Content uploaded by Jens C. Hegg
Author content
All content in this area was uploaded by Jens C. Hegg on Mar 15, 2019
Content may be subject to copyright.
This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/authorsrights
Author's personal copy
Predicting strontium isotope variation and fish location with bedrock
geology: Understanding the effects of geologic heterogeneity
Jens C. Hegg
a,
⁎, Brian P. Kennedy
a,b
, Alexander K. Fremier
a,1
a
Department of Fish and Wildlife Sciences, University of Idaho, Moscow, ID 83843-1136, USA
b
Departments of Geological Sciences and Biological Sciences, University of Idaho, Moscow, ID 83843-1136, USA
abstractarticle info
Article history:
Received 17 April 2013
Received in revised form 9 October 2013
Accepted 10 October 2013
Available online 18 October 2013
Editor: David R. Hilton
Keywords:
Strontium isotopes
Isoscapes
Otolith microchemistry
Migration
Recent advances in using naturally occurring isotopes to reconstruct movement patterns haverevolutionized the
study of migration and spatial patterns across taxa. Isoscape approaches utilize isotopic variation in the
underlying geology to quantify migration pathways. Spatial patterns in the geology can be used to predict
isotopic variation, such as
87
Sr/
86
Sr; however, previous attempts to create predictive models have had mixed
results. Our primary objective was to investigate the relationship between bedrock lithology and
87
Sr/
86
Sr ratio
as a tool to extend the spatial resolution of animal migration studies. Secondly, we investigated the ability to
use geologic prediction as an aprioritool for determining chemically distinct watersheds. We first developed a
regression model to relate known stream water
87
Sr/
86
Sr to rock information from geologic maps, then used
model outputs to classify adult fall Chinook salmon to their juvenile rearing location from
87
Sr/
86
Sr signatures
recorded in their otoliths (ear bones). We discuss the effect of scale and geologic heterogeneity on our ability
to determine
87
Sr/
86
Sr ratios within the study area. Our results indicate that the relationship between
87
Sr/
86
Sr
values and bedrock lithology can be used to accurately determine the rearing location of fish using otolith
87
Sr/
86
Sr signatures. The scale at which geology can be used as a predictor of
87
Sr/
86
Sr values is constrained
by geologic heterogeneity and inherent variability in
87
Sr/
86
Sr ratios within major rock categories. Further,
our results indicate that geological data alone can be used to quantitatively investigate which watersheds are
likely to be distinguishable using this method within a basin. Geologic prediction also has the potential to
improve the scale and resolution of isotopic studies and the development of isoscapes. By applying measures
of spatial heterogeneity we will be better able to quantitatively place limits on the accuracy of geologic
predictions of
87
Sr/
86
Sr ratios.
© 2013 Elsevier B.V. All rights reserved.
1. Introduction
Recent advances in reconstructing location and movement patterns
using naturally occurring isotopes have revolutionized the study of
migration and spatial patterns of habitat use across taxa (Hobson
et al., 2010).Isotopic methods have allowed researchers to link breeding
and overwintering grounds of butterflies (Wassenaar and Hobson,
1998)andbirds(Marra et al., 1998; Wassenaar and Hobson, 2000;
Hobson et al., 2012), to quantify the natal origins and movement
patterns of fish (Harrington et al., 1998; Thorrold et al., 1998;
Kennedy et al., 2002; Hogan et al., 2007), whales and bats (Hobson,
1999), and to identify the forensic location of marijuana origin and
growing conditions (Hurley et al., 2010). Recent investigations have
reconstructed movement patterns in unprecedented temporal and
spatial detail (1–10 km) (Hamann and Kennedy, 2012) and at larger
regional scales (Barnett-Johnson et al., 2010) that would be impossible
using traditional tagging techniques.
The precision and power of landscape isotope, or isoscape,
approaches rely on the underlying isotopic variation in the landscape
(West et al., 2010). The ratio of strontium isotopes (
87
Sr/
86
Sr) can
exhibit fine scale environmental variation in river systems, making
it useful in studying, origin, migration and species distribution at
both large and small scales (Kennedy et al., 1997; West et al., 2009;
Barnett-Johnson et al., 2010; Hamann and Kennedy, 2012; Muhlfeld
et al., 2012). Also, in contrast to most other isotope systems,
87
Sr/
86
Sr
values are tightly linked to the underlying geology (Faure, 1977; Bain
and Bacon, 1994; Stewart et al., 1998). This significant relationship
between bedrock geology and watershed chemistry may allow stream
water
87
Sr/
86
Sr ratios to be directly predicted from geology, potentially
leading to more accurate
87
Sr/
86
Sr isoscapes and increasing the reso-
lution and extent of research with less sampling effort. Lastly, because
biological fractionation of Sr isotopes does not occur, a precise signature
of provenance of organisms or habitat use is possible if water chemistry
can be characterized or predicted (Graustein, 1989; Kennedy et al.,
1997, 2000).
Chemical Geology 360–361 (2013) 89–98
⁎Corresponding author. Tel.: +1 509 592 8867.
E-mail address: hegg1432@vandals.uidaho.edu (J.C. Hegg).
1
Present address: School of the Environment, Washington State University, Pullman
WA 99164-2812, USA.
0009-2541/$ –see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.chemgeo.2013.10.010
Contents lists available at ScienceDirect
Chemical Geology
journal homepage: www.elsevier.com/locate/chemgeo
Author's personal copy
Previous attempts to predict
87
Sr/
86
Sr using geologic variation, both
across the landscape and in surface water, have been met with varying
degrees of success (Chesson et al., 2012). Humston et al. (2006)
introduced a qualitative, a priori tool for researchers to determine the
feasibility of isotopic and elemental studies within a study reach.
Barnett-Johnson et al. (2008) reported that the majority of
87
Sr/
86
Sr
variation in the California Central Valley could be explained using felsic
and old sedimentary rock within the basin but did not extend this
predictive relationship to predict
87
Sr/
86
Sr values within the basin or
determine the location of salmon in the study. Bataille and Bowen
(2012) created large scale isoscapes of
87
Sr/
86
Sr values with landscape
geology, age, and weathering rates which explained 70% of the variation
in surface water
87
Sr/
86
Sr across the United States. They then applied
this model to Caribbean watersheds and extended it to include
additional sources of
87
Sr/
86
Sr (Bataille et al., 2012). While contributing
to our understanding of spatial variation in Sr isotope signature, prior
studies have been hampered by an inability to generalize predictions
and difficulty in predicting across large differences in geologic makeup
or spatial scale.
We hypothesize that understanding how geology varies across
the landscape using metrics of landscape scale and heterogeneity will
improve our ability to generalize isotopic predictions from bedrock
geology in the future. We further hypothesize that geologic predictions
can be used to increase the resolution and extent of migration studies,
and that watershed geologic makeup alone can be used as an apriori
tool to isolate landscapes that are conducive to such studies.
The objective of this paper is to investigate the feasibility of
87
Sr/
86
Sr
prediction from bedrock as a tool to extend the spatial resolution
of animal migration studies. We first present a regression-based
approach that relates surface water
87
Sr/
86
Sr measurements to bedrock
lithology in the Snake River basin of Idaho and Washington. We then
demonstrate that outputs from this model can be used to correctly
classify the location of juvenile salmon. Next, we apply our regression
model to unsampled watersheds within the basin and discuss the effect
of watershed scale and geologic heterogeneity in creating generalized
predictions of
87
Sr/
86
Sr prediction in the future. Finally, we analyze the
geologic differences needed to distinguish watersheds isotopically and
use geologic data alone as an a priori tool to determine the whether
watersheds are likely to be isotopically distinct.
2. Methods
Our primary goal was to develop a statistical relationship between
87
Sr/
86
Sr values within the Snake River basinand the associated bedrock
geology to test whether it could be applied to answer ecological
questions. Secondly, we examined this geologic relationship to deter-
mine under what geologic conditions and watershed scales this type
of geologic modeling could be used to improve isoscape modeling of
87
Sr/
86
Sr ratios.
To develop our geologic relationship we used water samples
collected seasonally to measure dissolved
87
Sr/
86
Sr isotope ratios from
13 sites throughout the Snake River basin by Hegg et al. (2013) as the
dependent variable to create a multiple linear regression. All samples
were analyzed using Thermal Ionization Mass Spectrometry and the
seasonal values for each site were averaged. Refer to Hegg et al.
(2013) for specific analysis methods. The primary rock types in the
watershed upstream of each sample point were quantified and used
as the independent variable (Table 1 in the Online Appendix).
This regression relationship was then applied to a recent dataset of
fish otolith
87
Sr/
86
Sr values to test whether it could be used to extend
the resolution or scale of migration studies. To do this, a linear dis-
criminate function was developed using the predicted outputs of our
regression equation with the geology of watersheds within the Snake
River basin as inputs. We used
87
Sr/
86
Sr signatures of the rearing stage
for 127 adult and juvenile salmon from Hegg et al. (2013) to test the
performance of our regression relationship as a method for determining
the rearing location of fish of fish. We used this fish dataset because
the original classification had been completed using the same water
samples as we used to create our regression relationship. Therefore,
by comparing these two approaches, we can attribute any differences
in classification accuracy to the effects of geologic prediction.
Understanding the scale at which geologic predictions can be made
is important for developing future isoscape models. Therefore, we
analyzed the ability of our regression model to determine
87
Sr/
86
Sr
ratios from geology at various scales. First we applied our original
regression model to three additional, small watersheds within the
basin to determine if the relationship can be generalized to smaller
watershed scales. Next, we created a second regression model that
includes
87
Sr/
86
Sr values for these three additional watersheds to
determine if prediction accuracy can be improved. We then quantified
the effects of watershed scale and heterogeneity on the accuracy of
regression outputs using the percentage area of rock types within
basin watersheds and the Shannon index of diversity and evenness.
Finally, the relationship between geology and
87
Sr/
86
Sr chemistry
offers the possibility that researchers could quantify whether a study
area is amendable to isotopic research before valuable time and
resources are expended in sample analysis. We tested our ability
to use geologic maps directly, without water sampling, to determine
whether watersheds are distinguishable, and thus amendable to
isotopic study. We used watershed geologic data as the independent
variable in a linear discriminate function with major watersheds
as the dependent variable. Our ability to distinguish watersheds using
this method was then compared with the results based on water
sampling. Finally, we analyzed the geologic differences required for
two watersheds to be distinguishable using logistic regression on the
pairwise comparisons of the major Snake River watersheds.
All statistical analyses were conducted in R statistical package
(versions 2.10.1 and 2.15.1, http://www.r-project.org/).
2.1. Study site
The Snake River, the largest tributary to the Columbia River, drains
an area of 280,000 km
2
encompassing six states (Fig. 1). Fall Chinook
salmon, which spawn in the lower reaches of the major tributaries in
the basin, are listed as endangered under the Endangered Species Act
(April 22 1992, Federal Register, Vol 57, No 78, p 14653). Fall Chinook
salmon in the Snake River inhabit a river system that has been
significantly altered by hydropower construction, blocking upstream
access to the majority of fall Chinook salmon spawning grounds and
impounding a large portion of the downstream habitat behind eight
hydropower dams.
The Snake River tributaries can be grouped broadly by geology
(Fig. 1). The Clearwater and Salmon Rivers flow over felsic rocks of the
Idaho batholith, with the Clearwater being influenced most heavily by
the older metamorphic rock (Foster and Fanning, 1997). The Tucannon,
Grande Ronde and Imnaha Rivers flow primarily over the Columbia
River Basalts (Hooper et al., 2007), with the Grande Ronde and Imnaha
Rivers being influenced in their headwaters by the more diverse
Wallowa terrane (Hales et al., 2005). The upper Snake River begins
in the geologically more diverse and older Teton Range (Love et al.,
1978), and the basalt and rhyolites of the Snake River Plain (Leeman,
1982). The unique geologic conditions through which each group of
rivers flows lead to detectable differences in geochemical fingerprints
between rivers in the Columbia River Basin.
Geochemical variationamong basin tributaries has allowedprevious
classification of salmon spawning and rearing areas based upon
geochemical signature (Hegg et al., 2013). Hegg et al. (2013) classified
the main tributaries and Upper and Lower Snake River into four groups
based upon
87
Sr/
86
Sr ratio in stream water using a linear discriminate
function. Otoliths from returning adult fall Chinook salmon were then
analyzed and the
87
Sr/
86
Sr signatures from their natal, rearing and
overwintering stages were recovered using laser ablation inductively
90 J.C. Hegg et al. / Chemical Geology 360–361 (2013) 89–98
Author's personal copy
coupled plasma mass spectrometry. The location of each fish at these
life history points was then determined by classifying the
87
Sr/
86
Sr
signatures using the linear discriminate function developed from
water sampling across major basins.
In the research presented here, we aimed to determine the location
of these fish using geologic predictions of streamwater
87
Sr/
86
Sr ratio
developed from a regression of known
87
Sr/
86
Sr samples and geologic
data, then examine whether the spatial scale or resolution of fish
location could be increased using prediction. Therefore, we used the
stream water
87
Sr/
86
Sr signatures and the juvenile rearing phase of
the above otolith dataset to test our ability to determine fish origin
using predictions of
87
Sr/
86
Sr values from bedrock geology.
2.2. Quantifying bedrock geology
Our geological analyses were based upon current bedrock geologic
maps from the Preliminary Integrated Geologic Map Database of the
United States (Ludington et al., 2005). These maps detail the bedrock
geology of the Snake River basin at 1:500,000 scale with consistent
lithology across the region. The primary rock type within each map
N
Salmon
Palouse
Grande
Ronde
Clearwater
Imnaha
Tucannon
Snake
0240120
Kilometers
Snake basin geology
Mafic
Felsic
Carbonates
Metamorphic
Sedimentary/other
Water
Fall Chinook salmon
spawning range
Montana
Utah
Idaho
California Nevada
Oregon
Washington
Legend
River Courses
Water Sampling Points
Prediction Watersheds
Upper
Snake R.
Salmon R.
Upper
Clearwater R.
Lower
Clearwater R.
Bear Cr.
Big Cr.
Palouse R.
Imnaha R.
Tucannon R.
Grande
Ronde R.
Upper
Snake R.
Lapwai Cr.
A
B
Fig. 1. The location (A) of the Snake River basin and the watershed boundaries and water sample points used to predict
87
Sr/
86
Sr ratios are shown. Watersheds are nested; downstream
watersheds include all upstream watersheds. Lithology of the Snake River watershed (B) shows rock type categories with strong impacts on
87
Sr/
86
Sr ratio based primarily on protolithic
composition (Ludington et al., 2005; Stoeser et al., 2006).
91J.C. Hegg et al. / Chemical Geology 360–361 (2013) 89–98
Author's personal copy
polygon is recorded in the attribute table of the map layer according to
LithClass 6.1 standard (http://www.nadm-geo.org).
While quantifying geologic variability within our study area we
focused our analyses on percent abundance of rock types within
a basin rather than rock age. Rock age has an effect on
87
Sr/
86
Sr due
to the evolution of radiogenic
87
Sr from
87
Rb over time (half life
of 48.8 billion years); however, quantifying this relationship is
complicated for two reasons. First, the covariation of rock age with
rock type within the Snake River basin would not allow us to distinguish
causality. For example, the Idaho Batholith, a granitic structure that
wouldbeexpectedtohavehigh
87
Sr/
86
Sr, is also generally older
and associated with extremely old, high
87
Sr/
86
Sr, metasedimentary
deposits, making it difficult to determine the relative contribution of
age and rock type in the
87
Sr/
86
Sr values. Conversely, the Columbia
River basalts are relatively young, but would also be expected to have
low
87
Sr/
86
Sr based upon their mafic composition. Secondly, acquiring
reliable and comparable geologic ages for watersheds across the large
Snake River basin proved challenging. There can be large variation in
the range of ages reported for a given rock type polygon and rock age
varies in its precision between different map units. The addition of age
did not significantly improve our predictive ability, likely due to the
factors mentioned above, and therefore we only used geologic type in
this analysis.
Our aim in reclassifying rock types was to capture the geologic
variation relevant to
87
Sr/
86
Sr differences in stream water. Our rock
type classification was primarily based upon distinguishing between
mafic and felsic rock types since these broad rock types typically display
different strontium isotopic chemistry based upon the composition of
their magma source (Faure and Mensing, 2004). Thus, all igneous
rocks were classified by either maficorfelsicrocktype.Wedistinguished
all non-igneous rocks by their protolithic rock type when the protolith
was obvious from the map entry. Many non-igneous rock types have
no obvious protolithic composition. When no obvious protolith could
be determined the rock types were then classified into categories of
metamorphic, carbonate, or sedimentary/other.
The potentially high rates of weathering in carbonate rocks can have
large impacts on
87
Sr/
86
Sr ratios within a watershed, so carbonate rocks
weregiventheirownclassification (Blum et al., 1998). We designated
a metamorphic rock category based on prior knowledge that high
87
Sr/
86
Sr values might be present in very old metamorphic systems
associated with the Idaho batholith (Martignole et al., 2010; Jansen,
2011). The sedimentary/other classification included rocks and uncon-
solidated sedimentary deposits of indeterminate origin as well as
chemical and biogenic deposits other than carbonates. The details of
rock reclassification are listed in Table 2 of the online Appendix.
2.3. Surface area calculation
Surface area may be important when calculating the relative
contributions of geologic types. Differential weathering and the
propensity for more resistant rock types to form vertical faces may
bias the flat map area toward the least resistant rock types. We
used 3-dimensional (3D) surface area alongside the conventional
2-dimensional (2D) map area to determine whether 3D area might
be a better predictor of
87
Sr/
86
Sr.
Three-dimensional surface area was calculated for the entire Snake
River basin using 90meter elevation rasters available from the National
Elevation Dataset (Gesch et al., 2009) and the DEM surface tools
package from Jenness Enterprises (Jenness, 2012). The total 2D and 3D
surface area of each rock type within each sample watershed was then
calculated using the Zonal Statistics tool available in the Spatial Analyst
toolbox in ArcMap. Thus, two geologic datasets were produced, one
recording the flat 2D area of each rock type within the basin and another
which recorded the calculated 3D surface area.
The watershed above each water sample point was delineated
using the ArcHydro extension for ArcMap 9.3 (ESRI). In ArcMap, the
intersection between the watershed boundaries and the geologic map
layer was used to calculate the 2D and 3D area of each rock type within
the watershed above each sample point. The percent area of each rock
type for both 2D and 3D maps was then calculated from the attribute
table of the intersected map.
2.4. Model selection and prediction of
87
Sr/
86
Sr from geology
Multiple linear regression was used to develop a relationship
between
87
Sr/
86
Sr watershed geology within the spawning area of
Snake River Fall Chinook salmon. To develop this regression, percent
geologic area was the independent variable, while the
87
Sr/
86
Sr values
from 13 watersheds in the Snake River from Hegg et al. (2013)
were used as the dependent variable. Percent rock area followed an
exponential relationship with
87
Sr/
86
Sr and was transformed using
log(x+ 1) to meet the assumption of linearity and account for real
zero values within the data (Bartlett, 1947).
The most likely candidate models were compared (Table 3 online
Appendix) using AICc, a version of Akaike's Information Criterion modi-
fied for small datasets (Burnham and Anderson, 2002). This statistical
technique selects the model that best explains variation in the data
while rewarding parsimony by penalizing over-parameterization.
Candidate models were constructed using rock types that were well
represented in the basin, and rock types that were expected to have
asignificant influence on dissolved
87
Sr/
86
Sr values within the study
area. Based upon the large areas of mafic and felsic rock within the
basin these rock types were considered good candidate variables.
Similarly, the age and high
87
Sr/
86
Sr composition of metamorphic rock
in the study area indicated that it might be an important variable.
All other rock types were considered to have a representation that
was either of minor significance or too variable in composition to be
effective predictors.
Our ultimate aim was to re-classify the fish from (Hegg et al., 2013)
using a linear discriminate function and training set based on
87
Sr/
86
Sr
predictions from bedrock geology. To do this, we digitized prediction
points at 10 river-kilometer (10-rkm) intervals (Fig. 3) within the
spawning area of Fall Chinook based on spawning site surveys (Garcia
et al., 2008). The watershed upstream of each prediction point was
then delineated using ArcHydro and the percent area of each rock
type was calculated. The
87
Sr/
86
Sr value for each point was predicted
by applying our regression equation to the geologic data at each point.
Prediction points were then grouped according to the four river
groupings (Table 1)fromHegg et al. (2013), and a linear discriminate
function was created using these points as training sets.
2.5. Rearing origins classification from otoliths
We used the regression relationship developed in the previous
section to test whether geologic predictions can be used to extend
ecological studies using
87
Sr/
86
Sr as a tracer. We used results from our
regression to classify 127 salmon otoliths to rearing location based on
the
87
Sr/
86
Sr signature of their otoliths. We then compared these
classifications to prior classifications from Hegg et al. (2013) who used
stream water
87
Sr/
86
Sr samples as a training set. Fish were classified to
rearing location using the linear discriminate function we developed
previously using geologic predictions as the training set. We then
calculated the Kappa statistic (Fleiss et al., 2004) to compare the
classification accuracy using geologic prediction to those of Hegg et al.
(2013).
Hegg et al. (2013) estimated an 86% classification accuracy based on a
single misclassification of six known-origin fish, with a cross validation
error rate of 0% for the original training set. The regression equation
used to create geologic predictions is based on the
87
Sr/
86
Sr values
from Hegg et al. (2013) and is thus not completely independent.
Therefore, we would expect that any classification error when comparing
92 J.C. Hegg et al. / Chemical Geology 360–361 (2013) 89–98
Author's personal copy
our fish geologic regression based classifications to those of Hegg et al. is
due to the error in predicting
87
Sr/
86
Sr from geology.
2.6. Effects of scale and heterogeneity
To test the effect of watershed scale on our ability to predict
87
Sr/
86
Sr
ratios, we applied our regression model from Section 2.4 to three smaller
basins within the Snake River, each with varying geologic makeup
(Table 1, Online Appendix). These watersheds were chosen as a test of
scale due to their smaller size. They were not included in the original
geologic regression because they are outside the spawning range of
Snake River Fall Chinook salmon. They were also chosen due to the
existence of
87
Sr/
86
Sr values for their tributaries, allowing additional
analysis of the relationship of
87
Sr/
86
Sr to geologyat these smaller scales.
GPS locations and
87
Sr/
86
Sr values for 19 sites within the Big Creek
watershed, a tributary of the Middle Fork Salmon River, were obtained
from Hamann and Kennedy (2012) (Fig. 1, Online Appendix). Six
87
Sr/
86
Sr values and GPS locations were collected from Lapwai Creek,
a tributary to the Clearwater River (Fig. 2, Online Appendix). Twelve
87
Sr/
86
Sr water samples and GPS coordinates were also used from Bear
Valley Creek, in the headwaters of the Middle Fork Salmon River, as
part of a prior feasibility study in cooperation with the US Forest Service
(Kennedy, unpublished data) (Fig. 3, Online Appendix). All
87
Sr/
86
Sr
values were determined using Thermal Ionization Mass Spectrometry
(TIMS) except Lapwai Creek, which was analyzed using Multi-Collector
Inductively Coupled Plasma Mass spectrometry (MC-ICPMS).
The geologic makeup of these additional watersheds was deter-
mined using the same techniques as for the larger Snake River basin
in the preceding sections. Then,
87
Sr/
86
Sr ratios were predicted using
the regression model created in Section 2.4 to test the extent to which
this regression was scale independent. We then fit a second regression
model to the combined
87
Sr/
86
Sr data for all water samples (Snake
River, Lapwai Creek, Bear Valley Creek and Big Creek) to test whether
a predictive model could be created which encompassed a larger
range of watershed scales. Candidate models using the combined data
were constructed using the same independent variable combinations
of 2D mafic, felsic and metamorphic rock and selected using AICc.
To analyze how the varying representation rock types within
watersheds affected our ability to predict
87
Sr/
86
Sr ratios, we calculated
the Shannon index of diversity for rock types within each sampled
watershed (Shannon and Weaver, 1949). In this case the Shannon
diversity index (H′) is used to describe the variability of rock types
present within a watershed, in contrast to its more familiar application
to biodiversity.
H′¼−X
s
i¼1
pilnpi
ðÞ ð1Þ
H′takes into account the relative abundance (Pi)ofanindividual
rock type and the number of rock types (S) to calculate the relative
evenness and diversity of rock types within a basin. The metric varies
between 0 and ln(S) with low values indicating low diversity and
uneven representation of rock types.
2.7. Exploring a priori prediction from geology
We tested the ability of geology alone to be used as an a priori
classifier of location, without first predicting
87
Sr/
86
Sr values using
linear regression. This was done with the intent of providing a
quantitative method for determining whether
87
Sr/
86
Sr ratio may be
useful as a tracer, before
87
Sr/
86
Sr values are known within the system
or before a research project is undertaken. We used the percent area
of mafic and metamorphic rock types from the watershed upstream of
each 10-rkm point predictions as the explanatory variable to create a
linear discriminate function. We used the four chemically distinct
river groups from Hegg et al. (2013) as the classification variable in
this discriminate function. In this way we were able to test the ability
of geologic data alone to determine which watersheds are likely to be
distinguishable within a basin. We then calculated the cross-validation
error rate to determine whether geology alone could distinguish rivers
within the basin.
We examined the change in
87
Sr/
86
Sr ratio for a given amount of
change in geology within a watershed, and the probability that two
watersheds can be distinguished by
87
Sr/
86
Sr ratio given the difference
in their geologic makeup. To do this we first calculated the pairwise
differences in
87
Sr/
86
Sr values, geologic diversity, mafic, metamorphic
and felsic rock for all combinations of the watersheds used in the original
13 watersheds from Hegg et al. (2013). Then the difference in
87
Sr/
86
Sr
value was regressed against each geologic variable to understand
thechangein
87
Sr/
86
Sr value for a given change in
87
Sr/
86
Sr. Finally,
each pairwise watershed comparison was assigned as distinguishable
or indistinguishable based on whether the watersheds were part of the
same or distinguishable river groups in Hegg et al. (2013).Wethen
used logistic regression to determine the probability that two watersheds
are distinguishable given their difference in each geologic variable. Error
was estimated for both linear and logistic regression as a 95% confidence
interval using the standard error output of the regression.
3. Results
3.1. Model selection for prediction of
87
Sr/
86
Sr from geology
Model fit for multiple linear regression of the Snake River water
samples and bedrock geology was assessed using AICc weights
(Table 3, online appendix). All models were significant (p N0.05, α=
0.05) and accounted for greater than 70% of the variation in the data.
Models containing predictors calculated using 3D area outperformed
the same models using 2D area. The model with the highest AICc weight
was constructed using 3D area,
87Sr=86 Sr ¼0:0710050−0:005571•log %Mafic þ1ðÞ
þ0:015923 •log %Metamorphic þ1ðÞþεð2Þ
followed closely by the same model,
87Sr=86 Sr ¼0:0710131−0:005723•log %Mafic þ1ðÞ
þ0:015871•log %Metamorphic þ1ðÞþεð3Þ
Table 1
Comparison of classificationaccuracy to classify fish to theirjuvenile rearing location based upon
87
Sr/
86
Sr signatures in their otoliths.Linear discriminatefunctions were developed using
training sets of
87
Sr/
86
Sr ratios predicted from bedrock geology using the regression equation in Section 2.4 (rows) and from water samples (columns) collected by Hegg et al. (2013).
Classification using water samples
(Hegg et al., 2013)
Tucannon, Grande Ronde,
Imnaha Rivers
Clearwater
River
Lower Snake
River
Upper Snake
River
User accuracy
Classification using geologic regression Tucannon, Grande Ronde, Imnaha Rivers 1 0 0 0 100%
Clearwater River 0 37 1 0 97%
Lower Snake River 0 3 83 0 97%
Upper Snake River 0 0 0 2 100%
Producer accuracy 100% 93% 99% 100%
93J.C. Hegg et al. / Chemical Geology 360–361 (2013) 89–98
Author's personal copy
using 2D area. Both of the top models explained 97% of the variation in
the data. While the 3D models had higher AICc scores, the difference
in prediction accuracy between 2D and 3D was less than 1%. Since our
goal was to create a relationship between
87
Sr/
86
Sr and geology which
can be easily utilized by researchers, this slight increase in accuracy
did not justify the extra calculations involved in determining 3D area.
Therefore, we used the 2D model for subsequent analyses. The results
of geologic reclassification and watershed delineation are described in
Table 1 of the online Appendix.
3.2. Rearing location classification from otoliths
Developing a linear discriminate function using
87
Sr/
86
Sr calculated
at 10-rkm intervals using the geologic regression from Section 2.4
resulted in a 3% classification error rate (3 misclassified out of 93 total
prediction points) using leave-one-out cross validation. This same
discriminate function, when used to classify fish, classified 97% (124 of
127) of fish to the same rearing location as Hegg et al. (2013) with a
Cohen's Kappa of 0.93 (st. error = 0.03, C.I. = 0.86–1) indicating high
agreement between methods when accounting for random error
(Table 1,Fig. 2). Proportions of fish classified to each source group are
not significantly different between methods (Monte Carlo Chi-Square,
α= 0.05) indicating that predictions of
87
Sr/
86
Sr are capable of
predicting fall Chinook natal origins.
The three misclassified fish had a signature intermediate between
the Clearwater and Lower Snake River, similar to the single known-
origin juvenile misclassified in Hegg et al. (2013). This known-origin
juvenile originated in the Clearwater, but was captured in the Lower
Snake River during the rearing period. Hegg et al. (2013) speculated
that this misclassification may have been due to incomplete equili-
bration of the otolith signature to the Lower Snake environment,
potentially due to the rate at which juveniles outmigrate from their
natal streams.
3.3. Effects of scale and heterogeneity
Geologic diversity, as a function of both the representation and
evenness of the lithology within a watershed, was greatest for the
Snake River watershed. As expected, the largest watersheds displayed
the highest overall diversity of rock types and geologic diversity
decreased with watershed size. The three small watersheds used to
test the effects of scale had the lowest geologic diversity (Fig. 3A).
Error increased significantly when predicting
87
Sr/
86
Sr within the
three smallest watersheds using the regression developed in Section 2.4
relating
87
Sr/
86
Sr from water samples to watershed geology. This is
expected when extrapolating to finer scales than the data originally
used to create the regression. While residuals for the Snake River
watershed samples were quite small and centered on zero (mean =
6.9 × 10
8
± 0.000457 1-st. dev.), the combined residuals for the three
small watersheds were much higher (Mean = −0.001439 ± 0.002259
1-st. dev.) and tended to be positively skewed, indicating that the
relationship of
87
Sr/
86
Sr to geologic makeup does not follow the same
relationship at smaller scales. Of the small watersheds, Lapwai Creek
best fit the model, followed by Big Creek and Bear Valley Creek.
When dissolved
87
Sr/
86
Sr samples and geologic results from the
Snake River, Big Creek, Lapwai Creek and Bear Valley Creek were
combined to create a comprehensive linear regression, the model with
the highest AICc weight included mafic and felsic rock as independent
variables (Table 4, online Appendix). The model was significant
(p b0.0001, α= 0.05), however variance explained by the model
(R
2
= 0.64) was less than the original regression model.
3.4. Exploring a priori prediction from geology
The discriminate functiondeveloped using percent area of maficand
metamorphic geology as explanatory variables, and river groups as the
dependent variable, resulted in a 3% error rate using leave-one-out
cross validation. This error rate was identical to the discriminate
function from Section 2.4 which was based upon the regression
relationship between
87
Sr/
86
Sr and watershed geology.
Analysis of pairwise differences between watersheds showed that
increasing differences in watershed geology could be used to determine
whether two watersheds were likely to be distinguishable. Regression
of the difference in
87
Sr/
86
Sr for a given difference in watershed geology
showed a significant positive relationship between
87
Sr/
86
Sr and
the difference in mafic, felsic, and metamorphic geology between
watersheds (Fig. 4A). Geologic diversity, as measured by Shannon
diversity, also showed a significant positive relationship (Fig. 4B).
Rearing Classification
From Water Samples
Rearing Classification
From Geologic Prediction
0.706
0.708
0.710
0.712
0.714
Tucannon
G. Ronde
Imnaha
Clearwater
Salmon
Lower
Snake
Upper
Snake
0.706
0.708
0.710
0.712
0.714
Clearwater
Salmon
Lower
Snake
Upper
Snake
Tucannon
G. Ronde
Imnaha
Classification
Clearwater,
Salmon
Lower-
Snake
Upper-
Snake
Tucannon,
Grande Ronde,
Imnaha
AB
87Sr / 86Sr
Misclassified
Misclassified
Fig. 2. Classification of adult fish to their juvenile rearing location based on
87
Sr/
86
Sr signatures in their otoliths is shown. Classification using a linear discriminate function with water
samples as the training set ( A) from Hegg et al. (2013) resulted in one misclassified fish, a juvenile of known origin. Classification using a linear discriminate function based on
87
Sr/
86
Sr
ratios calculated from geology using the regression equation in Section 2.4 (B) misclassified four fish to the Clearwater group, including the fish originally misclassified in Hegg et al.
(2013). Points are coded by color and shape according to statistical classification to river groups. Misclassified fish appear as a different color and shape from the classification column.
Lighter colored points indicate juvenile fish of known origin. Points are jittered on the x-axis to avoid overplotting.
94 J.C. Hegg et al. / Chemical Geology 360–361 (2013) 89–98
Author's personal copy
Logistic regression of the pairwise differences between watersheds
showed a significant relationship between difference in watershed
geology and the probability that two watersheds were distinguishable
using
87
Sr/
86
Sr (Fig. 5A). Geologic diversity also showed a significant
relationship with the probability that watersheds are distinguishable
(Fig. 5B). Results from the pairwise comparisons of watersheds are
shown in Table 2.
4. Discussion
Naturally occurring chemical and isotopic tracers are proving to
be useful and powerful methods for studying ecological questions at
multiple scales (Harrington et al., 1998; Hobson, 1999; Kennedy et al.,
2000; Hamann and Kennedy, 2012). Strontium isotope ratios can
provide a powerful tracer of animal movement at varying scales, but
baseline isoscapes are not well developed, hindering its widespread
use (Hobson et al., 2010). Isoscapes for other isotope systems such as
hydrogen and oxygen, which are based on large scale precipitation
trends, are well developed but useful only at regional and continental
scales (Bowen, 2010). Because
87
Sr/
86
Sr can be used to track animal
movement at local, regional and continental scales,
87
Sr/
86
Sr isoscapes
can be used link ecological processes across scales, a major challenge
of traditional ecological studies (Hobbs, 2003). Much work is still
needed to create a generalizable model of
87
Sr/
86
Sr variation that
accounts for a majority of
87
Sr/
86
Sr variation across a broad range of
geologies and scales. Our method for classifying bedrock geology by
mafic or felsic protolith may help to standardize future models by
providing a repeatable method that accounts for the major geologic
drivers of
87
Sr/
86
Sr variation across geologic settings.
This study shows that the relationship between
87
Sr/
86
Sr and
bedrock geology can be used to accurately reconstruct the location of
aquatic organisms and extend the resolution of strontium isotopic
studies beyond baselinewater sampling points. This work also provides
a relatively simple method by which investigators can determine a
priori the feasibility of
87
Sr/
86
Sr ratio as an investigative tool within
a given population or watershed. By creating a linear discriminate
function with the important reaches as the grouping variable and
the mafic and felsic geology as the explanatory variables, the cross
validation error rate can be used to give an indication of how well
87
Sr/
86
Sr might distinguish between important reaches. Thus, it is
possible to quantify the degree to which unsampled watersheds are
likely to be isotopically distinct using geology alone. Further, analysis
of the differences between watersheds provides relationships between
geology and
87
Sr/
86
Sr, which may be useful to parameterize future
predictive models of
87
Sr/
86
Sr variation across the landscape.
4.1. Effects of scale and heterogeneity
The regression model between geologic and strontium ratio was
constructed with data from a regional watershed (Snake River Basin)
and performed well at that scale. However, our results indicate that
the ability to predict
87
Sr/
86
Sr from bedrock geology decreases when
extrapolated to watersheds below 1000 km
2
. Two factors may play
a role in this loss of predictive ability. Further, creating a second
regression model incorporating
87
Sr/
86
Sr water samples from these
smaller watersheds did not improve the predictive ability at smaller
scales.
One factor may be that rock type classifications are broad, and
can potentially obscure large underlying variation in signature. Any
generalizable model predicting
87
Sr/
86
Sr from bedrock must create
broad classifications of rock types, effectively forcing rocks of variable,
but similar,
87
Sr/
86
Sr isotope chemistry to be modeled as a unit. This
variation in
87
Sr/
86
Sr values within a rock type classification can be
significant, particularly for felsic rocks whose geochemical makeup,
age and origin can lead to large variations in
87
Sr/
86
Sr isotope signature
(Faure and Mensing, 2004).
Second, as the size of watersheds decreases, they are increasingly
dominated by single rock units rather than an average of upstream
inputs (Fig. 3A), exposing the variation in
87
Sr/
86
Sr within the felsic
rock type classification. At coarse scales the model can make accurate
predictions based upon broad rock types because each river is an
average of the diverse signatures of the upstream watershed. Smaller
watersheds however (Fig. 6A), and typically dominated by a single
rock type (Fig. 6B), and therefore prediction accuracy decreases.
Further, the more a watershed is dominated by felsic rock the lower
the prediction accuracy of the model (Fig. 6C). This is expected since
individual felsic rock units should vary widely from the mean
87
Sr/
86
Sr
of the basin based on their age and chemical makeup. Thus, as scale
decreases, local variations in
87
Sr/
86
Sr values become increasingly
important for accurate prediction, and rock types containing large
Watershed
Snake
River
Big
Creek
Bear Valley
Creek
Lapwai
Creek
% of Dominant Rock Type
40%
50%
60%
70%
80%
90%
Shannon Diversity
0.2
0.4
0.6
0.8
1.0
1.2
Watershed Area (km2)
102103104105
A
B
Fig. 3. (A) Geologic diversity increases with watershed scale. (B) The percentage area of
the dominant rock type within a watershed decreases with watershed scale. These
relationships illustrate that individual rock types begin to dominate as scale decreases,
making the unique
87
Sr/
86
Sr ratio of a given geologic formation more important at smaller
scales. Dotted lines are fitted linear models to indicate the direction of the trend.
95J.C. Hegg et al. / Chemical Geology 360–361 (2013) 89–98
Author's personal copy
variation in
87
Sr/
86
Sr signatures begin to be particularly poorly
predicted. Therefore, we would expect that generalized models would
perform best at the largest scales, where stream water
87
Sr/
86
Sr
signatures are an average of many rock units throughout a basin and
effectively integrate the overall variation within each rock type
classification. It is possible, however, that at smaller scales the difference
in a given rock type could provide an indication that watersheds are
distinguishable, even when the absolute
87
Sr/
86
Sr cannot be predicted.
Results from our analysis of pairwise differences between watersheds
give an indication of how this could be determined.
The scale and geologic makeup at which predictions of
87
Sr/
86
Sr
ratios can be made from bedrock will be different for every watershed.
Our study indicates, however, that in order to effectively apply
predictions from bedrock geology or produce generalizable isoscape
models, it is important to constrain the scale and geologic heterogeneity
at which predictions are made. Measures of heterogeneity such as
Shannon diversity and the percentage area of the dominant rock type
are independent of scale and geologic makeup, and can be applied to
any watershed. These metrics couldpotentially be used to quantitatively
determine, based on the geologic heterogeneity of the watershed, a
lower spatial threshold above which predictions are appropriate. In
the case of the Snake River, with more than 55% of the watershed
dominated by a single rock type and geologic Shannon diversity below
1.0 (H′), prediction accuracy begins to decrease substantially (Fig. 6).
Spatially this corresponds to watersheds of 1000 km
2
or smaller. Further
work is needed to understand if heterogeneity measures can be
generalized across differing watersheds with differing geologies, and
to develop quantitative methods for applying these metrics across
watersheds.
4.2. Exploring a priori prediction from geology
One difficulty for researchers planning to use
87
Sr/
86
Sr isotope
ratios as a tracer is determining whether a study system contains
enough isotopic variation between the important watersheds to
provide meaningful results. Our results indicate that it is possible
to use geologic variation as inputs to a linear discriminate function
to determine a priori whether important watersheds are likely to
be isotopically distinct. This provides a feasible a priori method
that requires minimal time input or prior knowledge. The data
requirements are minimal, only geology and watershed layers are
needed. Using GIS, a researcher can then reclassify the geologic map,
define the relevant watersheds, and intersect the layers. Calculating
the percentage area of each rock type is then a simple calculation.
Using the cross-validation error rate of a discriminate function to
determine likely isotopic differentiation is not, however, a definitive
test. It assumes that the major rock types within a study area are
isotopically distinct but it is possible for rocks to be isotopically similar
regardless of their rock type classification, which would confound the
results. Therefore a more mechanistic understanding of the isotopic
Logistic Regression
% Felsic Rock
% Mafic Rock
% Metamorphic Rock
Shannon Diversity
0 0.25 0.50 0.75 1.0
B
Difference in Shannon Diversity
0.00
0.25
0.50
0.75
1.00
0% 10% 20% 30% 40% 50%
Difference in % Rock Type
Probability of Distinguishable
Watersheds
A
Fig. 5. As the difference in geologic makeup between two watersheds increases, the probability that they can be distinguished using
87
Sr/
86
Sr values increases (A). Small differences in
metamorphic rock in particular increase the probability that watersheds are isotopically distinct, as do mafic and felsic geology. The difference in geologic heterogeneity between two
watersheds, as measured usingShannon diversity,is also a good predictor of thatwatersheds are isotopically distinct (B). Shaded area is the95% confidence intervalaround the fitted line.
Regression
% Felsic Rock
% Mafic Rock
% Metamorphic Rock
Shannon Diversity
0 0.25 0.50 0.75 1.0
Difference in Shannon Diversity
B
A
0.000
0.002
0.004
0.006
0.008
0% 25% 50% 75%
Difference in % Rock Type
Difference in 87Sr / 86 Sr
Fig. 4. The pairwise difference in
87
Sr/
86
Sr between watersheds of the Snake River varies predictably with the difference in geologic makeup. The absolute difference in
87
Sr/
86
Sr ratio
between two watersheds shows a positive relationship with the difference in mafic, felsic and metamorphic makeup of the watersheds (A). The absolute difference in
87
Sr/
86
Sr ratio
betweenwatersheds alsofollows a positive relationship with the absolutedifferencein geologic diversityas measured by Shannondiversity (B). Shadedarea is the 95% confidence interval
around the fitted line.
96 J.C. Hegg et al. / Chemical Geology 360–361 (2013) 89–98
Author's personal copy
variation within rock types is needed to provide a truly predictive a
priori method.
Our regression results provide a more quantitative method of
determining whether watersheds might be distinct. In the Snake
River, watersheds with a difference of 35–40% of maficorfelsicrock
type have a 95% probability of being isotopically distinct (Fig. 5A). Linear
regression shows a positive relationship between the difference in
percentage of rock type and the change expected in
87
Sr/
86
Sr values,
which gives a measure of the difference in
87
Sr/
86
Sr ratios between
distinguishable watersheds (Fig. 4A).
The logistic regression of metamorphic rock type provides insight
into how a single, relatively rare, geology can exert an important
impact on watershed differentiability. A difference of only 10% in meta-
morphic rock within a watershed provides an 85% probability that
the watersheds are distinguishable and only 15% difference results in a
95% probability (Fig. 5A). This is an important consideration when
creating future generalized models, as some rock types may exert
a larger influence than their representation on the landscape would
indicate.
Interestingly, our results show that a metric of geologic heteroge-
neity can be used to relate watershed geology to variation in
87
Sr/
86
Sr
isotopic values. A difference in Shannon diversity of 0.56 indicates a
95% probability that two watersheds are distinguishable and the linear
regression shows a significant positive relationship between diversity
and
87
Sr/
86
Sr ratios (Figs. 4Band5B). This provides additional evidence
that metrics of geologic diversity may provide a means to improve
87
Sr/
86
Sr isoscape methods.
This study shows that the relationship between
87
Sr/
86
Sr and
bedrock geology has the potential to be a useful ecological tool for the
study of migration and distribution of organisms across a landscape.
Though more work is needed to generalize these results outside the
Snake River, they provide some insight into what might be expected
in similar watersheds. Applying these methods to additional watersheds
may uncover patterns in geologic heterogeneity which could place
bounds on the scale and predictive ability of future models. Additionally,
these analyses provide a starting point for understanding the isotopic
differences we expect for a given change in geologic makeup, which
could be used to parameterize future isoscape models.
Our study highlights two areas in which methods must improve
for
87
Sr/
86
Sr isoscape predictions to be generalizable. First, geologic
age should be included as a variable in future models in order to
explain smaller scale variation in
87
Sr/
86
Sr, especially within felsic
rocks. Second, geologic heterogeneity should be explored as a method
to guide future modeling efforts, particularly as a tool to determine
the lower threshold of prediction accuracy. Past efforts at determining
stream water
87
Sr/
86
Sr values using geology (Barnett-Johnson et al.,
2008; Bataille and Bowen, 2012; Bataille et al., 2012; Chesson et al.,
2012) have focused on intrinsic characteristics of rock types. This
study shows that this approach is valid, but limited by our under-
standing of how
87
Sr/
86
Sr signatures vary in relation to the distribution
of geologic features within watersheds and at varying scales. By
applying tools of landscape ecology to understand how the spatial
occurrence of geologic features affects our ability to predict
87
Sr/
86
Sr
ratios we may be able to improve predictions and better constrain the
spatial limits of future models.
Acknowledgments
Thanks to members of the CIFEES lab at University of Idaho,
including E. Hamann, J. Reader and S. Bourret for help in otolith
collection, analysis, and method development. Thank you to R. Lewis
with the Idaho Geological Survey for discussions of Snake River geology
and critique of the rock type classifications. Thanks to B. Connor with U.
S. Fish andWildlife, B. Arnsberg from Nez Perce Tribal Fisheries,D. Milks
and staff at Lyons Ferry Hatchery, and the Washington Department of
Fish and Wildlife for fish sample collection. Thanks to the Washington
State University Geoanalytical Lab, J. Vervoort, C. Knaack, and G. Hart
for help with isotopic analysis and feedback on the model. This work
was funded by grants to R. Zabel and B. Kennedy from the USACE and
to B. Kennedy from NMFS–NWFSC. Additional funding came from an
NSF GK-12 award, #0841199, and by the NSF Idaho EPSCoR program,
Table 2
Results of regression analysisfor pairwise watershedcomparisons from Hegget al. (2013)
show significant positive relationships between the difference in mafic, felsic and
metamorphic geology and the difference in
87
Sr/
86
Sr ratio using linear regression. All
models and coefficients were significant (pb0.05). Error is reported a ± 95% confidence
interval.
Linear regression Intercept Coefficient
Model
87
Sr/
86
Sr = % mafic+ε0.0012 ± 0.0006 0.0064 ± 0.0016
87
Sr/
86
Sr = % felsic + ε0.0006 ± 0.0005 0.0101 ± 0.0016
87
Sr/
86
Sr = % metamorphic + ε0.0009 ± 0.0004 0.0167 ± 0.0020
87
Sr/
86
Sr = Shannon diversity+ ε0.0023 ± 0.0008 0.0023± 0.0018
Logistic regression Log odds
Model Intercept Coefficient
87
Sr/
86
Sr = % mafic+ε−1.2488 ± 1.1430 13.0494± 6.8689
87
Sr/
86
Sr = % felsic + ε−1.3093 ± 1.0953 15.1571± 7.1456
87
Sr/
86
Sr = % metamorphic + ε−0.4771 ± 0.8750 22.7080 ± 14.8427
87
Sr/
86
Sr = Shannon diversity+ ε−0.4835 ± 0.9639 6.1581± 3.6998
0.2
0.4
0.6
0.8
1.0
1.2
Geologic
Heterogeneity
(Shannon Diversity)
Snake River
Lapwai Creek
Big Creek
Bear Valley Creek
Watershed
Area of Dominant Rock Type
40% 50% 60% 70% 80% 90%
B
Watershed Area (km2)
102103104105
A
87Sr / 86Sr Residual
-0.006
-0.004
-0.002
0.000
0.002
0.004
0.006
0.008
0% 20% 40% 60% 80%
Felsic Rock
C
Fig. 6. The ability to predict
87
Sr/
86
Sr ratio using bedrock lithology increaseswith watershed size(A). As watersheds decrease in size andare increasingly dominated by a singlerock type,
prediction accuracy decreases (B). As the percentage of felsic rock increaseswithin a watershed,prediction ability decreases (C)showing that the lossof prediction ability is largely driven
by felsicgeology. Valuesfor each watershedare based on watersampling and analysis using TIMS,except Lapwai creekwhich was analyzed using MC-ICPMS. Values forthe Snake River are
seasonal averages from Hegg et al. (2013), all others are single water samples.
97J.C. Hegg et al. / Chemical Geology 360–361 (2013) 89–98
Author's personal copy
award number EPS-0814387. Thanks also to the anonymous reviewers
who helped improve the focus and scope of the manuscript.
Appendix A. Supplementary data
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.chemgeo.2013.10.010.
References
Bain, D., Bacon, J., 1994. Strontium isotopes as indicators of mineral weathering in
catchments. Catena 22 (3), 201–214.
Barnett-Johnson, R., Pearson, T., Ramos, F., Grimes, C., MacFarlane, R., 2008. Tracking natal
origins of salmon using isotopes, otoliths, and landscape geology. Limnol. Oceanogr.
53 (4), 1633–1642.
Barnett-Johnson, R., Teel, D.J., Casillas, E., 2010. Genetic and otolith isotopic markers
identify salmon populations in the Columbia River at broad and fine geographic
scales. Environ. Biol. Fish 89 (3), 533–546.
Bartlett, M.S., 1947. The use of transformations. Biometrics 3 (1), 39–52.
Bataille, C.P., Bowen, G.J., 2012. Mapping
87
Sr/
86
Sr variations in bedrock and water for
large scale provenance studies. Chem. Geol. 304, 39–52.
Bataille, C.P., Laffoon, J., Bowen, G.J., 2012. Mapping multiple source effects on the
strontium isotopic signatures of ecosystems from the circum-Caribbean region.
Ecosphere 3 (12), 1–24.
Blum, J.D., Gazis, C.A., Jacobson, A.D., Page Chamberlain, C., 1998. Carbonate versus silicate
weathering in the Raikhot watershed within the high Himalayan crystalline series.
Geology 26 (5), 411.
Bowen, G.J., 2010. Statistical and Geostatistical Mapping of Precipitation Water Isotope
Ratios, Isoscapes. Springer Verlag 139–160.
Burnham, K., Anderson, D., 2002. Model Selection and Multimodel Inference: A Practical
Information-Theoretic Approach. Springer Verlag.
Chesson, L.A., et al., 2012. Strontium isotopes in tap water from the coterminous USA.
Ecosphere 3 (7), 1–17.
Faure, G., 1977. Principles of isotope geology. Related Information: Smith and Wyllie
Intermediate Geology Series, Medium: X; Size. (Pages: 475 pp.).
Faure, G., Mensing, T., 2004. Isotopes: Principles and Applications. John Wiley & Sons Inc.
Fleiss, J.L., Levin, B., Paik, M.C., 2004. Statistical Methods for Rates and Proportions. Wiley-
Interscience.
Foster, D., Fanning, M., 1997. Geochronology of the northern Idaho batholith and the
Bitterroot metamorphic core complex: magmatism preceding and contemporaneous
with extension. Geol. Soc. Am. Bull. 109 (4), 379.
Garcia, A.P., Bradbury, S., Arnsberg, B.D., Groves, P.A., 2008. Fall Chinook salmon spawning
ground surveys in the Snake River basin upriver of lower granite dam. 2007 Annual
Report (P109172; TRN: US200916%%271).
Gesch, D., Evans, G., Mauck, J., Hutchinson,J., Carswell Jr., W.J., 2009. The National Map —
Elevation: U.S. Geological Survey Fact Sheet 2009–3053.
Graustein, W.C., 1989.
87
Sr/
86
Sr Ratios Measure the Sources and Flow of Strontium in
Terrestrial Ecosystems, Stable Isotopes in Ecological Research. Springer 491–512.
Hales, T.C., Abt, D., Humphreys, E., Roering, J.J., 2005. A lithospheric instability origin for
Columbia River flood basalts and Wallowa Mountains uplift in northeast Oregon.
Nature 438 (7069), 842–845.
Hamann, E.J., Kennedy, B.P., 2012. Juvenile dispersal affects straying behaviors of adults in
a migratory population. Ecology 93 (4), 733–740.
Harrington, R., Kennedy, B., Chamberlain, C., Blum, J., Folt, C., 1998. 15 N enrichment in
agricultural catchments: field patterns and applications to tracking Atlantic salmon
(Salmo salar). Chem. Geol. 147 (3–4), 281–294.
Hegg, J., Kennedy, B., Chittaro, P., Zabel, R., 2013. Spatial structuring of an evolving life-
history strategy under altered environmental conditions. Oecologia 1–13.
Hobbs, N.T., 2003. Challenges and opportunities in integrating ecological knowledge
across scales. For. Ecol. Manag. 181 (1–2), 223–238.
Hobson, K.,1999. Tracing origins and migration of wildlife using stable isotopes: a review.
Oecologia 120 (3), 314–326.
Hobson, K.A., Barnett-Johnson, R., Cerling, T., 2010. Using Isoscapes to Track Animal
Migration, Isoscapes: Understanding Movement, Pattern, and Process on Earth
through Isotope Mapping. Springer Verlag 273–298.
Hobson, K., et al., 2012. A multi-isotope (d13C, d15N, d2H) feather isoscape to assign
Afrotropical migrant birds to origins. Ecosphere 3 (5), 1–20.
Hogan, Z., Baird, I.G., Radtke, R., Vander Zanden, M.J., 2007. Long distance migration and
marine habitation in the tropical Asian catfish, Pangasius krempfi.J.FishBiol.71,
818–832.
Hooper,P.R., Camp, V.E.,Reidel, S.P.,Ross, M.E., 2007.The origin of the Columbia River flood
basalt province: plume versus nonplume models. Geol. Soc. Am. Spec.Pap. 430, 635.
Humston, R., et al., 2006. Geologic Analyses for Evaluating Watershed Heterogeneity:
Implications for Otolith Chemistry Studies. Southeastern Association of Fish and
Wildlife Agencies, 7221 Covey Trace Tallahassee FL 32308 USA.
Hurley, J.M., West, J.B., Ehleri nger, J.R. , 2010. Stable isotope models to predict geographic
origin and cultivation conditions of marijuana. Sci. Justice 50 (2), 86–93.
Jansen, A.C., 2011. Precambrian Basement Rocks of the Clearwater Metamorphic Core
Complex: A New Piercing Point along the Western Laurentian Margin, GSA Annual
Meeting, Minneapolis, MN.
Jenness, J., 2012. DEM Surface Tools. Jenness Interprises.
Kennedy, B., Folt, C., Blum, J., Chamberlain, C., 1997. Natural isotope markers in salmon.
Nature 387 (6635), 766–767.
Kennedy, B., Blum,J., Folt, C.,Nislow, K.,2000. Using natural strontiumisotopicsignaturesas
fish markers: methodology and application. Can. J. Fish. Aquat. Sci. 57 (11), 2280–2292.
Kennedy, B., Klaue, A., Blum, J., Folt, C., Nislow, K., 2002. Reconstructing the lives of fish
using Sr isotopes in otoliths. Can. J. Fish. Aquat. Sci. 59 (6), 925–929.
Leeman, W., 1982. Development of the Snake River Plain-Yellowstone Plateau province,
Idaho and Wyoming: an overview and petrologic model. Cenozoic Geol. Ida. 26,
155–177.
Love, J.D., Leopold, E.B., Love, D., 1978. Eocene Rocks, Fossils, and Geologic History, Teton
Range, Northwestern Wyoming. US Govt. Print. Off.
Ludington, S., et al., 2005. Preliminary Integrated Geologic Map Databases for the United
States —Western States: California, Nevada, Arizona, Washington, Oregon, Idaho,and
Utah [Version 1.3]. U.S. Geological Survey (pp. Open-File Report OF-2005–1305).
Marra, P., Hobson, K., Holmes, R., 1998. Linking winter and summer events in a migratory
bird by using stable-carbon isotopes. Science 282 (5395), 1884.
Martignole, J.M.J., Zirakparvar, N.A.Z.N.A., Vervoort, J.D.V.J.D., McClelland, W.M.C.W.,
Lewis, R.S.L.R.S., 2010. Insights into the metamorphic evolution of the Belt–Purcell
basin; evidence from Lu–Hf garnet geochronology. Can. J. Earth Sci. 47 (2), 161–179.
Muhlfeld, C.C., Thorrold, S.R., McMahon, T.E., Marotz, B., Gillanders, B., 2012. Estimating
westslope cutthroat t rout (Oncorhynchus clarkii lewisi) movements in a river network
using strontium isoscapes. Can. J. Fish. Aquat. Sci. 69 (5), 906–915.
Shannon, C.E., Weaver, W., 1949. TheMathematical Theoryof Communication. University
of Illinois Press, Urbana.
Stewart, B., Capo, R., Chadwick, O., 1998. Quantitative strontium isotope models for
weathering, pedogenesis and biogeochemical cycling. Geoderma 82 (1–3), 173–195.
Stoeser, D.B., et al., 2006. Preliminary Integrated Geologic Map Databases for the United
States: Central States: Montana, Wyoming, Colorado, New Mexico, North Dakota,
South Dakota, Nebraska, Kansas, Oklahoma, Texas, Iowa, Missouri, Arkansas, and
Louisiana. U.S. Geological Survey.
Thorrold, S., Jones, C., Campana, S., McLaren, J., Lam, J., 1998. Trace element signatures
in otoliths record natal river of juvenile American shad (Alosa sapidissima). Limnol.
Oceanogr. 1826–1835.
Wassenaar, L.I., Hobson, K.A., 1998. Natal origins of migratory monarch butterflies at
wintering colonies in Mexico: new isotopic evidence, Proceedings of the National
Academy of Sciences of the United States of America, 26, 95, pp. 15436–15439.
Wassenaar, L.I., Hobson, K.A., 2000. Stable-carbon and hydrogen isotope ratios reveal
breeding origins of red-winged blackbirds. Ecol. Appl. 10 (3), 911–916.
West, J.B., Hurley, J.M., Dudás, F.Ö., Ehleringer, J.R., 2009. The stable isotope ratios of
marijuana. II. Strontium isotopes relate to geographic origin. J. Forensic Sci. 54 (6),
1261–1269.
West, J.B., Bowen, G.J., Dawson, T.E., 2010. Isoscapes: Understanding Movement, Pattern,
and Process on Earth through Isotope Mapping. Springer Verlag.
98 J.C. Hegg et al. / Chemical Geology 360–361 (2013) 89–98
