Heavy isotope enrichment is widely observed in lakes and other surface waters undergoing evaporation because of mass-dependent differences in the equilibrium vapor pressures and gas-phase molecular diffusivities among the naturally occurring water isotopomers, including common light water containing only 16O and 1H (1H1H16O) and two of the rare heavy isotopomers containing either 18O (1H1H18O) or 2H (1H2H16O). An evaporating moisture flux is preferentially depleted in 18O and 2H, leading to progressive enrichment of these heavy isotopes in the residual liquid. In contrast, transpiration does not generally produce a fractionation between the moisture flux and residual soil water, since the transfer of moisture to the atmosphere by plants is essentially quantitative, and heavy-isotope enrichment is restricted to internal plant waters [Jacob and Sonntag, 1991; Wang and Yakir, 2000]. The absolute isotopic enrichment of a surface water body is controlled by both atmospheric and water balance processes [Gat, 1995]. The principal atmospheric factors are humidity, temperature, and the isotopic composition of ambient atmospheric moisture, which collectively control the relative rates and ultimate limits of heavy-isotope enrichment, whereas water balance (i.e., the evaporation/inflow ratio of the reservoir, or E/I) determines the degree to which isotopic enrichment is attenuated by dilution. Evaporation processes and the results of experimental studies to characterize turbulent and diffusive mass transfer mechanisms from an isotopic perspective have been extensively addressed in a number of key publications, including those by Gat , Gonfiantini , Merlivat and Jouzel , and Stewart .
 Valuable insight into the isotopic behavior of lakes in the strongly seasonal, continental climate of northern Canada has been gained through field-based investigations of lake evaporation and water balance over the past decade in the Northwest Territories, Nunavut, and northern Alberta [Gibson et al., 1993, 1996a, 1996b, 1998, 1999, 2002; Gibson, 2001, 2002]. Isotopic labeling of waters at catchment to regional scale is commonly manifested by the existence of two linear trends on a conventional 2H-18O crossplot (Figures 1a and 1b) (the δ notation expresses the relative abundances of 18O and 2H as deviations in per mil (‰) from a given standard, normally Vienna Standard Mean Ocean Water, such that δsample = 1000(Rsample/RVSMOW) − 1, where R is 18O/16O or 2H/1H), allowing differentiation of “meteoric” waters that retain their original isotopic composition (i.e., local precipitation, including that stored in the snowpack, and derived groundwaters) from waters that have undergone subsequent heavy-isotope enrichment due to evaporation (i.e., surface waters and derived stream and groundwaters). The former typically cluster along a local meteoric water line (MWL) having a slope of ∼8, reflecting the pervasive influence of mass-dependent isotope exchange processes during the transport and progressive distillation of atmospheric moisture, whereas evaporatively enriched waters generally plot to the right of the MWL, as a consequence of additional kinetic effects during the evaporation process related to variations in both overall mass and its distribution within the water molecules [Dansgaard, 1964; Merlivat and Jouzel, 1979]. As shown in Figure 1b, the isotopic composition of precipitation in northern Canada generally plots close to the global meteoric water line (MWL) of Craig , defined by δ2H = 8δ18O + 10, which is the locus of weighted monthly precipitation sampled at several hundred stations worldwide [see Rozanski et al., 1993].
 Evaporated waters within a given region often cluster along a local evaporation line (LEL) having a slope in the range 4–6, variably well defined as a function of the range of E/I ratios and degree of hydraulic connectivity within a system. The intersection point of an LEL with the local MWL commonly affords an excellent approximation of the weighted mean isotopic composition of local precipitation (δP), which is a key datum for isotope-mass balance investigations, since offset from the MWL along the LEL increases in proportion to the cumulative fraction of water lost by evaporation upstream of a given sampling station. Basic knowledge or assumptions about the hydrologic status of a lake and its catchment can be coupled with this isotopic information to quantify or constrain both oxygen and hydrogen isotope-mass budgets.
 The main objective of this study is to explore the potential application of isotope tracers as indicators of water balance systematics and variability at the large scale and to gain a better understanding of the role of lakes in the regional runoff regime of a boreal-arctic transition zone. Here we present and discuss a previously unreported stable isotope data set acquired from a regional survey of lake water quality across a remote 275,000 km2 region of the continental Arctic and subarctic encompassing northern tree line (Figure 2). Water samples were collected by the Department of Indian and Northern Affairs Canada as part of a study to investigate baseline controls on water chemistry and potential impacts related to mining and recreation [Puznicki, 1996]. Archived samples were separately analyzed for δ18O and δ2H by the authors at the University of Waterloo (Figure 1b), revealing systematic spatial variability in evaporative isotopic enrichment in the lakes consistent with previous point observations at localities within the region and affording the opportunity to gain a better understanding of regional variations in water budgets. As outlined in section 1.1.1, we apply a steady state isotope mass balance approach to the data set to quantify regional trends in evaporation loss as a fraction of total outflows, evaporation losses from lakes in proportion to that from upstream reservoirs, and evaporation losses as a fraction of the total evapotranspiration flux. Supporting information for the calculations is obtained from a geographic information system (GIS) and map-based analysis of isotopic and hydroclimatic parameters interpolated for each sampling site and basin. Use of the steady state approach is justified by the large volume of the lakes, most of which are larger than 109 m3, with average depths of >14 m. Such lakes are known from previous studies to have subdued seasonal isotopic cycles [see Gibson, 2001]. The analysis offers a unique, broad-scale perspective of regional water balance using models that have been intensively evaluated and refined in field-based studies and provides valuable insight into the role of lakes in the runoff regime of Precambrian Shield terrain underlain by permafrost.
1.1.1. General lake balance
 The water and isotope mass balances of a well-mixed lake undergoing evaporation while maintaining a long-term constant volume (assuming constant density of water) are
where IL is combined surface and subsurface inflow, QL is combined surface and subsurface outflow, EL is lake evaporation, and δI, δQ, and δE are the isotopic compositions of inflow, outflow, and evaporative flux, respectively. Substituting QL = IL − EL from (1) and δQ = δL (acknowledging that average outflow will be isotopically similar to the isotopic composition of lake water δL), equation (2) can be rearranged as
which assumes no long-term storage changes in the reservoir. As noted earlier, the evaporation flux δE is typically depleted in the heavy isotopes relative to lake water δL. Although impossible to measure directly, the magnitude of isotopic separation between lake water and the isotopic composition of the evaporation flux has been shown to be dependent on evaporation temperature, boundary layer state (i.e., laminar, turbulent, or static), and ambient atmospheric conditions (relative humidity and isotopic composition of atmospheric moisture). The standard approach for estimating δE from these boundary characteristics is by the Craig and Gordon  model assuming negligible resistance to mixing in the liquid phase [see Gat, 1995] (note that the Craig and Gordon  equation is modified to directly utilize isotopic data in per mil rather than as a decimal fraction, i.e., −15‰ rather than −0.015 should be used),
in parts per mil, where α* is the equilibrium liquid-vapor isotope fractionation (α* = 1 + ε*), h is the atmospheric relative humidity (ranging from 0 to 1) normalized to the saturation vapor pressure at the temperature of the air-water interface, δA is the isotopic composition of ambient moisture, and
in parts per mil, where ε is the total isotopic separation factor including both equilibrium ε* and kinetic εK components. The equilibrium separations can be evaluated using the empirical equations determined experimentally by Horita and Wesolowski  given by
in parts per mil, where T is the interface temperature (K). These relatively new equations yield values only slightly different than those reported by Majoube  and Bottinga and Craig  in the range of temperatures expected for average evaporation conditions (0°–25°C).
 Kinetic enrichment factors εK are dependent on both the boundary layer conditions and the humidity deficit evaluated according to
in parts per mil, where constant, experimentally-determined CK values of 14.2‰ for oxygen and 12.5‰ for hydrogen are used as representative of typical lake evaporation conditions [Gonfiantini, 1986; Araguás-Araguás et al., 2000].
 Introducing the term XL, which is the fraction of water loss by evaporation (E/I ratio),
where δL is the steady state isotopic composition of the lake, m is the enrichment slope
 Residence time of a reservoir in long-term isotopic and hydrologic steady state is given by
in years, which assumes that the evaporative enrichment signal that defines XL is derived entirely from the lake itself. If appreciable isotopic enrichment also occurs in upstream reservoirs, then this must be viewed as an upper limit.
1.1.2. Headwater lakes, nonheadwater lakes, and catchments
 Definition of the input signal is the most important distinction between headwater and nonheadwater basins. For headwater lakes, the assumption that inflow is close to the isotopic composition of precipitation (δI = δP) is a reasonable first approximation, as demonstrated in previous studies [Gibson et al., 2002]. Because of the so-called “string-of-lakes” effect (adopting the terminology of Gat and Bowser ), inflow to nonheadwater lakes is expected to be variably enriched in the heavy isotopes by evaporation from upstream lakes. This makes it impossible to distinguish isotopically the evaporation occurring directly from a lake from that occurring upstream, without additional information on the isotopic composition of inflows. Nevertheless, evaporation losses from the entire catchment can be evaluated by substitution of δI = δP into equation (10), acknowledging that input integrated over the catchment as a whole must be isotopically similar to the isotopic composition of precipitation, as demonstrated by Gibson . This provides an index of evaporation loss for a catchment area (or headwater lake) given by
 One important application, as suggested by Gibson et al.  is for tracing of long-term runoff from the catchment area. Assuming that runoff from the catchment is equal to inflow to the lake minus precipitation on the lake, then the runoff can be calculated as
and LA is the lake area, DBA is the drainage basin area, and CA is the catchment area such that CA = DBA + LA.
 For nonheadwater systems, it is possible to distinguish isotopically the evaporation losses from successive reservoirs [Gat and Bowser, 1991], although the data for inflows are not available for the present study. The evaporation losses in a nonheadwater lake consist of evaporation occurring from the lake and evaporation from upstream lakes,
where EL is evaporation from the lake, is the sum of evaporation from upstream lakes and reservoirs, PL is the precipitation on the lake, and QDBA is the runoff from the catchment area. Partitioning of evaporation losses from upstream lakes can therefore be evaluated by incorporating hydrologic data from other sources. In this case,
 The contribution of evaporation (E) to the total evapotranspiration flux (ET) in the catchment is likewise evaluated using
where et is derived evapotranspiration from the land surface, also interpolated from the Hydrological Atlas of Canada [denHartog and Ferguson, 1978b]. Although the latter two indices rely heavily on nonisotopic information, the isotope data assist in constraining the mechanisms of water transfer and offer a new perspective on the role of lakes in the regional hydrologic regime.
1.2. Study Area and Methods
 The study was conducted in the continental Arctic and subarctic of the Northwest Territories and Nunavut (Figure 2), within the ranges of 62°N–68°N and 107°W–118°W, in an area characterized by low relief and myriad lakes. Bedrock consists mainly of fractured Canadian Shield rocks of Precambrian age with a thin to discontinuous overburden of glacial till, moraine, and lesser organic soils. Permafrost extent ranges from thin, discontinuous occurrences in bogs and peaty soils near Yellowknife to thick, continuously frozen ground up to several hundred meters deep in the north and northeast. Apart from the extreme northwestern region along the Coppermine River, where river channels may be deeply incised, drainage is disorganized, and large areas of up to 20,000 km2 may be drained through individual lakes. Lakes in this setting can be considered as nodes in a regional string-of-lakes drainage network. The region straddles the divide between the Mackenzie River basin and the Arctic coastal drainage, producing a prominent southwest-northeast bidirectional drainage pattern. Steep southwest-to-northeast hydroclimatic gradients also exist in the area, with mean annual temperatures ranging from −6° to −14°C. Precipitation and evaporation range from about 350 and 400 mm yr−1, respectively, in the southwest to <200 mm yr−1 for both parameters in the northeast and along the coast of Coronation Gulf, according to mapped compilations presented in the Hydrological Atlas of Canada [denHartog and Ferguson, 1978a, 1978b].
 The study area spans the northern limit of trees, and regional vegetation patterns are strongly influenced by the steep climatic gradients, leading to a steady transition from high-boreal forest to subarctic forest-tundra to low-arctic shrub tundra toward the northeast [Ritchie, 1993; see also Ruhland and Smol, 1998].
 Water samples were collected in 1993 and 1994 as part of a water quality survey of the Slave Structural Province of the Canadian Shield [see Puznicki, 1996]. Random and thorough coverage of the study area was ensured by selecting lakes closest to intersections of a 25-km grid on 1:250,000 topographic map sheets that were large enough to accommodate the landing and takeoff of a single-engine float plane. Additional lakes were also included in specific catchments prone to potential impact from mining or recreational development. Lakes were sampled during the open water season (July to September 1993 and 1994), with sampling conducted as close as possible to the lake center. Irregularly shaped lakes were sampled more than once to ensure that a representative set of samples was obtained. All samples were collected using a 3-L horizontal Van Dorne water sampler, with collection typically at 4 m depth (or middepth in the case of shallower lakes). Measurements were made of lake and sample depths, water temperature, pH, and conductivity on site and were reported with results of major ion and trace element analyses performed in the laboratory by Puznicki . Limnological characteristics of a subset of lakes were also evaluated by Ruhland and Smol . The 182 filtered, but otherwise untreated, samples from the 1993 survey were analyzed for δ18O and δ2H. Sampling in 1994 was carried out on a smaller set of different lakes overlapping the same area. Seventy of the latter samples were also analyzed for stable isotopes, and the results are presented here to test the reproducibility of the regional trends discerned from the larger sample set.
 Isotopic ratios were determined by conventional mass spectrometric techniques in the University of Waterloo Environmental Isotope Laboratory, with δ18O and δ2H values reported with respect to Vienna standard mean ocean water (VSMOW) on a scale normalized such that Standard Light Arctic Precipitation (SLAP) has values of −55.5 and −428‰, respectively [seeCoplen, 1996]. Maximum analytical uncertainties are ±0.1‰ for δ18O and ±2‰ for δ2H.
1.3. Potential Incomplete Mixing or Temporal Isotopic Enrichment
 Prior to analyzing spatial variations the isotopic data were scrutinized to test whether potential “noise” from within-lake heterogeneity or seasonal isotopic enrichment of lake water over the 50-day sampling interval strongly influenced primary isotopic water balance signals. To assess the former, isotopic analyses were obtained on an additional 29 samples from 12 lakes, with two or three samples collected per lake, either in a vertical profile at a given site or at bottom or middepth at several different locations within the same lake. On average, the standard deviation of the δ18O and δ2H values for the 12 lakes was found to be 0.14 and 1.2‰, respectively, which is similar to the analytical uncertainty for each tracer. This shows that lakes are generally well mixed, consistent with results from a survey of lakes in northern Alberta involving extensive comparative sampling of epilimnion, hypolimnion, and euphotic zone waters [Gibson et al., 2002]. The only significant within-lake variability (0.5 and 3.8‰ for δ18O and δ2H, respectively) was apparent for an unusually large and irregular lake having several inflows, which was atypical for water bodies included in the present survey. Full consideration of water balance in such lakes would require much more detailed basin-specific assessment than the current regional-scale analysis permits.
 On the basis of extensive prior observations and modeling studies in the region [Gibson, 2001], seasonal evaporative isotopic enrichment is expected to be subdued in this set of lakes, affording strong potential to preserve longer-term spatial water balance information. Although analysis of isotope data over the course of the 50-day sampling campaign does reveal an apparent overall shift with time (Figure 3), the systematic relation with latitude suggests that this is primarily an artifact of the sampling strategy, imposed by logistical constraints, rather than a reflection of progressive seasonal enrichment. Nevertheless, this could be an important factor in surveys of highly responsive water bodies (especially small headwater lakes), possibly necessitating explicit compensation or detrending if sampling extended over a protracted time period.