We analyzed the oxygen isotope composition of 29 samples spanning 15 localities to build on existing data from Takeuchi (2007). Clay-rich paleosol samples were gently ground with a mortar and pestle and centrifuged to separate the <0.5 µm size fraction. The separated material was dried and gently ground with a mortar and pestle into a powder. Samples with indication of organic matter (dark coloration) were treated with a 3% hydrogen peroxide (H₂O₂) solution for 24–48 h before being rinsed at least 5 times with de-ionized water in a centrifuge.

At least one sample from each sedimentary package at each locality was run for X-ray diffraction to identify the clay mineralogy and screen for quartz. X-ray diffractometry was run at the Stanford University Environmental Measurements Facility using a Rigaku MiniFlex 600 Benchtop X-ray Diffraction System. The diffractometer is equipped with a copper (Cu) anode set at the maximum power of 600 W. Powdered samples were first suspended in isopropanol and left to air-dry on a zero-background quartz sample holder. Once dried, each sample was run twice with the 2-theta angle ranging from 2 to 90°. Samples were untreated in the first run and glycolated in the second run. To glycolate the samples, dried powders were placed in a sealed desiccator with 1 cm of ethylene glycol at the base and left overnight in an oven set to 65°C. The Rigaku PDXL software was used to aid in mineral identification for each sample. Samples identified to have quartz and/or illite peaks were not analyzed for oxygen isotopes.

After screening for quartz and illite, 1–2 mg of sample powder was mixed with lithium flouride ( L i F ; ∼ 1:1 by mass) and pressed into pellets. The L i F pellets prevent dispersion of clay powder during laser ablation. Samples were brought to a vacuum to remove atmospheric vapor in the L i F . Any remaining vapor and any sorbed water in smectite minerals was removed through 2–4 pre-flourinations where samples were exposed to bromine pentaflouride ( B r F 5 ) for ∼90 s. Afterward, samples were laser ablated with a New Wave Research MIR10-25 infrared laser ablation system in the presence of B r F 5 to liberate O 2 gas (Sharp, 1990; Sjostrom et al., 2006; Mix and Chamberlain, 2014; Mix et al., 2016). Oxygen isotope analyses on O 2 gas were performed in dual-inlet mode at the Stanford University Stable Isotope Biogeochemistry Laboratory on a Thermo Finnigan MAT 252 or 253+, depending on the date of acquisition (Abruzzese et al., 2005; Hren et al., 2009; Mix et al., 2016; Chamberlain et al., 2020). We purified O 2 gas from samples run on the MAT 252 with two liquid nitrogen cold traps and one potassium bromide ( K B r ) trap before freezing O 2 on a liquid nitrogen-temperature zeolite and subsequently equilibrating it with the mass spectrometer sample bellows. On the MAT 253+, O 2 gas was purified with three liquid nitrogen traps and one sodium chloride ( N a C l ) trap, frozen on a zeolite at liquid nitrogen temperature, passed over a room temperature flow-through zeolite with high purity He as the carrier gas, and finally frozen in a zeolite trap and equilibrated with the mass spectrometer sample bellows. Repeated analyses of in-house standard DS069 were made during each day of analysis to correct for drift. Precision of DS069 replicates was 0.26‰ (n = 43). All isotopic ratios are reported relative to Vienna Standard Mean Ocean Water (VSMOW).

In order to map out the spatial pattern of authigenic clay δ 18 O through the Cenozoic we calculate the residual of all data points relative to the Cenozoic trend. We build this Cenozoic trend by taking the average of the eastern and western facies trends, each defined by a loess smooth curve. This approach ensures that the Cenozoic trend is not biased toward one facies or the other when its sample density is relatively higher for a given point in time. After establishing the Cenozoic trend we calculate the residual δ 18 O value for each sample ( δ 18 O s a m p l e − δ 18 O t r e n d ). This represents a detrended record of δ 18 O that allows for a spatial comparison of samples from different times of the record.

Lower oxygen isotope values in the Blue Mountains province (eastern facies) suggests this local topographic feature has influenced regional precipitation patterns for at least the last ∼50 million years (Dickinson and Thayer, 1978; Dickinson, 1979; Schwartz et al., 2010; LaMaskin et al., 2015). Based on spatial interpolation of oxygen isotope residuals (Figure 2B) we find that the boundary between negative (lower δ 18 O ) and positive (higher δ 18 O ) residuals closely tracks the modern boundary of the eastern and western facies (which is, itself, defined by the extent of the Blue Mountains) (Figure 3). This close correspondence with the modern extent of the Blue Mountains suggests the possibility that authigenic clay δ 18 O values reflect modern conditions due to diagenetic alteration or overprinting by modern waters instead of reflecting changes in past precipitation δ 18 O . However, δ 18 O values from the eastern and western facies both increase by ∼3‰ between 30 and 20 Ma, and this shift likely would not be coherent if all samples have been similarly altered by modern waters. Thus, we interpret δ 18 O values to represent the oxygen isotope composition of water at the time of clay mineral formation.

Today, the windward slopes of the Blue Mountains capture more annual precipitation than the adjacent high plains (see Figure 1C) and receive a larger fraction of annual precipitation during the winter months (Supplementary Figure S2). Both of these observations support lower δ 18 O values in the Blue Mountains than in the adjacent plains. First, increased rainout over mountains is often explained by topography forcing air upward, leading to adiabatic cooling and moisture condensation and precipitation (Aristotle, 1931; Smith, 1979; Smith and Barstad, 2004; Roe, 2005). Higher precipitation due to Blue Mountains orography implies more rainout and a decrease in δ 18 O values as precipitation preferentially removes δ 18 O . Alternatively, lower δ 18 O values in the Blue Mountains region may be related to precipitation seasonality as, compared to the plains, the Blue Mountains receive a greater fraction of total precipitation in the winter months when precipitation δ 18 O values are lower. It is possible that upslope rainout and precipitation seasonality both contribute to lower δ 18 O values in the Blue Mountains. However, disentangling the effect of seasonality from upslope rainout on precipitation δ 18 O is difficult because most modern δ 18 O data in the John Day region are sampled in warmer months or from rivers that integrate seasonal precipitation (e.g., Bershaw et al. (2019)).

Despite the close link between residual δ 18 O values and the extent of the Blue Mountains, δ 18 O residuals do not correlate strongly with modern elevation or fraction of winter precipitation (Supplementary Figure S4). Residual δ 18 O values generally decrease with elevation by ∼3.3‰ per kilometer, similar to the regional lapse rate of ∼3.2‰ Bershaw et al. (2020) per kilometer, but the correlation is weak ( R 2 = 0.17 ). There is no significant correlation between residual δ 18 O and the fraction of annual precipitation occurring in winter. The weak relationships between past clay δ 18 O and modern conditions is not unexpected since some clays likely formed in contact with local meteoric water (tracking local elevation and precipitation seasonality) while others formed in floodplains in contact with upstream runoff (tracking the upstream hypsometric mean elevation and precipitation seasonality). Without reliable constraints on past floodplain extents and local drainage divides it is difficult to determine the relative importance of elevation, precipitation seasonality, or other factors in eastern and western facies δ 18 O values. Additionally, modern winter precipitation and topography may differ from that of most samples in our record. Thermochronometry and geochronology data from the western Cascades in the latitude range of our samples points to exhumation between 20 and 10 Ma (Pesek et al., 2020), and there is abundant evidence that the timing of Cascades uplift and exhumation varied north-to-south (Reiners et al., 2002; Takeuchi and Larson, 2005; Pesek et al., 2020). If spatially variable topographic changes affected the spatial pattern of δ 18 O , a direct comparison of past John Day data with modern conditions may be invalid.

Our results suggest that a regional gradient in precipitation amount and/or seasonality has existed around the Blue Mountains for most of the last 50 million years. If the Blue Mountains did not increase rainout or were not elevated above the plains, we would expect no difference in clay δ 18 O values between the eastern and western facies. While the Blue Mountains are generally wetter than the adjacent plains today, lower δ 18 O values in the eastern facies does not require that the Blue Mountains were wetter in the past. For example, it is possible that the Blue Mountains and the plains received similar amounts of annual precipitation, but summer (winter) precipitation made up a larger fraction of annual precipitation in the plains (mountains). However, summer and winter precipitation both increase with elevation in the Blue Mountains (and across the western United States; Supplementary Figure S5) today, and there is no clear reason why topography would not have the same effect on precipitation in the past. We suggest the Blue Mountains have received more precipitation than the surrounding plains for at least the last ∼50 million years. Thus, a possible Eocene boundary between tropical lowland and temperate highland vegetation (Bestland and Retallack, 1994; Bestland et al., 2002) is likely driven by colder temperatures or topographic relief, but not drier conditions in the Blue Mountains.

Despite a long-lived precipitation gradient associated with the Blue Mountains, both the Blue Mountains and the plains record an increase in δ 18 O values that tracks independent evidence for regional aridification. This δ 18 O increase is probably not driven by temperature since the shift is unidirectional while global climate both warmed and cooled between 30 and 20 Ma (Zachos et al., 2001). Meanwhile, the onset of drier conditions between 30 and 20 million years ago is evidenced by the expansion of open-habitat grasslands and mammals adapted to running and eating tougher vegetation like grasses (Woodburne and Robinson, 1977; MacFadden and Hulbert, 1988; Jacobs et al., 1999; Janis et al., 2002; Retallack, 2004), although this transition may also be explained by grassland-grazer coevolution, independent of climate (Retallack, 2001; Retallack, 2013). The ∼3‰ increase in authigenic clay δ 18 O may be related to drying, but the interpretation is not straightforward. For example, Retallack (2004) suggests that drying is related to 1) the uplift of the Cascades and 2) a decrease in summer (high- δ 18 O ) precipitation. But both of these effects would likely decrease δ 18 O values rather than increase them. Further, marine δ 18 O values vary by just 1‰ from 30 to 20 Ma and do not show a systematic increase capable of overprinting the effects of Cascades uplift and less summer precipitation (Zachos et al., 2001). Increased upstream (westward) moisture recycling in drier climates can also increase δ 18 O (Salati et al., 1979; Sonntag et al., 1983; Mix et al., 2013; Chamberlain et al., 2014; Winnick et al., 2014; Kukla et al., 2019), but this effect is negligible in near-coastal settings where upstream land area is minimal (Mix et al., 2013; Winnick et al., 2014; Kukla et al., 2019).

Instead, we speculate that the δ 18 O increase in our data reflects the onset of the Cascades rainshadow driving a decrease in winter (low- δ 18 O ) precipitation. Today, the Cascades represent a much stronger rainshadow in the winter than in the summer (e.g., Siler and Durran (2016); Supplementary Figure S6), suggesting the possibility that their uplift would disproportionately decrease winter precipitation inland. Still, it is not clear that the increase in δ 18 O from less winter precipitation could outweigh any decrease in δ 18 O from Cascades uplift. It is possible that other factors associated with drier conditions, like surface water and subcloud evaporation, could also increase δ 18 O . Thus, more work is needed to test the hypothesis that drying in the John Day region is owed to disproportionate drying in the winter.

The distinct isotopic signature of the Blue Mountains emphasizes the importance of local elevation in regional tectonic reconstructions. Even though the eastern and western facies lie leeward (east) of the Cascades, lower δ 18 O values in the eastern facies (the Blue Mountains) may be misinterpreted to reflect the height of the Cascades instead of local topography. For example, using hydrogen isotopes of a volcanic glass sample from the Blue Mountains, Bershaw et al. (2019) suggests the Cascades may have been higher than present as early as ∼32 million years ago. Alternatively, we suggest that low hydrogen isotope ratios in the Oligocene reflect the local topography of the Blue Mountains rather than upstream elevation of the Cascades. Interestingly, the hydrogen isotope data of Bershaw et al. (2019) reveal an increase in δ D of ∼16‰ (∼2‰ in δ 18 O ) that happens near the mid-Miocene, between ∼16 and 8 Ma, significantly later than the increase of ∼3‰ between 30 and 20 Ma found in our data. Due to sparse hydrogen isotope data east of the Cascades older than 20 Ma, it is not clear whether volcanic glass δ D also records an increase from 30 to 20 Ma. Further, while there are few coeval glass samples from the eastern and western facies to test for a spatial pattern, the existing data do not show a systematic difference in δ D between the two regions. If the spatial and temporal trends of authigenic clay and volcanic glass isotopes differ, it is likely that the proxies are forming under different conditions and may be seasonally biased relative to one another. Such information would provide increasingly nuanced insight to the uplift history of the Cascades and its effect on leeward precipitation seasonality. However, at present there is not enough overlap between our authigenic clay data and the volcanic glass data to determine whether the results truly reflect distinct trends.