr from bedrock weathering is low, the atmospheric contribution can become the dominant source of soluble Sr to soils and ecosystems. Comprehensive modeling of the soluble Sr mass balance in soil requires detailed soil models accounting for (1) inputs of Sr varying with relative weathering rates and/or deposition rates of each source, (2) outputs of Sr varying with leaching rates and/or erosion associated with each source. Although the theoretical formulation of such models exist (Hilley and Porder 2008, Hilley et al. 2010, Porder and Hilley 2011), not enough data are available to apply them at large scales.

Here, we develop a simple model that estimates the ⁸⁷Sr/⁸⁶Sr of bioavailable Sr in the circum‐Caribbean region as a function of Sr inputs from a limited number of sources (Fig. 1). We assume that the bioavailable Sr pool is well represented by the mixing of soluble Sr from weathering of primary minerals (F_w→bio), weathering of deposited Saharan mineral dust (F_d→bio), and deposition of dissolved sea salt in rainfall (F_ss→bio). We neglect: (1) bedrock sources other than the major surficial bedrock; (2) local and regional atmospheric sources of Sr, such as recycling of local minerals by erosion and/or deposition of burned or dead biomass (Nakano and Tanaka 1997) and volcanic ash (Muhs and Budahn 2009); and (3) other continental atmospheric sources, e.g., North American mineral dust (Muhs et al. 2007). This assumption is supported by the low deposition rates of these other atmospheric sources of Sr in comparison with sea salt and Saharan dust in the circum‐Caribbean region (Muhs et al. 1990, Bern et al. 2005, Muhs et al. 2007) despite their importance in other geographic settings (Graustein and Armstrong 1983, Naiman et al. 2000).

Representation of the box model with fluxes of dissolved Sr mixing in the soluble bioavailable Sr pool. Lbio represents the losses of Sr from the bioavailable pool to other pools.

We model the bioavailable Sr pool as a well‐mixed reservoir at steady state. Both assumptions are likely to be invalid for some systems, especially those affected by disturbance in which changes in Sr cycling and vertical or small‐scale spatial gradients exist, but are necessary simplifications for this investigatory large‐scale regional modeling. We test the sensitivity of our models to several different Sr sources using three model versions.

The first is a “bedrock model” (as used in Bataille and Bowen 2012), in which the isotopic composition of bioavailable Sr is equal to the isotopic composition of Sr released from local bedrock weathering. Eq. 1 from Bataille and Bowen (2012) encapsulates the theory for both silicate and carbonate rocks: $$$$ where (⁸⁷Rb/⁸⁶Sr)_parent and (⁸⁷Rb/⁸⁶Sr)_rock refer to the ⁸⁷Rb/⁸⁶Sr of the parent material and the modern rock unit, respectively, t₁ is an approximate age of crustal differentiation (3000 Ma), and t₂ the age of the modern rock unit. ⁸⁷Rb/⁸⁶Sr can be expressed as: $$$$ where W refers to the atomic weight of an element and Ab to the abundance (%) of an isotope.

In Eq. 1, the parent rock term is treated differently in the silicate and carbonate models. Silicates parent rock term is calibrated by assuming that the silicon content of the rock type is an indicator of magma source composition whereas carbonates parent rock term was calibrated using seawater ⁸⁷Sr/⁸⁶Sr variations throughout geological time (for details see Bataille and Bowen 2012).

The second is a “two source mixing model” describing the mixing of sea salt and bedrock derived Sr: $$$$ where 2s indicates the two source mixing model, F_ss→bio and F_w→bio and (⁸⁷Sr/⁸⁶Sr)_ss and (⁸⁷Sr/⁸⁶Sr)_b are the magnitudes and Sr isotope ratios of the sea salt and bedrock weathering fluxes, respectively.

The third is a “three source mixing model” describing the mixing of sea salt, mineral dust, and bedrock weathering Sr: $$$$ where 3s indicates the three source mixing model, F_d→bio and (⁸⁷Sr/⁸⁶Sr)_d, are the magnitude and Sr isotope ratio, respectively, of the dust deposition flux.

Model parameterization

Bedrock weathering flux

To calculate F_w→bio, we applied the continental‐scale, process‐oriented, empirical weathering model developed by Jansen et al. (2010) on the Caribbean USGS geodatabase. The model calculates the rate of Si dissolution from different rock types as: $$$$ where F_DSi is the flux of dissolved SiO₂ to river water (t·km⁻²·a⁻¹) from the bedrock lithology at a given map location having mean annual runoff q (L·a⁻¹·m⁻²), lithology‐specific parameters b_n and b₀. The latter two parameters have been calibrated by Jansen et al. 2010 using measurements of dissolved silica loads in rivers from Hartmann et al. 2010. We were not able to compile enough measured F_DSi data to recalibrate this model for the circum‐Caribbean region and instead opted to use an existing calibration. Two calibrations exist, one for the conterminous USA (Jansen et al. 2010) and one for Japan (Hartmann et al. 2010, Jansen et al. 2010). We use the Japan calibration because it reflects a range of environmental conditions (coastal and insular region), climate (high temperature and precipitation), and geological settings (subduction area) that are reasonably good analogs for the circum‐Caribbean region. We obtained b₀ and b_n values by matching each rock type in the Caribbean bedrock geology geodatabase with an equivalent lithological category given in Jansen et al. (2010) (see Appendix: Table A1). The Japan calibration does not offer values for mafic plutonic rocks and carbonates, and for both we used the calibrated values from the conterminous USA (Jansen et al. 2010).

In order to obtain a high‐resolution gridded runoff dataset for our study area we calculated q as: $$$$ where MAP is the mean annual precipitation (mm·a⁻¹) obtained from the WorldClim website (Hijmans et al. 2005) and MAAET the mean annual actual evapotranspiration in mm·a⁻¹ obtained from the Global High‐Resolution Soil‐Water balance from the CGIAR‐CSI website (Trabucco and Zomer 2010). Previous applications of the weathering model have used modeled runoff grids from the Global Runoff Data Center (GRDC) (Fekete et al. 2002), but these data provide insufficient spatial resolution for the island regions we examine here. Our runoff estimates compare well with the GRDC grids for catchments where they overlap: by resampling our estimates to the 0.5° GRDC resolution we found that at 55% of the GRDC gridcells our runoff estimates are within 20% of the GRDC values and at only 10% do the two estimates differ by more than 40%.

Modeled F_DSi values were used to obtain Sr weathering fluxes for each lithology using a Sr/SiO₂‐normalization technique similar to Hartmann and Moosdorf (2011a): $$$$ Here F_DSr is the flux of soluble Sr from the bedrock lithology at a given map location to the river and (Sr/SiO₂) is the molar ratio of Sr to SiO₂. Sr/SiO₂ ratios are approximated as the median value for each rock type based on 121,253 analyses available through the Earthchem Portal (www.earthchem.org; query by “chemistry”: all Sr AND SiO2, “database”: Georoc results). Calculated medians and associated standard deviations for each rock category descriptor used in Bataille and Bowen (2012) are available in Appendix: Table A1.

Our application of this model to estimate contributions of Sr from bedrock to the bioavailable pool involves two fundamental assumptions. First, by adopting a model calibrated using data from rivers, we are effectively assuming that river SiO₂ loads depend solely on the flux of silicon (Si) from bedrock (F_DSi) and that Si from other sources is negligible. This assumption is justified for sea salt aerosol because sea salt contains only small amounts of Si. This assumption is not always justified for mineral dust can participate in the Si flux to river water (Chadwick et al.1999). Second, Eq. 7 assumes that the bedrock weathering flux of Sr is approximately equal to the flux into the bioavailable pool. This may not be the case in many systems because plants cycle cations primarily from the upper soil and some Sr may be routed directly to stream systems via groundwater without interacting with the bioavailable zone (Nakano et al. 2001, Pozwa et al. 2002). To the degree that they are incorrect, both assumptions would lead to overestimation of F_w→bio, and this possibility is considered in the interpretation of our results.

Bedrock Sr isotope flux

We applied the bedrock Sr isotope mapping method developed by Bataille and Bowen (2012) to calculate (⁸⁷Sr/⁸⁶Sr)_b as a function of rock age and lithology (Fig. 2). The bedrock‐only model from Bataille and Bowen (2012) represents an averaged ⁸⁷Sr/⁸⁶Sr prediction of the dominant bedrock limited by the resolution of geological maps. Consequently, the bedrock‐only model does not account for (⁸⁷Sr/⁸⁶Sr)_b variations due to contribution of non dominant lithologies and/or (⁸⁷Sr/⁸⁶Sr)_b variations related to compositional variation within rock units (e.g., van Soest et al. 2002) or to difference in the length of bedrock exposure to weathering (Lasaga 1984, Lasaga and Blum 1985). This lack of resolution in the geological map is particularly significant for carbonates because their high Sr/SiO₂ ratio and their high weatherability (see Appendix Table A1) causes them to be the dominant source of Sr in many catchments even when they are only present in traces (Anderson et al. 2000). The optimized model parameters are subject to uncertainties and biases related to the incompleteness of the databases used for calibration and the non‐random distribution of samples therein, as discussed by Bataille and Bowen (2012).

Bedrock‐only model 87Sr/ 86Sr variations calculated using equations from Bataille and Bowen (2012).

We developed our bedrock Sr isotope map using a geodatabase describing the age and lithology of rock units throughout the Caribbean region (French et al. 2004). The lithological descriptors used in this database are different than those in the database previously used by Bataille and Bowen (2012) for the conterminous USA. For our new application we matched each lithological descriptor in the Caribbean geodatabase with its closest analogous descriptor found in Bataille and Bowen (2012) (see Appendix: Table A1). Calculations for carbonate lithologies were conducted as described in Bataille and Bowen (2012).

Limitations arise from applying this ⁸⁷Sr/⁸⁶Sr mapping method to the Caribbean geodatabase because the lithological descriptors used are highly generalized. Intrusive igneous rocks are particularly poorly characterized. As a result, we selected all of the polygons having intrusive rock types and reclassified each of these as granite, granodiorite, quartz diorite, or gabbro based on local geological maps of Puerto Rico (Reed et al. 2005), Hispanolia (Draper et al. 1995), Trinidad and Tobago (Saunders and Snoke 1998), and Cuba (Pushcharovski 1989). New lithological maps for this area will soon be available (Moosdorf et al. 2010) and will help resolve these issues (Moodsdorf, personal communication).

Atmospheric sources

To calculate F_d→bio, we assumed that the Sr content of mineral dust, and its ⁸⁷Sr/⁸⁶Sr, were constant over the study area. Mineral dust mineralogy and geochemistry is highly variable seasonally but long term average deposition is relatively homogeneous in the Caribbean (Prospero et al. 1970, Trapp et al. 2010). We assigned a concentration and ⁸⁷Sr/⁸⁶Sr of mineral dust equal to the average Sr content and ⁸⁷Sr/⁸⁶Sr of mineral dust collected over the Caribbean region: 195 ppm (Grousset et al. 1992, Rognon et al. 1996, Grousset and Biscaye 2005) and 0.71788, respectively (Grousset and Biscaye 2005, Formenti et al. 2011).

We obtained long‐term average mineral dust deposition rates at 1° × 1° spatial resolution from a synthesis (Mahowald et al. 2005) of results from three reanalysis models, each run for 10+ years (Luo et al. 2003, Ginoux et al. 2004, Tegen et al. 2004). The results have been shown to compare well with available satellite observations (Mahowald et al. 2005). However, aerosol modeling is a relatively recent field and large uncertainties remain in the models because the physics of aerosol deposition are not fully understood (Huneeus et al. 2011).

The low‐resolution dust product does not represent variation in deposition rates driven by fine‐scale variation in dust scavenging by precipitation, which may be an important control on dust deposition across our study region (Rea 1994). We downscaled the low resolution (1° × 1°) dataset as: $$$$ where (Sr)_d is the abundance of Sr in mineral dust (in percent), D_d is the deposition rate of mineral dust from the reanalysis dataset in t·km⁻²·a⁻¹, 0.75 is the proportion of wet deposition of mineral dust in the circum‐Caribbean region (Jickells et al. 1998, Prospero et al. 2010), 0.25 the proportion of dry deposition of mineral dust, and PPF is a “precipitation corrective factor” used to account for the enhancement of atmospheric deposition over land areas (relative to the open ocean) due to higher precipitation rates (Rea 1994). The PPF is defined as: $$$$ where MAP_hr is the high resolution (30 arc‐second) mean annual precipitation on land obtained from the WorldClim dataset (Hijmans et al. 2005) and MAP_lr is the estimated mean annual precipitation amount for the same location calculated by bilinear interpolation of the low resolution (2.5° × 2.5°) Global Precipitation Climatology Project Version 2.2 (GPCP).

This downscaling calculation acts primarily to enhance the fine‐scale structure of the dust deposition field over island regions, where the coarse‐resolution GPCP and dust model output do not represent high‐frequency variation in precipitation rates and dust scavenging by precipitation. Local deposition rates are affected by a maximum of a factor of three due to this calculation. The calculation is not mass‐conservative, but should provide a first‐order approximation of the relative rates of dust deposition as a combination of large‐scale circulation features represented in the reanalysis dataset and fine‐scale scavenging processes related to regional variation in rainfall rates.

Our calculations estimate the rate of delivery of Sr from dust to the surface of soils, but this will only be equal to the flux of Sr to the bioavailable pool if Sr present in mineral dust is soluble enough to be released before mineral dust is removed by erosion (Kennedy et al. 1998). In the circum‐Caribbean region, chemical weathering and release of Sr form Saharan mineral dust is likely rapid relative to rates of removal of dust by surface erosion because: (1) mineral dust deposits on acidic soils; (2) dust reaching the Caribbean is finely‐grained (∼2 μm) and has a high exchange surface (Prospero et al. 1970); and (3) dust Sr is mostly contained in easily weatherable minerals such as calcite, dolomite, and plagioclase (Glaccum and Prospero 1980, Schütz and Sebert 1987, Kandler et al. 2007, Formenti et al. 2011). To the degree that dust is lost to erosion prior to dissolution of Sr, this will cause our model to overestimate the relative contribution of dust Sr to the bioavailable pool. To estimate F_ss→bio and (⁸⁷Sr/⁸⁶Sr)_ss, we obtained the long term annual deposition of wet and dry sea salt aerosols from a Community Climate System Model 3 (CCSM3) simulation (1.4° × 1.4°) in current climate conditions (Mahowald et al. 2006). This dataset shows good agreement with available satellite observations, and associated details and limitations are discussed in Mahowald et al. (2006). We assigned fixed values for Sr concentration and isotopic composition (0.04% and 0.7092) corresponding to the average abundance of Sr and ⁸⁷Sr/⁸⁶Sr in bulk sea salt. We downscaled the coarse resolution grid (1.4° × 1.4°) to a 30 arc‐second grid using a formulation equivalent to that used for dust deposition: $$$$ where (Sr)_ss is the abundance of Sr in sea salt (in percent) and wetD_ss and dryD_ss are the wet and dry deposition rates of sea salt in t·km⁻²·a⁻¹ as given by CCSM3. Our calculation may overestimate the contribution of sea salt Sr to the bioavailable pool if a significant fraction of deposited Sr is rapidly leached to surface or groundwater, bypassing the bioavailable pool. Although controls on sea salt Sr retention in the bioavailable pool are uncertain, several authors have demonstrated that across a wide range of conditions a significant fraction of the soluble Sr from sea salt is incorporated in the soil exchangeable cation pool and retained in the bioavailable zone (Stewart et al. 1998, Nakano et al. 2001, Stewart et al. 2001, Pozwa et al. 2002), rather than being leached rapidly from soils.

Model evaluation

Bedrock‐only model evaluation

We evaluate the accuracy of the bedrock Sr isotope model using a dataset of ⁸⁷Sr/⁸⁶Sr ratios of igneous rocks from the Caribbean region (n = 920) from the Earthchem Portal (www.earthchem.org; Query by “chemistry”: ⁸⁷Sr/⁸⁶Sr, “Location”: circum‐Caribbean region, “Age” = Age exists:). The parameterized silicate model was applied to predict the ⁸⁷Sr/⁸⁶Sr of samples represented in this database, and the predicted and observed values were compared. Data from 11 samples (1.1% of the samples) were removed from the igneous rock validation dataset. These samples were very old felsic rocks (granites, rhyolites, gneisses) displaying exceptionally high ⁸⁷Sr/⁸⁶Sr, for which we have previously shown the model to perform poorly (Bataille and Bowen 2012). The bedrock model reproduces the pattern of ⁸⁷Sr/⁸⁶Sr variations in bedrock (mod = 0.45obs + 0.39; R² = 0.46) and predicts the absolute ⁸⁷Sr/⁸⁶Sr values of the validation samples with MAE = 0.000249 and RMSE = 0.00113 (where MAE = mean absolute error and RMSE = root mean square error).

We also qualitatively evaluated the spatial patterns of bedrock‐model predicted ⁸⁷Sr/⁸⁶Sr variation against patterns documented by observational studies in the region. At a regional scale, predicted bedrock ⁸⁷Sr/⁸⁶Sr variation is controlled by the signatures of three dominant lithologies, and corresponds well with observations from: (1) tertiary mafic to intermediate volcanic rocks which border the eastern (Antilles) and western (Central America volcanic front) limits of the Caribbean plate, displaying modeled ⁸⁷Sr/⁸⁶Sr from 0.7041 to 0.705 and measured values between 0.703 and 0.708 (van Soest et al. 2002, Vogel et al. 2006); (2) Cretaceous to modern carbonates present either as carbonate blocks such as the Chorotega block, an over‐thickened oceanic crust block (e.g., Yucatan Peninsula) or as marine terraces, with modeled values ranging from 0.707 to 0.7092, similar to reported values (Hodell et al. 2004); and (3) felsic plutonic (modeled values from 0.730 to 0.767) or old metamorphic rocks (modeled values from 0.704 to 0.767) of the Guiana shield, for which the model estimates are also in the range of the observations (www.earthchem.org; Query by “chemistry”: ⁸⁷Sr/⁸⁶Sr, “database”: Georoc results, “location”=northern South America).

At local scales, modeled ⁸⁷Sr/⁸⁶Sr variations are more difficult to validate because of the scarcity of observations. Modeled ⁸⁷Sr/⁸⁶Sr values do correlate well with existing regional geological features driven by: (1) differences in carbonate age, which drive slight ⁸⁷Sr/⁸⁶Sr variations in both Mesoamerica and the Antilles; and (2) small scale geological processes such as metamorphism around the Motagua shear or local plutonism in both the Antilles and Mesoamerica (French et al. 2004). However, as discussed in Bataille and Bowen (2012), the bedrock‐only model gives ‘smoothed' ⁸⁷Sr/⁸⁶Sr predictions and does not account for local geological processes causing ⁸⁷Sr/⁸⁶Sr to vary within lithological units having internal heterogeneity in age or composition. In the circum‐Caribbean region, this limitation affects the accuracy of the prediction for mafic volcanic rocks, which display highly variable ⁸⁷Sr/⁸⁶Sr (0.703–0.708) depending on the time of interaction between the magma and more radiogenic wall rock (van Soest et al. 2002, Vogel et al. 2006).

Weathering model evaluation

We evaluated the performance of the bedrock weathering model, applied using the calibration developed for Japan, by comparing its predictions with F_DSi measurements for our study region. The validation dataset (Table 1) for this area is likely biased towards high F_DSi areas because most of the observed F_DSi values reported in this region are associated with research on extremely high chemical weathering rates in volcanic highlands (Rad et al. 2007, Allegre et al. 2010). For comparison, in 516 catchments of the Japanese Archipelago, the calibrated weathering model explains more than 70% of the F_DSi variance (Hartmann 2009).

Chemical weathering model validation. Modeled and observed (Obs) FDSi and FDSr in the circum‐Caribbean region in t·km−2·a−1.

Several significant factors related to weathering rate are not accounted for in the model, such as topography, land cover, and temperature (Hartmann 2009). We expected the chemical weathering model to underestimate weathering rates and solute fluxes because of both high topographic relief and high mean annual temperatures in the circum‐Caribbean region (White and Blum 1995). However, the presence of thick tropical soils may counteract this effect (Stallard and Edmond 1983). Other important sources of error for F_DSi estimates come from inaccuracies in the correspondence between lithological descriptors of the Caribbean geodatabase and the classes of Jansen et al. (2010; see Appendix: Table A1).

Despite these theoretical limitations, the error in our F_DSi predictions across a range of catchment types spanning an order‐of‐magnitude range in observed F_DSi values does not exceed a factor of two at any site (Table 1). As expected, F_DSi for the young volcanic‐dominated catchment (e.g., Guadeloupe, Martinique) is underestimated. Observed F_DSi values are more closely approximated by the model in catchments dominated by other lithologies. Few data are available to validate the weathering model, calibrated using values from Japan, on the circum‐Caribbean region, but future work by Moodsdorf et al. (personnal communication) to calibrate the model to tropical regions should improve the performance of our model. In the absence of other existing data available to re‐calibrate the weathering model, the F_DSi predictions cannot be assumed to be accurate to within better than a factor of two in the circum‐Caribbean region.

Further uncertainties arise from using the Sr/SiO₂ normalization technique and a Si‐specific weathering model because we do not take into account Sr‐specific dissolution kinetics. However, Hartmann and Moosdorf (2011b) demonstrated good performance using a similar method to estimate the flux of phosphorus to rivers in Japan by re‐scaling results from a silicate weathering model. When compared to the few observed F_DSr measured in this region (Table 2), our predicted F_DSr underestimates the observed F_DSr in all the catchments likely due to presence of trace quantities of non‐siliciclastic minerals within siliciclastic units as suggested by Hartmann and Moosdorf (2011b). The underestimation of F_DSr propagates in our mixing model (Eqs. 3 and 4).

Mixing model validation. Modeled and observed (Obs) contributions of the different sources of Sr to the bioavailable Sr.

Mixing model evaluation

To semi‐quantitatively evaluate the performance of the three source mixing model in representing the relative contributions of Sr from different sources to the bioavailable Sr pool, we compared our model output with data from three studies which quantified the contribution of each source to soil water. For each of these studies, we reported the contribution from bedrock weathering, sea salt, and dust deposition and compared them to the results of our model.

Although numerous assumptions were made to develop the three source mixing model, this simple formulation reproduces the pattern of variation in the relative contribution of different sources of Sr to the bioavailable Sr pool (Table 2). In both catchments of the Osa Peninsula the model matches the observations well, with bedrock weathering being the dominant source of Sr to ecosystems. In the Guyana (analog) catchment, bedrock weathering is slow and atmospheric deposition becomes dominant. In the Luquillos Mountains, our model underestimates the contribution of mineral dust weathering to the bioavailable Sr pool. This underestimation is transmitted from the mineral dust deposition dataset which has been shown to underestimate dust deposition in this watershed (Pett‐Ridge et al. 2009a). Overall, however, the analysis suggests that our Sr fluxes estimates for different sources can be confidently used to give an order of magnitude estimate of each Sr source's contribution to bioavailable Sr. Moreover, in most circum‐Caribbean ecosystems, one source of Sr (either atmospheric or weathering), is largely dominant. In areas where Sr fluxes from the different sources are very different, such as the Guyana and Osa Peninsula sites (Table 1), the mixing model prediction uncertainty decreases whereas the uncertainty increases when the relative contribution of Sr from the different sources are at the same order of magnitude, such as in the Luquillos Mountains (Table 1).

Evaluation of bioavailable Sr isotope models

We evaluated the performance of the different model formulations against published datasets reporting bioavailable Sr isotope measurements. Because different sample materials reflect different spatial scales of integration, we limited our comparison to ⁸⁷Sr/⁸⁶Sr measurements made on sample materials likely to reflect average Sr inputs from local (∼1 km² or less) areas. We thus excluded from the analysis animals with large home ranges and river water, focusing our analysis mainly on data from plants and animals with small home ranges (Fig. 3).

Sample locations for data included in the validation datasets. The Mesoamerica dataset (n = 99) is a compilation of 87Sr/86Sr measurements of modern plants, and modern and archeological animal remains from several sources (Price et al. 2000, Buikstra et al. 2004, Hodell et al. 2004, Wright 2005, Price et al. 2006, Price et al. 2007, White et al. 2007, Price 2008, Price et al. 2010, Wright et al. 2010, Thornton 2011). The Antillean dataset (n = 287) is a compilation of 87Sr/86Sr measurements of modern plants, and modern and archeological animal remains from Laffoon et al. (2012).

Bedrock is a significant source of Sr to ecosystems in the circum‐Caribbean region, as evidenced by the significant correlation between the bedrock‐only modeled ⁸⁷Sr/⁸⁶Sr and observed values in both validation datasets. All three model versions perform similarly well in predicting the observed variation in ⁸⁷Sr/⁸⁶Sr values within the Mesoamerican dataset, explaining more than 80% of the observed variation. Models that include atmospheric deposition (the two and three source mixing models) show similar ⁸⁷Sr/⁸⁶Sr predictability (Fig. 4B, MAE = 0.00031, RMSE = 0.00079 and Fig. 4C, MAE = 0.00040, RMSE = 0.00087, respectively) in comparison with the bedrock‐only model (Fig. 4A, MAE = 0.00011, RMSE = 0.00073). While the correlation between predicted ⁸⁷Sr/⁸⁶Sr and observed ⁸⁷Sr/⁸⁶Sr values is strong, it is dominated by a few individuals, and modeled ⁸⁷Sr/⁸⁶Sr does not represent well the high variability in observed ⁸⁷Sr/⁸⁶Sr (Fig. 4A–C). The modeled ⁸⁷Sr/⁸⁶Sr approximates well the regional ⁸⁷Sr/⁸⁶Sr signature driven by relatively non‐changing factor such as geology and atmospheric deposition but the models do not account for more variable local factors such as Sr recycling influencing the local observed ⁸⁷Sr/⁸⁶Sr signature. The poor resolution of the geological map for Mesoamerica also adds some uncertainty. Samples collected in Mesoamerica are primarily underlain by two lithological types: young mafic volcanic rocks (with ⁸⁷Sr/⁸⁶Sr around 0.703–0.704) or marine carbonates (with ⁸⁷Sr/⁸⁶Sr around 0.707–0.709) which explain the bimodal distribution in the observed ⁸⁷Sr/⁸⁶Sr values in Fig. 4 A–C. However, in the geological map the age of carbonates is often coarsely defined (e.g., “Tertiary”) which leads to uncertainty in all models (Fig. 4A–C) as ⁸⁷Sr/⁸⁶Sr of carbonates varied rapidly throughout geological time (Veizer et al. 1999). Similarly, the level of details of lithological description is not consistent: sometimes the map describes units as “volcanic rocks” which results in modeled ⁸⁷Sr/⁸⁶Sr around 0.705 and some other time it gives “mafic volcanic rocks” resulting in a modeled ⁸⁷Sr/⁸⁶Sr values around 0.703. In reality, most volcanic units of the volcanic front in Mesoamerica are mafic volcanics with ⁸⁷Sr/⁸⁶Sr around 0.703 with an observed bioavailable Sr signature ranging from 0.703 to 0.705. All models slightly overestimate the bioavailable ⁸⁷Sr/⁸⁶Sr signature for these rocks (Fig. 4A–C).

Bioavailable Sr isotope model validation results. Modeled and measured 87Sr/86Sr for the Mesoamerican (A–C) and Antillean (D–F) validation datasets. Circles show results from the bedrock‐only model (A, D), triangles show results from the two source mixing model (sea‐salt, and bedrock weathering; B, E), and squares show results from the three source mixing model (sea‐salt, dust, and bedrock weathering age‐only water model; C, F). Filled symbols represents individuals sampled on silicates dominated areas and open symbols represent individuals sampled on carbonates dominated areas. Linear regression models are calculated on silicates only. Dashed lines show the 1:1 relationship.

In contrast, more of the observed variation in the bioavailable Sr isotope dataset from the Antilles is explained by the two models that include atmospheric deposition (the two source and three source mixing models, each explaining about 50% of the observed variation) than by the bedrock‐only model (R² = 0.09). The models including atmospheric deposition also provide more accurate predictions of the observed values (Fig. 4E, two‐source model MAE = 0.00063 and RMSE = 0.0013, and three source model Fig. 4F, MAE = 0.00014, RMSE = 0.0010) in comparison with the bedrock‐only model (Fig. 4D, MAE = 0.0021, RMSE = 0.0027). The improvement of the MAE and RMSE for the three source mixing model in comparison with the two source mixing model demonstrates that consideration of both types of atmospheric sources are necessary to accurately predict ⁸⁷Sr/⁸⁶Sr in the Antilles. In this region, sea salt deposition is often the dominant atmospheric source in terms of Sr flux, but dust deposition is an important contributor to the bioavailable ⁸⁷Sr/⁸⁶Sr signature because Saharan dust is highly radiogenic.

Spatial pattern of bioavailable ⁸⁷Sr/⁸⁶Sr

Analysis of the relative magnitude of modeled fluxes in the fully integrated, three source model, shows that, although the fluxes vary substantially across the study domain, bedrock is usually the dominant contributor of Sr to circum‐Caribbean ecosystems (Fig. 5A). On a continental scale (1000 km), we observe a general trend of decreasing contribution of atmospheric derived Sr from east to west, which can be attributed to decreasing deposition rates of Saharan dust (Fig. 5A). On a more regional or local scale (i.e., 100 km–1 km), variation in the bedrock contribution is controlled by variation in weathering rates due to differences in both lithology and runoff. Even when atmospheric deposition rates are large, ⁸⁷Sr/⁸⁶Sr of ecosystems developing on highly weatherable Sr‐rich carbonate substrates (e.g., the Maya block or marine deposits in the Antilles and Bahamas) resemble their carbonate parent (Fig. 5B). ⁸⁷Sr/⁸⁶Sr of ecosystems developing on highly weatherable but Sr‐poor volcanic rocks (e.g., Antillean island arc and the Central American volcanic front) are predicted to be more variable at small spatial scales, with Sr isotope ratios and relative contributions of Sr to the bioavailable pool depending on the local interaction of lithology and climate (Fig. 5B). Only ecosystems developing on slowly weathering parent material (e.g., felsic Precambrian rocks of the Guyana shield and Chortis block) show a strong influence of atmospheric deposition (e.g., F(Sr)_w→bio < 0.5).

(A) Contribution of bedrock weathering to the bioavailable Sr pool calculated as F(Sr)w→bio/(F(Sr)w→bio + F(Sr)ss→bio+ F(Sr)d→bio). (B) Modeled Sr isotope ratios for the circum‐Caribbean region from the three source mixing model.

Both types of aerosols, sea salt and dust deposition, contribute significantly to the predicted bioavailable ⁸⁷Sr/⁸⁶Sr in parts of the study region (Fig. 6A,  B). Sea salt deposition is relatively ubiquitous and constant throughout the circum‐Caribbean and contributes to the modeled ⁸⁷Sr/⁸⁶Sr by increasing ⁸⁷Sr/⁸⁶Sr in ecosystems developing on slowly weathering mafic (felsic) rocks in all regions (Fig. 6A). The contribution of mineral dust is more variable spatially, both over large scales (e.g., declining from the east to west across the region) and regionally (e.g., due to variation in precipitation scavenging rates). Because dust Sr in this region is relatively radiogenic, its relative influence on the modeled Sr isotope ratios is greatest in areas of high deposition rate that are also characterized by mafic bedrock, where the differences between bedrock and dust ⁸⁷Sr/⁸⁶Sr are largest (Fig. 6B).

(A) Difference between predicted 87Sr/86Sr from the bedrock‐only model and 87Sr/ 86Sr from the two source mixing model including both sea salt and bedrock weathering. (B) Difference between predicted 87Sr/86Sr from the bedrock‐only model and 87Sr/ 86Sr from the three source mixing model including sea salt, mineral dust, and bedrock weathering fluxes.

To better illustrate the processes considered in the two and three source mixing models, we analyzed the modeled pattern of ⁸⁷Sr/⁸⁶Sr variation at finer scales on the Guadeloupe Islands (inset panels in Fig. 2; Fig. 5A,  B, Fig. 6A,  B). The islands of Guadeloupe are an interesting location to study bioavailable Sr because they present varied geological and climatic conditions. Guadeloupe consists of several different islands with bedrock geology dominated by either Tertiary marine carbonates with ⁸⁷Sr/⁸⁶Sr around 0.7085 or Tertiary intermediate and mafic volcanic rocks with ⁸⁷Sr/⁸⁶Sr ranging from 0.703 to 0.705 (www.earthchem.org). Geomorphology and erosion rates also differ between the islands, with the relatively flat Grande‐Terre Island characterized by quasi constant MAP (2000 mm/yr) and the mountainous Basse‐Terre Island characterized by highly variable temperature and precipitation.

For Grande‐Terre, our three source model suggests that despite high aerosol deposition rates the bioavailable ⁸⁷Sr/⁸⁶Sr is equal to the ⁸⁷Sr/⁸⁶Sr of the carbonate bedrock. The modeled ⁸⁷Sr/⁸⁶Sr values are quite similar to those of sea salt aerosols due to the young age and similar Sr isotopic composition of the local carbonate bedrock, but analysis of the model results shows that bedrock weathering contributes at least 90% of the bioavailable Sr and thus dominates the absolute isotopic value of the bioavailable Sr pool and gives highly invariant values across the island. This result is consistent with observations from older carbonate terrains in the region, where bioavailable Sr isotope values tend to follow bedrock values rather than the sea salt value (Hodell et al. 2004, Laffoon et al. 2012). Despite the relatively homogenous ⁸⁷Sr/⁸⁶Sr of the intermediate volcanic rocks on Basse‐Terre, the modeled bioavailable ⁸⁷Sr/⁸⁶Sr is highly variable and can diverge significantly from bedrock. Bedrock weathering contributes between 50% and 90% of the bioavailable Sr depending mostly on bedrock weathering rates. At the top of La Soufriere (the main volcano on Basse‐Terre, Guadeloupe), modeled weathering rates are high because of high runoff. Despite large deposition rates of atmospheric Sr, bedrock weathering contributes up to 90% of the bioavailable Sr in this area. In contrast, in the lower lands around the volcano, runoff and weathering rates are lower and atmospheric deposition is predicted to contribute significantly to the bioavailable pool. This is also visible on the island of Saba where, despite the presence of a highly weatherable volcanic substrate, the relatively low rainfall amount (<2000 mm·yr⁻¹) does not favor high chemical weathering rates and induces a dominance of atmospheric derived Sr to the bioavailable pool (Laffoon et al. 2012).

Most of the sampling sites (Fig. 2) are characterized by roughly similar tropical humid climates, although some areas, for example the Leeward Antilles (Aruba, Bonaire, Curacao and the Venezuelan archipelago), are characterized by more semi‐arid conditions. In the Leeward Antilles, our three source mixing model predicts a relatively invariant ⁸⁷Sr/⁸⁶Sr value around 0.708–0.709 due to a dominance of sea salt derived Sr in comparison with the other sources. Bedrock contribution to the modeled ⁸⁷Sr/⁸⁶Sr is low because low runoff limits chemical weathering of bedrock. In the Leeward Antilles, in spite of a variable lithology with bedrock ⁸⁷Sr/⁸⁶Sr ranging from 0.703–0.709 (www.earthchem.org), the modeled ⁸⁷Sr/⁸⁶Sr compare well to observations of bioavailable ⁸⁷Sr/⁸⁶Sr which remain relatively invariant and range from 0.7075–0.709 (Laffoon et al. 2012). The good correlation between model and observations for this region was surprising because in semi‐arid conditions soils are usually thin (∼1 m) due to water and wind erosion and average rooting depth is usually deep (McCulley et al. 2004) both of which should favor strong contribution from bedrock derived Sr. However, Capo and Chadwick (1999) demonstrated that in semi‐arid to arid conditions most of the exchangeable Sr originates from atmospheric deposition (either sea salt or dust) because bedrock weathering rates are slow. Geochemical analysis of soils from the Leeward Antilles (Vries 2000) showed an accumulation of sea salt and carbonate mineral dust which tend to confirm the results of our model simulation.

Discussion

Our model shows that across most of the circum‐Caribbean ecosystems, bedrock weathering is the dominant source of Sr to ecosystems.

The contribution of atmospheric deposition is significant in many areas but is rarely dominant. This pattern is somewhat surprising because previous research on Sr cycling in a similar tropical climate in Hawaii showed dominant to exclusive atmospheric contribution to ecosystems when soils are older than 20 ka (Chadwick et al. 2009). While this pattern could be an artifact of the model, it has been observed in other studies in the circum‐Caribbean region (e.g., Bern et al. 2005). Several mechanisms can be advanced to explain the maintenance of bedrock as the main source of Sr to ecosystems in the circum Caribbean region such as: (1) immobilization and preferential recycling of nutrients in upper soils and biomass, which can concentrate bedrock derived Sr in the bioavailable zone (Jobbagy and Jackson 2001, Porder and Chadwick 2009); (2) deposition of locally eroded fresh primary minerals by rivers and landslides (Bern et al. 2005) and/or atmospheric deposition of bedrock‐like Sr such as local dust, biomass, and volcanic ash (Muhs and Budahn 2009); (3) preferential loss of atmospheric Sr relative to bedrock Sr due to processes such as rapid leaching of soluble sea salt Sr and surficial erosion of mineral dust Sr (Porder and Hilley 2011); (4) ecological characteristics of plants that favor uptake of bedrock derived nutrients, such as deep rooting (Jobbagy and Jackson 2001, Pozwa et al. 2002, Pozwa et al. 2004); and (5) geomorphological and hydrological processes coupling stream and soil water (Nakano et al. 2001).

Our results suggest that bedrock dominance is a wide‐spread pattern in the circum‐Caribbean region. Among the mechanisms proposed, immobilization and preferential recycling of nutrients and deposition of locally eroded fresh primary minerals are more likely to affect circum‐Caribbean ecosystems at large scales than are the other factors related to local soil type, plant species, and/or geomorphology. Recycling of bedrock‐derived Sr by ecosystems could maintain bedrock dominance for long periods of time by renewing the stock of fresh primary mineral. However, Porder and Chadwick (2009) showed that at MAP greater than 1,400 mm·a⁻¹ plant recycling of bedrock‐derived Sr was limited. Type and rate of erosion favors the contribution of bedrock‐derived Sr by lowering the bedrock depth to plants, decreasing the residence time of primary minerals, and favoring redeposition of freshly eroded primary minerals. Both mechanisms should be further investigated to understand Sr cycling in the circum‐Caribbean region.

Our two‐ and three source mixing models show relatively good power to predict bioavailable Sr isotope patterns at large spatial scales throughout the circum‐Caribbean region. However, these models are highly simplified representations of Sr cycling within these systems and do not consider a number of processes that can contribute to bioavailable ⁸⁷Sr/⁸⁶Sr. Our models estimate the mixing ratios of soluble Sr in soil as a simple function of climate and lithology specific chemical weathering rates and atmospheric deposition rates. Other variables such as erosion, pedology, geomorphology, hydrological flowpaths, or plant ecology, which have been shown in local studies of Sr systematics, are not explicitly considered in these models. Some of these variables are correlated with others incorporated in our models (e.g., runoff and/or lithology), helping to explain the relatively good performance of the models. Higher precipitation rates, for instance, increase runoff and bedrock weathering rates in our models, and often lead to a dominance of bedrock Sr. In reality, although increasing precipitation increases runoff it can also increase erosion. Both processes (runoff and erosion) increase the bedrock‐derived Sr flux to the bioavailable zone. Significant potential remains for refining our models through explicit incorporation of some of these additional variables, which may help to predict the 30–50% of bioavailable Sr isotope variation remaining unexplained in our analysis. The results of this first analysis, however, suggest that relatively simple multi‐source models can explain a large fraction of regional Sr isotope variation and provide relatively accurate predictions of bioavailable ⁸⁷Sr/⁸⁶Sr in the circum‐Caribbean region.

Implications for Sr isotope provenance applications

Despite significant progress in the mapping of bioavailable ⁸⁷Sr/⁸⁶Sr in many regions, applications that rely exclusively on empirical data are still relatively few in number and generally limited in scale because of the substantial investments of time, energy, and resources required. Integrated approaches that combine empirical and theoretical modeling will be of great benefit to the further development of biosphere ⁸⁷Sr/⁸⁶Sr mapping and to provenance studies more generally, especially if they can provide reliable predictions of bioavailable ⁸⁷Sr/⁸⁶Sr under a broad array of geographic settings and environmental conditions. One potential contribution of these efforts is that models which explicitly consider multiple factors that influence the spatial variation of bioavailable ⁸⁷Sr/⁸⁶Sr can help to guide sample selection strategies. The choice of both appropriate sample materials and methods for provenance studies cannot be independent of a clear understanding of the Sr cycle at the scale studied.

In areas where bedrock is easily weatherable and the dominant source of Sr, ⁸⁷Sr/⁸⁶Sr should remain fairly constant in the different pools of the Sr cycle (e.g., groundwater, soil water, plants and animals). In such cases, the development of bioavailable ⁸⁷Sr/⁸⁶Sr maps through focused sampling of any of these sample materials combined with the bedrock‐only model (Bataille and Bowen 2012) should be sufficient to predict accurately the bioavailable ⁸⁷Sr/⁸⁶Sr. In environments where multiple sources of Sr interact, different sample materials may cycle Sr differently and display highly variable ⁸⁷Sr/⁸⁶Sr ratios. In these areas, defining the different processes interacting in the Sr cycle of local soils is important to correctly interpret the magnitude, pattern and scale of variability of ⁸⁷Sr/⁸⁶Sr and how these relate to the spatial scale of integration represented by different sample materials. This is well illustrated by the contrasting spatial patterns of ⁸⁷Sr/⁸⁶Sr variation from Grande‐Terre and Basse‐Terre, Guadeloupe. Despite the fact that both islands are characterized by relatively uniform geologies, biosphere ⁸⁷Sr/⁸⁶Sr on Grande‐Terre is similar over large spatial scales (essentially the entire island) while biosphere ⁸⁷Sr/⁸⁶Sr on Basse‐Terre is heterogeneous and highly variable at small localized spatial scales. For Grande‐Terre the marine limestone substrate is rich in easily weatherable, Sr‐rich minerals and ⁸⁷Sr/⁸⁶Sr of the local biosphere is dominated by a single isotopically homogenous source of Sr. In contrast, the intermediate volcanic lithology of Basse‐Terre is less weatherable and less rich in Sr and thus local variations in the conditions influencing the proportional contribution of bedrock/soil Sr to local bioavailable Sr budgets also strongly influence the spatial variation of ⁸⁷Sr/⁸⁶Sr. Therefore, a higher sampling density is required to empirically map, or calibrate and validate models for, the relatively localized spatial variation of ⁸⁷Sr/⁸⁶Sr on Basse‐Terre, whereas a reduced sampling density should suffice for Grande‐Terre.

Comparison between bedrock, flux‐weighted catchment water and bioavailable Sr isoscape predictions and observed ⁸⁷Sr/⁸⁶Sr in plants, rocks and river waters (Bataille and Bowen 2012, Chesson et al. 2012) highlights the necessity of developing specific isoscapes for each substrate (e.g., rock, river water, biomass) because each substrate can cycle Sr from different sources. For instance, Chesson et al. (2012) showed large divergence between the local bedrock Sr and tap water ⁸⁷Sr/⁸⁶Sr. This is not surprising because US tap waters generally originate from large rivers or subsurface aquifers which reflect ⁸⁷Sr/⁸⁶Sr from the weathering of large drainage basins and/or subsurface rock units. When substrates sample Sr from local sources or ecosystems, then mapping the bioavailable ⁸⁷Sr/⁸⁶Sr requires explicit considerations of local variables such as presented in this study or in Capo et al. (1998). In contrast, when the dominant source of Sr of the substrate is river water, modeling efforts will be better focused on regional variables such as weathering and catchment hydrology. When both local (bioavailable) and regional (river waters) Sr sources participate in governing the ⁸⁷Sr/⁸⁶Sr values of Sr assimilated by a sample substrate (e.g., humans), a quantification of the contribution from each source may be required.

Conclusions

We present new models and isoscapes for large‐scale patterns of bioavailable ⁸⁷Sr/⁸⁶Sr in the circum‐Caribbean region that include consideration of both bedrock and atmospheric Sr sources. In spite of the relative simplicity of the models and the limitations discussed throughout this paper, this new mapping method demonstrates good predictive power and can contribute to future provenance studies and inform further data collection. Our results suggest that throughout this region bioavailable ⁸⁷Sr/⁸⁶Sr is generally dominated by bedrock Sr, but that in some areas atmospheric deposition is significant and must be considered when interpreting ⁸⁷Sr/⁸⁶Sr datasets for provenance studies.

The following steps are the most critical to continue the development of Sr isoscapes for different substrates: (1) focus on improving, simplifying and validating modeling strategies for three relevant pools of Sr for provenance studies: bedrock, bioavailable Sr, and river water; (2) account for local processes influencing the Sr cycling, such as local Sr recycling through erosion and dust, pedology and surficial deposits; (3) expand modeling in each representative pool to broader geographic coverage using the new lithological world map (Moosdorf et al. 2010). Ultimately a Sr isoscape of the world could be applied to a variety of fields including large scale provenancing studies in ecology and archeology (e.g., Bentley 2006), dust modeling (e.g., Nakai et al. 1993) or refining Sr budget in seawater (e.g., Vance et al. 2009).

Acknowledgments

This research was supported by NSF Award DBI‐0743543. GCM outputs for sea salt and mineral dust were kindly provided by Natalie Mahowald. We thank Nils Moosdorf for advice in applying and interpreting the chemical weathering model to the circum‐Caribbean region. The dataset of bioavailable ⁸⁷Sr/⁸⁶Sr from the Antilles was generated by JEL in collaboration with Gareth Davies of the Department of Petrology, VU, Free University Amsterdam and was financially supported by an NWO‐funded research grant under the supervision of Corinne Hofman and Menno Hoogland of the Faculty of Archaeology, Leiden University, The Netherlands.

Supplemental Material

Appendix A

Table A1. Parameterization of Eqs. 5 and 7 for each lithological descriptor present in the Caribbean geodatabase (French et al. 2004). Sr_ratio is calculated using the analogous descriptor from Bataille and Bowen (2012) and Eq. 1; b and b0 are obtained from analogous descriptor and calibrations recalculated in Jansen et al. (2010); Sr/SiO2 ratios is parameterized using median and standard deviation values from 121,253 analyses available through the Earthchem Portal for each descriptor from Bataille and Bowen (2012; www.earthchem.org; Query by “chemistry”: all Sr AND SiO2, “database”: Georoc results).