The values for rainwater δ18O range from −22‰ to −5‰, have a mean of −11‰ ± 4‰ (1σ), and are negatively skewed. Because of the strong precipitation seasonality in SW Oregon, our sampling is biased toward the cool-rainy season when most of precipitation falls at our site. Skewness is associated with a small population of highly depleted values with δ18O of less than −18‰, a surprising level of depletion in this midlatitude site near the ocean. These highly depleted values were analyzed in replicate and confirmed. At least one such value was measured in each of four winters.
3.1. Effect of Parcel Trajectories and Temperatures
 Studies that assessed the influence of parcel trajectories on rainfall δ18O found a strong connection between the path of the storm and the isotopic composition of precipitation [Barras and Simmonds, 2008; Burnett et al., 2004; Gedzelman and Lawrence, 1982]. Our analysis of back trajectories shows that parcel tracks span an area from 26°N to 55°N and from ∼130°W to 150°W (Figure 2 and auxiliary material Figure S1). Because of this large spatial spread in parcel tracks, we focus our discussion on the end-members of the rainwater samples.
Figure 2. Back trajectories for (a) low δ18O and (b) high δ18O rain events arriving at OCNM at an altitude of 500 m above ground level. The color scale of the parcel trajectories represents the hourly temperature as extracted from the HYSPLIT model, and the colored background represents the gridded sea surface δ18O from LeGrande and Schmidt . The rest of the trajectories are shown in auxiliary material Figure S1.
Download figure to PowerPoint
 Back trajectories for high δ18O events (values ≥ −8‰) span over ∼20 degrees of latitude, from 26°N to 46°N (Figure 2b). Seawater δ18O over this latitude band ranges from −1.0 to +0.2‰ [LeGrande and Schmidt, 2006]. Back trajectories for low δ18O events (values ≤ −18‰) span over ∼10 degrees of latitude of the same latitude band as the high δ18O events (32°N to 42°N) (Figure 2a), with correspondingly similar seawater δ18O values (−1 to 0‰). Overall, the trajectory analysis thus indicates that, at least on the synoptic timescales analyzed here, the large differences in δ18O of rainwater collected at our site cannot be attributed to systematic changes in air mass trajectory.
 Local surface temperature explains 29% of the variability in the raw rainwater δ18O (r = 0.54, P value = 0.0001), with a T: δ18O slope of 0.7‰/°C (Figure 3a). The isotopic composition of rain within an individual storm can vary greatly due to continual depletion of the air mass or changes in the degree of isotopic exchange between raindrops and the surrounding air [McDonnell et al., 1990; Pionke and DeWalle, 1992]. To remove some of these meteorological effects and gain insight into the underlying climatology that influences the isotopic composition of rainfall, we average all values within a calendar month and perform a linear regression of average δ18O against the corresponding monthly averaged temperatures. Each monthly data point is weighted in the regression analysis by the monthly average precipitation. In this case, monthly temperature explains 65% of the variance in amount-weighted monthly δ18O (P value = 0.005), and the T: δ18O slope increases to 0.9‰/°C (Figure 3b). In this analysis the average leverage of a single data point is 0.2 while the leverage for June is more than three times larger than the average value. June rainfall represents only ∼1% of total annual precipitation at OCNM, however, indicating that this leverage is unjustified in evaluating the temperature influence on precipitation δ18O. Accordingly, if we exclude the values for June, monthly surface temperature explains 80% of monthly averaged δ18O variability.
Figure 3. (a) Temperature versus precipitation δ18O and (c) precipitation amount versus δ18O on synoptic timescales and (b and d) the corresponding monthly averaged values and the standard error of the mean for the monthly averages. Black regression lines and text show the linear regressions and the corresponding statistics while the red line and text in Figures 3a and 3b indicate the logarithmic fit and the related regression statistics. Shaded regions in Figures 3a and 3b indicate the area bounded by Rayleigh distillation curves of air masses originating in the tropical Pacific (source temperature 28°C) and North Pacific (source temperature 10°C). The most depleted samples in winter are best explained by distillation of water vapor that originated at relatively high temperatures, consistent with a low-latitude source of water vapor reaching western Oregon.
Download figure to PowerPoint
 A logarithmic fit explains a higher percentage of variance in δ18O values than a linear fit, both in the case of raw (r2 = 0.39) and monthly averaged data (including June, r2 = 0.74). This is not surprising since both the Rayleigh distillation equation, which expresses changes in isotopes as a function of fraction of vapor remaining, and the equations for water vapor saturation as a function of temperature would predict a curvilinear relationship between oxygen isotopes and temperature.
 Most of the measured δ18O values fall within the range predicted by Rayleigh distillation for reasonable oceanic moisture sources in the North Pacific. Figures 3a and 3b include the field of Rayleigh-predicted rainfall δ18O from air masses initially equilibrated with the seawater in the tropical Pacific (T = 28°C) and in the north Pacific (T = 10°C) then cooled during transport. In both scenarios we set the seawater δ18O at 0‰ and the relative humidity at 85%. Most points fall between the two Rayleigh curves suggesting that a Rayleigh process with a combination of low and high-latitude moisture sources could approximate the isotopic fractionation at our location. Some of the most depleted samples fall outside the two Rayleigh curves suggesting that these depleted samples could have formed at high levels in the clouds. Similarly, Friedman et al.  found the dominance of Rayleigh processes on rainfall δ18O in the Great Basin area with varying moisture sources.
 While OCNM is in a near-coastal setting where slopes are generally lower [Rozanski et al., 1993], the T: δ18O slopes that we observe at our site are higher than the average slope of 0.5 commonly observed for the midlatitudes [Dansgaard, 1964]. In general, lower T: δ18O slopes are characteristic of precipitation falling during warmer seasons due to a combination of lower fractionation coefficients for the liquid-water transition versus solid-vapor phase change, the mixing of both temperature and mixing ratios during convection [Jouzel, 1986] and higher evapotranspiration during the plant growing season [Rozanski et al., 1993]. These effects are less important in Oregon, where the summers are generally dry, and we therefore infer that the high T: δ18O slope in the Klamath Mountains reflects the predominance of winter precipitation. Further, OCNM is situated at high altitude and models using Rayleigh adiabatic condensation processes show that the δ18O:altitude slope increases with altitude due to lower temperatures and the associated increase of the condensation rate [Gonfiantini et al., 2001]. To obtain additional insight into this issue we use rainfall δ18O data from Vancouver Island (British Columbia), the nearest GNIP station that has both δ18O and temperature information. This station has data collected from 1975 to 1982. The GNIP Database, 2006, from IAEA/WMO (available at http://isohis.iaea.org) and the monthly precipitation δ18O is well correlated with monthly temperature (r = 0.8), but with a low slope of 0.2 ‰/°C. However, if the Vancouver Island data from the cool half of the year (October–March) are considered, the slope increases to 0.5 ‰/°C (r = 0.9) while during the warm season (April–September) the slope is 0.1 ‰/°C (r = 0.5).
 In Figure 4 we consider the temporal relationship between temperature changes along the parcel track and rainwater δ18O. Although there is some overlap in distributions, three out of the five isotopically depleted events clearly have colder storm temperatures than the enriched events for most of the 48 h prior to rainfall collection. In most cases, temperatures associated with depleted trajectories are rising in the period between 48 and 8 h before arrival, while for enriched trajectories temperatures tend to remain approximately constant or vary within a narrow range over the same time interval. Both enriched and depleted trajectories show a sharp decrease in temperature during the last 8–4 h before arrival as the air masses rise over the Klamath Mountains. The cooling of air masses shortly before precipitation collection was also calculated for parcel trajectories arriving in Melbourne, Australia [Barras and Simmonds, 2009].
Figure 4. Calculated (top) specific humidity, (middle) rainfall rate, and (bottom) temperatures for the last 48 h before arrival at OCNM for depleted and enriched trajectories.
Download figure to PowerPoint
3.2. Effect of Rainfall Amount
 The amount of precipitation can influence the isotopic composition of rainfall due to a high degree of removal of 18O from the cloud layer during intense precipitation events, incomplete equilibration of large raindrops with water vapor near the ground, evaporation below the cloud base during less intense storms [Dansgaard, 1964], and continuous isotopic exchange of the water vapor with raindrops below the cloud base [Miyake et al., 1968; Rozanski et al., 1993].
 We find no statistical significant relation between rainfall amount integrated 24 h prior to sample collection and δ18O at 95% confidence (r = −0.05, P value = 0.7) (Figure 3c). This lack of correlation is also true regardless of the time interval for which precipitation amount is considered (ranging between 1 and 48 h prior to collection). To assess whether the low correlation is caused by competing factors that act in opposite directions, we performed a linear regression between the residuals of T versus δ18O and precipitation amount, but also find no significant correlation between the residuals and precipitation amount. If we remove the weather noise by averaging precipitation and δ18O into monthly values, the R-squared statistic indicates that the model explains 11% of the variability in δ18O (Figure 3d), but the P value (0.34) indicates that there is not a statistically significant relationship between precipitation amount and δ18O at the 95.0% or higher confidence level.
 Large or rapid changes in specific humidity along the storm tracks can indicate condensation or entrainment of moisture, which in turn will modify the isotope ratios in the air mass [Barras and Simmonds, 2008]. The calculated specific humidities along the trajectories suggest that isotopically depleted trajectories start entraining moisture 30–16 h before the rainfall was collected, while enriched trajectories have relatively constant humidities, indicating water vapor exchange of the air mass for the much of the 48 h before arrival (Figure 4). This phenomenon may partly counteract the continuous isotopic depletion of air masses through Rayleigh distillation and contributes to the higher δ18O values measured in rainwater.
3.3. Multivariate Analysis
 To obtain a semiquantitative distinction between the factors that influenced the most depleted and most enriched rainwater samples, we conducted a Q-mode factor analysis [Klovan and Imbrie, 1971; Miesch, 1980] using the parameters obtained for parcel trajectories in the HYSPLIT model. Included in this analysis are latitude, longitude, altitude, pressure, temperature, precipitation and specific humidity. The first three extracted varimax factors in the Q-mode analysis for depleted storms explain 98% of the variance in rainfall δ18O. Factor 1 explains 64% of the variance and is dominated by temperature (Figure 5a), factor 2 has the highest scores for altitude and explains 24% of the variance, while factor 3 explains 10% of the variance and is clearly dominated by precipitation. In the case of enriched storms the first three factors explain 98% of the rainfall δ18O variance. Factor 1 has 76% of the total variance and shows similar factor scores for all of the variables, except rainfall and altitude, factor 2, which explains 12% of the variance, is dominated by altitude, and factor 3 is clearly dominated by precipitation, explaining 10% of the variance (Figure 5b). It would therefore appear that while storm temperature is the dominant control over δ18O values in depleted storms, the situation is more complex in warmer storms, although temperature still plays a major role in determining the isotopic composition of rainfall.
Figure 5. Score loading for the first three varimax factors from Q-mode factor analysis. (a) Factor scores for depleted events. (b) Factor scores for enriched events. The labels for the x axes are: La, latitude; Lo, longitude; Al, altitude; Ps, pressure; Te, temperature; Pr, precipitation; and Sh, specific humidity.
Download figure to PowerPoint