The cause of the seasonal variation in the oxygen isotopic composition of precipitation along the western U.S. coast

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 117,, doi: /2012jd018050, 2012 The cause of the seasonal variation in the oxygen isotopic composition of precipitation along the western U.S. coast Nikolaus H. Buenning, 1 Lowell Stott, 1 Kei Yoshimura, 2 and Max Berkelhammer 3,4 Received 3 May 2012; revised 3 August 2012; accepted 15 August 2012; published 26 September [1] This study seeks to find the primary influence on the seasonal cycle of the oxygen isotopic composition of precipitation (d 18 O p ) along the western U.S. coast. Observed long-term mean seasonal variations of d 18 O p from 16 different stations along the west coast are presented. The most robust features in the observations are high values in the summer and a drop in d 18 O p during the winter. The Isotope-incorporated Global Spectral Model (IsoGSM) also simulates this wintertime drop in d 18 O p along the west coast of the U.S. Sensitivity experiments are performed with IsoGSM where individual oxygen isotope fractionation processes are turned off. These simulations reveal that the primary control on the seasonal variations is equilibrium oxygen isotopic fractionation during vapor condensation. There is almost no influence of the temperature dependence of equilibrium fractionation on the seasonal d 18 O p cycle for both evaporation and condensation. Additional experiments (including tagging simulations) are performed to better understand why Rayleigh distillation causes the seasonal variation in d 18 O p. The tagging simulations and budget calculations reveal that vertical oxygen isotope gradients and variations in condensation height cause the seasonal cycle in d 18 O p. This results from seasonal changes in the polar jet, and subsequent changes to divergence and vertical velocities, which affects the uplift of moisture. These findings suggest that d 18 O p in the western U.S. is a tracer of condensation height on seasonal timescales. The large influence of condensation height on d 18 O p seasonality complicates interpretations of interannual climate proxy records based on isotopes in precipitation as the seasonality is likely not static. Citation: Buenning, N. H., L. Stott, K. Yoshimura, and M. Berkelhammer (2012), The cause of the seasonal variation in the oxygen isotopic composition of precipitation along the western U.S. coast, J. Geophys. Res., 117,, doi: /2012jd Introduction [2] The 18 O/ 16 O and D/H compositions of precipitation are commonly used in studies of hydrology [Lee et al., 1999; Kendall and Coplen, 2001; Vachon et al., 2007] and past climate variability [Dansgaard, 1964; Dansgaard et al., 1969; Lorius et al., 1979; Grootes et al., 1993; Thompson et al., 1995]. Such studies make use of observed correlations associated with precipitation that carries lower 18 O/ 16 O and D/H ratios with increasing precipitation amount (the amount effect) and higher ratios with increasing temperature 1 Department of Earth Sciences, University of Southern California, Los Angeles, California, USA. 2 Atmosphere and Ocean Research Institute, University of Tokyo, Kashiwa, Japan. 3 Department of Atmospheric and Oceanic Sciences, University of Colorado Boulder, Boulder, Colorado, USA. 4 Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, Colorado, USA. Corresponding author: N. H. Buenning, Department of Earth Sciences, University of Southern California, Zumberge Hall of Science, 3651 Trousdale Pkwy., Los Angeles, CA , USA. (buenning@colorado.edu) American Geophysical Union. All Rights Reserved /12/2012JD (the temperature effect). Studies have shown that the amount effect is related to diffusive exchanges of re-evaporated precipitation with background vapor and recycling of vapor within convective fluxes [Lee et al., 2007; Worden et al., 2007; Risi et al., 2008]. The temperature effect comes about from the influence of temperature on the saturation vapor pressure and preferential rainout of the heavier isotopologue [Dansgaard, 1964; Hendricks et al., 2000; Kavanaugh and Cuffey, 2003]. However, it is possible that for some regions the correlation between temperature and the isotopic composition of precipitation may be an artifact of other processes that are either indirectly related to or in phase with temperature variations [Dansgaard, 1964;Noone, 2008]. [3] Several studies have suggested that the statistical correlation between the isotopic composition of precipitation (hereafter denoted as d, where d = R/R VSMOW 1, R is the heavy to light isotope ratio, and R VSMOW is the Vienna Standard Mean Ocean Water) and temperature and precipitation amount could arise from variations in the seasonal timing and spatial characteristics of moisture advection [Fricke and O Neil, 1999; Werner et al., 2000; Alley and Cuffey, 2001; Brown and Simmonds, 2004; Noone, 2008; Feng et al., 2009]. These include studies focused on the western coast of the U.S., where the variations in precipitation 1of17

2 Figure 1. Map showing station locations for the GNIP sites (green), the west coast NADP sites used by Vachon et al. [2010a] (blue), and the NADP sites used here and by Berkelhammer et al. [2012] (red). Sites used by both NADP studies are represented as a blue circle surrounded by a red square. d values have been linked to changes in moisture source on weekly to seasonal timescales [Friedman et al., 1992, 2002a, 2002b; Berkelhammer et al., 2012] and to condensation height [Coplen et al., 2008] and rainfall re-evaporation [Yoshimura et al., 2010] on shorter (sub-hourly) timescales. Examining 6 month integrated measurements of precipitation d values; Friedman et al. [1992] concluded that storm trajectories were the primary driver of seasonal variations in d values in the western U.S. On the other hand, in southwestern Oregon, Ersek et al. [2010] found no connection between d values and storm trajectories, but high correlations with temperature. Similarly, Vachon et al. [2010a] noted the possibility that variations in source water temperatures and/or the temperatures of condensation might explain the observed seasonality in precipitation d values within the U.S. mainland. However, they also noted that these factors require use of atmospheric General Circulation Models to confirm this hypothesis. [4] The present study seeks to understand whether or not the temperature of condensation, temperature of evaporation (i.e., the temperature of the marine source region), or other processes (such as storm trajectory, condensation height, or kinetic effects) control the seasonal cycle in precipitation d values along the western U.S. coast. This study also aims to quantify the contribution that the various fractionation processes have on the regional isotopic seasonal cycle. [5] In the following section both the observational data and the isotope-incorporated General Circulation Model (GCM) used to investigate the isotope seasonality are described (section 2). The observations are then presented and used to validate the GCM (section 3). Sensitivity experiments are conducted with the model to understand which fractionation process has the largest influence on the seasonal isotope variations within the model (section 4). Additional simulations and calculations are then presented to better understand the exact cause of the modeled seasonal cycle (section 5). Finally, the paper concludes with a summary of the results and a discussion of the implications of the findings (section 6). 2. Methods 2.1. Observations [6] To better quantify the observed seasonal cycle in precipitation d values, archived precipitation samples were obtained from 9 stations along the western U.S. coast (7 in California, one in Oregon, and one in Washington, Figure 1). The samples were collected from 2001 through 2006 and come from the National Atmospheric Deposition Program (NADP). Previous work has demonstrated that the NADP collection and archiving protocol is sufficient for isotopic analysis [Harvey and Welker, 2000; Welker, 2000; Harvey, 2001; Vachon et al., 2007, 2010b, 2010a; Berkelhammer et al., 2012]. A total of 848 precipitation samples were used for this study, which include the measurements made by Berkelhammer et al. [2012]. The previous water samples were analyzed by continuous flow IRMS using a Thermofinnigan TC/EA and Delta Plus XP mass spectrometer. The water samples not included by Berkelhammer et al. [2012] were measured at the Stott Paleoclimate Laboratory at the University of Southern California using a Picarro Cavity Ring-down Spectrometer. Monthly mean isotopic seasonal cycles in precipitation d values are presented here, along with the values reported by Vachon et al. [2010a] (hereafter V10a) and data from two stations within the Global Network of Isotopes in Precipitation (GNIP). GNIP is a network run by the International Atomic Energy Agency and the World Meteorological Organization ( Thus, this study has assembled the most complete set of 2of17

3 Table 1. Name and Information of Collection Sites Station Name Longitude ( W) Latitude ( N) Elevation (m) Source Victoria GNIP, ( WA14, Olympic NP This study and Vachon et al. [2010a] WA24, Palouse Conserv. Farm Vachon et al. [2010a] OR18, Starkey Exp. Forest Vachon et al. [2010a] OR97, Hyslop Farm This study OR02, Alsea Guard Ranger Vachon et al. [2010a] OR10, H. J. Andrews Exp. For Vachon et al. [2010a] CA50, Sagehen Creek This study CA45, Hopland This study and Vachon et al. [2010a] CA99, Yosemite Vachon et al. [2010a] CA95, Death Valley NP This study CA75, Sequoia NP This study CA66, Pinnacles NM This study Santa Maria GNIP, ( CA42, Tanbark Flat This study CA67, Joshua Tree NP This study precipitation isotope measurements along the western U.S. coast to date. A complete list of the stations used in this study is given in Table 1. [7] For brevity, this study focuses on the 18 O/ 16 O composition of precipitation (d 18 O p is used for both observed and modeled values hereafter). V10a report mean monthly d 18 O p for samples collected from 1989 to The additional NADP data reported here come from weekly samples. The measurements are converted to monthly means by weighting each sample within a month according to precipitation amount. The d 18 O p values are applied to both months when a week begins in one month and ends in another, though the precipitation amount is divided based on how many of the days fall in each month for that given week. In other words, the precipitation is assumed to have fallen equally during each day of the week Model [8] The principal tool used to interpret the seasonal isotopic variability of precipitation is the Experimental Climate Prediction Center s Global Spectral Model fit with isotope tracers (IsoGSM) [Yoshimura et al., 2008]. The atmosphere in IsoGSM is represented by 28 vertical s layers (unit-less) with s = representing the uppermost layer. The horizontal resolution of the model is given by triangular truncation of the spherical harmonic spectrum at wave number 62, which corresponds to a Gaussian grid of about 1.85 degree longitude x 1.85 degree latitude. The simulations use a relaxed Arakawa-Schubert scheme [Moorthi and Suarez, 1992] for parameterizations of moist convection. The model is forced with prescribed sea surface temperatures and sea-ice conditions from the optimal interpolation weekly data set, downloaded from the Experimental Climate Prediction Center database [Reynolds and Smith, 1994]. Simulations were spectrally nudged at six-hour intervals to wind and temperature fields from the National Center for Environmental Prediction and National Center for Atmospheric Research Reanalysis version 1 [Kalnay et al., 1996]. The spectral nudging technique converts spectrum base meteorological data to Fourier series, and nudges the Fourier coefficients to that of the reanalysis (the specific details are described in Yoshimura and Kanamitsu [2008]). This approach constrains the dynamic wind and temperature fields in the model to closely match reanalysis fields. Furthermore, the technique results in simulated precipitation that is sufficiently accurate to allow direct comparison between model simulations and historical isotopic observations on event timescales [Berkelhammer et al., 2012]. The present study investigates the model s ability to capture the long-term mean seasonal variations in d 18 O p along the western U.S. coast. [9] Water isotopoogues are accounted for in IsoGSM in all three phases within the simulated atmosphere, and fractionation factors (a = R cd /R g, where subscripts cd and g refer to condensed phase and gas) are calculated and applied when phase changes occur. IsoGSM estimates equilibrium oxygen isotopic fractionation during condensation in the atmosphere (a eq-con ) and both ocean (a eq-ev ) and raindrop evaporation (a eq-rev ) according to the equations given by Majoube et al. [1971a, 1971b]. Ocean water in IsoGSM is assumed to have a constant d value of 0. Evaporation and isotopic exchange from falling raindrops in IsoGSM is estimated following Stewart [1975]. It is assumed that 95% of falling rain isotopically equilibrates with surrounding vapor for stratiform precipitation and 45% for convective precipitation. The formulation of Stewart [1975] for kinetic fractionation during raindrop evaporation is implemented into IsoGSM. The model also accounts for kinetic fractionation during ocean evaporation (a k-ev ) [Merlivat and Jouzel, 1979] and during vapor deposition onto ice crystals (a k-con ) under supersaturated conditions [Jouzel and Merlivat, 1984]. [10] An unperturbed control simulation (CTRL) is first compared to observations. The model is then used to quantify contributions to the simulated seasonal d 18 O p cycle from certain isotope fractionation processes, similar to the experiments shown by Noone and Sturm [2010] (a complete list of each simulation s name and description is given in Table 2). To this end, a total of 8 simulations were performed (name in all capitals and in parenthesis), where fractionation factors (the a variables defined above) are set to one. Two simulations remove the influence of equilibrium oxygen isotopic fractionation processes [Majoube, 1971a, 1971b] in order to quantify how each isotope effect contributes to the simulated seasonal d 18 O p cycle. These processes are equilibrium oxygen isotopic fractionations associated with ocean water evaporation (a eq-ev, NOFEQ1) 3of17

4 Table 2. Name and Description of IsoGSM Simulations Simulation Name Description CTRL Unperturbed control simulation NOFEQ1 Equilibrium oxygen isotopic fractionation during ocean water evaporation is turned off (a eq-ev =1) NOFEQ2 Equilibrium oxygen isotopic fractionation during condensation is turned off (a eq-con =1) NORNEV All oxygen isotopic fractionation associated with raindrop evaporation is turned off CONFEQ1 Equilibrium oxygen isotopic fractionation during ocean water evaporation is set to constant, removing the temperature dependence CONFEQ2 (a eq-ev = , T = 293 K) Equilibrium oxygen isotopic fractionation during condensation is set to a constant, removing the temperature dependence (a eq-con = , T = 274 K) NOFKI1 Kinetic oxygen isotopic fractionation during ocean water evaporation is turned off (a k-ev =1) NOFKI2 Kinetic oxygen isotopic fractionation during vapor deposition onto ice is turned off (a k-con =1) NOLOC37 Equilibrium oxygen isotopic fractionation associated with condensation is turned off (a eq-con = 1) at one grid-cell: 37.1 N and 124 W. NOLOC41 Equilibrium oxygen isotopic fractionation associated with condensation is turned off (a eq-con = 1) at one grid-cell: 41.1 N and 126 W. NOLOC45 Equilibrium oxygen isotopic fractionation associated with condensation is turned off (a eq-con = 1) at one grid-cell: 45.1 N and 126 W. NOLOC49 Equilibrium oxygen isotopic fractionation associated with condensation is turned off (a eq-con = 1) at one grid-cell: 49.1 N and 126 W. TAGLAT Tagging simulation where tag1 is applied within 10 N 30 N and W. Tag 2 is applied within 40 N 60 N and 200 W 230 W. TAGLEV Tagging simulation where both tags are applied within 25 N 55 N and 200 W 245 W. Tag 1 is applied below the 0.85 s level and tag 2 is above the 0.85 s level. and condensation in the atmosphere (a eq-con, NOFEQ2). Because setting a eq-ev and a eq-con equal to 1 removes both the temperature dependence and the isotope effect (i.e., preferential rainout or evaporation), additional simulations are conducted in which the equilibration temperature is assigned a constant value, globally (CONFEQ1 and CONFEQ2 with a eq-ev = and a eq-con = , respectively). Another simulation is configured so that raindrops do not evaporate in isotopic equilibrium [Stewart, 1975]. In this case, both equilibrium and kinetic oxygen isotopic fractionations (a eq-rev and a k-rev,respectively) and isotopic exchanges that occur during evaporation from droplets are turned off in the model (NORNEV) [Bony et al., 2008], similar to the experiments of Wright et al. [2009], Yoshimura et al. [2010], and Field et al. [2010a]. Simulations that remove kinetic isotopic fractionation are also conducted, which included isotopic fractionation during ocean evaporation (a k-ev NOFKI1) [Merlivat and Jouzel, 1979] and vapor deposition onto ice crystals (a k-con, NOFKI2) [Jouzel and Merlivat, 1984]. Discussed in more detail in section 5.2, two tracer simulations are also performed (TAGLAT, TAGLEV), where moisture is tagged within predefined regions (see Figure 5a) for the purpose of understanding how changes in moisture source and changes in condensation height influence the seasonal variations in d 18 O p. The tagging is done such that the amount of tagged vapor is always equal to the simulated vapor amount within a predefined boxed region. Each IsoGSM simulation runs from 1953 through 2010 with a 10-min time step, and each simulation is nudged to the same reanalysis wind fields [Kalnay et al., 1996]. 3. Observed and Simulated Seasonal Cycle [11] Long-term mean seasonal cycles in precipitation d 18 O values are presented here to characterize the amplitude and phase of monthly d 18 O p variability along the western U.S. coast (all monthly means presented here can be downloaded at: Figures 2a and 2p show the observed average cycles from the two GNIP stations, which are located on a middle latitude (Victoria, N) and a subtropical latitude (Santa Maria, 34.9 N). At both stations the highest d 18 O p values occur during warm summer months and the lowest values during cool winter months, a middle latitude observation that is consistent with many other studies [Gonfiantini and Picciotto, 1959; Dansgaard, 1964; Rozanski et al., 1982; Feng et al., 2009; Ersek et al., 2010]. These types of variations would be expected if evaporation temperatures were the primary control on seasonal d 18 O p variations (i.e., cooler Pacific sea surface temperatures occurring in the winter). However, the phases and the amplitudes of the seasonal cycle differ between the two sites. For example, at Santa Maria the highest d 18 O p occurs in July and then d 18 O p drops sharply to a minimum value in September, whereas at Victoria d 18 O p increases more or less progressively from early spring through the summer months, reaching a maximum in October and then decreases to a minimum in February (Figure 2a). The seasonal amplitude in d 18 O p is also larger at Santa Maria, where the mean seasonal range is 7.3, compared to 4.0 at Victoria. It is important to note that this discrepancy in seasonal amplitudes is inconsistent with a regional temperature effect, as seasonal temperature ranges are much higher at Victoria than at Santa Maria (not shown). [12] Vachon et al. [2010a] reported mean seasonal d 18 O cycles for 7 stations located in Washington, Oregon, and California. Here we combine the data from V10a with data for 9 additional NADP stations (2 of which were also reported by V10a), including the data from Berkelhammer et al. [2012] (Figure 2). Two of the NADP stations (CA45 and WA14) are shown twice to account for the different years that V10a and this study measured rain samples (Figures 2b, 2c, 2j, and 2k). Consistent with the Victoria data, each of the northern NADP stations exhibits their highest d 18 O p values in late summer and lowest d 18 O p during the winter months. Many of the northern stations also have a semi-annual cycle, with two peaks. The stations in California have highest d 18 O p values earlier than those in Oregon and Washington, typically in July. One feature of the seasonal cycle that is consistent among each of the western U.S. stations is a seasonal drop in d 18 O p during winter months (Figure 2). The Hopland, California station (CA45) from V10a is the exception to this pattern (Figure 2j), with the lowest d 18 O p values in September, followed by a sharp 4of17

5 Figure 2. Long-term mean observed seasonal cycles of d 18 O p (black dashed line) and simulated seasonal cycles from ISOGSM (blue). Error bars indicate one standard deviation. 5of17

6 Figure 3. Long-term mean ( ) simulated seasonal cycles of (a) d 18 O p and (b) d 18 O PW for grid cells located along the western North American coast (Figure 3a). The southernmost grid cell (red) is located in southern California, and the northern most grid cell (violet) is located in British Columbia. increase in values in October. Winter values at this site are close to the annual mean from 1989 to [13] The mean seasonal cycle for the CTRL IsoGSM simulation is also calculated (Figure 2) using only the years when samples were collected. At many of the stations the mean simulated d 18 O p values are less negative compared to the observations. Berkelhammer et al. [2012] found that the model s coarse topographic resolution was responsible for the overall high values of d 18 O p simulated by IsoGSM, which has been found in other models [Schmidt et al., 2005] and can be seen at most high elevation locations (Table 1) in Figure 2. The course resolution causes the elevation in the model to be lower than the station elevation. The lower elevation causes higher d 18 O p because the falling raindrops are subject to more enrichment in 18 O via evaporation and equilibration with surrounding vapor that have enriched d values relative to vapor at higher elevations. The relatively coarse resolution of IsoGSM can also cause problems when geographical structures have large influences on regional d 18 O p, such as coastline resolution for coastal grid cells and vegetation classification for inland grid cells. [14] To quantify the model s ability to capture the mean seasonal d 18 O p cycle, correlation coefficients between the modeled and observed values are calculated for each panel in Figure 2. Error bars in Figure 2 represent the standard deviation of the individual monthly means that went into calculating the long-term averages for both the model and observations. Despite the model limitation listed above, the model simulates the seasonal variations at most west coast stations very well, while failing to simulate the variations at others. For example, at Hyslop Farm, Oregon (OR97) and Sagehen Creek, California (CA50) the correlation coefficient between mean seasonal d 18 O p observations and the model is about 0.9. The model also accurately simulates the sub-seasonal d 18 O p variations at the Hopland station (CA45), especially the low d 18 O p in September that was reported by V10a (Figure 2j). IsoGSM does not capture the seasonal cycles reported by V10a at Olympic, Washington (WA14) and Yosemite, California (CA99). The model also performs poorly at Death Valley. The data/model discrepancies at Yosemite and Death Valley are likely a result of complex topography in the region that is not resolved in the model. The discrepancy at Olympic appears to be due to one month (September) when the model simulates anomalously low d 18 O p. [15] Figure 3a shows the 1953 to 2010 simulated mean seasonal cycles for five grid cells along the coast of western North America. At the southernmost grid cell there are large d 18 O p minima in July and September. The minimum in July should be viewed with caution as there is no southern California station data with such a drop in d 18 O p during July and the low value occurs during the region s dry season. The low simulated d 18 O p value in September is consistent with low values observed at Santa Maria, California. Some of the general features of the observational record are also found in the model simulation. For example, the seasonal amplitude in d 18 O p is larger and the maximum occurs earlier at the southern-most stations and grid cells. Also, the model simulates a seasonal drop in d 18 O p during winter months along the west coast, a feature that is also observed at all but one of the observational stations. This study now turns to understanding the cause of the simulated seasonal drop in d 18 O p. 4. Sensitivity of Isotope Seasonality [16] In this section a set of sensitivity experiments (Table 2) are presented to evaluate the processes that cause the simulated wintertime drop in d 18 O p along the west coast 6of17

7 Figure 4. Simulated long-term mean ( ) seasonal cycles of d 18 O p anomalies (relative to the annual mean) for a grid-cell located at coastal Oregon, centered near Glendale, Oregon ( W, N). The control simulation is represented by the solid blue curves. (a and b) Comparison of the control simulation with sensitivity experiments that turn off and set constant the equilibrium fractionation during ocean evaporation (Figure 4a) and vapor condensation (Figure 4b). (c and d) Comparison of the control simulation with sensitivity experiments that removes isotope effects associated with rain evaporation and post condensation exchange (Figure 4c) and kinetic fractionation processes (Figure 4d). Each seasonal cycle has the annual mean removed for better comparison. of North America. The sensitivity experiments test whether the seasonal change in d 18 O p is due to temperature of evaporation and/or temperature of condensation (i.e., the temperature dependence of equilibrium fractionation), a hypothesis put forth by V10a. These simulations are also used to assess whether the seasonality is due to isotopic fractionation during Rayleigh distillation (rainout), raindrop evaporation, or kinetic processes. Each model experiment is compared to the control simulation (CTRL) to assess which experiment results in a reduced seasonal cycle (i.e., Does the removal of a fractionation process remove the simulated seasonal d 18 O p cycle?). [17] For brevity, results of the sensitivity experiments are shown for only one location along the western U.S. coast, though the same general results were consistently found north and south of this one location (complete model results can be downloaded at Figure 4 illustrates the model results for the grid cell centered at N and W, near Glendale, Oregon. This location is selected because it is centrally located on the west coast. Removing the equilibrium oxygen isotopic fractionation associated with ocean water evaporation (NOFEQ1) only slightly modifies the simulated wintertime drop in d 18 O p (Figure 4a). Also, removing the temperature dependence of the equilibrium 7of17

8 Figure 5. Annual mean d 18 O PW for (a) the control (CTRL) simulation and (b) the simulation that turns off equilibrium fractionation during vapor condensation. Color contour intervals are 1, and black contour intervals and 2. Boxed regions in Figure 5a indicate the tagging regions for the TAGLAT simulation. isotopic fractionation factor (CONFEQ1) results in almost no change to the d 18 O p seasonality. These results indicate that both the equilibrium isotopic fractionation during ocean water evaporation and the temperature dependence of the fractionation factor did not substantially contribute to the seasonal wintertime drop in d 18 O p within IsoGSM. [18] The largest model response is the removal of equilibrium isotopic fractionation during vapor condensation (NOFEQ2, Figure 4b). In particular, the seasonal drop in d 18 O p during winter months does not occur for the NOFEQ2 simulation. These results illustrate how the modeled seasonal cycle in the western U.S. is almost completely caused by equilibrium isotopic fractionation associated with condensation of vapor (i.e., Rayleigh distillation during rainout), and Figure 4 illustrates how the other fractionation processes have only a small contribution to the simulated seasonal d 18 O p cycle. Like evaporation, this isotopic fractionation factor is temperature-dependent. Results from the CONFEQ2 (which sets the fractionation factor to a constant value) results in a slight increase in the seasonal amplitude. This is not a surprising result since warmer atmospheric temperatures will result in a decrease in the d 18 O value of condensed phase water (under equilibrium conditions) [Majoube, 1971a]. Thus, seasonal variations in temperature and the temperature dependence of equilibrium fractionation during condensation act to decrease the simulated seasonal amplitude of d 18 O p in the western U.S. (and likely most nontropical regions). The exact reason why Rayleigh distillation causes the large seasonal variations in d 18 O p will be explored further in section 5. [19] Removing isotope effects associated with raindrop evaporation (NORNEV) only slightly changes the modeled seasonal amplitude in d 18 O p along the western U.S. coast (Figure 4c). The largest difference between the NORNEV 8of17

9 Figure 6. Simulated annual mean vertical profile of d 18 O v for the control simulation (CTRL, blue) and the simulation that turns off equilibrium fractionation during vapor condensation (NOFEQ2, yellow). The profiles are taken from a grid cell near Santa Cruz, California. and CTRL simulations occurs during fall and spring months (this cannot be seen from Figure 4c because the annual mean is removed) when d 18 O p is lower for the NORNEV simulation. This result indicates there is significant evaporative enrichment of raindroplets prior to and after the wet season. These results contrast with other studies [Wright et al., 2009; Field et al., 2010a; Noone and Sturm, 2010; Yoshimura et al., 2010] who performed similar experiments to the NORNEV simulation and found that post condensation processes had a large influence on the spatial and temporal variation of d values, though these studies were not focused on isotope seasonality in the western U.S. [20] Removing the kinetic isotopic fractionation associated with ocean evaporation (NOFKI1) slightly reduces the simulated seasonal cycle of d 18 O p (Figure 4d). The seasonal cycle of d 18 O p is also slightly reduced when kinetic isotope fractionation during vapor deposition onto ice crystals is turned off (NOFKI2, Figure 4d). These results suggest that both kinetic effects contribute slightly to the observed seasonal d 18 O p variations. As such, the residual seasonal cycle in the NOFEQ2 experiment (Figure 4b) is partially a result of these kinetic effects. 5. Reconciling the Sensitivity Experiments 5.1. Additional Simulations [21] The main result of the sensitivity experiments (the NOFEQ2 experiment) was the near complete dampening of the simulated seasonal d 18 O p cycle along the western U.S. coast when removing equilibrium isotopic fractionation during condensation. Four mechanisms could be responsible for this result. First, the seasonal variations could be a result of local rainout. Second, the NOFEQ2 simulation changes the horizontal spatial patterns of the isotopic composition of water vapor (d 18 O v ), such that the equator-to-pole gradient is greatly reduced (Figure 5). If the winter drop in d 18 O p was a result of more moisture advection from the middle latitudes during winter, a reduced latitudinal isotopic gradient would reduce the seasonal d 18 O p amplitude. Third, the NOFEQ2 simulation also changes the vertical gradient of d 18 O v. In the CTRL simulation, d 18 O v sharply decreases away from the surface; however, this surface-to-tropopause gradient becomes almost flat when Rayleigh distillation is removed in the NOFEQ2 simulation (Figure 6). Thus, a seasonal change in the condensation height could cause the simulated drop in d 18 O p. Fourth, the seasonal variations could be the result of the temperature dependence of a eq-con and the seasonal change to the temperature of condensation. However, this hypothesis was disproved by results of the CONFEQ2 experiment. [22] To investigate which of these mechanisms have the greatest influence on the simulated seasonal d 18 O p cycle, additional model simulations are performed (Table 2). To assess the role of local rainout, 4 simulations are conducted where condensation equilibrium isotopic fractionation is turned off at a single grid cell for each simulation (NOLOC37, NOLOC41, NOLOC45, NOLOC49, corresponding to grid cells at latitudes of 37 N, 41 N, 45 N, and 49 N, respectively). [23] Two tracer simulations are performed where water vapor is tagged within predefined boxed regions, in order to assess how horizontal and vertical d 18 O v gradients influence the seasonal d 18 O p cycle. For each of these simulations the tagged mixing ratios are set to the normal vapor mixing ratio at each grid cell at each time step within the predefined boxed region. Once the vapor tags are added to the atmosphere they are treated like normal water molecules (i.e., they are allowed to advect, mix, condense, and rainout in the atmosphere), though they are not factored into energy budget calculations. Thus, the model simulates and outputs tagged vapor concentrations and tagged precipitation. For the first tagging simulation (TAGLAT), two water vapor tags are added within two regions: one within the tropics/ subtropics near Hawaii and another in the Gulf of Alaska (Table 2 and Figure 5a). The second tracer simulation (TAGLEV) tags vapor within the western U.S. (tagging region: 150 W 115 W, 25 N 67 N) according to the vertical level. One tag is applied below the 0.85 s level (typically higher vapor d values) and one above (lower d values). Above the ocean (mean sea level pressure), the vertical tags are located at the surface to 854 mb for the lower level tags and 854 mb to 2 mb for the upper level tag. The 0.85 s level was chosen after a trial-and-error process for maximizing the seasonal variation in tagged precipitation fraction along the western U.S. coast. For the TAGLEV simulation, tagged vapor that advects across the 0.85 s level or outside of the western U.S. coast region is immediately removed. This ensures that the precipitated tags reflect the condensation height. However, slight errors may result from horizontal flow of condensed phase water into the tagging region, which could cause the sum of tagged precipitation at a given point to be less than normal precipitation. [24] Removing the condensation equilibrium isotopic fractionation at single grid cells (process 1) causes the simulated mean d 18 O p to decrease by about 10. This demonstrates that (in the model) most of the condensation and precipitation occurs locally. However, each of the NOLOC simulations results in a slight increase in the d 18 O p seasonal amplitude (Figure 7). Even when the region of no fractionation was symmetrically expanded (by as much as 10 degrees of longitude and latitude), the simulated seasonal cycle remains nearly unchanged (not shown). These findings imply that the 9of17

10 Figure 7. Changes to the long-term mean seasonal cycle when local equilibrium fractionation during vapor condensation is removed. Each seasonal cycle has the annual mean removed. The grid cells shown here are located at (a) N W; (b) N, N; (c) N, W; and (d) N W. Each time series has the annual mean removed. simulated seasonal d 18 O p cycles are not caused by local rainout or rainout to the west of the U.S. coast, but are likely a consequence of either the horizontal or vertical gradients in d 18 O v (Figures 5 and 6) that result from Rayleigh distillation occurring at all locations. [25] The TAGLAT simulation is designed to quantify how the source of vapor affects the simulated seasonal d 18 O p cycle (process 2). Recall both observed and modeled d 18 O p drop during winter months, so if seasonally varying winds transport tropical moisture with higher d values toward the west coast in summer and lower d values from middle latitudes in winter [Field et al., 2010b], it could explain the seasonal d 18 O p cycle. However, the TAGLAT simulation reveals an opposite source-related influence on the average simulated seasonal d 18 O p cycle. The fraction of precipitation with tropical tags has a maximum along the west coast during winter, rather than summer months (Figure 8a). Conversely, the fraction of precipitation advected from the Gulf of Alaska to the western U.S. coast reaches a maximum in summer (Figure 8b) and decreases in winter. Both of these variations are inconsistent with the seasonal source effect described above. Additional tagging simulations, with slightly modified tagging region boundaries, confirm these results are robust (not shown). This seasonal shift in moisture advection is related to the clockwise flow around and position of the North Pacific high (descending branch of the Hadley Cell), as the Hadley Cell migrates south during the Northern Hemisphere winter. [26] Simulated seasonal variations in the oxygen isotopic composition of column-integrated vapor in the troposphere, or precipitable water (d 18 O PW ), are consistent with the variations in tagged precipitation fraction (and tagged precipitable water fraction) from the TAGLAT simulations (compare Figure 3b with Figures 8a and 8b). For grid cells along the west coast of the U.S., IsoGSM simulates a maximum in d 18 O PW in the winter/early spring (Figure 3b), which is when tropical (middle latitude) tagged precipitation fraction is highest (lowest) along the coast. Similarly, the minimum in d 18 O PW that occurs during summer coincides with the minimum (maximum) in tropical (middle latitude) tagged precipitation fraction. Satellite data have shown clear north-south gradients in vapor d values over the North Pacific [Worden et al., 2007; Brown et al., 2008; Risi et al., 2012], and the model results presented here underscore the large influence the source region has on the isotopic composition of vapor advected to the west coast. 10 of 17

11 Figure 8. Long-term mean ( ) results from tagging simulations, showing mean seasonal variations of fraction of precipitation with (a) tropical, (b) middle latitude, (c) lower level, and (d) upper level tags. Results are from grid cells that run north-south along the western U.S./Canadian coast; the same grid cells in Figure 3. This relationship is important for interpretations of climate proxies that are partially dependent on the isotopic composition of vapor (e.g., tree cellulose d 18 Ovalues[Berkelhammer and Stott, 2008, 2009]). [27] This seasonal source effect, however, does not translate over to the simulated isotopic composition of precipitation. The reason why the source effect is not apparent in the simulated seasonal d 18 O p cycle stems from the much greater isotopic effect that condensation height has on the isotopic composition of precipitation (process 3). Previous studies [Ehhalt, 1974; Ehhalt et al., 2005; Sayres et al., 2010; Risi et al., 2012] have shown the existence of vertical gradients in vapor d values (as in Figure 6), where vapor is more depleted higher in the troposphere. Results from the TAGLEV simulation reveal that along the west coast (with the exception of southern California), the fraction of precipitation derived from the upper troposphere peaks during the winter months and is the lowest during summer months (Figure 8d). On the other hand, the fraction of precipitation with the lower troposphere tag is lowest during the winter, and peaks during the summer months (Figure 8c). The variations shown in Figures 8c and 8d are 11 of 17

12 Figure 9. Long-term mean ( ) seasonal cycles of (a) 500 mb vertical pressure wind and (b) 200 mb divergence. Each monthly mean is weighted by daily precipitation totals. Results are from grid-cells that run north-south along the western U.S./Canadian coast. consistent with observed and modeled vertical gradients in d 18 O v and a seasonally varying condensation height effect. These results suggest that wintertime storms are tapping into vapor that is depleted in 18 O from high in the atmosphere, resulting in a simulated drop in d 18 O p during winter months. [28] In southern California (red curve in Figures 3 and 8), IsoGSM simulates the lowest d 18 O p during July and September, which is also observed in some stations in southern California (e.g., Santa Maria). These seasonally low values in simulated d 18 O p coincide with months when a larger fraction of precipitation carries the upper level tags. Indeed, summer precipitation in southern California, though not as frequent, is largely a result of local convective plumes that result from intensified surface heating and vertical instability. These convective plumes are not as frequent in regions further to the north, where low level marine layer precipitation dominates summer months. Nonetheless, the low d 18 O p during July and September coincide with months where the fraction of precipitation with upper (lower) level tags is high (low) First Order Budget Calculations [29] In order to assess whether or not the vertical isotopic gradient and temporal changes in tagged fraction are large enough to explain the seasonality of d 18 O p within IsoGSM, an isotopic budget is calculated. The isotopic composition of precipitation is approximated by: d p ¼ d v1p 1 þ d v2 p 2 p 1 þ p 2 þ ɛ; ð1þ where p 1 and p 2 are precipitation totals from tags 1 and 2 (corresponding to two different locations), d v1 and d v2 are the isotopic compositions of vapor within the first and second tagged regions, respectively, and ɛ is an oxygen isotopic equilibrium fractionation factor (assumed in this calculations to be a constant 10 ). Equation (1) can be simplified by defining the fractions: f 1 = p 1 /(p 1 + p 2 ) and f 2 = p 2 /(p 1 + p 2 ), yielding: d p ¼ d v1 f 1 þ d v2 f 2 þ ɛ: The intent of this calculation is to isolate the contribution of condensation height on d 18 O p. As such, the isotopic composition of vapor is assumed to be constant in time; thus removing the influence of horizontal moisture advection on d 18 O v and subsequently d 18 O p (which will be addressed in another set of calculations). However, it is likely that d v1 and d v2 have seasonal variations associated with moisture transport [Berkelhammer et al., 2012]. [30] Values of d v1 and d v2 are calculated using tagweighted averages of d 18 O v from the CTRL simulation, and f 1 and f 2 are calculated by dividing tagged precipitation by the total of tagged precipitation (making the two values sum to one) from the TAGLEV simulation. At a grid cell located near Santa Cruz, California (orange curve in Figures 3 and 8), the values of d v1 and d v2 are 16 and 31, respectively, while July values of f 1 and f 2 are 0.75 and 0.25, respectively. These values result in an estimated isotopic composition of 9.75 in July when d 18 O values are highest. In January, when isotopic values are lowest, f 1 and f 2 change to 0.33 and 0.67, respectively, yielding an isotopic estimate of d p = Thus, equation (2) estimates a seasonal variation at 37.1 Nof 6.3, based on variations in condensation height alone with no seasonal change to the isotopic composition of vapor. The overestimation relative to observed and simulated amplitudes is due to factors that are not represented in equation (2), such as seasonal variations in d 18 O v. Additional calculations were carried out for grid cells located near Glendale, Oregon and Sooke Lake, British Columbia (green and blue curves in Figures 3 and 8), which resulted in an estimated seasonal change of 4.7 ð2þ 12 of 17

13 Figure 10. (a)long-term mean ( ) precipitation-weighted vertical profiles of vertical pressure velocity for two locations during January (blue) and July (red) with units of Pa s 1. (b) The same plot for horizontal divergence with units of s 1. Solid curves represent the west coast at the N, latitude and dashed curve show the N latitude. and 3.6, respectively. These additional calculations show that the large influence of seasonal condensation height variations on simulated precipitation d values is not unique to California, but the influence (though slightly reduced) extends northward up the coastline. [31] Equation (2) can also be applied to the TAGLAT simulation to understand the influence of variations in moisture source on the simulated seasonal d 18 O p variations. For this case, d v1 and d v2 are non-local and estimated by the annual average of d 18 O PW within the two tagging regions (Figure 5a), which are 17 and 23, respectively. Using the same Santa Cruz location, the values of f 1 and f 2 (calculated from the TAGLAT simulation) for July are 0.27 and 0.73, respectively, which results in an estimate of The same respective values of f 1 and f 2 for January are 0.59 and These January values yield a first order estimate of 9.5 and a seasonal isotopic increase of 1.9. Thus, the role of seasonal variations in tropical versus middle latitude moisture advection on the simulated seasonal cycle acts to increase d p from summer to winter (via changes to local d v ). The change in moisture source can be seen in the seasonal cycle of the oxygen isotopic composition of precipitable water (d 18 O PW ) along the western U.S. coast, shown in Figure 3b. These results reveal that horizontal moisture advection slightly counteracts the role of condensation height on the simulated seasonal d 18 O p variations. However, the influence of condensation height outweighs that of moisture advection, and the result is a seasonal summer-to-winter decrease in d 18 O p of 2 to 4. These model results imply that the increase in upper level condensation/precipitation (where vapor is most depleted in 18 O) is the main cause of the simulated seasonal drop in d 18 O p during winter months along the western U.S. coast Vertical Winds, Diverging Flow, and the Polar Jet [32] Increased condensation and rainout of upper/middle tropospheric vapor that is isotopically more depleted in 18 O relative to near surface vapor, is found here to be the primary influence on the simulated seasonal d 18 O p variations. However, it is not immediately obvious what would cause the seasonal variations in condensation height, which is the intent of this section. These findings are consistent with Lin et al. [2009], who found from satellite data that off the California coast cloud top and cloud base heights are lower during summer months than winter months. Similarly, Kubar et al. [2012] using MODIS-aqua data and ECMWFinterm reanalysis data found that the fraction of low-level clouds decreases from summer to winter months along the west coast of the U.S. [33] Figure 9a shows the mean simulated seasonal cycle of the 500 mb vertical pressure velocity (w 500 ) weighted by daily precipitation totals (negative values indicating upward motion). Indeed, values are the most negative during the winter months for all locations, except the southern California grid cell, an indication that air is still rising and condensing above the 500 mb level. These vertical motions are primarily a result of the horizontal winds, which are tightly constrained by the NCEP/NCAR Reanalysis wind fields via the spectral nudging technique. For central and northern California, Oregon, and southern Washington, summer w 500 are positive, an indication that air is moving downward and likely not condensing above 500 mb. Indeed, the seasonal profiles of simulated vertical motion (Figure 10a shows the amount weighted profile for two locations) indicate that during winter months most of the tropospheric column is rising during rain events. Vertical motion for summer rain events only occurs in the lower troposphere (Figure 10a). However, this is not the case for the southern California grid cell (red curve in Figure 9a), where modeled w 500 is on average negative during July and September rain events. [34] Figure 9b shows seasonal cycles of 200 mb divergence, weighted by daily precipitation amount in IsoGSM (as were the curves in Figures 9a and 10a). These curves 13 of 17

14 Figure 11. Spatial pattern of the long-term mean ( ) 200 mb wind speeds (filled contours) and geopotential heights. Each field is weighted by daily precipitation totals from one western U.S. location, indicated by the red asterisk: (a and c) central California and (b and d) southern Oregon. Figures 11a and 11b are for July, and Figures 11c and 11d are for January. Wind speed contours are in unit of 2 m s 1, and height contours are in units of 0.1 km. show a sharp increase in upper level divergence above the western U.S. coast during early winter. Not surprisingly, 1000 mb divergence is negative (i.e., air is converging) and the lowest during the winter months (Figure 10b). Maintaining hydrostatic balance, the wintertime vertical motion (shown in Figures 9a and 10a) is a response to divergence aloft and convergence near the surface. This is not the case for summer months, when the vertical divergence profile is almost reversed, indicating that diverging flow is suppressing summertime rising motion. The slight rising motion that occurs in the lowest levels is likely a result of enhanced summer heating and vertical instability. The seasonal variations in divergence is in part associated with the wintertime intensification of the polar jet stream [Reiter, 1963], which can be seen in the seasonally varying 200 mb wind speeds (precipitation weighed), as the velocities increase from summer to winter (Figure 11). Wintertime upper level divergence is also a result of seasonal changes in the wind curvature. Figure 11 shows 200 mb wind speeds and geopotential height contours weighted by precipitation for two locations along the western U.S. coast for July and January. In July, the distance between geopotential height contours decreases moving eastward toward the coast, indicating convergence. In January, the flow is more zonal, and the geopotential contours slightly diverge from one another toward the coast. These model results reveal a seasonal link between the isotopic composition of precipitation and the strength and curvature of the polar jet stream via upper tropospheric divergence, vertical vapor transport, and condensation height. 6. Summary and Discussion [35] The long-term mean seasonal cycles of precipitation d 18 O values from 13 stations along the western U.S. coast share three common features: (1) highest d 18 O p in summer and lower values during winter [Ersek et al., 2010]; (2) reduced 14 of 17