See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267866324

Stable isotope estimates of evaporation: Inflow and water residence time for
lakes across the United States as a tool for national lake water quality
assessments

Article in Limnology and Oceanography · November 2014

DOI: 10.4319/lo.2014.59.6.2150

CITATIONS READS

139 843

6 authors, including:

J Renee Brooks John J Gibson

United States Environmental Protection Agency, Corvalls Oregon InnoTech Alberta
125 PUBLICATIONS 7,771 CITATIONS 222 PUBLICATIONS 9,185 CITATIONS

SEE PROFILE SEE PROFILE

Sandra Jean Birks Marc H Weber

Alberta Environment and Protected Areas (and University of Calgary) United States Environmental Protection Agency
123 PUBLICATIONS 4,463 CITATIONS 49 PUBLICATIONS 1,705 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by J Renee Brooks on 29 April 2019.

The user has requested enhancement of the downloaded file.

Limnol. Oceanogr., 59(6), 2014, 2150–2165
E 2014, by the Association for the Sciences of Limnology and Oceanography, Inc.
doi:10.4319/lo.2014.59.6.2150

Stable isotope estimates of evaporation : inflow and water residence time for lakes across
the United States as a tool for national lake water quality assessments
J. Renée Brooks,1,* John J. Gibson,2 S. Jean Birks,3 Marc H. Weber,1 Kent D. Rodecap,1 and
John L. Stoddard 1
1 Western Ecology Division, National Health and Environmental Effects Research Laboratory, U.S. Environmental Protection Agency,
Corvallis, Oregon
2 Alberta Innovates–Technology Futures, Victoria, British Columbia, Canada
3 Alberta Innovates–Technology Futures, Calgary, Alberta, Canada

Abstract
We used water d2H and d18O from ca. 1000 lakes sampled in the 2007 U.S. Environmental Protection Agency’s
National Lakes Assessment (NLA) to assess two hydrological variables—evaporation as a percentage of inflow
(E : I) and water residence time (t) for summer 2007. Using a population survey design, sampled lakes were
distributed across the conterminous U.S., and results were scaled to the inference population (,50,000 U.S.
lakes). These hydrologic variables were related to lake nutrients and biological condition to illustrate their
usefulness in national water quality monitoring efforts. For 50% of lakes, evaporation was , 25% of inflow, with
values ranging up to 113% during the 2007 summer. Residence time was , 0.52 yr for half of the lakes and
, 1.12 yr for 75% of lakes. Categorizing lakes by flow regime, 66.1% of lakes were flow-though lakes (60% or
more of the water flows through the lake, E : I , 0.4), 33.6% were restricted-basin lakes (40% or more of the lake
inflow evaporates, 0.4 , E : I , 1), and , 0.3% were closed basin (all water entering the lake leaves through
evaporation, E : I . 1). While climate patterns drove some of the spatial patterns of E : I and t, variation in lake
depth and watershed size (influencing precipitation volume) were also significant drivers. Lake hydrochemistry
was strongly correlated to E : I and more weakly related to t. Lakes in poor biological condition (based on a
predictive model of planktonic taxa) were significantly more evaporated than lakes in good biological condition.

The importance of lentic freshwater ecosystems in understanding stressors and causes of impairment such as
global-scale biogeochemical and hydrologic cycling has excess nutrients, and including more detailed hydrological
become increasingly apparent (Downing et al. 2006; Cole et indicators in the national survey could help in this latter
al. 2007; Tranvik et al. 2009). Globally, lakes have been goal. The water quality and biological condition of lakes
estimated to number . 64 million and cover . 3.8 million depend not only on local land use and disturbance to the
km2 in area (Downing et al. 2006; McDonald et al. 2012). lake and shores, but also on lake hydrological processes
The role of lakes in these cycles has changed with human that link the surrounding landscape and climate to the lake
activities and is predicted to continue to change with (Fraterrigo and Downing 2008). Hydrological characteris-
climate change and further human activity. However, the tics such as residence time and the proportion of water
data needed to describe the current status of lakes at this flow-through in a lake compared with evaporation have
scale (continent or larger) are limited. One source of been linked to chemical stressors (Pham et al. 2008; Romo
detailed lake information at the national scale comes from et al. 2013). One way to improve the NLA and our
the U.S. Environmental Protection Agency’s (U.S. EPA’s) understanding of lentic systems at the continental scale
National Lakes Assessments (NLAs), which began in 2007 would be to include more detailed measures of lake
with plans to repeat the lake assessment every 5 yr. hydrological characteristics in national surveys.
During the 2007 NLA assessment, a wide variety of Many of the biogeochemical functions that affect
biological, recreational, chemical, and physical indicators stressors and lake biology are strongly affected by lake
were measured at . 1000 lakes across the U.S.A. and were hydrological characteristics (Tranvik et al. 2009). For
used to evaluate the condition of the nation’s lakes (U.S. example, evaporative concentration may increase the
EPA 2009). Using a probabilistic-based survey design, concentration of conservative ions, sometimes with dra-
NLA results were scaled to represent the entire inference matic effects on salinity (Anderson et al. 2001), and is
population of nearly 50,000 lakes (Olsen et al. 2009; Peck et generally associated with higher nutrient concentrations
al. 2013). While the information collected was extensive for (Wolfe et al. 2007; Sokal et al. 2008). Longer residence
water quality and biological diversity, hydrological indica- times increase the cycling and potential retention of
tors were initially limited to lake area and depth, basin biologically active ions, including nutrients like phosphorus
area, and annual precipitation and other climate data. U.S. (Brett and Benjamin 2008; Koiv et al. 2011) and nitrogen
EPA defines lake ecological condition in terms of biological (Kaste et al. 2003; Harrison et al. 2009). Longer residence
indicators; however, U.S. EPA is also interested in times may also increase the sedimentation rates of toxic
metals, reducing the concentrations of total mercury
* Corresponding author: [email protected] (Selvendiran et al. 2009) and heavy metals (Rippey et al.
2150
 U.S. lakes: Evaporation and residence time 2151

2004). The effects that hydrologic functions may have on annual precipitation values compared with rivers, which
potential stressors to biological assemblages underline the were more influenced by isotopic seasonality in precipita-
importance of assessing them in regional and national lake tion. Additionally, extensive surveys such as the NLA are
assessments. This importance will likely increase in the not trying to predict the condition of any particular lake
future as climate change affects rates of lake evaporation but are assessing the distribution of conditions across the
and hydraulic transit time (Malmaeus et al. 2006). Yet, continent for the assessment period: the summer of 2007. In
direct measurements of many hydrologic processes cannot the case of the NLA, a statistical survey design was utilized
be conducted in a spatially extensive survey, where sites are so that inferences across the population of lakes in the U.S.
randomly distributed and therefore often remote and where (e.g., the proportion of lakes with residence times . 1 yr
data collection is constrained by the number of measure- during the assessment period) could be made with known
ments that can be conducted in a single day. confidence from sampling on a single day (Peck et al. 2013).
One approach that has been useful for assessing Thus, point measurements for dynamic parameters such as
hydrological processes in lakes from broad-scale lake lake nutrients and hydrological characteristics are reflective
surveys was to measure the lake water d18O and d2H of a broad range of conditions that exist during the
(Gibson and Edwards 2002; Gibson et al. 2002; Wolfe et al. sampling period but are not meant to characterize any
2007). The theory for estimating evaporation : inflow (E : I) particular lake. Some broader scale regional lake studies
and lake water residence time from stable isotopes has been have successfully estimated hydrological parameters from
well documented and refined over time (Gat 1995; Gibson water isotopes based on single samples from lakes to assess
et al. 2002; Gat 2010). The theory is based on isotopic mass the regional status of lakes (Gibson and Edwards 2002;
balance and a detailed knowledge of isotopic fractionation Wolfe et al. 2007). Using the single isotopic sample
during evaporation of lake water. If the isotopic value of approach, Gibson and Edwards (2002) were able to
water flowing into a lake is known (average isotopic value describe distinct hydroclimatic regimes of lakes adequately
of surface water, groundwater, and direct precipitation), from three distinct biomes in northern Canada: tundra,
then the proportion of water that is evaporated from the boreal forest, and the ecotone between the two. Lakes
lake can be estimated from the evaporative increase in d18O within the boreal forest evaporated significantly greater
and d2H values of the lake water. Most studies where this proportions of the lake inflow compared with lakes in the
isotopic approach has been used involve relatively few lakes arctic tundra. Wolfe et al. (2007) used E : I estimates from
with detailed sampling over time (Froehlich 2000; Gibson single samples obtained from a survey of 57 lakes in
et al. 2002; Stets et al. 2010). Using water isotopes, Pham et northern Alberta to classify lakes into drainage categories
al. (2008) determined how solute concentrations within and found a strong relationship between these categories
lakes were related to lake evaporation and land use and lake chemistry. Using this approach of assessing
activities. Gibson and Reid (2010) illustrated how the lake hydrological status using an isotope mass balance model
isotope models could determine annual changes in the and the hydrogen and oxygen isotopic composition of lake
proportion of water entering a lake that leaves through water determined on a single sample is well-suited to the
evaporation (E : I) and further separate water loss mecha- logistical constraints of the NLA program and can provide
nisms between the lake and the watershed. The isotopic information about the hydrological characteristics of lakes
method for determining basic lake hydrological character- that are contemporaneous with the biological, chemical,
istics is a practical alternative to classical hydrological and physical indicators collected during the survey.
monitoring for understanding water balance effects on Here, we use lake water evaporation theory for water
water quality and biological condition of lakes where isotopic ratios to predict E : I and residence time (t) during
detailed lake hydrological data are not available (Gibson et the 2007 summer assessment period for 1000+ lakes
al. 2002; Wolfe et al. 2007). distributed across the contiguous U.S., and extrapolate
One potential challenge in using the water isotope those results to predict the distribution for the entire
approach for estimating lake hydrological parameters in inference population of lakes (nearly 50,000 lakes) on the
an extensive survey such as the NLA is that only one water basis of an NLA statistically weighted survey design (Peck
sample was collected during the summer, with 10% of lakes et al. 2013). The 2007 NLA survey represents the largest
being sampled twice. The steady state models for estimating spatially extensive survey of lakes in the U.S. and the only
E : I and residence time assume a constant and well-mixed lake survey to have a statistically based design for
volume of water, both of which are not accurate for many extrapolating the results to assess condition of lakes at
lakes. However, the steady-state model predictions provide the continental scale. Our goal was to assess the potential
estimates of E : I and residence time that are representative for isotopic analysis conducted on one water sample from
of conditions at the time of sample collection, even though each lake to be used to generate hydrological data at the
these conditions are dynamic and will change seasonally. continental scale and to provide hydrologic insights to lake
The water isotope signal within a lake represents an water quality and biological condition status.
integrated signal of lake water evaporation over the
residence time of water within the lake and thus is more Methods
reflective of the average lake hydrologic condition with
longer residence time (Gat 1995). Henderson and Shuman National Lakes Assessment—In the summer of 2007, the
(2010) surveyed 100 lakes in the U.S. western mountains U.S. EPA conducted its first NLA as part of the National
and found that lake isotopic values reflected weighted Aquatic Resource Surveys (U.S. EPA 2009). A total of
2152 Brooks et al.

1028 lakes across the contiguous U.S. were selected using a prevent evaporation. Vials were filled to avoid headspace,
probability-based survey design, with 10% being randomly and samples were stored cap side down until analysis.
selected for a second sampling later in the summer (Peck et Laboratory duplicates were also split on 5% of the samples
al. 2013). The survey design means that the results from the upon arriving from the field. In 2010, all samples were
1028 lakes can be extrapolated to the entire inference analyzed for water isotope ratios (d2H, d18O) on a laser
population of lakes found within the U.S. that fit the absorption water vapor isotope spectrometer (Model 908-
following criteria. The inference population included both 0004, Los Gatos Research) located at the Integrated Stable
natural and man-made lakes if they were . 0.04 km2 in Isotope Research Facility at the Western Ecology Division
area, at least 1 m deep, and with a minimum of 0.001 km2 of the U.S. EPA, Corvallis, Oregon. All d2H and d18O values
of open water. Each lake was designated as man-made or were expressed relative to Vienna Standard Mean Ocean
natural on the basis of visual examination of maps; all Water (VSMOW) in parts per thousand (%),
impounded lakes were designated as man-made. The survey
excluded the Great Lakes, the Great Salt Lake, commercial Rsample
d2 H or d18 O~ {1 ð1Þ
treatment ponds, disposal ponds, brackish lakes, and Rstandard
ephemeral lakes. Using the National Hydrography Data-
where R is the ratio of 2H to 1H atoms or 18O to 16O atoms
set, Peck et al. (2013) estimated that 68,223 lakes in the
in the sample and the standard VSMOW. Measurement
contiguous U.S. met these criteria; however, a certain
precision estimates (6 1 standard deviation) were determined
percentage were inaccessible, leaving 49,546 lakes in the
on both field and lab duplicates and were 0.35% and 0.11%,
inference population that the sampled lakes represent.
and 0.27% and 0.10%, for d2H and d18O, respectively. The
Inference lakes were stratified by state and lake size. Each
precision on these duplicate samples were similar to the
lake randomly selected for sampling from the inference
population was assigned a weight that indicated the variance on repeated measures of our internal standards:
number of lakes it represented based on its probability of 0.26% and 0.14% for d2H and d18O, respectively, represent-
being selected from the stratified state-size group. These ing analytical precision of the instrument. The similarity of
weights were applied to our isotope results to scale the precision for field and lab duplicates with analytical precision
results to the entire inference population (49,546 lakes). of the instrument indicates that samples did not evaporate
The NLA was focused on assessing biological condition of during shipping or storage before analysis.
lakes. Two biological indices were developed for NLA: a We calculated deuterium excess (d-excess) as an index of
planktonic observed : expected (O : E) taxa loss model how much evaporation has affected the isotopic value of
(Hawkins 2006) and a sediment diatom index of biological each surface water sample using the following equation
integrity (Stoddard et al. 2008). Details of both of these (Dansgaard 1964; Clark and Fritz 1997):
biological indicators can be found in the NLA technical
d-excess~d2 H{8d18 O ð2Þ
appendix (U.S. EPA 2010). U.S. EPA (2009) classified lakes
into three biological condition categories based on the The d-excess value indicates the influence of kinetic
plankton O : E taxa loss model: good (, 20% taxa loss), fair fractionation (evaporation) compared with equilibrium
(20–39% taxa loss), and poor (. 40% taxa loss). We compare fractionation on water isotopic ratios. The ratio of d2H
our hydrological indicators with these indices to help assess to d18O equilibrium fractionation factors is approximately
the usefulness of our indicators in these national assessments. 8 under standard atmospheric conditions. Ocean water has
an average d-excess of 0, whereas the average d-excess for
Water isotope measurements—Lakes were sampled be- precipitation is 10. In terrestrial environments, surface
tween 08 May 2007 and 18 October 2007, with three lakes waters with d-excess values , 10 are presumed to have
sampled in May, 175 lakes in June, 409 lakes in July, 375 undergone some evaporation. Lower d-excess values in
lakes in August, 157 in September, and 9 lakes in October. lakes generally indicate more evaporation.
At the deepest part of the lake (where lake depth was
measured), a 1-liter sample of water was collected from an Isotopic estimation of E : I and lake water residence time—
integrated sample from the upper 2 m of the lake (or to lake We estimated hydrological metrics from lake water isotopes
depth if shallower than 2 m) into a completely filled by applying steady-state mass balance equations that
cubitainer with a tightly closed cap. This integrated summer assume that lakes were well mixed and maintain a long-
sample of the epilimnion in deep lakes, or integrated sample term constant volume (Gibson and Edwards 2002; Gat
of the whole or nearly whole water column in shallow lakes, 2010). Because lakes are dynamic systems, these assump-
was designed to provide a good representation of the well- tions are usually not valid. However, applying steady-state
mixed portion of the lake for water quality monitoring (U.S. equations to dynamic systems such as lakes provides useful
EPA 2010). Samples were shipped overnight to the U.S. quantitative indicators of hydrological conditions at the
EPA water chemistry laboratory in Corvallis, Oregon, for time of sampling and is more representative of average
standard chemical analysis, including total nitrogen and hydrological characteristics for lakes with longer residence
phosphorus concentration and chlorophyll a (Chl a) times (Gat 2010). We used the following equations:
concentration (U.S. EPA 2009). For 10% of the lakes,
duplicate field samples were collected at the same time and IL ~QL zEL ð3Þ
shipped. After filtering, a subsample was collected in 20-mL
glass vials with polyseal conical inserts within the cap to IL dI ~QL dL zEL dE ð4Þ
 U.S. lakes: Evaporation and residence time 2153

where IL is inflow (combined surface water, groundwater, one that minimized the difference between E : I estimated
and direct precipitation), QL is outflow (combined surface independently from the two isotopes (d18O and d2H), which
water and groundwater), and EL is lake evaporation. The minimized the error associated with estimating dI. We also
isotopic values of inflow, outflow, and evaporation fluxes conducted sensitivity analyses for the variation between
are dI, dL, and dE respectively (which assumes that outflow estimates.
values are equivalent to measured lake values). dL was The isotopic value of evaporating water, dE, was
measured from the lake water sample as described above. estimated using the Craig–Gordon model for open-water
We used precipitation isotopes for dI from Water- evaporation (Craig and Gordon 1965):
Isotopes.org, which uses global precipitation oxygen and
hydrogen isotope data and empirically calculates the long- ðdL {ez Þ=az {hdA {eK
dE ~ ð6Þ
term average monthly and annual isotopic composition of 1{hzeK
precipitation at any location based on algorithms devel- where h is atmospheric relative humidity, dA is the isotopic
oped by Bowen and Wilkinson (2002) and refined by composition of atmospheric vapor, a+ is the equilibrium
Bowen and Revenaugh (2003) and Bowen et al. (2005). fractionation between liquid and vapor (a+ 5 1 + e+). We
This assumes that annual averages of precipitation isotopes assumed that dA was in isotopic equilibrium with evapo-
represent dI, which includes all flow paths of water to the ration flux-weighted precipitation (i.e., annual estimates
lake (surface and groundwater flows and direct precipita- derived from monthly precipitation isotopes weighted by
tion). Clark and Fritz (1997) describe the strong connection monthly potential evapotranspiration [PET]; see below for
between isotopic composition of locally recharged ground- PET details).
water and mean annual precipitation. Using the weighted Combining Eqs. 4 and 6 and substituting QL with IL 2
mean annual isotopic composition of precipitation to EL, we derive the following equation for lake E : I (Gibson
approximate these combined surface and groundwater and Edwards 2002; Gibson and Reid 2010):
inflows is appropriate for most hydrogeological settings,
except perhaps where the lake is sustained by deep regional EL dI {dL
~ ð7Þ
aquifers with isotopic values representative of paleowaters IL m(d {dL )
or is controlled by distant recharge areas with significantly
different isotopic values. Using precipitation isotopes for dI where m~ðh{eK {ez =az Þð1{hzeK Þ{1 and d ~ðhdA z
also assumes all shifts in water isotopes resulting from eK zez =az Þðh{eK {ez =az Þ{1 . We estimated lake water t
evaporation occur in the lake, and not along the flow path from the E : I estimates and annual estimates of PET from
to the lake, which can overestimate E : I values in areas with the lake surface for E in the following equation (Gibson
significant wetlands and other lakes within the watershed. et al. 2002):
We used three methods to determine dI. Using Water-  
Isotopes.org, precipitation isotopes were estimated using E V
t~ ð8Þ
both the lake location (point estimate) and the watershed I E
centroid and mean watershed elevation (watershed esti-
mate). Additionally, we estimated precipitation isotope where V is lake volume. We used the method described by
values by calculating the dual isotope slope of the local Hollister and Milstead (2010) for estimating lake volume,
evaporation line (LEL, slope estimate) and extrapolating which uses maximum depth measurement and a geographic
back from the lake value to the global meteoric water line information system (GIS) layer of lake shoreline. Monthly
(GMWL, d2H 5 8d18O + 10). This LEL slope method climate data (precipitation, dew point, mean maximum and
allows us to estimate inputs for lakes in which dI might not minimum temperatures) for 2007 were obtained for each
be well represented by precipitation isotope, as mentioned lake using the Parameter-elevation Regressions on Inde-
above. The LEL slope (SLEL) was estimated from the pendent Slopes Model algorithm (Daly et al. 2008).
following equation assuming that input waters and Monthly PET was estimated from temperature data using
atmospheric moisture are in isotopic equilibrium (Gat the Hamon equation (mm month21) according to Wolock
2010): and McCabe (1999). E : I and t values were estimated for
both d18O and d2H, and the average value was used.
½eþ zek 2 Because very low values of E : I can lead to unrealistically
SLEL ~ ð5Þ large estimates of I, we substituted annual precipitation for
½eþ zek 18
I when estimates of I were greater than annual precipitation
where e+ is the equilibrium enrichment factor and eK is the in the watershed (, 100 lakes). As an independent estimate
kinetic enrichment factor. The values in the numerator are of t, we used runoff estimates from McCabe and Wolock’s
for d2H, and the denominator for d18O. We estimated e+ (2011) water balance model generated from 2007 precipi-
using temperature data and the equations from Horita and tation and temperature data (tm). The isotope-derived t will
Wesolowski (1994). eK is influenced by relative humidity be denoted tE:I.
and boundary layer conditions and can be estimated by eK All statistical analyses were conducted using R (version
5 CK(1 2 h), where CK was set to 14.2% for oxygen and 2.15.2, 2012-10-26, The R Foundation for Statistical
12.5% for hydrogen (Gibson and Edwards 2002). Because Computing, www.r-project.org). Weighted analysis was
no single method for estimating dI was uniformly good used to analyze and scale the sampled lake data (1028
across the entire range of lakes, the dI value used was the lakes) to the entire inference population of lakes (49,546
2154 Brooks et al.

Fig. 1. Values of d18O and d2H for all lakes sampled in the 2007 National Lakes Assessment
along with frequency distributions for the range of isotope values. The white points are the 10%
of lakes sampled a second time. The line in the dual-isotope plot is the GMWL.

lakes) using the lake weights assigned in the NLA was 24.2% and 239.9%, respectively, for d18O and d2H
probability-based survey design (see above for details). and 23.45% for d-excess. Ninety-five percent of lake
We summarized the data using lake weights with the R waters plotted below the GMWL (d2H 5 10 + 8d18O;
packages plyr and Hmisc. A combination of weighted Fig. 1) indicating the evaporated nature of lakes during the
analysis of variance and weighted linear regressions was summer (d-excess 95th weighted percentile 5 9.6%). All
used to determine trends and significant groupings within lakes with d18O values greater than 23% and d2H values
the data and is specified within the results section. We also greater than 210% were below the GMWL. Lake water
used Kruskal–Wallis and Theil–Sen nonparametric tests in isotopic values were more enriched than estimated local
the R packages pgirmess and zyp to confirm patterns, but precipitation isotopic values, with median LEL of 5.1.
these tests are not available using weighted statistics. The spatial distribution of lake water d18O values
Nitrogen fertilizer loading to the lake watersheds was resembles the spatial pattern of precipitation isotopes
estimated using the data synthesized by Sobota et al. (2013) (Bowen and Wilkinson 2002; Bowen and Revenaugh
based on average (1990–2001) annual U.S. county-level N 2003), with most heavy isotope–enriched lakes located in
fertilizer applications. We provide spatial maps of the the southeast and the most heavy isotope–depleted lakes
isotopes and E : I largely to show general trends rather than located in the western mountain regions (Fig. 2A). The
to model the spatial variation explicitly. Therefore, we d-excess values were highest (indicating low evaporation)
simply used ordinary kriging in ArcGIS (five nearest in the northeastern U.S. and other mountainous regions,
neighbors and no underlying trend) to interpolate values. whereas the lowest (high evaporation) were in the upper
When values are aggregated by region, we used nine Midwest and southwestern U.S. (Fig. 2B). Surprisingly, d-
aggregated ecoregions used in U.S. EPA’s NLA and excess values were also high in the lower Midwest, but 2007
described in the NLA technical appendix (U.S. EPA was a particularly wet summer for that region.
2010; Peck et al. 2013). Whereas most lakes were sampled only once during the
summer, 10% (95 lakes) were sampled twice on different
Results dates and were used to evaluate the influence of seasonal
variation on our hydrologic estimates. The d18O and d2H
In the summer of 2007, lake water isotopes ranged from values of lakes fluctuated seasonally in response to
4.2% to 217.6% for d18O and 21.3% to 2134% for d2H variations in the isotopic composition of precipitation
with d-excess values ranging from 15.6% to 241.2% and snowmelt and because of evaporative enrichment that
(Fig. 1). The median, weighted by sampling probability, occured over the open-water season. In snow-dominated
 U.S. lakes: Evaporation and residence time 2155

Fig. 2. Isocapes of (A) d18O and (B) d-excess of lake waters. Each circle represents a
sampled lake in the 2007 National Lakes Assessment.

portions of the country, this seasonality included minimum In other regions of the country, the seasonal cycle in lake
d18O and d2H values after the influx of snowmelt following isotope values was likely dominated by evaporative
the spring freshet, followed by progressive evaporative enrichment during the summer and early fall. All of the
enrichment over the open-water season (Stets et al. 2010). 2007 samples were collected between May and October,
2156 Brooks et al.

Fig. 3. A comparison between the first and second visit to a lake for (A) d-excess and (B) E : I. The line is the 1 : 1 line. Shading of
circles indicates variation in residence time.

with the majority (70%) collected during July and August. residence times , 0.25 yr. Moreover, the E : I difference
Consequently, these samples should represent intermediate between visits tended to get larger as the time between
evaporative enrichment for each lake. As expected, the repeat sampling dates increased: lakes repeated 100 d after
repeat sampling produced d18O and d2H values that were the initial sampling had a mean difference of 0.1 in E : I.
significantly greater by an average of 0.6% and 3.0%, These repeat measurements indicate that although evapo-
respectively, during the second visit, and d-excess values ration does tend to increase through the summer and
were significantly lower by an average of 1% (Wilcoxon increase more for lakes with short residence times, these
signed-rank test, p , 0.001 for all tests) consistent with the within-lake changes were relatively small compared with
greater evaporative enrichment that would occur as the the range of E : I values for the lake population as a whole.
summer season progressed. However, these shifts between The precision (RMSE) of E : I values was 6 0.065. The
visits were very small relative to the range of values and the signal to noise ratio for E : I was 11.5, indicating that the
variance among sites (Fig. 1). Because all samples were variance among lakes is 11.5 times greater than the
integrated over the upper 2 m at the deepest portion of the variance between visits. Residence time also tended to
lake, within lake variability related to sampling location increase during the summer, with a median increase of
should be minimized. Signal to noise ratios (variance 0.06 yr with a RMSE of 0.23 yr, and a signal to noise ratio
among lakes : mean variance between visits) were 29.3, 72.6, of 10.6. Thus, although both E : I and t tended to increase
and 8.7 for d18O, d2H, and d-excess, respectively, where during the summer, estimating E : I and t from one visit
values . 10 are considered excellent indicators for during the summer was a viable approach for assessing the
detecting differences among sites relative to the variance distribution of the summertime hydrologic status of the
within a site (Kaufmann et al. 2014). Using root mean nation’s lakes for 2007 using the statistical survey
squared errors (RMSE) between visits as a measure of approach.
precision, precision values were 0.8%, 3.8%, and 3.1%, Besides the measured lake water isotope compositions,
whereas the ranges are over 20%, 150%, and 55% for d18O, estimates of E : I are also sensitive to the estimated
d2H and d-excess, respectively. For d-excess, a mean shift parameters used in Eqs. 6 and 7 and, in particular, to
of only 1% was found with repeat values clustering closely estimates of the isotopic value for lake inflows dI, which
around the 1 : 1 line (Fig. 3A). Additionally, the rank order includes all inflows into the lake (surface water, ground-
of lakes between the first and second visit was highly water, and direct precipitation) and assumed that no
correlated (Spearman’s rank correlation r . 0.9 for d- evaporation occurred before the water entered the lake.
excess, E : I, and tE:I). In most cases, these repeat values We estimated the isotopic value of lake water inflow using
were too isotopically similar to determine a reasonable three different methods outlined above. The point and
LEL to compare with the LEL slope estimated by Eq. 5. watershed method estimates of precipitation isotopes from
E : I differences between visits were also small, with a WaterIsotopes.org gave very similar dI estimates of E : I
median increase in E : I of 0.03 and clustering around the (median 0.21 vs. 0.20, mean 0.30 vs. 0.28, standard
1 : 1 line (Fig. 3B). The median E : I difference tended to deviations 0.37 vs. 0.38, respectively), whereas the slope
increase with shorter residence times, increasing from 0.02 method estimates were generally higher (median 0.26, mean
for lakes with residence times . 1 yr to 0.04 for lakes with 0.36, standard deviation 0.31), because dI estimates from
 U.S. lakes: Evaporation and residence time 2157

Table 1. Spearman’s rank correlation (r, upper matrix) and

Pearson’s correlation coefficients (r, lower matrix) between E : I
values estimated using three different methods for selecting the
isotope values for lake inflow. Point method uses the lake location
and elevation; the watershed method uses the watershed centroid
and mean elevation in WaterIsotopes.org for predicting
precipitation isotopes. The slope method uses the estimated
local evaporation line predicted from climate data (see Methods
for details).

Point E : I Watershed E : I Slope E : I

estimates estimates estimates
Point E : I
estimates — 0.94 0.87
Watershed E : I
estimates 0.96 — 0.80
Slope E : I
estimates 0.73 0.84 —

the LEL slope were generally lower than those from

WaterIsotopes.org. Nevertheless, the E : I correlations
between the three estimates were all very high (Table 1),
indicating that although the E : I values may shift some- Fig. 4. Rank correlation between E : I values calculated using
what from method to method, the relative ranking of a either d18O or d2H values. The line is a 1 : 1 line.
particular lake is very similar. Equations 6 and 7 were
solved independently for each isotope. Although the E : I variable explaining variation in E : I after lake depth. The
estimates based on d2H tended to be higher than those spatial distribution of E : I resembled that of d-excess and
based on d18O (median 0.21 vs. 0.18), the overall rank order was largely driven by climate factors, whereas the variation
of lakes was very similar between isotopic estimates within a region was driven by lake-specific factors, such as
(Fig. 4). E : I values used throughout were those based on depth and watershed size (Fig. 7). Low values of E : I were
dI values that gave the most similar E : I values between found in mountain ecoregions with high precipitation and
oxygen and hydrogen isotopes, and dI values were not low evaporation rates: the northern Appalachians ecore-
higher than the measured lake isotope values. We averaged gion with a weighted median E : I value of 0.11, the western
the estimates from the two isotopes for the final E : I values. mountains median of 0.19, and a southern Appalachians
During the summer of 2007, lake E : I values ranged from median of 0.17. Additionally, low values of E : I were found
0 to 1.13, and the distribution was positively skewed with a in the southern and coastal plains ecoregions, with median
median of 0.25 and a mean of 0.32 (Fig. 5A). For half of values of 0.16 and 0.23, respectively. These areas were
the lakes, , 25% of the water entering the lake leaves lower than one might expect as a result of high
through evaporation. Only 25% of lakes have E : I values precipitation in that area for 2007. The xeric ecoregion
. 0.5. The proportion of water that leaves through was also not as high as might be expected based on
evaporation decreases with increasing lake depth (Fig. 6A). precipitation and evaporative demand, with a median value
Shallow lakes (, 2 m measured at the deepest point in the of 0.21 and a mean of 0.28, because basin sizes were an
lake, 22% of lakes) have the widest range of E : I, with a order of magnitude larger in this ecoregion and most xeric
weighted median of 0.42, and have significantly greater E : I region lakes received waters from the southwestern
values than deeper lakes. Lakes with maximum depth mountains, so inflow tends to be high. The three ecoregions
. 10 m (14% of lakes) have a weighted median E : I of 0.10, with the highest weighted median E : I values were the
which was significantly lower than shallower lakes. temperate plains, with a median of 0.51, the northern
Maximum lake depth was more strongly correlated with plains, with 0.42, and the upper Midwest, with 0.32—all
E : I than was lake volume or lake area (r 5 20.41 for ecoregions with relatively high evaporative demand and
depth vs. r 5 20.25 and 20.12 for volume and area, low precipitation.
respectively, Spearman’s rank; all p , 0.001), or even It is important to note that isotope mass balance cannot
volume per area, although this was closely correlated with be used to estimate lake water t for lakes with no detectable
maximum depth (r 5 237, Spearman’s rank; p , 0.001). evaporative enrichment signal in their isotopic composi-
Other important factors determining the variation in E : I tion. Additionally, when E : I estimates were very small,
were climate (precipitation, temperature, and relative calculations of I could be unrealistically large (larger than
humidity) and watershed size. Not surprisingly, the precipitation inputs). When I was greater than precipitation
precipitation volume that falls within a lake’s watershed inputs to the watershed, we substituted annual precipita-
was inversely correlated with E : I because it represented a tion volume for I in estimating t (V:I). For the summer of
rough approximation of I (r 5 234, p , 0.001, Spearman’s 2007, estimates of tE:I range from essentially 0 to just
rank). Precipitation volume was the second most important . 10 yr with two outlier lakes being . 20 yr (21 and 146 yr),
2158 Brooks et al.

Fig. 5. Cumulative and frequency distributions of (A) E : I and (B) lake water residence time scaled to the entire NLA inference
population of lakes within the U.S.

again with a positively skewed distribution (Fig. 5B). We natural and man-made lakes (Fig. 8). E : I alone explains
estimated that 50% of lakes have a residence time of 24% of the variation in total nitrogen (log scale) within
, 0.52 yr and a mean value of slightly , 1 yr. Ninety-five lakes across the U.S. (weighted linear regression, R2adjusted
percent of lakes had water residence times , 3.7 yr in the 5 0.24, F1,1027 5 334), which was a similar amount of
summer of 2007. Estimates of tE:I correlated with tm variance explained as fertilizer loading onto the landscape.
estimates based on McCabe and Wolock’s mass balance Together, fertilizer loading and E : I explain 41% of the
runoff model (r 5 0.70, Spearman’s rank; p , 0.001). variation in lake nitrogen concentration (weighted linear
However, tm had a lower mean and median (0.8 yr and regression, R2adjusted 5 0.41, F1,1026 5 356). The relation-
0.26 yr, respectively). ship between lake nitrogen concentration (log scale) and
Of the driving factors that could explain the variance in E : I was stronger for natural lakes, with a slope of 3.3 (3.0–
lake water residence time (tE:I), lake depth and the volume of 3.6, 95% confidence interval, Theil–Sen estimate of slope)
precipitation that falls within the watershed had the compared with man-made lakes with a slope of 2.1 (1.7–
strongest correlation (r 5 0.50 and 20.31, respectively, 2.4, 95% confidence interval, Theil–Sen estimate of slope).
Spearman’s rank; p , 0.001). We excluded driving factors E : I was also positively correlated with total phosphorus
that were used to estimate tE:I: E : I, lake volume, and lake concentration and Chl a content (Table 2). Lake water
evaporation (Eq. 8). Precipitation volume is a close residence times (both tE:I and tm) were negatively
approximation of I and is influenced by the size of the correlated with chemical concentrations, but the correla-
watershed and precipitation amount. Interestingly, the tions were much weaker than for E : I. The negative
correlation between tE:I and lake depth was the strongest correlations were also stronger with the isotope-derived
of all climate or lake parameter correlations. Lake water tE:I compared with the modeled tm.
residence time increased with maximum lake depth We also found that lakes in poor biological condition, as
(Fig. 6B). Lakes shallower than 2 m had a mean and median determined by the NLA, had higher E : I values than lakes in
tE:I of 0.37 and 0.34 yr, respectively, whereas the deepest good biological condition (Fig. 9; Table 2). As shown earlier
lakes (. 20 m) had a mean tE:I of 3.8 yr and a median of (Fig. 7), E : I varied by region, with the plains and coastal
2.25 yr. Because of the strong influence of lake depth and areas being more isotopically enriched than either the
watershed area (through its effect on precipitation volume) western region or eastern highlands. Consequently, we split
on tE:I, the spatial patterns for tE:I were not as robust as for these three aggregated ecoregions for comparing lakes with
E : I, but some trends were noted. Lake water residence times different biological condition. Within each of the three
were the lowest in the coastal plains ecoregion, with a ecoregions, lakes in poor biological condition were signifi-
median of 0.17 yr and a mean of 0.27 yr. The longest cantly more evaporated than lakes in good condition where
residence times were located in the upper Midwest, with a condition was based on O : E plankton species (Fig. 9). E : I
median value of 1 yr and a mean value of 1.6 yr. was negatively correlated with both indicators of biological
The chemical condition of lakes was strongly related to condition used in the NLA (Table 2). Lake water residence
the isotopic estimates of E : I (Table 2). Total nitrogen times were not consistently correlated with the biological
concentration increased with increasing E : I for both indices, but tended to be weakly positively correlated.
 U.S. lakes: Evaporation and residence time 2159

water residence time, both variables are dynamic within

lakes; thus, these estimates provide a snapshot of the
hydrological condition of lakes within the U.S. for the
summer of 2007. While it would be problematic to estimate
the annual average hydrologic condition of individual lakes
based on single samples used in the NLA, the seasonal
variability in E : I and t becomes part of the error around
the estimated population distribution using the NLA
survey design (Peck et al. 2013). The summertime
distributions of both E : I and tE:I were positively skewed
(Fig. 5). Seventy-five percent of lakes have greater flow-
through than evaporation (E : I,0.5), and residence time of
, 1 yr. Regional differences in E : I and tE:I were driven
primarily by climate (Fig. 7), and we also found variation
relating to depth of the lake (Fig. 6), with deeper lakes
having lower evaporation and longer residence times than
shallower lakes. Watershed area was also an important
driver of both variables through its influence on the volume
of water entering a lake. Lake nitrogen concentration was
strongly related to E : I (Fig. 8; Table 2), with higher
nitrogen concentrations found in lakes with greater E : I.
Lake nitrogen concentration was negatively correlated to
tE:I, although more weakly than E : I. We also found an
association of E : I with biological condition of a lake
(Fig. 9), likely related to the nutrient relationship because
nutrients can be stressors to lake biology (Van Sickle 2013).
Our results demonstrate that including lake water isotope
measurements in large-scale, spatially extensive monitoring
programs is a practical and inexpensive way to improve our
understanding of lake hydrological characteristics at the
time of assessment and aids in understanding the reasons
for lake impairment at the national scale.
Regardless of the many assumptions, estimating E : I
from isotopes measured in a single representative sample
provides a good first approximation for lake E : I in the
context of the national assessment, and represents the
distribution of hydrologic conditions for lakes in the
Fig. 6. The box plots of (A) E : I values and (B) lake water summer of 2007 within the U.S. It is important to note
residence time (tE:I) separated by maximum lake depth. The box that these E : I and t distributions do not represent mean
represents the 25th to 75th percentile, and the center band is the annual values, but summertime values in 2007 when the
median value. Whiskers represent the 5th and 95th percentiles.
assessment was made. E : I values do tend to increase
The distributions have been scaled to the entire inference
population of lakes in the U.S. (, 50,000) by using the sampling seasonally as evaporative demand and precipitation change
probability weights (see Methods for details). Depth categories (Gibson et al. 2002; Gibson and Reid 2010), but the degree
with different letters are significantly different from each other (a of change we found within a lake over the summer
5 0.05, weighted analysis of variance, partial F-test, and sampling period was relatively small compared with the
nonweighted, nonparametric Kruskal–Wallis rank sums test; distribution of E : I in lakes across the nation (Fig. 3), with
results were the same for both tests). a signal to noise ratio of 11.5. With E : I precision of 0.065
(RMSE), the minimum E : I difference needed between two
lakes based on one sample before the lakes can be
Discussion considered different in summertime E : I was estimated as
1.96(RMSE)(2n)K 5 2.77(RMSE) (Kaufmann et al. 2014),
By analyzing d18O and d2H of lake water samples which gives the 95% confidence interval (0.18). Thus, we
collected as part of the U.S. EPA’s 2007 NLA, we found that the relative ranking of a lake’s E : I within the
estimated summertime E : I and residence time for . 1000 national assessment was robust over the summer sampling
lakes distributed across the nation. We scaled those results season. Another sensitive parameter was the isotopic value
to estimate the 2007 summertime distribution of these of water flowing into a lake (Wolfe et al. 2007; Yuan et al.
parameters for the entire inference population of nearly 2011). The best method might be to have direct isotopic
50,000 lakes based on the NLA probabilistic survey design measures of surface waters feeding a lake, but this was not
(Olsen et al. 2009; Peck et al. 2013). Although isotopes do part of the 2007 NLA sampling and would not account for
provide estimates of E : I and tE:I that are integrated over potential groundwater inflow. Use of weighted mean
2160 Brooks et al.

Fig. 7. Map of lake E : I.

annual precipitation not only represents the immediate WaterIsotopes.org produced similar E : I values, the LEL
direct inputs to lakes through precipitation and surface slope estimation method using Eq. 5 produced greater E : I
flows, but should also be representative of groundwater values, because this method produced more depleted dI
inputs in most cases because groundwater is mainly derived values than WaterIsotopes.org. Nevertheless, the relative
from local recharge by precipitation (Clark and Fritz 1997). rank of a lake was very similar regardless of the method
Estimates of precipitation isotopes was the most logical used for dI (Table 2). Another approach to reduce
way to estimate dI, and the estimates generated by uncertainty was making input parameter assumptions that
algorithms like those used by WaterIsotopes.org are reduced the difference between E : I estimated for d18O and
continuing to improve as more precipitation datasets d2H. For example, we assumed that atmospheric moisture
become available and as conceptual and numerical models was in isotopic equilibrium with evaporative flux-weighted
of climate controls on the isotopic composition of precipitation because this assumption lowered the differ-
precipitation continue to improve. The LEL slope method ence between the oxygen and hydrogen E : I estimates,
did help account for lakes in which estimates of precipita- compared with when atmospheric moisture was assumed to
tion isotopes from WaterIsotopes.org did not adequately be in isotopic equilibrium with precipitation-weighted
reflect dI. However, using precipitation isotope estimates precipitation. Other studies have found similar results
assumes that no evaporation occurs prior to the water concerning dA equilibrium (Wolfe et al. 2007; Gibson et al.
entering the lake, which would overestimate E : I values in 2008). Although these assumptions may shift E : I estimates
areas with wetlands or other lakes along the flow path. We for a given lake, they did little to the rank order of lakes
approached estimating dI by using three methods (see across the nation (Fig. 4), giving us confidence in these E : I
Methods for details). Whereas the two estimates from estimates based on a single representative and integrated

Table 2. Spearman’s rank correlation coefficients (%) between hydrological measures and
water quality parameters. tE:I and tm represent residence time estimated using stable isotopes and
runoff modeling, respectively. Correlations in bold are significant (p , 0.01).

N(mg L21) P(mg L21) Chl a(mg L21) Plankton O : E Diatom IBI*
E:I 53.0 38.8 32.0 226.1 228.9
Residence time (tE:I) 28.8 225.7 227.5 6.3 5.0
Residence time (tm) 26.2 220.0 221.8 23.4 7.9
* Diatom IBI, index of biological integrity.
 U.S. lakes: Evaporation and residence time 2161

Fig. 8. The relationship between lake E : I values and the natural log of total nitrogen content in the lake for man-made and natural
lakes. Lakes were categorized on the basis of nitrogen fertilizer loading to the watersheds (based on data synthesized by Sobota et al.
[2013]). The lines are linear regression using lake weights to scale results to the entire population of lakes. Grey areas around lines
represent 95% confidence intervals for the regression.

Fig. 9. Box plot distributions of E : I between lake condition classes based on O : E plankton
species for three combined ecoregions. The distributions have been scaled to the entire lake
population in the U.S. by using the sampling probability weights (see Methods for details). E : I
values for condition classes with different letters are significantly different from each other (a 5
0.05, weighted analysis of variance, partial F-tests, and nonweighted, nonparametric Kruskal–
Wallis rank sums test; results were the same for both tests).
2162 Brooks et al.

sample from a lake for the purposes of a spatially extensive that . 60% of lakes are flow-through, and the restricted
national lakes assessment. flow category varies around 35% of lakes for the summer of
Estimates of residence time introduce more potential 2007. One reason for the high proportion of flow-through
sources of error because we needed to estimate lake volume lakes was that 42% of lakes were man-made lakes and
and annual evaporation from the lake (Eq. 8). Lake volume reservoirs, of which 80% were flow-through. For natural
can be particularly challenging to estimate (Sobek et al. lakes, 54% were flow-though, and 45% had restricted flow.
2011). The method developed by Hollister and Milstead This difference in natural and man-made lakes could also
(2010) used maximum depth measurements made in the be influencing our geographical distribution of E : I because
field (maximum measurable depth was 50 m) and GIS the southern half of the U.S. is dominated by man-made
shoreline information, so although these estimates are more lakes (U.S. EPA 2009). Similar to our findings that E : I
accurate than techniques that assume a conical shape, they varied with climatic region, Gibson and Edwards (2002)
still introduce a level of error. Our estimate of precision found E : I differences between ecoclimatic regions of
based on repeat visits was 6 0.23 yr (RMSE), and the northern Canada, with the coldest tundra zone having the
minimum difference needed to detect a difference between lowest E : I values and the warmer boreal forest region
two lakes based on one sample was 0.64 yr. Our isotope- having the highest. Within the U.S., although temperature
based estimates of t were correlated with more traditional was an important factor, it was not as important as was
estimates of t using modeled runoff, but the isotope aridity, with relative humidity and annual precipitation
method tended to predict longer residence times than the being better predictors of E : I and tE:I variation than
runoff method. In comparing tE:I and tm with previous temperature. The correlation of E : I with relative humidity
regional estimates of t calculated from measured lake and precipitation was quite logical because these were the
bathymetry and regional runoff values (Lindthurst et al. major drivers of evaporation and inflow to a lake.
1986), the values are similar for the northeastern U.S., with Lake depth was the most important lake characteristic
median tE:I and tm values of 0.27 and 0.21 yr, respectively, that related to variation in E : I and tE:I (Fig. 6), followed
compared with 0.20 yr for U.S. EPA’s 1986 regional closely by the volume of precipitation that falls within a
estimate. Median values diverged more for t estimates for lake’s watershed. We are not aware of other studies that
the upper Midwest, with median tE:I and tm values of 1.04 report on the influence of lake depth on E : I, but Noges
and 0.69 yr, respectively, compared with the 1986 estimate (2009) found that t and depth were positively correlated
of 0.48 yr. for European lakes. It does seem logical that lake depth
We found that lake water isotopes, E : I and tE:I values would have a strong influence on these parameters,
varied with climatic region across the U.S. (Figs. 2, 7). Our particularly for residence time, because lake volume was
lake d18O and d2H values ranged approximately two thirds included in the calculation (Eq. 8) and lake depth was a
the range of a global survey of large lakes (Jasechko et al. parameter used to estimate volume (Hollister and Milstead
2013), with our survey being more limited on the enriched 2010). Shallow lakes have shorter residence times and
end of the range. Not surprisingly, the d18O spatial pattern greater evaporation compared with deeper lakes (Fig. 6).
superficially reflected the spatial pattern of precipitation We were surprised that lake area and lake volume were
and river waters across the U.S. (Kendall and Coplen 2001; not better associated with E : I and that depth was a better
Bowen and Wilkinson 2002; Bowen and Revenaugh 2003), predictor than lake volume per unit area, although they
but with significantly enriched values in the southwest and were closely correlated. Lake area and depth were
central to upper Midwest, where E : I was highest. As might similarly important for tE:I. The observation that deep
be expected given the spatial extent of our survey, the range lakes had the lowest level of E : I indicated that lake
of E : I values estimated for the contiguous U.S. (0–1) was stratification might not have had a large affect on water
greater than those observed in regional studies such as for isotope values, even though the water samples were
northern Canada, where 255 lakes ranged from 0 to 0.7 collected as a composite of the upper 2 m at the deepest
with a median of 0.11 (Gibson and Edwards 2002) and point in the lake. Stratification would tend to concentrate
from the Tibetan Plateau, were values ranged from 0.1 to the water enriched through evaporation in the epilimnion
0.75 with a mean of 0.52 for 27 lakes (Yuan et al. 2011). and thus overestimate E : I relative to nonstratified lakes.
However, Wolfe et al. (2007) found a much broader range We found that 85% of lakes deeper than 10 m were flow-
in the Peace–Athabasca Delta in Alberta, Canada, ranging through lakes (E : I , 0.4), whereas for lakes shallower
from 0 to . 2 with a median of 0.7. They classified their than 2 m, only 40% were flow-through. Precipitation
lakes into flow-through lakes, where 60% or more of the volume is negatively correlated with both E : I and tE:I
lake inflow flows back out of the lake (E : I , 0.4); because it is closely related to lake inflow (I), which is in
restricted-drainage lakes, where 40% or more of lake inflow the denominator of both variables.
leaves through evaporation (0.4 # E : I , 1); and closed- Total nitrogen (TN, mg L21) has been found to have a
basin lakes, where all lake inflow leaves through evapora- high relative risk to lake biological condition (U.S. EPA
tion (E : I $ 1). Lakes with E : I values . 1 are losing lake 2009; Van Sickle 2013), so hydrologic mechanisms that
volume. Using those definitions, our survey of lakes found alter TN concentration within lakes would be of interest to
66.1% of lakes within the U.S. were flow-though lakes, the U.S. EPA. We found a strong positive correlation
33.6% were restricted-basin lakes, and , 0.3% were closed between lake E : I and TN (Fig. 8), which was as strong as
basin. Although these proportions shift slightly depending fertilizer loading within the watershed for explaining
on the method used to predict dI, all three methods predict variation in lake TN. E : I integrates many important
 U.S. lakes: Evaporation and residence time 2163

climate drivers, such as precipitation and evaporation, and groups (Becker et al. 2010; Xiao et al. 2011; Rigosi and
hydrological processes, such as catchment runoff, ground- Rueda 2012). This link between biological condition and
water inflows, and lake discharge, all of which can E : I could be through the water chemistry connection
influence the biological activity and geochemical processes described above or directly related to water mixing and
that can occur within the lake to alter the concentrations of flow-through, but because the two factors covary, it would
biologically active dissolved solutes like carbon, nitrogen, be difficult to separate in the NLA (Van Sickle 2013).
and phosphorus (Fraterrigo and Downing 2008). Changes Factors such as lake depth, outflow regime, and lake
in hydrological cycling that drive E : I can have profound mixing have all been directly related to plankton assem-
effects on lake carbon budget (Cardille et al. 2009) and blages (Becker et al. 2010; Xiao et al. 2011; Rigosi and
likely nitrogen budgets (Jeppesen et al. 2011). Changes in Rueda 2012) and were all related to E : I within a lake.
the proportion of lake inflow that leaves as evaporation While we found weak correlations between residence time
indicate not only evaporative effects of concentrating and indicators of biological condition (Table 2), others
solutes, but also reflect lake energy balances that would have found links between residence time and bacterial
influence internal processing of biologically active nutri- contaminants that influence human health (Romo et al.
ents. We found total phosphorus and Chl a were also 2013) and that were used as recreation indicators within the
correlated with E : I (Table 2). Other studies have also NLA. For example, Romo et al (2013) noted that longer
found a close relationship between evaporation and water water residence times caused by dryer conditions were
chemistry in lakes (Wolfe et al. 2007; Pham et al. 2008; associated with increased toxic cyanobacteria biomass.
Sokal et al. 2008). Yuan et al. (2011) found evaporation, as Thus, lake hydrological parameters can be helpful in
indicated by d-excess, was strongly correlated to total understanding the causes behind poor biological condition
dissolved solutes in lakes from the Tibetan Plateau (r 5 and water quality in lakes.
20.69), although in our study, d-excess was not as powerful Lake hydrologic processes are fundamental in under-
a predictor as E : I for lakes in the U.S. Pham et al (2008) standing lake biogeochemical cycling and the biological
also found evaporation to be a strong driver of solutes in condition in lakes (Sokal et al. 2008; Tranvik et al. 2009;
lakes in the northern Great Plains of Canada. Wolfe et al. Becker et al. 2010). Our study indicated that E : I values
(2007) found that flow-through lakes had lower concentra- estimated from water isotope ratios determined on a single
tions of a wide range of solutes including TN, dissolved representative water sample collected from each lake were
organic carbon, and phosphorus. It is also possible that very useful indicators of hydrological condition for lakes
nutrient inputs via groundwater exchange may be more evaluated in the 2007 NLA and were related to both
important in lakes with a high degree of closure, which nutrient concentrations and biological condition. The
might explain the increase in TN with E : I, although this derived values of E : I were more clearly correlated with
remains to be investigated. Although more research should lake chemistry and biological condition than were the
be done to determine exact linkages, our results clearly parent isotope values from which they were derived, and
indicate that E : I does indicate processes that influence lake E : I values are more easily interpreted in a hydrologic
nitrogen concentrations (Fig. 8). sense, justifying the need for quantitative modeling used to
Both estimates of t were negatively correlated with lake derive E : I. Using the NLA survey design, these results
chemistry (Table 2), but the correlations were not as strong have been scaled to represent the entire inference popula-
as they were with E : I. (Table 2). Other studies have found tion of 50,000 lakes included in the 2007 NLA, thus
residence time to be an important factor controlling lake representing the largest survey of E : I and tE:I ever
chemistry (Noges 2009; Koiv et al. 2011; Finlay et al. 2013). conducted. Our results illustrate that E : I and tE:I modeling
The negative correlation with nutrient concentrations is based on isotopic composition provide an excellent tool
expected, arising from the increase in retention rates for approximating those hydrological characteristics across
associated with longer residence times, as observed in a wide range of lake and reservoir types and climate
several other studies (Kaste et al. 2003; Harrison et al. settings.
2009). Total phosphorus concentration was also found to
be negatively correlated with hydraulic residence time Acknowledgments
(Brett and Benjamin 2008). Because lake hydrologic We thank Gabriel Bowen for precipitation isotope estimates
processes can either raise (i.e., by increasing E : I) or lower from WaterIsotopes.org, Jeff Hollister for estimating lake
(i.e., through longer residence times) nutrient concentra- volumes, and William Rugh for overseeing the Integrated Stable
tions, they should be considered carefully when evaluating Isotope Research Facility at the Western Ecology Division of the
potential causes for elevated lake nutrient levels. U.S. EPA, Corvallis, Oregon. We also thank the many dedicated
Our results indicate E : I behaved similarly to chemical people who designed, collected, analyzed, and interpreted data
and physical stressors, at least in its relationship to lake from the National Lakes Assessment, in particular David Peck,
Tony Olsen, Steve Paulsen, and Amina Pollard. Thanks for
biological condition. We found that lakes in poor
reviews from Emily Standley, Robert Ozretich, and three
biological condition had greater E : I values than lakes in anonymous reviewers.
good biological condition (Fig. 9). The NLA used an index This work was supported by the U.S. EPA. This manuscript
of phytoplankton and zooplankton taxon loss (O : E) as its has been subjected to the U.S. EPA’s peer and administrative
primary biological indicator for determining lake condition review, and it has been approved for publication as a U.S. EPA
(U.S. EPA 2009), and hydrological controls have often document. Mention of trade names or commercial products does
been linked with plankton assemblages and functional not constitute endorsement or recommendation for use.
2164 Brooks et al.

References ———, AND T. W. D. EDWARDS. 2002. Regional water balance

trends and evaporation–transpiration partitioning from a
ANDERSON, N. J., R. HARRIMAN, D. B. RYVES, AND S. T. PATRICK. stable isotope survey of lakes in northern Canada. Glob.
2001. Dominant factors controlling variability in the ionic Biogeochem. Cycles 16: 1026, doi:10.1029/2001GB001839
composition of West Greenland Lakes. Arct. Antarct. Alp. ———, E. E. PREPAS, AND P. MCEACHERN. 2002. Quantitative
Res. 33: 418–425, doi:10.2307/1552551 comparison of lake throughflow, residency, and catchment
BECKER, V., L. CAPUTO, J. ORDONEZ, R. MARCE, J. ARMENGOL, runoff using stable isotopes: Modelling and results from a
L. O. CROSSETTI, AND V. L. M. HUSZAR. 2010. Driving factors regional survey of Boreal lakes. J. Hydrol. 262: 128–144,
of the phytoplankton functional groups in a deep Mediter- doi:10.1016/S0022-1694(02)00022-7
ranean reservoir. Water Res. 44: 3345–3354, doi:10.1016/j. ———, AND R. REID. 2010. Stable isotope fingerprint of open-
watres.2010.03.018 water evaporation losses and effective drainage area fluctu-
BOWEN, G. J., AND J. REVENAUGH. 2003. Interpolating the isotopic ations in a subarctic shield watershed. J. Hydrol. 381:
composition of modern meteoric precipitation. Water Resour. 142–150, doi:10.1016/j.jhydrol.2009.11.036
Res. 39: 1299, doi:10.1029/2003WR002086 HARRISON, J. A., AND OTHERS. 2009. The regional and global
———, L. WASSENAAR, AND K. A. HOBSON. 2005. Global application significance of nitrogen removal in lakes and reservoirs.
of stable hydrogen and oxygen isotopes to wildlife forensics. Biogeochemistry 93: 143–157, doi:10.1007/s10533-008-9272-x
Oecologia 143: 337–348, doi:10.1007/s00442-004-1813-y HAWKINS, C. P. 2006. Quantifying biological integrity by
———, AND B. H. WILKINSON. 2002. Spatial distribution of d18O taxonomic completeness: Its utility in regional and global
in meteoric precipitation. Geology 30: 315–318, doi:10.1130/ assessments. Ecol. Apps. 16: 1277–1294, doi:10.1890/1051-
0091-7613(2002)030,0315:SDOOIM.2.0.CO;2 0761(2006)016[1277:QBIBTC]2.0.CO;2
BRETT, M. T., AND M. M. BENJAMIN. 2008. A review and HENDERSON, A. K., AND B. N. SHUMAN. 2010. Differing controls
reassessment of lake phosphorus retention and the nutrient on river- and lake-water hydrogen and oxygen isotopic values
loading concept. Freshw. Biol. 53: 194–211. in the western United States. Hydrol. Process. 24: 3894–3906,
CARDILLE, J. A., S. R. CARPENTER, J. A. FOLEY, P. C. HANSON, doi:10.1002/hyp.7824
M. G. TURNER, AND J. A. VANO. 2009. Climate change and lakes: HOLLISTER, J., AND W. B. MILSTEAD. 2010. Using GIS to estimate
Estimating sensitivities of water and carbon budgets. J. Geophys. lake volume from limited data. Lake Reservoir Manag. 26:
Res. Biogeosci. 114: G03011, doi:10.1029/2008JG000891 194–199, doi:10.1080/07438141.2010.504321
CLARK, I. D., AND P. FRITZ. 1997. Environmental isotopes in HORITA, J., AND D. WESOLOWSKI. 1994. Liquid–vapour fraction-
hydrogeology. Lewis. ation of oxygen and hydrogen isotopes of water from the
COLE, J. J., AND OTHERS. 2007. Plumbing the global carbon cycle: freezing to the critical temperature. Geochim. Cosmochim.
Integrating inland waters into the terrestrial carbon budget. Acta 58: 3425–3437, doi:10.1016/0016-7037(94)90096-5
Ecosystems 10: 171–184, doi:10.1007/s10021-006-9013-8 JASECHKO, S., Z. SHARP, J. J. GIBSON, S. J. BIRKS, Y. YI, AND P. J.
CRAIG, H., AND L. I. GORDON. 1965. Deuterium and oxygen-18 FAWCETT. 2013. Terrestrial water fluxes dominated by
variation in the ocean and the marine atmosphere, p. 9–130. transpiration. Nature 496: 347–350, doi:10.1038/nature11983
In E. Tongiorgi [ed.], Proceedings of a conference on stable JEPPESEN, E., AND OTHERS. 2011. Climate change effects on
isotopes in oceanographic studies and paleotemperatures. nitrogen loading from cultivated catchments in Europe:
Lab. di Geologia Nucleare. Implications for nitrogen retention, ecological state of lakes
DALY, C., AND OTHERS. 2008. Physiographically sensitive mapping and adaptation. Hydrobiologia 663: 1–21, doi:10.1007/
of climatological temperature and precipitation across the s10750-010-0547-6
conterminous United States. Int. J. Climatol. 28: 2031–2064, KASTE, O., J. L. STODDARD, AND A. HENRIKSEN. 2003. Implication
doi:10.1002/joc.1688 of lake water residence time on the classification of
DANSGAARD, W. 1964. Stable isotopes in precipitation. Tellus 4: Norwegian surface water sites into progressive stages of
436–468, doi:10.1111/j.2153-3490.1964.tb00181.x nitrogen saturation. Water Air Soil Pollut. 142: 409–424,
DOWNING, J. A., AND OTHERS. 2006. The global abundance and size doi:10.1023/A:1022015814800
distribution of lakes, ponds, and impoundments. Limnol. KAUFMANN, P. R., R. M. HUGHES, J. VAN SICKLE, T. R. WHITTIER,
Oceanogr. 51: 2388–2397, doi:10.4319/lo.2006.51.5.2388 C. W. SEELIGER, AND S. G. PAULSEN. 2014. Lakeshore and
FINLAY, J. C., G. E. SMALL, AND R. W. STERNER. 2013. Human littoral physical habitat structure: A field survey method and
influences on nitrogen removal in lakes. Science 342: 247–250, its precision. Lake Reservoir Manag. 30: 157–176,
doi:10.1126/science.1242575 doi:10.1080/10402381.2013.877543
FRATERRIGO, J. M., AND D. J. DOWNING. 2008. The influence of KENDALL, C., AND T. B. COPLEN. 2001. Distribution of oxygen-18
land use on lake nutrients varies with watershed transport and deuterium in river waters across the United States.
capacity. Ecosystems 11: 1021–1034, doi:10.1007/s10021-008- Hydrol. Process. 15: 1363–1393, doi:10.1002/hyp.217
9176-6 KOIV, T., T. NOGES, AND A. LAAS. 2011. Phosphorus retention as a
FROEHLICH, K. 2000. Evaluating the water balance of inland seas function of external loading, hydraulic turnover time, area
using isotopic tracers: The Caspian Sea experience. Hydrol. and relative depth in 54 lakes and reservoirs. Hydrobiologia
Process. 14: 1371–1383, doi:10.1002/1099-1085(20000615)14: 660: 105–115, doi:10.1007/s10750-010-0411-8
8,1371::AID-HYP141.3.0.CO;2-T LINDTHURST, R. A., D. H. LANDERS, J. M. EILERS, D. F. BRAKKE,
GAT, J. 1995. Stable isotopes of fresh and saline lakes, p. 139–165. W. S. OVERTON, E. P. MEIER, AND R. E. CROWE. 1986.
In A. Lerman, D. Imboden, and J. Gat [eds.], Physics and Characteristics of lakes in the eastern United States.
chemistry of lakes. Springer-Verlag. Volume I. Population descriptions and physico-chemical
———. 2010. Isotope hydrology: A study of the water cycle. relationships. EPA/600/4-86/007a. U.S. Environmental Pro-
Imperial College Press. tection Agency.
GIBSON, J. J., S. J. BIRKS, AND T. W. D. EDWARDS. 2008. Global MALMAEUS, J. M., T. BLENCKNER, H. MARKENSTEN, AND I.
prediction of dA and d2H-d18O evaporation slopes for lakes PERSSON. 2006. Lake phosphorus dynamics and climate
and soil water accounting for seasonality. Glob. Biogeochem. warming: A mechanistic model approach. Ecol. Model. 190:
Cycles 22: GB2031, doi:10.1029/2007GB002997 1–14, doi:10.1016/j.ecolmodel.2005.03.017
 U.S. lakes: Evaporation and residence time 2165

MCCABE, G. J., AND D. M. WOLOCK. 2011. Independent effects of STETS, E. G., T. C. WINTER, D. O. ROSENBERRY, AND R. G.
temperature and precipitation on modeled runoff in the STRIEGL. 2010. Quantification of surface water and ground-
conterminous United States. Water Resour. Res. 47: W11522, water flows to open- and closed-basin lakes in a headwaters
doi:10.1029/2011WR010630 watershed using a descriptive oxygen stable isotope model.
MCDONALD, C. P., J. A. ROVER, E. G. STETS, AND R. G. STRIEGL. Water Resour. Res. 46: W03515, doi:10.1029/2009WR007793
2012. The regional abundance and size distribution of lakes STODDARD, J. L., A. T. HERLIHY, D. V. PECK, R. M. HUGHES, T. R.
and reservoirs in the United States and implications for WHITTIER, AND E. TARQUINIO. 2008. A process for creating
estimates of global lake extent. Limnol. Oceanogr. 57: multimetric indices for large-scale aquatic surveys. J. N. Am.
597–606, doi:10.4319/lo.2012.57.2.0597 Benthol. Soc. 27: 878–891, doi:10.1899/08-053.1
NOGES, T. 2009. Relationships between morphometry, geogra- TRANVIK, L. J., AND OTHERS. 2009. Lakes and reservoirs as
phic location and water quality parameters of European regulators of carbon cycling and climate. Limnol. Oceanogr.
lakes. Hydrobiologia 633: 33–43, doi:10.1007/s10750-009- 54: 2298–2314, doi:10.4319/lo.2009.54.6_part_2.2298
9874-x [U.S. EPA] U.S. ENVIRONMENTAL PROTECTION AGENCY. 2009.
OLSEN, A. R., L. L. STAHL, B. D. SNYDER, AND J. L. PITT. 2009. National Lakes Assessment: A collabotative survey of the
Survey design for lakes and reservoirs in the United States to nations’s lakes. EPA 841-R-09-001. U. S. Environmental
assess contaminants in fish tissue. Environ. Monit. Assess. Protection Agency, Office of Water and Office of Research
150: 91–100, doi:10.1007/s10661-008-0685-8 and Development.
PECK, D. V., A. R. OLSEN, M. H. WEBER, S. G. PAULSEN, C. ———. 2010. National Lakes Assessment: Technical appendix
PETERSON, AND S. M. HOLDSWORTH. 2013. Survey design and
data analysis approach. EPA/841/R/09/001a. U.S. Environ-
extent estimates for the National Lakes Assessment. Freshw.
mental Protection Agency, Office of Water and Office of
Sci. 32: 1231–1245, doi:10.1899/11-075.1
Research and Development.
PHAM, S. V., P. R. LEAVITT, S. MCGOWAN, AND P. PERES-NETO.
VAN SICKLE, J. 2013. Estimating the risks of multiple, covarying
2008. Spatial variability of climate and land-use effects on
lakes of the northern Great Plains. Limnol. Oceanogr. 53: stressors in the National Lakes Assessment. Freshw. Sci. 32:
728–742, doi:10.4319/lo.2008.53.2.0728 204–216, doi:10.1899/11-050.1
RIGOSI, A., AND F. J. RUEDA. 2012. Hydraulic control of short- WOLFE, B. B., AND OTHERS. 2007. Classification of hydrological
term successional changes in the phytoplankton assemblage in regimes of northern floodplain basins (Peace–Athabasca Delta,
stratified reservoirs. Ecol. Eng. 44: 216–226, doi:10.1016/ Canada) from analysis of stable isotopes (d18O, d2H) and water
j.ecoleng.2012.04.012 chemistry. Hydrol. Process. 21: 151–168, doi:10.1002/hyp.6229
RIPPEY, B., C. L. ROSE, AND R. W. DOUGLAS. 2004. A model for WOLOCK, D. M., AND G. J. MCCABE. 1999. Estimates of runoff
lead, zinc, and copper in lakes. Limnol. Oceanogr. 49: using water-balance and atmopsheric general circulation
2256–2264, doi:10.4319/lo.2004.49.6.2256 models. J. Am. Water Resour. Assoc. 35: 1341–1350, doi:10.
ROMO, S., J. SORIA, F. FERNANDEZ, Y. OUAHID, AND A. BARON- 1111/j.1752-1688.1999.tb04219.x
SOLA. 2013. Water residence time and the dynamics of toxic XIAO, L. J., T. WANG, R. HU, B. P. HAN, S. WANG, X. QIAN, AND J.
cyanobacteria. Freshw. Biol. 58: 513–522, doi:10.1111/j.1365- PADISAK. 2011. Succession of phytoplankton functional groups
2427.2012.02734.x regulated by monsoonal hydrology in a large canyon-shaped
SELVENDIRAN, P., C. T. DRISCOLL, M. R. MONTESDEOCA, H. D. reservoir. Water Res. 45: 5099–5109, doi:10.1016/j.watres.
CHOI, AND T. M. HOLSEN. 2009. Mercury dynamics and 2011.07.012
transport in two Adirondack lakes. Limnol. Oceanogr. 54: YUAN, F. S., Y. W. SHENG, T. D. YAO, C. J. FAN, J. L. LI, H. ZHAO,
413–427, doi:10.4319/lo.2009.54.2.0413 AND Y. B. LEI. 2011. Evaporative enrichment of oxygen-18 and
SOBEK, S., J. NISELL, AND J. FOLSTER. 2011. Predicting the volume deuterium in lake waters on the Tibetan Plateau. J. Paleolim-
and depth of lakes from map-derived parameters. Inland nol. 46: 291–307, doi:10.1007/s10933-011-9540-y
Waters 1: 177–184, doi:10.5268/IW-1.3.426
SOBOTA, D. J., J. E. COMPTON, AND J. A. HARRISON. 2013. Reactive
nitrogen inputs to US lands and waterways: How certain are
we about sources and fluxes? Front. Ecol. Environ. 11: 82–90,
doi:10.1890/110216 Associate editor: George W. Kling
SOKAL, M. A., R. I. HALL, AND B. B. WOLFE. 2008. Relationships
between hydrological and limnological conditions in lakes of
the Slave River Delta (NWT, Canada) and quantification of Received: 25 October 2013
their roles on sedimentary diatom assemblages. J. Paleolim- Accepted: 31 August 2014
nol. 39: 533–550, doi:10.1007/s10933-007-9128-8 Amended: 02 September 2014

View publication stats