Isotopes Environ. Health Stud.

Vol. 40, No. 4, December 2004, pp. 291300

GIS-BASED MODEL OF STABLE HYDROGEN ISOTOPE RATIOS IN NORTH AMERICAN GROWING-SEASON PRECIPITATION FOR USE IN ANIMAL MOVEMENT STUDIES
TIMOTHY D. MEEHAN3 , J. TOMASZ GIERMAKOWSKI and PAUL M. CRYAN

Department of Biology and Museum of Southwestern Biology, The University of New Mexico, 167A Castetter Hall, Albuquerque, NM 87131-1091, USA
(Received 21 July 2003; in nal form 23 March 2004)
Stable hydrogen isotope ratios of precipitation (dDp) show distinct geographic patterns across North America. Over the last decade, ecologists have utilized growing-season dDp patterns to study the movements of migratory animals. The accuracy and precision of such studies is, in part, contingent upon the accuracy and precision of growing-season dDp maps. Previous mapping efforts have employed simple kriging procedures to produce smooth contor maps of growing-season dDp. We attempted to improve these maps by incorporating the effects of altitude on both dDp values and growing season length. This involved producing elevation-corrected monthly dDp, temperature, and precipitation amount values for 1-km grid cells across the continental United States and Canada using recently developed interpolation procedures. We used a geographic information system (GIS) to calculate a weightedaverage growing-season dDp value for each grid cell using dDp and precipitation amount values for all months in which the mean temperature was greater than 0 8C. We used seven independent data sets to compare the precision of the resulting altitude-corrected map with another that did not account for altitude. Overall, predicted dDp values from the altitude-corrected map more closely matched observed values, and correspondence was more pronounced at ner spatial scales. Digital versions of the GIS-based maps generated during this effort are available via the Internet at http://biology.unm.edu/wolf/precipitationD.htm. These dDp layers can be combined with other types of spatial information, such as species' geographic ranges and habitat associations, to further improve our understanding of animal movements.

Keywords: Animal migration; Deuterium; Growing-season; Hydrogen; Modeling; Precipitation

INTRODUCTION
Stable hydrogen isotope ratios of precipitation (dDp) show distinct geographic patterns across North America. The dDp decreases as moist air masses travel from low to high latitudes (latitude effect), from low to high altitudes (altitude effect), and from the coasts to inland portions of the continent (coast effect) [14]. Recently, ecologists have begun to utilize isotopic patterns in North American growing-season precipitation to study the movements of migratory animals such as butteries, birds, and bats [58].

Corresponding author. Tel.: 1-505-277-9173; Fax: 1-505-277-0304; E-mail: tdmeehan@unm.edu

ISSN 1025-6016 print; ISSN 1477-2639 online DOI: 10.1080/10256010410001731404

# 2004 Taylor & Francis Ltd

292

TIMOTHY D. MEEHAN et al.

There are two general principles that make possible the tracking of animal movements using growing-season dDp. First, the dD values of animal tissues such as insect chitin [9, 10], mammal bone [11], mammal fur [7, 8, 12], and bird feathers [13, 14] reect those of their food during tissue synthesis. Second, the dD value of a terrestrial animal's food is a function of the dD values of local growing-season dDp. Animal tissue dD values are often between 20 and 30 lower than local dDp due to fractionation that occurs during plant carbohydrate synthesis [15, 16]. Once this stable-hydrogen isotope ratio is xed into the carbon-bound hydrogen of chitin, fur, or feathers, it remains unchanged due to the metabolically inert nature of these tissues. Thus, when migratory animals are observed away from the location of tissue growth, their movements can be inferred by relating the dD values of their tissues to maps of growing-season dDp [6, 13, 16]. Although stable hydrogen isotopes provide an effective means of studying animal migration, the accuracy and precision of the technique is, in part, contingent upon the accuracy and precision of growing season dDp maps [6, 13]. Hobson and Wassenaar published the rst map of continental patterns in growing-season dDp to be used widely by ecologists [13, 17]. This map was created using the following four-step process. First, dDp data were gathered for 39 sites across North America from the International Atomic Energy Agency's global network for isotopes in precipitation (GNIP) database and from published literature. Months of the growing season were dened as those during which the average temperature exceeded 0 8C [11, 13]. The mean amount of precipitation per month was used to weight monthly dDp values when calculating the growing-season average. Finally, localities of the 39 sampling sites were geo-referenced, and the spatial trends of growing-season dDp were modeled using a geo-statistical procedure known as `kriging', where values of unsampled areas are interpolated using data from known localities. This seminal effort resulted in a map showing smooth contour lines of growingseason dDp on a continental scale. Simple kriging of dDp data from precipitation sampling stations leads to interpolated estimates that take into account the coarse-scale (1001,000 km) effects of latitude, distance from the coast, and altitude. However, this approach does not incorporate the ne-scale (1100 km) effects of altitude on dDp and growing season length. We suggest that the accuracy and precision of growing-season dDp maps can be improved by incorporating ne-scale altitude effects. Climatologists [18] and geologists [4] have developed interpolation methods that improve the accuracy of temperature and dDp estimates by taking into account the ne-scale effects of altitude. Herein, we use these methods along with monthly dDp, monthly climate, and altitude data to create a geographic-information-system (GIS) based model of altitude-corrected growing-season dDp. We treat this model as a working hypothesis and compare model predictions with seven independent data sets. We also discuss how it can be applied to migration research, accessed over the Internet, and further tested and rened.

MATERIALS AND METHODS Data Acquisition

Our primary objective was to produce a ne-scale map of amount-weighted-average growing-season dDp for North America that was corrected for the effects of altitude on both dDp values and growing season length. To meet this objective, we needed monthly data on dDp, temperature, and precipitation amount, as well as information on altitude. We obtained data on monthly dDp (VSMOW) for 40 sites in North America. The majority

H IN ANIMAL MOVEMENT STUDIES

293

of these data came from the GNIP database (http://isohis.iaea.org) [19]. Additional data on monthly dDp were obtained for Iowa, Nebraska, and New Mexico [2022]. Although monthly dDp was measured during at least 12 months at each station, the years during which sampling occurred and the total number of years sampled differed between sites (see the earlier citations for details). We decided to use all available data across several years because restricting our analyses to certain years would drastically reduce the information content of the models. We obtained average monthly temperature (8C) and precipitation (mm) data for 1931 and 1418 sites, respectively, in the continental United States and Canada from the US National Climate Data Center's global historical climatology network (GHCN) (http://lwf.ncdc. noaa.gov/cgi-bin/res40.pl?page ghcn.html). These summaries were derived from all available data, collected between the years 16971990, depending on the sample site [23]. We derived elevation information from a 30 arc-second digital elevation model (DEM) ( 1 km + 30 m cell size) available from the Earth Resources Observation Systems Data Center, National Mapping Division, US Geological Survey (GTOPO30 dataset, http://edcdaac.usgs.gov/gtopo30/ gtopo30.html).

Interpolations
After acquiring the appropriate data, we produced monthly estimates of altitude-corrected dDp, ambient temperature, and precipitation amount for 1-km grid cells across North America. We produced altitude-corrected estimates of monthly dDp for each grid cell using interpolation methods described by Bowen and Wilkinson [4]. Briey, for each month we regressed dDp from the 40 North American stable isotope monitoring stations against station latitude (decimal degrees), latitude squared, and altitude (m) (monthly regression equations are available upon request from the corresponding author). After generating a monthly regression equation, we computed monthly dDp residuals. Then, we modeled the remaining spatial trends in the residuals using simple kriging and interpolated dDp residuals for each grid cell. Next, we added back the effects of latitude and altitude to the residual values in each grid cell using the appropriate monthly regression formula and elevation data from the DEM. This process resulted in 12 monthly GIS layers of altitudecorrected dDp with 1-km2 grid cell resolution. We calculated altitude-corrected estimates of monthly temperature for each grid cell using interpolation methods described by Willmott and Matsuura [18]. Briey, we calculated a `sea-level' temperature for each month for the 1931 GHCN temperature stations. This was done by adding the product of 6.5 8C 2 station elevation (adiabatic lapse rate based on the 1976 US Standard Atmosphere [18]) to each database record. Then, we modeled the spatial trends in monthly sea-level temperature using simple kriging and interpolated monthly sealevel temperatures for each grid cell. Finally, we added back the effect of altitude to each monthly sea-level temperature layer by subtracting the product of 6.5 8C 2 grid cell DEM elevation from the sea-level value interpolated for each grid cell. This process yielded 12 monthly GIS layers of altitude-corrected temperature with 1-km2 grid cell resolution. Finally, we modeled the spatial trends in the monthly precipitation amount data from the 1418 GHCN precipitation stations using simple kriging and interpolated monthly precipitation amounts for each grid cell. This process yielded 12 monthly GIS layers of precipitation amount with 1-km2 grid cell resolution. We did not correct precipitation amount for altitude because the relationship between altitude and precipitation amount is complex and varies across seasons and regions of the continent. All kriging was carried out using the geostatistical analyst extension of ArcGIS version 8.3 (ESRI, Inc., Redlands, California, USA)

294

TIMOTHY D. MEEHAN et al.

following methods outlined by Johnston et al. [24]. For all kriging procedures, we used spherical covariance models and default parameter values determined by optimization routines built into the geostatistical analyst extension.

Grid Calculations
Interpolation procedures resulted in a monthly estimate of altitude-corrected dDp, ambient temperature, and precipitation amount for each 1-km grid cell. We calculated a nal amount-weighted-average, growing-season dDp for each grid cell by averaging dDp for all months for which ambient temperature was . 0 8C and weighting this average using precipitation amount per month. The process can be formalized as
dD p

P (f (Ti)dDpiPi) P (f (Ti)Pi)

(1)

where dDpi is the monthly dDp value, Pi is monthly precipitation amount, and f(Ti) is a step function where for Ti  0 W C f (Ti ) 1 for Ti . 0 W C

0

(2)

All grid calculations were performed using ERDAS Imagine version 8.6 (Leica Geosystems GIS & Mapping, LLC, Atlanta, Georgia, USA), rst by using the Model Maker module, and subsequently by converting modeling operations into a script.

Validation
To assess the utility of ne-scale altitude corrections, we compared predicted altitudecorrected annual dDp with the observed annual dDp values reported in the hydrology literature. Abbott et al. [25] reported d18O values of precipitation along an elevational transect in northwestern Vermont. Ingraham et al. [26] reported d18O values of precipitation from a topographically diverse locality in southern Nevada. In both of these studies, local meteoric water lines were reported; we used these local relationships to convert d18O values to dD values. Friedman et al. [27] reported dD values of precipitation from a topographically diverse region of southeastern California. Welker [28] reported d18O values of precipitation from across the continental USA; we converted these values to dD using the global meteoric water line [29]. Altitude-corrected annual dDpvalues were calculated by summing the products of the monthly dDp and precipitation layers and then dividing this sum by the sum of the monthly precipitation amount layers for each 1-km2 raster cell. This was essentially the same process as the one described earlier for creating the altitude-corrected growing-season dDp map, except here we eliminated the condition that average monthly temperature was . 0 8C, i.e., Eq. (1) without f(Ti). Once we obtained observed and predicted annual dDp values, we used linear regression to assess their relationship. We expected the relationship between predicted and observed dDp values to be linear with a slope  1 and an intercept  0 . After regressing observed dDp values against predictions from the altitude-corrected model, we regressed observed dDp values against predictions from a model that was not corrected for altitude. This model was produced by kriging annual dDp values from the 40 dDp sampling stations. Once again, we expected the relationship between predicted and observed dDp values to be linear with a slope  1 and an intercept  0 .

H IN ANIMAL MOVEMENT STUDIES

295

We were able to assess the utility of our efforts by comparing the statistics from the two regressions described earlier. We assessed the precision of models by comparing r 2 values; we assumed that the model with the higher r 2 had a higher degree of precision. We assessed the accuracy of the models by comparing the slope and intercept parameter estimates. We assumed that a model was more accurate if it had parameter estimates closer to those expected. We also compared data from the model of altititude-corrected growing-season dDp to dD values of bird feathers (dDf) reported in the ecology literature. Kelly et al. [30] reported dDf values for Wilson's Warblers (Wilsonia pusilla) from throughout western North America. Lott et al. [31] reported dDf values for nine species of diurnal raptors from throughout North America. Data collected by Lott et al. included a subset of samples from an elevational gradient spanning from eastern to western North Carolina. We also acquired unpublished data from DeLong on dDf values of ammulated owls (Otus ammeolus) from southwestern North America. We used linear regression to assess the relationship between observed dDf and altitudecorrected growing-season dDp values. For this comparison, we expected the relationship between dDp and dDf values to be linear with a slope  1 and an intercept  2 25 , reecting precipitation-to-plant-carbohydrate hydrogen isotope fractionation. After regressing dDf values against dDp from the altitude-corrected growing-season dDp model, we regressed dDf values against dDp values derived from a growing-season dDp model that was not corrected for elevation. This map was produced by simply kriging growing season dDp values from the 39 sites reported by Hobson and Wassenaar [9]. Once again, we expected the relationship between dDp and dDf values to be linear with a slope  1 and an intercept  2 25 , reecting precipitation-to-plant-carbohydrate hydrogen isotope fractionation. We assessed the utility of our efforts to correct for the effects of elevation on dDp and growing-season length by comparing the statistics from the two regressions, as described earlier.

RESULTS
The maps of elevation-corrected weighted-average dDp (Fig. 1) depict the general tendency for dDp to decrease with latitude, distance from the coast, and altitude. These maps do not show smooth variation across the continent, but illustrate the effects of abrupt topographic changes on dDp. The degree to which the altitude-explicit model of annual dDp predicted observed annual dDp values is shown in Figure 2(a). With three of the four test data sets, the relationship between observed and predicted annual dDp had a slope not signicantly different from 1 and an intercept not signicantly different from 0 (Tab. I). However, in southern California, the dDp values predicted by this model were signicantly different from those reported in the literature. In this case, the slope between observed and expected values was signicantly greater than 1 and the intercept was greater than 0 . Figure 2(b) depicts the degree to which the kriged-only model of annual dDp predicted variation in observed annual dDp values. In three of the four data sets, the relationships between observed and predicted dDp had slopes signicantly different from 1 and intercepts signicantly different from 0 (Tab. I). Overall, r 2 values of observed vs. predicted dDp regressions increased when elevation was taken into account. Such increases ranged from 0.12 at coarse spatial scales to 0.89 at ne spatial scales. The relationships between altitude-corrected growing-season dDp predictions and observed dDf values were close to those expected (Fig. 2(c), Tab. I). With two of the four test data sets, regressions between observed dDf and predicted dDp revealed slopes that did not differ signicantly from 1 and, in all cases, slopes were between 0.6 and 0.9. In

296

TIMOTHY D. MEEHAN et al.

FIGURE 1 Altitude-corrected annual dD () and altitude-corrected growing-season dD of North American precipitation. Circles are dDp sampling stations used in map building.

three out of four cases, the intercepts were not signicantly different from 2 25 and the intercepts of all four regression lines were between 2 8 and 2 37 . Growing-season dDp values predicted by the kriged-only model exhibited slightly weaker relationships with observed dDf values (Fig. 2(d), Tab. I). In all four cases, regression of observed dDf and predicted dDp had slopes signicantly different from the expected value of 1. The slopes ranged from 0.4 to 7.8. In three of four instances, intercepts were signicantly different from the expected value of 2 25 , ranging from 2 27 to 168 . In general, use of the kriged-only model of growing season dDp led to greater deviation in slope and intercept values as the spatial extent of dDf data decreased. Overall, the r 2 values of the predicted dDp vs. observed dDf regressions increased slightly when elevation was accounted for. Improvements in r 2 ranged from 0.02 at coarse spatial scales to 0.21 at ne spatial scales.

H IN ANIMAL MOVEMENT STUDIES

297

FIGURE 2 (a) Comparison of annual dDp () values from the literature with those predicted by the altitudeexplicit annual dDp model. (b) Comparison of annual dDp values from the literature (same as (a)) with those predicted by an annual dDp model constructed by kriging only (no elevation correction). In (a) and (b), diamonds are Welker (2000), triangles are Friedman et al. (1992), squares are Ingraham et al. (1991), and circles are Abbott et al. (2000). (c) Comparison of dDf values from the literature with dDp values predicted by the altitudeexplicit growing-season dDp model. (d) Comparison of dDf values from the literature (same as (c)) with growing-season dDp values predicted by a model constructed by kriging only (no elevation correction). In (c) and (d), diamonds are Lott et al. (2003, North Carolina only), triangles are Kelly et al. (2002), squares are Lott et al. (2003, all data), and circles are DeLong (unpublished). Regression coefcients are reported in Table I.

DISCUSSION
Tracking the long-distance movements of animals remains a challenge in ecological research. Stable-isotope analysis of animal tissues offers a means of answering questions about migration and dispersal that were previously intractable. However, the success of this method is inextricably tied to our understanding of the natural variation of isotope ratios in the environment. In this study, we used published methods to generate altitudeexplicit estimates of growing-season dDp and improved upon kriged-only growing-season dDp maps. We found that altitude-explicit estimates of both annual and growing-season dDp were more closely related to observed precipitation and bird feather dD values than esti-

.detcepxe morf tnereffid yltsav era seulav retemarap esuaceb lufgninaem ton si 2 r 3 .)d( )a(2 erugiF ni shparg ot dnopserroc scitsitatS :etoN

TIMOTHY D. MEEHAN et al.

66.0 81.0 15.0 84.0 79.0 98.0 56.0 65.0

3

)52( 8 2 )23( 52 2 )8( 73 2 )91( 12 2 )3.0( 9.0 )9.0( 7.0 )1.0( 6.0 )2.0( 7.0 )2.0( 2.1 )7.0( 4.1 )5.0( 8.1 )6.0( 3.1 58.0 399.0 350.0 80.0
2r

54.0 374.0 64.0 64.0 )12( 41 2 )1.0( 47 2 )11( 15 2 )3( 15 2

)41( 95 2 )531( 861 )8( 83 2 )81( 72 2

)2.0( 4.0 )8.4( 8.7 )1.0( 6.0 )2.0( 7.0 )3.0( 7.0 )4000.0( 500.0 )1.0( 1.0 2 )40.0( 20.0

22 41 19 26 6 4 03 41

aciremA htroN nretsewhtuoS aniloraC htroN aciremA htroN aciremA htroN nretseW

srehtaef lwO srehtaef rotpaR srehtaef rotpaR srehtaef dribgnoS

)dehsilbupnu( gnoLeD )3002( .la te ttoL )3002( .la te ttoL )2002( .la te ylleK

)51( 2 2 )65( 12 )63( 17 )65( 61

Dd devresbO

aciremA htroN tnomreV ainrofilaC nrehtuoS adaveN nrehtuoS

noitatipicerP noitatipicerP noitatipicerP noitatipicerP

)0002( rekleW )0002( .la te ttobbA )2991( .la te namdeirF )1991( .la te mahargnI

Dd devresbO )IC( tpecretnI

pDd

2r pDd

)IC( tpecretnI detcerroc-noitavelE

)IC( epolS

)IC( epolS ylno-degirK

noiger ydutS

ecnatsbus tseT

tes atad tseT

.seulav fDd dna pDd devresbo tsniaga sledom ylno-degirk dna detcerroc-noitavele morf pDd detciderp fo snoisserger morf stluseR I ELBAT

298

H IN ANIMAL MOVEMENT STUDIES

299

mates from kriged-only models. Although the altitude-explicit models showed improvements at all spatial scales, they were noticeably more precise than the kriged-only models at ner spatial scales. We have made the annual and growing-season dDp maps shown in Figure 1 available for download at http://biology.unm.edu/wolf/precipitationD.htm. The layers are offered in ERDAS Imagine (3 .img) raster-based format. As such, the layers can be easily incorporated into a GIS and used in conjunction with other spatial data (e.g., species range, yway dimensions, or habitat type) to generate more precise estimates on the origins of migratory animals [32]. Although the altitude-explicit map of growing-season dDp developed during this study appears to be an improvement over kriged-only maps, we view it as a work in progress. One way in which growing-season dDp maps could be further improved would be to incorporate dD data from more precipitation sampling stations across the continent. The elevation-explicit growing season dDp map was created using data from 40 dDp sampling stations in North America, and model estimates will undoubtedly improve as data from more precipitation sampling sites become available. In addition, future GIS-based mapping efforts could incorporate more sophisticated models of atmospheric circulation patterns, topographic slope and aspect, and land cover. The map created during this study, as well as those generated by future efforts, should be further evaluated using independent data sets. By testing models with data gathered in different regions and at various spatial scales, geographic areas of model weakness may be discernable. It will be of particular interest to test these models using dD data from animal and plant tissues collected along ne-scaled elevational transects. We encourage others to test and improve upon the altitude-corrected growing season dDp map, and to extend modeling efforts to other parts of the world.

Acknowledgements
We thank the following people for their assistance: Z. D. Sharp, V. N. Atudorei, and J. P. DeLong allowed us to use their unpublished data. C. A. Lott and J. F. Kelly sent us electronic versions of their published data. V. A. Atudorei, C. A. Lott, Z. D. Sharp, B. O. Wolf, and two anonymous reviewers offered advice on improving previous versions of the maps and manuscript. T.D.M. was supported by an assistantship provided by the Department of Biology at the University of New Mexico.

References
[1] Dansgaard, W. (1964). Stable isotopes in precipitation. Tellus 16, 436468. [2] Rozanski, K., Araguas-Araguas, L. and Geoantini, R. (1993). Isotopic patterns in modern global precipitation. Geophys. Monogr. 78, 136. [3] Araguas-Araguas, L., Froehlich, K. and Rozanski, K. (2000). Deuterium and oxygen-18 isotope composition of precipitation and atmospheric moisture. Hydrol. Processes 14, 13411355. [4] Bowen, G. and Wilkinson, B. (2002). Spatial distribution of delta O-18 in meteoric precipitation. Geology 30, 315318. [5] Wassenaar, L. I. and Hobson, K. A. (1998). Natal origins of migratory monarch butteries at wintering colonies in Mexico: New isotopic evidence. Proc. Natl. Acad. Sci. U.S.A. 95, 1543615439. [6] Meehan, T. D. et al. (2001). Using hydrogen isotope geochemistry to estimate the natal latitudes of immature Cooper's Hawks migrating through the Florida Keys. Condor 103, 1120. [7] Cryan, P. M. Migration and thermoregulation strategies of hoary bats (Lasiurus cinereus) in North America. PhD dissertation. University of New Mexico, Albuquerque 2003. [8] Cryan, P. M., Bogan, M. A., Rye, R. O., Landis, G. P. and Kester, C. L. (2004). Stable hydrogen isotope analysis of bat hair as evidence of seasonal molt and long-distance migration. J. Mamm. (in press)

300

TIMOTHY D. MEEHAN et al.

[9] Miller, R. F., Fritz, P. and Morgan, A. V. (1988). Climatic implications of D/H ratios in beetle chitin. Paleogeog. Paleoclimatol. Paleoecol. 66, 277288. [10] Hobson, K. A., Wassenaar, L. I. and Taylor, O. R. (1999). Stable isotopes (delta D and delta C-13) are geographic indicators of natal origins of monarch butteries in eastern North America. Oecologia 120, 397404. [11] Cormie, A. B., Schwarcz, H. P. and Gray, J. (1994). Relation between hydrogen isotopic ratios of bone collagen and rain. Geochim. Cosmochim. Acta 58, 377391. [12] Sharp, Z., Atudorei, V., Panarello, H., Fernandez, J. and Douthitt, C. (2003). Hydrogen isotope systematics of hair: Archeological and forensic applications. J. Archaeol. Sci. 30, 17091716. [13] Hobson, K. A. and Wassenaar, L. I. (1997). Linking breeding and wintering grounds of Neotropical migrant songbirds using stable hydrogen isotopic analysis of feathers. Oecologia 109, 142148. [14] Hobson, K. A., Atwell, L. and Wassenaar, L. I. (1999). Inuence of drinking water and diet on the stablehydrogen isotope ratios of animal tissues. Proc. Natl. Acad. Sci. U.S.A. 96, 80038006. [15] White, J. W. C. Stable Isotopes in Ecological Research (Eds. P. W. Rundel, J. R. Ehleringer and K. A. Nagy). Springer-Verlag: New York 1988; pp. 142162. [16] Wassenaar, L. I. and Hobson, K. A. (2001). A stable-isotope approach to delineate geographical catchment areas of avian migration monitoring stations in North America. Environ. Sci. Tech. 35, 18451850. [17] Hobson, K. A. (1999). Tracing origins and migration of wildlife using stable isotopes: a review. Oecologia 120, 314326. [18] Willmott, C. J. and Matsuura, K. (1995). Smart interpolation of annually averaged air temperature in the United States. J. Appl. Metorol. 34, 25772586. [19] IAEA (2001). GNIP maps and animations, International Atomic Energy Agency, Vienna. Accessible at http: // isohis.iaea.org. [20] Yapp, C. J. (1985). D/H variations in meteoric waters of Albuquerque, New Mexico, USA. J. Hydrol. 76, 6384. [21] Simpkins, W. W. (1995). Isotopic composition of precipitation in central Iowa. J. Hydrol. 172, 185207. [22] Harvey, F. and Welker, J. (2000). Stable isotopic composition of precipitation in the semi-arid north-central portion of the US Great Plains. J. Hydrol. 238, 90109. [23] Peterson, T. C. and Vose, R. S. (1997). An overview of the global historical climatology network temperature database. Bull. Am. Meteorol. Soc. 78, 28372849. [24] Johnston, K., Ver Hoef, J. M., Krivoruchko, K. and Lucas, N. Using ArcGIS Geostatistical Analyst. Environmental Systems Research Institute Inc.: Redlands, California 2001. [25] Abbott, M. D., Lini, A. and Bierman, P. R. (2000). d18O, dD and 3H measurements constrain groundwater recharge patterns in an upland fractured bedrock aquifer, Vermont, USA. J. Hydrol. 228, 101112. [26] Ingraham, N. L., Lyles, B. F., Jacobson, R. L. and Hess, J. W. (1991). Stable isotopic study of precipitation and spring discharge in southern Nevada. J. Hydrol. 125, 243258. [27] Friedman, I., Smith, G. I., Gleason, J. D., Warden, A. and Harris, J. M. (1992). Stable isotope composition of waters in southeastern California 1. Modern precipitation. J. Geophys. Res. 97, 57955812. [28] Welker, J. M. (2000). Isotopic (delta O-18) characteristics of weekly precipitation collected across the USA: an initial analysis with application to water source studies. Hydrol. Processes 14, 14491464. [29] Craig, H. (1961). Isotopic variations in meteoric waters. Science 133, 17021703. [30] Kelly, J. F., Atudorei, V., Sharp, Z. D. and Finch, D. M. (2002). Insights into Wilson's Warbler migration from analyses of stable hydrogen isotope ratios. Oecologia, 130, 216221. [31] Lott, C. A., Meehan, T. D. and Heath, J. A. (2003). Estimating the latitudinal origins of migratory birds using hydrogen and sulfur stable isotopes in feathers: inuence of marine prey base. Oecologia 134, 505510. [32] Smith, R. B., Greiner, E. C. and Wolf, B. O. (2004). Migratory movements of sharp-shinned hawks in relation to prevalence, intensity, and biogeography of avian hematozoan parasites. Auk (in press)