The measurements reported in this study were made at the University of Minnesota tall tower trace gas observatory (TGO KCMP, Minnesota Public Radio tower, 44∘41′19′′ N, 93∘4′22′′ W; 290 m a.s.l.). The tall tower (244 m) is located about 25 km south of Saint Paul, Minnesota (Fig. 1). It was instrumented in spring 2007 with air sample inlets at 32, 56, 100, and 185 m. Three-dimensional sonic anemometer-thermometers (CSAT3, Campbell Scientific Inc., Logan, Utah, USA) are mounted at 100 and 185 m, with signals transmitted to data loggers and computers via fiber optic cables and modems . Scalars including carbon dioxide, water vapor, nitrous oxide, methane, isoprene, and other trace gases have been measured at the site since 2007 . Land use in the vicinity of the tall tower (extending from 10 to 600 km radius) consists of about 40 % agriculture (mainly corn and soybean) that is typical of the US Corn Belt . The concentration footprint of the tall tower (185 m sample inlet) when coupled to inverse model analyses has shown to be representative of the upper Midwest, United States, for a number of active and passive scalars . Here, we define the regional domain of the observations on the order of 80 km × 80 km, which is consistent with the numerical modeling described below.

The oxygen and hydrogen isotopes in water vapor were measured in situ using a tunable diode laser (model TGA200, Campbell Scientific Inc., Logan, Utah, USA) . These measurements were initiated in April 2010. A large diaphragm pump (1023-101Q-SG608X, GAST Manufacturing Inc., Benton Harbor, Michigan, USA) pulled air continuously at 3 L min-1 down sample tubing (Synflex Type 1300, Aurora, OH, USA) at the TGO to the analyzer that was maintained inside the climate-controlled radio broadcast building. The sample inlets used in this investigation were located at approximately 185 and 3 m above the ground surface. The tubing was heated from the base of the tower to the laser sample inlet, a distance of about 30 m, to prevent condensation. The sampling scheme consisted of a 10 min (600 s) cycle: (1) zero calibration with ultra dry air (110 s), (2) calibration with three span values (15 s/each) for the 3 m inlet, (3) sampling of the 3 m inlet (145 s), (4) zero calibration with ultra dry air (110 s), (5) calibration with three span values (15 s/each) for the 185 m inlet, and (6) sampling of the 185 m inlet (145 s). The three calibration span values dynamically tracked and bracketed the total ambient water vapor mixing ratios through time. The isotope composition of the span values was determined by the calibration dripper source water, which was maintained at approximately -60.0 and -8.5 ‰ for δ2H, δ18O, respectively . An omit time of 5 s was used on the calibration spans and air samples, and a 90 s omit time was used for the dry air calibration. Given the low pressure of the subsample inlets (40 kPa) and tunable diode laser sample cell (0.8 kPa), the equilibration time of the system was relatively fast, on the order of 5 s for the span calibrations and 30 s for the zero calibration. Further details regarding the measurement system and calibration techniques and uncertainties are described in . All raw data were recorded at 10 Hz using a data logger and then block-averaged into 1 h intervals. The hourly water vapor signals were filtered using an outlier detection algorithm based on the double-differenced time series that identifies outliers according to the median absolute deviation about the median values .

Precipitation samples have been collected from RROC, and at the University of Minnesota, Saint Paul campus, from January 2006 to present using a typical all-weather rain gauge with mineral oil added to eliminate evaporative fractionation effects. Samples were typically collected within 0–3 days of precipitation events and transferred to screw-top glass vials, sealed with Parafilm, and refrigerated until analysis. The timing and amount of rainfall was recorded using a tipping bucket rain gauge (6028-B, All Weather Inc., CA, USA) and snowfall was measured using a snow board provided by the Minnesota State Climate Office (http://climate.umn.edu/doc/journal/snowboard.doc). Leaf, stem, and soil samples were collected from within a 5 km radius of the tall tower during numerous campaigns and as part of the International Atomic Energy Agency's Moisture Isotopes in the Biosphere and Atmosphere (IAEA-MIBA) program. Vegetation sampling sites chosen for this analysis were representative of the local land cover characteristics, including corn (Zea mays L.), soybean (Glycine max), and big bluestem (Andropogon gerardii Vitman). The MIBA sampling protocol was followed. Sunlit leaves, non-green stems, and soil approximately 10 cm below the surface were collected near midday (12:00 local standard time, LST). Cryogenic vacuum distillation was used to extract water from the plant and soil samples. Surface (i.e., lake and river) water and ground water samples were also collected from within a 25 km radius of the tall tower.

All liquid water samples were analyzed for their isotope composition using an off-axis cavity ring-down infrared laser spectroscopy system (Liquid Water Isotope Analyzer, DLT-100, Los Gatos Research, Inc., Mountain View, California) coupled to an autosampler (HT-300A, HTA s.r.l., Brescia, Italy) for simultaneous measurements of 2H /1H and 18O /16O. This instrument has a precision of ±1.0 ‰ for 2H /1H and ±0.25 ‰ for 18O /16O. Precalibrated laboratory standards used to calibrate the unknown samples to the VSMOW scale were selected based on the expected isotope composition of the unknown samples, and were injected after every two unknown samples to correct for instrumental drift. Linear calibration equations were calculated using each set of standards throughout the autorun and used to correct unknown samples. Contamination of plant water samples by ethanol/methanol was corrected following the procedures described by .

Signals were analyzed using techniques based on the continuous wavelet transform (CWT). Wavelet-based techniques are particularly suited to analyzing non-stationary geophysical time series because signals are simultaneously decomposed into time–frequency space. See and for an overview of the theoretical background and practical application. Here, we use cross-wavelet analyses to help elucidate how different atmospheric processes influence the isotope composition of PBL water vapor and to better understand the patterns and timescales of those relations.

Briefly, all CWTs were calculated on the fluctuating component of the signal using the complex Morlet wavelet basis with the nondimensional frequency (ω0) set to 6 to obtain a good balance between time and frequency localization . Another desirable feature of the Morlet wavelet basis with ω0=6 is that the scales map closely to an analogous Fourier period (λ) according to λ=1.03s , where s is the scale, and the dimension of both λ and s is time. Scales were set to have a minimum of 2 h (i.e., twice the hourly averaging interval), and to have 12 suboctaves per octave. Calculating the CWT of the signal yields a set of wavelet coefficients, Wn(s), spanning all times (n) and scales. Here, we concern ourselves with disentangling the effects of different processes on PBL water vapor, and thus employ the multivariate technique known as wavelet coherence analysis to probe correlation and phase relationships between variables.

The cross-wavelet spectrum, SnXY(s), of two time series, Xn and Yn, is obtained from the wavelet coefficients calculated for the respective variables according to SnXY(s)=WnX(s)WnY(s)∗, where ∗ represents complex conjugation . The cross-wavelet spectrum identifies regions of high common power, but does not provide information regarding the coherency between the signals.

To examine the coherency of the cross-wavelet transform in time–frequency space, we made use of the wavelet coherence spectrum, Rn2(s), that is defined according to Rn2(s)=|Λs-1SnXY(s)|2Λs-1|SnX(s)|2Λs-1|SnY(s)|2, where Λ represents a smoothing operator and its definition can be found in (see their Eqs. 9 and 10). A useful interpretation of the coherence spectrum is that values of Rn2(s) represent local correlation coefficients in time–frequency space . Statistical significance testing was performed using the Monte Carlo approach described in . All wavelet analyses were implemented using the package of MATLAB functions developed by , which is available at http://www.glaciology.net/wavelet-coherence.

We used the National Center for Atmospheric Research (NCAR) Weather Research and Forecasting (WRF) model version 3.5 to simulate the regional surface latent heat flux, PBL height, and to examine other controls on the regional water vapor . The simulations made use of four nested domains (with a recommended 3 : 1 ratio for inner domains), with the innermost domain containing the location of the tall tower. The inner domain 4 occupied the smallest area (80 × 80 km) and employed a 1 km grid resolution (see Fig. S1 in the Supplement). In these simulations a two-way feedback among the nested domains was turned on. The NOAH land surface scheme option was selected for all WRF simulations for three reasons: (1) it has been used extensively in the literature; (2) we have been using WRF-NOAH to forecast evaporation for our region and have tested it extensively against eddy covariance flux observations; and (3) the WRF-NOAH system is computationally efficient compared to other options such as WRF-CLM (Community Land Model surface scheme option). The WRF-NOAH simulations used land cover information from the United States Geological Survey (USGS) land use product, which includes 24 land use categories. The WRF settings (namelist file) used to run these simulations are provided in the Supplement. Boundary and initial conditions were provided by the NCEP FNL Operational Global Analysis data product with a 1∘ × 1∘ resolution at 6 h intervals (National Centers for Environmental Prediction/National Weather Service/NOAA/U.S. Department of Commerce, 2000; http://rda.ucar.edu/datasets/ds083.2/). Further, the Stochastic Time-Inverted Lagrangian Transport (STILT) model was used to examine the water vapor concentration source footprint associated with an extreme dew point event at the tall tower. The meteorological fields required to drive STILT were obtained from the WRF simulations. Since water vapor is an active scalar, the STILT source footprints computed here likely represent the maximum spatial extent of influence with respect to the tall tower observations. All of these model simulations were run on an HP ProLiant BL280c G6 Linux Cluster at the University of Minnesota Supercomputing Institute (https://www.msi.umn.edu/).

The isotope composition of precipitation and water vapor is reported as δ=Rs-RstdRstd, where δ is the isotope ratio. All values are reported in parts per thousand (‰) by multiplying δ by 103. Rs is the sample molar ratio of the heavy (minor) to light (major) isotope (i.e., 18O /16O or 2H /1H) and Rstd is the standard molar ratio defined according to the VSMOW scale.

We make use of precipitation events to examine the isotope composition of water vapor in relation to the falling precipitation. In theory, if atmospheric humidity is at saturation below the cloud base, then thermodynamic equilibrium is expected for isotope exchange between the liquid water and atmospheric vapor : Rv=RLα, where Rv is the absolute isotope ratio of water vapor (18O /16O or 2H /1H), α is the equilibrium fractionation factor (isotope-specific), and RL is the isotope ratio of the liquid water (rain precipitation) . Under these conditions, the equilibrium relation can provide a useful diagnostic regarding the validity of the tall tower water vapor isotope ratios or the influence of evaporation of raindrops and humidification of the PBL.

The global meteoric water line (GMWL), δ2H=8δ18O+10, represents the linear relation between δ2H and δ18O for global precipitation and is a useful benchmark for examining the origin, modification, and history of other water sources . The GMWL parameters are derived from empirical observations and are related to Rayleigh distillation processes . The slope of ≈ 8 results from the equilibrium condensation conditions and the ratio of the equilibrium fractionation factors . The intercept of ≈ 10 is determined by the average equilibrium and kinetic fractionation factors for ocean–atmosphere exchange with a global evaporation-weighted mean relative humidity of ≈ 85 % . Sources of water undergoing evaporation result in isotope kinetic effects that cause δ2H–δ18O slopes less than 8 .

Three simple models were used to aid the interpretation of the tall tower δ18Ov data. These models were selected because their physics are well understood and they represent three idealized processes that influence the behavior of water vapor in the PBL . First, a classic Rayleigh model (RM1) assuming a closed system with no rainout was assessed : δRM1=1000(α-1)log⁡(χw)-log⁡(χo)+δo, where α is the equilibrium fractionation factor evaluated at a condensation temperature of -3 ∘C (this represents the mean adiabatically adjusted temperature at the lifted condensation level). Here, the initial air mass is assumed to have an oceanic source region with a water vapor mixing ratio (χo) of 35 mmol mol-1 and an oxygen isotope ratio (δo) of -10 ‰ . While these initial values are somewhat arbitrary, it is the variation in the response function relative to the observations that is of primary interest. Second, a Rayleigh model (RM2) with a rainout fraction (f) of 30 % was evaluated: δRM2=1000α1-f/α/(1-f)-1log⁡(χw)-log⁡(χo)+δo, where precipitation/condensate is removed, causing the isotope composition of the water vapor to become more depleted . Finally, a simple two-source evaporation mixing model (EM1, a Keeling plot, ) was examined: δEM1=χbχw(δb-δE)+δE, that considers surface evaporation into an air mass. χw and χb represent the air mass and background water vapor mixing ratios, respectively. Here, the oxygen isotope ratio of evaporation (δE) is taken as -6.2 ‰, which is based on the growing season (May to September) tall tower oxygen isotope flux-gradient measurements (Table 1).

Finally, we optimized the RM1 and EM1 models to determine the equilibrium fractionation factor and the isotope composition of surface evaporation, respectively, that best fit the tall tower data. These optimized models are referred to as BestFitRM and BestFitEM, respectively. These models were fit to the observed tall tower data using a nonlinear fitting algorithm (fitnlm) implemented using Matlab (Matlab Version 2013b, The Mathworks Inc., Natick, Massachusetts, USA).

The isoforcing (IF) approach was used to help interpret short-term (hourly) variations in the water vapor isotope observations: IF=ECa(δE-δv), where Ca is the molar density of water vapor, δE is the oxygen isotope composition of evaporation as determined from the tall tower flux-gradient measurements , and δv is the oxygen isotope composition of the water vapor in the PBL. The IF calculations are used to isolate the influence of surface evaporation on δv. Although the same approach can be applied using the deuterium isotopes, the atmospheric gradients are considerably smaller because the source strength is smaller, resulting in lower signal to noise ratios. As a result, we restricted our deuterium isoforcing calculations to May through August.

A simple two-member isotope mixing model was used to estimate the relative contribution of surface evaporation to the total water vapor concentration of the PBL: fv=δv-δbδE-δb, where fv is the fraction of vapor in the PBL derived from regional evaporation, δv is the oxygen isotope composition of the water vapor measured at 185 m, and δb represents the oxygen isotope ratio of the “background” vapor, which can vary depending on synoptic meteorological conditions. Further, this approach does not explicitly account for the influence of advection. Direct observations of the oxygen isotope composition of background vapor for the region do not exist. However, we make use of a unique set of aircraft observations collected by over New England, United States, in 1996. They obtained profiles of water vapor mixing ratio and δ18Ov at altitudes ranging from 195 m to 2851 m during three campaigns (15 June, 17 July, and 12 October 1996). We have plotted their data in Fig. 2 and demonstrate that δ18Ov follows a power law (Rayleigh) function with respect to water vapor mixing ratio (y=axb, where x is water vapor mixing ratio, r2=0.98, p<0.001, n=24) through the PBL. Here, we define the background signal assuming a power law relation for the tall tower site. In this approach, the theoretical background value was obtained by evaluating the power law relation with water vapor mixing ratio estimated at 700 hPa (i.e., above the PBL at a standard atmosphere height of approximately 3000 m) using reanalysis data provided by the National Centers for Environmental Prediction and the National Center for Atmospheric Research (NCEP/NCAR) Reanalysis-2 product. Over the 3-year period the mean annual water vapor mixing ratio at 700 hPa was 3.9 mmol mol-1, and was 5.9 mmol mol-1 during the growing season. With respect to the power law function, these mean values occur before the function reaches its vertical asymptote (i.e., where it becomes hypersensitive). However, as shown in Fig. 2, there are cases where the uncertainty in the background value will be large because of this sensitivity.

Constraints on the oxygen isotope composition of surface evaporation (δE) were provided from multiple studies conducted near the tall tower. The oxygen isotope composition of evaporation was determined over a corn canopy using the eddy covariance approach . These studies showed that δE ranged from -20 to -5 ‰ with a mean flux-weighted value of -7.7 ‰ for a 74-day period in 2009. The δE of soybean crops has also been estimated within the study domain using the flux-gradient approach with values ranging from about -30 to +20 ‰ with a mean flux-weighted value of -4.8 ‰ over the period June to September 2006. Regional δE has also been obtained from our tall tower flux-gradient observations. These values were similar to those reported for the above field-scale investigations with a mean flux-weighted value of -6.2 ‰ for the 2010 to 2012 growing season (Table 1). Further, based on plant stem water extractions, and assuming steady-state conditions for the mid-afternoon to late afternoon period, the oxygen isotope composition of transpiration can be approximated as stem water . Our data from plant sampling in the vicinity of the tall tower indicate a mean stem water oxygen isotope composition of -7.0 ‰ in 2010 .

Following the methodology of , we estimated the recycling ratio of growing season precipitation (fp) using the two-member mixing model approach: fp=dp-dadvvdE-dadvv, where dp is the deuterium excess of precipitation, dadvv is the deuterium excess of the advected moisture (approximated here by the large concentration footprint of the tall tower water vapor measurements at 185 m), and dE is the deuterium excess of evaporation estimated from the flux ratio measurements at the tall tower.

Here we describe the climatology of the isotope composition of precipitation, water vapor, and surface evaporation as observed at the tall tower (Tables 1 and 2 and Figs. S2–S3). The mean oxygen and hydrogen isotope composition of precipitation (weighted by amount) was -10.9 and -76.2 ‰, respectively, with a range of monthly means of 18.0 and 148.3 ‰, respectively. The isotope signature of precipitation showed peak enrichment of the heavier isotopes in August. The mean deuterium excess of precipitation was 11.3 ‰ with a range of 14.5 ‰. Peak values were observed during November. The oxygen and hydrogen isotope composition of water vapor (δ18Ov and δ2Hv) measured at the 185 m level had a mean annual value of -26.2 and -176.0 ‰, respectively, with a range of monthly means of 24.3 and 165.7 ‰, respectively. The isotope signature of water vapor showed relatively strong enrichment of the heavier isotopes in July when the water vapor mixing ratio reached its maximum value. The mean annual deuterium excess (dv) of water vapor was 28.4 ‰, with a range of 30.6 ‰. Deuterium excess of water vapor reached a minimum value in July.

The mean annual flux-weighted oxygen isotope ratio of surface evaporation (δE=-13.3 ‰) was in excellent agreement with the mean annual oxygen isotope ratio of the precipitation. There was strong seasonal variability in δE, with a mean growing season value of -6.2 ‰ over the 2010 to 2012 period, which was within the uncertainty of the oxygen isotope ratio estimates of evaporation and precipitation for the same period. The mean deuterium isotope composition of evaporation was -77.8 ‰ and was relatively depleted compared to precipitation. Over relatively long timescales (seasonal) we would expect there to be isotope mass balance between the inputs (precipitation) and outputs (evaporation, runoff, drainage). The relatively good agreement observed here suggests that our atmospheric measurements provide a reasonable constraint on the isotope composition of evaporation. The effect of surface evaporation on the δ18Ov and δ2Hv of the PBL was estimated using the isoforcing approach. The oxygen isoforcing associated with evaporation was relatively strong from May to September with a mean value of 0.0068 m s-1 ‰ (Table 1). The mean deuterium isoforcing was 0.0071 m s-1 ‰ from May through August. These calculations show that surface evaporation acts to enrich PBL water vapor in the heavier isotopes. We hypothesize that this contributes to the highly enriched values of convective precipitation observed during the growing season (discussed further below).

The observations reported here are in broad agreement with previous work conducted near New Haven and Great Mountain Forest, Connecticut, United States . However, the continental location of Saint Paul, Minnesota, exhibits a larger seasonal amplitude of δ18Ov associated with the Rayleigh distillation effect, and perhaps, higher rates of evaporation and isoforcing from crops during the middle of the growing period.

The observed isotope ratios in water vapor, δ18Ov and δ2Hv, measured at 3 and 185 m, were compared with those derived from the isotope equilibrium theory (δ18Ov,e and δ2Hv,e) for individual precipitation events to gain insights regarding the validity of the tall tower observations and the isotope fractionation of water vapor in the PBL. Figure 3 shows results for 35 rain events from the 2010 to 2011 growing seasons. Overall, there was good agreement between the measured isotope ratios in water vapor compared to those predicted from the equilibrium theory. The mean measured δ18Ov was lower by 1.4 ± 0.4 ‰ (uncertainty reported as the standard error) relative to the rain event δ18Ov,e values. The linear regression shown in Fig. 3a (y=0.54x-7.3, r2=0.42, p<0.001) supports that the derived equilibrium vapor values were modestly correlated (r=0.65) with the observed vapor values. A similar relation was observed for δ2Hv (y=0.73x-33.3, r2=0.50, p<0.001). The mean measured δ2Hv in water vapor was lower by 2.9 ± 2.3 ‰ relative to the rainwater δ2Hv,e values. These differences were magnified when calculating deuterium excess (d) (Fig. 3c). Derived equilibrium vapor dv,e values were lower by 7.8 ± 3.1 ‰.

It is well established that partial raindrop evaporation occurs below the cloud base because atmospheric humidity rarely achieves saturation through the entire depth over the course of an event . Partial raindrop evaporation acts to enrich the raindrop in heavy isotopes as the lighter isotopes preferentially escape to the atmosphere due to kinetic fractionation . This is especially true for short-duration and low-magnitude convective rain events . concluded that 20 to 50 % of rainfall evaporates near convective clouds over tropical locations, leading to strong isotopic signatures as observed from the Tropospheric Emission Spectrometer (TES). Further, recent work by has pointed out that the vertical structure of a cold front will tend to produce these observed differences, as warm air and water vapor that is relatively enriched in the heavier isotopes is lifted from the surface (warm sector), and as colder air and water vapor that is relatively depleted in the heavier isotopes is sinking and influencing the surface observations.

The results shown here are similar to other field-based studies. concluded that observed δ18Ov in water vapor and that derived from the equilibrium theory for a site in New Haven, Connecticut, United States, agreed to within -2.5 to 1.5 ‰. reported that values for a site in Beijing, China, were within -0.8 ± 1.9, 1.9 ± 9.9, and 7.7 ± 8.3 ‰ for δ18Ov, δ2Hv, and dv (uncertainty reported as 1 standard deviation), respectively. Precipitation data collected from 2006 to 2011 near the tall tower site also support that isotope ratios in precipitation tend to be more enriched in heavy isotopes for small rainfall events. Overall, the difference between observed isotope ratios in water vapor and the equilibrium values is small and partial raindrop evaporation likely contributes to this observed difference.

The relation between δ18Ov and water vapor mixing ratio measured at 185 m (2010 to 2012) is compared with the three isotope models (RM1, RM2, and EM1 defined above) for different time periods (Fig. 4) to gain further insights regarding the dominant processes influencing the tall tower observations. Given the large number of hourly water vapor observations, these data are displayed using a smoothed histogram technique . On an annual basis, the upper bound is defined by the simple two-source mixing models (EM1 and BestFitEM1) (Fig. 4a). A lower bound is defined by RM2 (a Rayleigh model that allows for a rainout fraction of 30 %). Assuming a simple closed system, RM1 provides an intermediate fit, and its curvature, relative to the data density contours, illustrates that Rayleigh processes have a predominant influence on the oxygen isotope composition of the PBL vapor.

Given the initial conditions of the air mass, described above, the best fit Rayleigh model yielded an r2 of 0.76 and an equilibrium fractionation factor of α=1.0103 (p<0.001) (equivalent to a condensation temperature of 15 ∘C). also reported a large warm bias in the condensation temperature when applying the same type of model to their annual data set in New Haven, Connecticut, United States. The best fit Keeling mixing model yielded an r2 of 0.37 and a very realistic estimate of the oxygen isotope composition of surface evaporation (-7.4 ‰, p<0.001) (Fig. 4a). Although the process of surface evaporation explained much less of the total variation in PBL vapor compared to the Rayleigh model, the relatively high coefficient of determination and statistical significance of the best fit parameters provides some evidence that surface evaporation within the region strongly modifies the oxygen isotope composition of vapor arriving at the tall tower.

Closer examination of the growing season data (Fig. 4b) indicates that the rainout fraction may exceed f=30 % as evidenced by the relatively large isotope depletion that occurs for water vapor mixing ratios between 15 and 20 mmol mol-1. It is also possible that these observations are associated with smaller convective summertime rain events when partial raindrop evaporation is favorable . The best fit Rayleigh and Keeling models explained 59 and 50 % of the variation, respectively. During the non-growing season the best fit Rayleigh and Keeling models explained 72 and 29 % of the variation, respectively. The density plot shows that the curvature of the data is similar to the Rayleigh model; however, the highest data density region (see bright yellow shaded contours) indicates a departure from this curvature that is consistent with evaporation effects.

The tall tower vapor data differ substantially from the GMWL and the Local Meteoric Water Line (LMWL, δ2H=7.8δ18O+6.9) (Fig. 4d). The growing season PBL Water Vapor Line (WVL, δ2H=6.2δ18O-15.3, r2=0.86, p<0.001), with a slope much less than 8, implies a relatively strong influence of evaporation. Analyses of local leaf water from agricultural plants (δ2H=2.7δ18O-37.1) and the soil (δ2H=5.3δ18O-21.6) provide strong evidence that evaporation was an important source of the PBL vapor. If the isotope composition of water vapor within the region were determined primarily by precipitation inputs (i.e., if the vapor were in isotope equilibrium with precipitation), then the δ2H–δ18O relation would be equal to the LMWL. If we make this assumption, a growing season water vapor equilibrium line can be calculated (WVLeq=δ2H=7.4δ18O-0.18). In this case, the slope and intercept of the WVL and WVLeq relations are statistically different (p<0.05 and p<0.1) and demonstrate that the isotope composition of water vapor is not simply derived from the precipitation, but is modified by other processes. came to a similar conclusion for field-scale measurements conducted within a few kilometers of the tall tower during the summer of 2006.

While the GMWL parameter values are determined primarily by the Rayleigh distillation effect, deuterium excess values (dv=δ2H-8δ18O) in water vapor are largely governed by non-Rayleigh distillation processes . Here, we observed large positive dv in vapor for all months. The mean annual values were 28.4 ‰ (Table 1) with mean monthly values (> 35 ‰) observed from November through January. The mean growing season dv value was 22.3 ‰. The mean monthly values showed negative relations with water vapor mixing ratio (y=-0.98x+43.6, r2=0.55), air temperature (y=-0.83x+41.0, r2=0.52), and precipitation amount (y=-0.09x+39.3, r2=0.37), and a very weak positive relation with relative humidity (y=1.28x-68.5, r2=0.08).

Based on an analysis of water vapor dv from several midlatitude locations, found that the diurnal variability was likely controlled by two dominant processes, including plant transpiration and PBL water vapor entrainment. also observed a strong influence of PBL entrainment on the early morning variations in dv in a Pacific west coast Douglas fir forest. have also examined the factors controlling dv over cropland in Zhangye, northwest China. In their analyses, they showed that variation in the deuterium excess of evaporation explained 94 % of the variation in daytime water vapor dv, implying that at some locations water vapor dv is an excellent tracer of surface evaporation. The recent work of suggests that plant transpiration has a dominant influence on vapor dv on diurnal timescales. At the longer timescales (monthly) examined here we expect that the variability and departure from the GMWL is influenced by synoptic conditions and air mass trajectories with strong modification by surface evaporation from within the region. For instance, the large dv values observed during the non-growing season, especially during November and December, suggest the important role of near-surface water evaporation (i.e., large kinetic fractionation effects associated with evaporation) within the region and probably reflect the dominant contributions of evaporation from bare agricultural soils and the Great Lakes, of which the latter reach peak evaporation rates in late fall and early winter . As noted by , the ability to simulate dv is highly sensitive to the isotope fractionation during soil evaporation. During the main growing season, dv, was less positive because plant transpiration is a non-discriminating process under equilibrium conditions and represents a substantial fraction of surface evaporation.

To further explore the influence of Rayleigh distillation, evaporation, and PBL growth processes on the isotope composition of the PBL, we performed cross-wavelet multivariate analyses for near-continuous time series observed in August 2010 (Figs. 5 and 6). Analyses for the Rayleigh modeled (model RM2 from Fig. 4) oxygen isotope composition of water vapor (δ18OR) vs. the tall tower δ18Ov observations (Fig. 5) demonstrate relatively strong in-phase coherence through the month of August 2010 across a broad range of periods. It is interesting to note when the Rayleigh relation fails to describe the observations. For example, at periods greater than 64 h and periods less than 8 h there are numerous days in August 2010 when the Rayleigh relation and observations show little or no coherence. Identifying the exact mechanisms that account for these discrepancies is challenging because many meteorological processes operating in the PBL are not independent (i.e., there is feedback between surface evaporation and PBL growth; ). For example, Fig. 6 shows there is strong coherence with a phase lag of about 3 h (90 degrees) between evaporation and PBL growth rate for diurnal cycles (periods ranging from 8 to 32 h) for nearly the entire month of August 2010. Figure 5 also shows the wavelet coherence between the evaporation isoforcing and the time derivative of δ18Ov as well as the PBL growth rate vs. the time derivative of δ18Ov. These analyses show that there are a number of more isolated periods when there is strong coherence, confirming that both surface evaporation and PBL growth are key forcing factors .

Similar analyses were also performed to examine the behavior of dv (Fig. 6). These analyses reveal the influence of synoptic/air mass effects and PBL effects on dv. For example, similar coherence was observed for wind direction vs. dv and water vapor mixing ratio vs. dv. The coherence was significant for periods ranging from 100 to 256 h or 4 to 10 days implying the importance of synoptic-scale air mass back trajectories. The effects of PBL growth and surface evaporation on dv clearly operate at different periods through the time series. The effects of PBL growth rate showed significant coherence at diurnal scales (periods ranging from 4 to 64 h), while the evaporation showed significant coherence with dv on diurnal (8 to 32 h) and synoptic (128 to 256 h) scales. In many cases, the phase lag between evaporation and dv implies that evaporation is leading the change in dv.

To probe this further, we focus our attention on the evaporation isoforcing (oxygen isotope) characteristics (Fig. 7). Using the WRF-modeled PBL heights we estimated the evaporation isoforcing effect over the depth of the PBL for each hour. The time derivative of the evaporation isoforcing was then compared to the time derivative of δ18Ov. The time series and distributions of these derivatives show that they are of similar magnitude. Here, the mean absolute values of both distributions indicate that evaporation can account for about 53 % of the variation in δ18Ov for August 2010, implying that surface evaporation is a dominant controlling factor.

A case study of high PBL water vapor concentration (defined here as ≥ 30 mmol mol-1) was carried out to further examine the underlying controlling factors. The extreme event of 14 July 2010 had a maximum dew point temperature of 26 ∘C at 13:00 LST. Local water vapor mixing ratios increased from about 22 to 39 mmol mol-1 over the 24 h period. The locally measured and modeled vapor fluxes were very high, ranging up to 10.6 mmol m-2 s-1 near midday. Over a 12 h period, starting at midnight, we calculated the change in water vapor concentration within the PBL that was associated with the average rate of evaporation for the tall tower domain (i.e., 80 × 80 km inner domain). These calculations indicate that evaporation could account for about 8.4 mmol mol-1 change (about 83 % of the observed variation) in the PBL water vapor concentration. The WRF-STILT source footprint analyses are shown for this case in Fig. 8. These results illustrate that the vapor source was associated with NNE to ESE flow the day before (13 July 2010), with flow switching to WNW the day after (15 July 2010) the extreme event. The highest water vapor concentrations were observed on 14 July 2010 when the flow was southerly before the passage of a cold front. The source footprint intensity was greatest in Minnesota, Iowa, and Indiana and was dominated by agricultural sources (59 %).

Additional evidence is provided by the tall tower isotope data and isoforcing (oxygen isotope) calculations. The tall tower observations during this period indicate that the δ18O of water vapor increased steadily from about -18 to -13 ‰. Further, for the same 12 h period as described above, the instantaneous IF averaged 0.08 m s-1 ‰. Therefore, over the 12 h period, the IF associated with evaporation accounted for a 3.8 ‰ variation in the PBL vapor and about 61 % of the observed variation. Thus, multiple lines of evidence support that this extreme dew point event was substantially enhanced by local/regional evaporation. These observations also support the general relationship described below in Fig. 9, indicating that a high fraction of the PBL water vapor was generated locally.

Although other approaches have been used to infer the impact of the US Corn Belt on regional humidity, the combined data, and the analytical and modeling approaches used here offer a unique and more direct quantification. The higher amplitude of crop transpiration rates during the middle of the growing season (Fig. S4) indicate that summertime humidity can be significantly amplified by crops and may, therefore, enhance convective precipitation.

WRF modeling and isotope mixing model analyses were used to help constrain the contribution of regional evaporation to PBL water vapor. The mean (2008–2011) growing season latent heat flux densities for each land use class within the study domain (i.e., the innermost domain of 80 × 80 km) were approximately 25 (0.57), 114 (2.6), 119 (2.7), 112 (2.5), 130 (2.9), and 14 (0.32) W m-2 (mmol m-2 s-1) for urban, dryland crops, dryland crops/grasslands, grasslands, evergreen needle leaf forest, and lakes, respectively (Fig. S5). The area-weighted contribution of each land use type to the total evaporative flux for the study domain was dominated by dryland crop (58 %) and dryland crops/grasslands (42 %), respectively. The growing season contributions to evaporation for all other land use types were insignificant according to the WRF-NOAH modeling (and given the spatial resolution for the domain) over the period 2008 to 2011.

The WRF land use evaporation analysis was combined with the oxygen isotope observations using a simple mixing model to help constrain the relative contributions of evaporation to PBL water vapor. Since the area-weighted flux densities indicate that evaporation is dominated by the agricultural land use, we make use of the key isotope signals from the agricultural component and a simple two-end-member isotope mixing model. Figure 9 shows the histogram of the fraction of local vapor (fv), estimated using the oxygen isotope mixing model for the daytime for June through August. The median fv was 34 % for the 2010–2012 growing seasons. The fraction of local vapor is also plotted as a function of the PBL water vapor mixing ratio observed at 185 m. The PBL vapor partitioning followed a saturation-type function (fv=0.66χw/(14.7+χw), r2=0.18, p<0.001). This relation indicates that the fraction of local water vapor increases asymptotically with water vapor mixing ratio. As expected, small changes in local evaporation can have a stronger effect on the fraction of water vapor in the PBL when mixing ratios are relatively low (< 10 mmol mol-1). At mixing ratios of 25 mmol mol-1, this relation implies that the locally generated vapor from evaporation accounts for about 42 % of water vapor in the PBL. However, the uncertainty is very large with prediction bounds, indicating a 1σ uncertainty range of 21 to 62 %. Also shown in Fig. 9 is the fraction of PBL water vapor derived from evaporation as simulated by WRF for June to August 2010. The WRF simulations indicate that on average daytime evaporation accounted for about 61 ± 18 % of the PBL water vapor. The median water vapor mixing ratios in 2010, 2011, and 2012 were 19.7, 18.1, and 15.9 mmol mol-1, respectively, indicating that the locally generated vapor accounted for 38, 36, and 34 % of the signal. Based on global analyses, best estimates indicate that approximately 40 000 km3 of water vapor are transported to the continents each year, with evaporation from terrestrial ecosystems accounting for 73 000 km3 . This global ratio of oceanic advection to terrestrial evaporation implies that 65 % of the vapor signal over the continents is derived from evaporation and is considerably larger than our median values obtained for the PBL in the upper Midwest, United States.

The different estimates of δE provide a way of evaluating the relative uncertainty of the mixing model approach. For example, a change in the mean flux-weighted isotope composition of evaporation by +3 ‰ would shift the relations observed in Fig. 9 lower. At mixing ratios of 25 mmol mol-1 the local contributions to PBL water vapor would be lower by approximately 6 %. Further, if the isotope composition of the background vapor were 3 ‰ lower, the sensitivity of the partitioning approach to the background estimate of the isotope composition of vapor would shift the relation observed in Fig. 9 higher. At mixing ratios of 25 mmol mol-1 the local contributions to PBL water vapor would be higher by approximately 2 %. This sensitivity is lower compared to changes in δE because δb appears in the numerator and denominator of Eq. (10).

As described above, the isotope composition of the annual (non-growing and growing season) precipitation for the period 2006–2011 closely followed the GMWL. Here we examine in more detail the isotope composition of precipitation during the growing season to gain new insights regarding source origin and regional recycling. As discussed by , numerical models tend to overestimate local-scale moisture recycling; therefore, additional constraints provided by empirical data may be used to help diagnose such biases.

Examination of growing season (1 May to 31 August) precipitation in δ2H–δ18O space indicated a near-identical slope (8.04) to the GMWL, and a smaller intercept (8.3) with r2=0.94. Figure 10 shows that fp ranged from close to 0 to 96 % over the period, with a median value of 26 %. Interestingly, Fig. 10 indicates that from DOY 121 to DOY 180, fp was approximately 10 % and increased significantly to 54 % for the period DOY 180 to DOY 240. This step change is coincident with high land surface evaporation during this period of peak growth for the agricultural region. Further, it has been shown that the Great Plains low-level jet (GPLLJ) has a strong influence on vapor transport into the region and can have an important effect on regional water recycling . Based on the model data presented by (his Table 2.5, the 100 strongest warm season precipitation events in the North Central U.S.) the median recycling ratio was 12.1 % with a range of 4.2 to 34.6 %. We re-examined these data and found that the recycling ratio increased as the GPLLJ weakened (y=-0.099x+0.18, r2=0.18), indicating that local evaporation becomes increasingly important as long-distant transport from the Gulf of Mexico weakens.

Because the dadv and dE are highly variable and subject to considerable noise, we performed a Monte Carlo simulation to provide a more robust growing season estimate of fp based on the observed precipitation data from 2006 to 2011 at the tall tower. Here we use the Monte Carlo approach to select values of dadv and dE based on the tall tower observations from 2010 to 2011. The Monte Carlo method selected median values within the 95 % confidence intervals. One thousand simulations were performed to evaluate Eq. (8) for each precipitation event from 2006 to 2011. Figure 10 shows the frequency distribution of values. Notice that we did not filter any of the fp estimates; therefore, there are values that fall outside of the realistic range. Overall, we find that the growing season fp value was 31 %, indicating that terrestrial evaporation significantly enhances the warm season precipitation.

Atmospheric water recycling is expected to be strongly linked to climate change with amplification anticipated during wet periods . used a general circulation model with water vapor tracers to follow their transport through the model atmosphere. They concluded that 14 % of the water precipitated within the US Midwest was derived from local evaporation. restricted their numerical modeling analyses to the growing season and US Corn Belt and estimated that the water recycling index ranged up to 45 %. In fact, they found that seasonal and monthly analyses masked the importance of recycling on short (daily) timescales. As discussed by the calculation of water recycling using numerical models is scale-dependent. In his analysis, annual moisture recycling in the Mississippi Basin was on the order of 7 and up to 21 % during the summertime when using a length scale of 1800 km. Further, also suggest that summertime water recycling within the Mississippi basin is on the order of 25 %. used stable isotope analyses of precipitation to estimate the contribution of evaporation from the Great Lakes to continental water vapor content. In their study they estimated a contribution of 5 to 16 %. These previous studies are in line with our own independent analyses and show that warm-season precipitation events have a relatively strong local signature, and that these rates are reasonably well constrained by models, at least on seasonal timescales.

The tall tower water vapor isotope data reported in this paper can be made available upon request and will be hosted at http://www.biometeorology.umn.edu/research/data-archives. The supporting NCEP NFL data used for the WRF simulations are available at http://rda.ucar.edu/datasets/ds083.2/.