Drier Winters Drove Cenozoic Open Habitat Expansion in North America

The shift from denser forests to open, grass‐dominated vegetation in west‐central North America between 26 and 15 million years ago is a major ecological transition with no clear driving force. This open habitat transition (OHT) is considered by some to be evidence for drier summers, more seasonal precipitation, or a cooler climate, but others have proposed that wetter conditions and/or warming initiated the OHT. Here, we use published (n = 2,065) and new (n = 173) oxygen isotope measurements (δ18O) in authigenic clays and soil carbonates to test the hypothesis that the OHT is linked to increasing wintertime aridity. Oxygen isotope ratios in meteoric water (δ18Op) vary seasonally, and clays and carbonates often form at different times of the year. Therefore, a change in precipitation seasonality can be recorded differently in each mineral. We find that oxygen isotope ratios of clay minerals increase across the OHT while carbonate oxygen isotope ratios show no change or decrease. This result cannot be explained solely by changes in global temperature or a shift to drier summers. Instead, it is consistent with a decrease in winter precipitation that increases annual mean δ18Op (and clay δ18O) but has a smaller or negligible effect on soil carbonates that primarily form in warmer months. We suggest that forest communities in west‐central North America were adapted to a wet‐winter precipitation regime for most of the Cenozoic, and they subsequently struggled to meet water demands when winters became drier, resulting in the observed open habitat expansion.

(<1 million years) may have triggered the OHT (Strömberg, 2005(Strömberg, , 2011. Drier conditions across the OHT are supported by the expansion of open, grassy vegetation dominated by dry-adapted pooids (Harris et al., 2017;Strömberg, 2005;Webb & Opdyke, 1995;Wing, 1998;Wolfe, 1985), but the survival of moisture-dependent taxa suggests that any aridification was minimal (Strömberg, 2005(Strömberg, , 2011. Instead, increased seasonal aridity (as opposed to annual) may explain the survival of moisture-dependent taxa during grassland expansion (Harris et al., 2017). Increasingly dry summers are supported by the OHT expansion of open woodland and savanna habitats presumed to be adapted to a warm dry season (Webb & Opdyke, 1995;Wing, 1998;Wolfe, 1985). Further, the shift from calcite to silica-rich paleosols in central Oregon across the OHT has been interpreted as showing a transition from summer-wet to winter-wet seasonality (Retallack, 2004b), but the link between precipitation seasonality and soil calcite/silica content remains tenuous. Additional, independent (non-floral) evidence for summer aridity is not available.
The modern relationship between vegetation and precipitation, however, suggests that wintertime moisture has a far greater influence on vegetation than the magnitude of summertime aridity (Clow, 2010;Hu et al., 2010;Knowles et al., 2017Knowles et al., , 2018Trujillo et al., 2012). The balance between winter precipitation supply (mostly snowpack) and spring/summer evaporative demand is closely correlated with gross primary productivity (GPP) in the western US (Hu et al., 2010;Knowles et al., 2018). Moreover, water isotope studies show that trees in mid-latitude North American and European forests generally use more winter moisture than summer during the growing season (Allen et al., 2019;Berkelhammer et al., 2020;Hu et al., 2010). Summer precipitation can drive montane forest GPP (Berkelhammer et al., 2017), especially when winter snowpack is already high (Berkelhammer et al., 2020), but the spatial pattern of open versus closed habitats is more sensitive to annual precipitation (Schimel et al., 2002) which, for most of WCNA, is dominated by winter ( Figure S1 in Supporting Information S1). Today, wetter wintertime climates in WCNA are typically characterized by higher tree cover, consistent with wooded, closed habitats, whereas low winter precipitation yields grassland-dominated ecosystems characterized by low tree cover (Figures 1d and 1e). If this relationship holds through time, then decreases in winter precipitation should lead to decreasing biomass and the expansion of open, grassy habitats.
Here, we test the hypothesis that drier winters-rather than drier summers-led to aridification across WCNA and prompted the expansion of grasslands and open habitats. Precipitation seasonality transitions from winterwet to summer-wet from west to east across WCNA today (Figures 1c and 1d), so testing this hypothesis requires data that cover the OHT in time and space. Oxygen isotopes in precipitation (δ 18 O p ; ‰) derived from proxy records can be used to address this hypothesis because δ 18 O p is sensitive to seasonality and aridity Dansgaard, 1964;Kukla et al., 2019;Mix et al., 2013;Salati et al., 1979;Winnick et al., 2014) and δ 18 O proxy coverage across WCNA is temporally comparable to, and often co-located with, paleobotanical archives ( Figure 1a). Oxygen isotopes are particularly useful because, while precipitation seasonality changes from west to east (Figures 1c and 1d), the seasonality of δ 18 O p does not. Precipitation δ 18 O is higher in the summer and lower in the winter across the entire WCNA (a correlation known as the "temperature effect" (Dansgaard, 1964); Figure 1b). Thus, a change in the relative contribution of winter versus summer moisture would likely be accompanied by the same direction of change in δ 18 O p across WCNA.
We compare oxygen isotopes of two independent proxy materials-authigenic clay and soil carbonate-to reconstruct past precipitation seasonality. Soil carbonates can form on seasonal (and shorter) timescales, preserving seasonally biased oxygen isotope signals (Breecker et al., 2009;Gallagher & Sheldon, 2016;Huth et al., 2019;Kelson et al., 2020;Peters et al., 2013). The timing of soil carbonate formation is strongly influenced by precipitation seasonality with winter-wet (summer-wet) climates generally favoring carbonate formation in the summertime (spring/fall) when soil CO 2 and moisture are declining (Gallagher & Sheldon, 2016;Huth et al., 2019;Kelson et al., 2020;Peters et al., 2013). Clay minerals, in contrast, form on much longer timescales making them less likely to record a seasonal bias (Maher et al., 2009;Palandri & Kharaka, 2004;White et al., 2008), and are usually interpreted to reflect precipitation-weighted oxygen isotope ratios (Gao et al., 2021;Lawrence & Taylor, 1971;Stern et al., 1997;Tabor et al., 2002). Since clays are more likely to record wet season δ 18 O p than carbonates, the difference between clay and carbonate δ 18 O depends on (a) how much total precipitation falls in the wet season (the magnitude of seasonality) and (b) whether the wet season is the high-δ 18 O or low-δ 18 O season (i.e., summer or winter in WCNA, respectively).

New Paleosol Carbonate and Authigenic Clay Data
Our new data help fill key spatial and temporal gaps in the west to east transect. The data comprise 173 samples spanning the Salmon Basin in eastern Idaho (Harrison, 1985;Janecke & Blankenau, 2003;Schwartz et al., 2019), the Muddy Creek Basin in southwestern Montana (Dunlap, 1982;Janecke et al., 1999;Schwartz et al., 2019), the Flagstaff Rim region of Wyoming (Emry, 1973(Emry, , 1992Evernden et al., 1964), and the Toadstool Park region of northwestern Nebraska (Terry, 2001;Zanazzi et al., 2007). Salmon Basin data include 21 authigenic clay samples spanning the middle-late Oligocene following the stratigraphy of Harrison (1985) and the age constraints compiled in Schwartz et al. (2019) (Axelrod, 1998;M'Gonigle & Dalrymple, 1993). We did not find any soil 10.1029/2021AV000566 4 of 20 carbonates of middle-late Oligocene age in the Salmon Basin, perhaps due to wetter conditions inhibiting carbonate formation. Muddy Creek Basin data span the Eocene and include both clay (n = 4) and carbonate (n = 29) samples pinned to the straigraphy of Dunlap (1982) with age constraints from Janecke et al. (1999). Most of the samples collected in the Muddy Creek Basin come from paleosols containing coeval gypsum indicative of evaporative conditions. Flagstaff Rim samples are all carbonates (n = 30) that span the late Eocene and are linked to the stratigraphy of Emry (1992) and the compiled age constraints therein (Evernden et al., 1964;Swisher & Prothero, 1990). Toadstool Park data cover the late Eocene-early Oligocene with both clay (n = 4) and carbonate (n = 59 data). Our Toadstool samples are pinned to the stratigraphy of Terry (2001) following the age model of Zanazzi et al. (2007).

Carbonate Stable Isotopes
Carbonate samples collected as part of this study (n = 140) were powdered using mortar and pestle or a handheld drill. Stable carbon and oxygen isotopes were measured at the Stanford University Stable Isotope Biogeochemistry Laboratory using a Thermo Finnigan Gasbench peripheral preparation system with isotope ratios measured in continuous flow (Thermo Finnigan ConFlo III) on a Finnegan MAT Delta + XL mass spectrometer. Based on carbonate content 250-800 μg of sample powder were reacted with 0.25 ml of 105% phosphoric acid for 1 hr at 72°C. External precision (1 σ) of oxygen and isotope data was ±0.1‰ based on repeat measurements of internal carbonate standards that have been calibrated against NBS-18, NBS-19, and LSVEC. We report all δ 18 O values relative to Vienna Standard Mean Ocean Water (VSMOW).

Clay Stable Isotopes
Stable oxygen isotopes of authigenic clay minerals collected as part of this study (n = 33) were measured at the Stanford University Stable Isotope Biogeochemistry Laboratory. About 250 g of bulk rock sample was suspended in DI water and the <0.2 μm size fraction was separated out via a Thermo IEC centrifuge. The <0.2 μm fraction was subsequently sequentially cleaned to remove contaminants: (a) 0.5 mol/L sodium acetate buffer solution to remove carbonates when necessary (Savin & Epstein, 1970); (b) 3% hydrogen peroxide solution to remove organic matter; and (c) ammonium citrate/sodium dithionite solution (80°C) to remove iron oxyhydroxides (Bird et al., 1993;Stern et al., 1997). Samples were then rinsed 5 times with deionized water and dried in oven at 60°C for at least 40 hr. Clay mineralogy was assessed via X-ray diffraction at the Environmental Measurements Facility at Stanford University. Analyses were conducted using a Rigaku MiniFlex 600 Benchtop X-ray Diffraction System equipped with a Cu anode set at the maximum power of 600 W. Measurements were repeated after glycolation with one cm of ethylene glycol at the base of a sealed desiccator left overnight in an oven set to 65°C (Poppe et al., 2001). Mineral identification was determined with the Rigaku PDXL software and the USGS X-ray Diffraction Lab manual (Poppe et al., 2001). Samples contained predominantly smectite with minor contributions of quartz ( Figure S2 in Supporting Information S1). To prepare for isotope analyses clay samples were mixed with LiF and pressed into pellets. Samples were dried at 80°C in a vacuum oven prior to analysis.
Oxygen isotopic composition was determined using a laser fluorination line and measured on a Thermo Finnigan MAT 252 mass spectrometer in a dual inlet configuration (e.g., Sharp, 1990;Sjostrom et al., 2006;Mix et al., 2016). Prior to laser fluorination samples were exposed to three BrF 5 prefluourinations at 30 millitorr and then fluorinated using a New Wave Research MIR 10-25 infrared CO 2 laser in a 130 millitor BrF 5 atmosphere. Oxygen gas (O 2 ) was purified through liquid nitrogen cold traps and heated KBr trap prior to being frozen on a 5Å mol sieve (zeolite) prior to equilibration of the purified O 2 with the sample bellows. Measurements were made at a 2V intensity on mass 32 O 2 and were standardized by measurements of NBS-28 quartz and bracketing using in-house smectite standard DS069, run in between samples to assess for drift. The external reproducibility was assessed based on corrected values of NBS-28 and the DS069 standard, giving values of ±0.1‰ for each standard. Isotopic ratios are reported relative to VSMOW.

Phytolith Data Compilation
Among paleobotanical sources of evidence, phytoliths stand out for providing information that is relevant for reconstructing the spread of open, grass-dominated habitats (Strömberg et al., 2018). Specifically, phytolith assemblages have been shown to reliably reflect the composition and dominance of grass communities in past vegetation on local to regional scales, as demonstrated by a body of modern analog work for example, Barboni et al. (2007); Iriarte and Paz (2009);Novello et al. (2012); Strömberg (2020, 2021). We therefore limited our compilation of paleovegetation data to phytolith studies and use these to define the OHT.
Phytolith-based data on the relative abundance of open-habitat grasses in the Cenozoic of Western North America were collected from the literature (Supporting Information Data; Chen et al., 2015;Cotton et al., 2012;Harris et al., 2017;Hyland et al., 2019;Miller et al., 2012;Strömberg, 2005;. The compilation includes 216 phytolith assemblages ranging from the middle Eocene (∼40 Ma) to the latest Miocene (∼5 Ma) from Montana, Idaho, Nebraska, Kansas, Colorado, and Wyoming. These studies employed phytolith-based plant functional groups (PFT) following Strömberg and McInerney (2011) and thereafter (see Strömberg et al., 2018), namely: (a) the total sum of forest indicator morphotypes (FI TOT) found in palms, woody or herbaceous dicotyledons, conifers, ferns, and tropical monocotyledonous herbs in the Order Zingiberales; (b) grass silica short cell phytoliths (GSSCP), exclusively produced by grasses (Poaceae), which can be subdivided into morphotypes typical of closed-habitat grasses (CH; e.g., bambusoid and early-diverging grasses, such as Anomochlooideae) and open-habitat (OH) grasses in the Pooideae (POOID-D, which are nearly exclusively produced by Pooideae, and POOID-ND, which are produced by many grass lineages but are most abundantly and commonly found in the C 3 Pooideae) and PACMAD clade (PACMAD TOT; C 3 and C 4 grasses in the subfamilies Panicoideae, Aristidoideae, Chloridoideae, Micrairioideae, Arundinoideae, and Danthonioideae (Grass Phylogeny Working Group II, 2012), but also contains forms that cannot (currently) be assigned to a specific grass PFT because they are widely produced, as well as GSSCP that cannot be identified because they are broken or obscured (OTHG); (c) morphotypes typical of certain plants often associated with wetlands (AQ), such as sedges and horsetails; and (d) non-diagnostic and unclassified forms (OTH). OTH contains both forms commonly or exclusively produced by grasses, but that are not suitable for indicating grass biomass (NDG), and forms that are found in such a broad range of plants they are non-diagnostic (NDO; . To assess the relative abundance of open-habitat grasses in plant communities, we focused on the relative abundance of open habitat (OH) grasses (OH = POOID-D + POOID-ND + PACMAD TOT) in the sum of all diagnostic phytoliths (FI TOT + all GSSCP). To account for the fact that OTHG undoubtedly contains some GSSCP attributable to open-habitat grasses (as opposed to, in effect, all being counted as closed-habitat grasses), we "scaled" the relative abundance of OH GSSCP following . Specifically, the proportion (%) OH in vegetation was calculated as % OH phytoliths out of GSSCP-OTHG scaled to the relative abundance of all GSSCP out of the diagnostic sum (FI TOT + GSSC). In doing so, we reasonably assumed that OTHG contains the same proportion of CH and OH phytoliths as does the sum of CH, POOID-D, POOID-ND, and PACMAD TOT GSSCP (see discussion in McInerney et al., 2011). We inferred 95% confidence intervals (unconditional case, using the total count as the sample size) for the %OH phytoliths using a bootstrap routine in the statistical software R (R Team, 2021; code available upon request). Note that in several Eocene-Oligocene samples from Montana/Idaho, GSSCP assemblages were dominated by forms that are reminiscent of morphotypes typical of open-habitat grasses; however, their exact affinity and ecology remains unclear (Retallack, 2004a;Strömberg, 2005). These assemblages are marked as transparent in Figure 2.

Isotope Data Compilation and Study Domain
We compile stable isotope data within WCNA spanning the last 50 million years and filter and process the data in three steps. Starting with 3400 data points (all of the compiled and new data), we first remove samples outside of our study domain (40-47°N; −123 to −100°E) and filter for smectite, kaolinite, and calcite minerals, excluding lacustrine carbonates, yielding 2238 data points. Of the clay samples, only five are kaolinite (of 268) and the rest are smectite as determined by X-ray diffraction by the original authors. We also eliminate Quaternary data (the last 2.6 Myr) to avoid confounding signals with glacial-interglacial cyclicity (especially the Laurentide ice sheet) which are known to have changed both mean annual precipitation and the pattern of atmospheric circulation across the western U.S (Amundson et al., 1996;Badgley & Finarelli, 2013;Oster et al., 2015;Oerter et al., 2016). Second, following the original authors' interpretations, we eliminate data that (a) do not reflect a primary meteoric signal (n = 61; 2.5% of all data, (this study; Chamberlain et al., 2012;Horton et al., 2004;McLean & Bershaw, 2021)); (b) were updated in a later publication (n = 5; 0.2%, (Kent-Corson et al., 2006)); or (c) record a transient isotope excursion such as a climate event that may not reflect long-term, background conditions (n = 6; 0.2%, ). Of the samples determined to not reflect primary meteoric conditions, most (n = 56) are from the Muddy Creek Basin data presented in this study where we sampled from strata with interlayered gypsum (indicative of strong evaporation) and δ 18 O values are generally high (although the evaporative designation of samples depends solely on the presence of gypsum). Finally, we average sample replicates so there is a single isotope value for each sample, yielding a total of 1851 samples. Data for each data processing step and an R script to conduct all data processing and statistical analyses are in the Supporting Information S1.
To produce the time series in Figure 2, oxygen isotope ratios are averaged by sampling site. Sampling sites are defined by distinct sampled sections as reported by the original authors or based on proximity of sample coordinates. Records that cover more than 2 million years and reveal a long-term trend were sub-sampled at 1 million year intervals so the trend is not averaged out. In the eastern domain, δ 18 O is positively correlated with longitude (see Figure 3) and we de-trend the data using a linear regression applied to all eastern domain clay and carbonate  (Miller et al., 2012;Strömberg, 2005). The OHT has also been identified in the western domain with different methods (e.g., Retallack, Wynn, & Fremd, 2004;Retallack, Orr, et al., 2004;Wheeler et al., 2006). data. De-trending these data ensures that trends in the eastern domain data through time are not attributable to changes in sampling density across the domain.
Finally, we calculate the pre-OHT δ 18 O value for each domain by taking the average δ 18 O of all data prior to 26 Ma (the OHT spans ∼26-15 Ma). This pre-OHT value is subtracted out from each domain so the domains can be directly compared in the time series of Figure 2. It was previously recognized that there are two sub-domains in the western domain clay δ 18 O data due to local topographic effects in Oregon  and we subtract a pre-OHT value for each Oregon sub-domain to account for the δ 18 O offset between them. Each sub-domain records the ∼3‰ increase in clay δ 18 O across the OHT . The carbonate data in the west come from the same sub-domain, except for one data point (McLean & Bershaw, 2021) that we remove because it is insufficient to analyze trends through time.  (Kim & O'Neil, 1997) and clays (Sheppard & Gilg, 1996) we can derive the following equations to predict co-equilibrium Δδ 18 O clay−carb as a function of temperature (see Text S1 in Supporting Information S1 for derivation): For the range of environmental temperatures, co-equilibrium Δδ 18 O clay−carb for smectite-and kaolinite-calcite is near −3.5‰ (±∼0.3‰ depending on the temperature; Figure S3 in Supporting Information S1). Because the co-equilibrium Δδ 18 O clay−carb is not very sensitive to temperature, we can interpret Δδ 18 O clay−carb values near −3.5‰ as clays and carbonates likely forming under similar conditions regardless of the formation temperature.

The Difference in Clay and
However, previous work comparing clay and carbonate δ 18 O has rarely found the minerals to be in co-equilibrium (Gao et al., 2021;Poage & Chamberlain, 2002;Stern et al., 1997;Tabor et al., 2002;Torres-Ruíz et al., 1994). Instead, Δδ 18 O clay−carb is consistently below −3.5‰ in places where the wet season is thought to be the low-δ 18 O season (Stern et al., 1997;Tabor et al., 2002) and above −3.5‰ when the wet season is the high-δ 18 O season (Gao et al., 2021;Poage & Chamberlain, 2002). These findings are consistent with soil carbonates forming outside of the wet season (Breecker et al., 2009;Gallagher & Sheldon, 2016;Huth et al., 2019;Kelson et al., 2020;Peters et al., 2013) and clays forming more slowly (Maher et al., 2009;Palandri & Kharaka, 2004;White et al., 2008) capturing precipitation-weighted conditions (Lawrence & Taylor, 1971). Soil carbonates are also known for predominantly forming in warm months, and some studies have found that seasonal biases in carbonate formation (determined via clumped isotope thermometry) do not necessarily correspond to the same seasonal bias in carbonate δ 18 O (e.g., Kelson et al., 2020). We consider the implications of carbonate δ 18 O and formation biases differing in the discussion. Differences in mineral formation temperature can also affect Δδ 18 O clay−carb , but Δδ 18 O clay−carb is about three times more sensitive to δ 18 O p seasonality than temperature seasonality ( Figure  S4 in Supporting Information S1). Taken together, since WCNA δ 18 O p is higher in the summer and lower in the winter, we expect Δδ 18 O clay−carb to be higher in a summer-wet climate, and lower in a winter-wet climate. While this basic theoretical framework is supported by independent clay (Lawrence & Taylor, 1971) and soil carbonate studies (Gallagher & Sheldon, 2016;Kelson et al., 2020;Peters et al., 2013), we are not aware of any modern work directly comparing clay and carbonate δ 18 O.

Calculating the Clay-Carbonate δ 18 O Difference From Data
In order to quantify Δδ 18 O clay−carb in each domain at each timeslice we conduct a two-sample Student's t-test (using the t.test function in R version 4.0.2) comparing clay and carbonate δ 18 O for each designation (west, central, east; pre-OHT, post-OHT). Pre-OHT and post-OHT are defined as all data before 26 Ma and after 15 Ma, respectively. We exclude data within the OHT in order to unambiguously compare pre-and post-OHT conditions, recognizing that the precise timing of the transition and whether it occurred synchronously or asynchronously across WCNA remains uncertain. The t-test returns the difference in clay and carbonate δ 18 O means and the 95% confidence interval around this difference. For the eastern domain, the data were de-trended using the same regression line noted in Section 2.5. For the western domain we do not consider clay or carbonate δ 18 O from the sub-domain that has only one carbonate value (see Section 2.5). We test whether Δδ 18 O clay−carb changes significantly (p < 0.05; null hypothesis that Δδ 18 O clay−carb does not change) in each domain from pre-to post-OHT using a two-way analysis of variance test (ANOVA) in R (the glm function, R version 4.0.2).
In order to further validate our Δδ 18 O clay−carb results, we repeat the above analysis for the mean δ 18 O at each sample site, rather than each individual measured sample ( Figure S5 in Supporting Information S1). If the site-averages give similar results for each domain, we can be confident that our results are not biased by sites with more samples. The p-value for the change in Δδ 18 O clay−carb in the site-average analysis is below 0.05 in each domain. Site averages also yield similar Δδ 18 O clay−carb results as individual samples. Therefore, we do not find any evidence that our data are biased by densely sampled sites.

Oxygen Isotope and Vegetation Data Through Time
We first compare trends in phytolith (plant biogenic silica) and isotope data across WCNA through time ( Figure 2). Phytolith data coverage is restricted to the central and eastern domains (existing data to the west and south are outside the bounds of isotope data for comparison 40-47°N; −123 to − 100°E; Figure 1; e.g., Dillhoff et al., 2014;Smiley et al., 2018;Loughney et al., 2020). However, the OHT has also been identified in the western domain from fossil megafloras and paleosol morphology and these sites are noted by "other fossil" in Figure 1a (Retallack, 2004a(Retallack, , 2004bRetallack, Orr, et al., 2004). Figure 2c shows the percent of phytoliths that indicate open habitat conditions in the updated WCNA phytolith compilation. The percent of open habitat phytoliths (and the percent of grass phytoliths; Figure S6 in Supporting Information S1) increases between 26 and 15 Ma from ∼20% to ∼73%, defining the OHT. Since the major trends in the proxy data occur across the OHT, not before or after, we directly compare pre-and post-OHT oxygen isotope data in the next section to quantify their relative differences.
Authigenic carbonate δ 18 O data vary by more than 10‰ with no uniform trends across the OHT (Figure 2b and adjacent box plot). Clay δ 18 O, in contrast, increases by ∼1%-3‰ across the OHT in each domain (Figure 2a). The increase in clay δ 18 O is statistically significant (p < 0.05) in the full and site averaged datasets in the western and central domains whereas, in the east, it is significant in the site averaged data (p = 0.03) and just above the significance level with the full data set (p = 0.08). We therefore consider each increase in clay δ 18 O to be significant. In contrast, when considering the full and site averaged carbonate data there is no significant change in the western and central domains and a significant decrease in δ 18 O in the east (see Table S1 in Supporting Information S1). Carbonate δ 18 O increases significantly in the western domain in the full data set, but this increase is not significant in the site averaged data, suggesting the signal is driven by a small number of densely sampled sites or site averaged data coverage is insufficient to resolve the signal. We note that the carbonate δ 18 O record in the western domain does not start until just prior to the OHT due to absence of soil carbonates in paleosol strata (Bestland et al., 1997(Bestland et al., , 2002. In general, the first soil carbonates to form in the western domain show features similar to groundwater carbonates, rather than nodules forming well within the vadose zone  Text S2 and Figure S7 in Supporting Information S1).

Oxygen Isotopes of Clay and Carbonate by Geographic Domain
Clay and carbonate δ 18 O generally decrease from the western to central domain, then increase through the eastern domain. This longitudinal pattern is similar to that of modern meteoric and surface water δ 18 O ( Figure S8 in Supporting Information S1), which reflects the balance between westerly rainout in the west and central domains, and increasing mixing with warm season, Gulf of Mexico moisture to the east (Kendall & Coplen, 2001;Z. Liu et al., 2010).
This spatial pattern of clay and carbonate δ 18 O is similar before and after the OHT, but the offset between clay and carbonate δ 18 O (Δδ 18 O clay−carb ; ‰) is not (Figures 3a and 3b). Within a given domain, clay and carbonate δ 18 O approach more similar values (Δδ 18 O clay−carb closer to zero) after the OHT than before (Figure 3c Figure S5 in Supporting Information S1).
In the western domain, the mean Δδ 18 O clay−carb is −7.2 ± 0.9‰ before the OHT and −5.6 ± 1.0‰ after (mean and 95% confidence interval). The clay-carbonate δ 18 O difference before and after the OHT is largest in the central domain where Δδ 18 O clay−carb is −5.6 ± 0.5‰ before and −2.2 ± 0.7‰ after. And in the east, Δδ 18 O clay−carb increases from −2.9 ± 0.4‰ to −0.2 ± 1.0‰ across the OHT. The increase in Δδ 18 O clay−carb in the three domains ranges from 1.6‰ to 3.3‰ with an average increase of 2.6‰.

Clay and Carbonate Oxygen Isotope Trends Across the OHT
Previous work speculated that the increase in western domain clay δ 18 O across the OHT can be explained by a greater fraction of annual precipitation occurring in summer, perhaps due to Cascades Range uplift disproportionately blocking winter moisture compared to summer . A disproportionate decrease in winter precipitation could also explain the increase in clay δ 18 O in the central and eastern domains. However, a gap in our central domain data across the OHT prohibits a direct link to the western domain data. Additionally, if the eastern domain increase in clay δ 18 O (and decrease in carbonate δ 18 O) is driven by less winter precipitation, the causal mechanism may differ from the western domain since the δ 18 O shift appears delayed (Figure 2a; we return to this point in Section 4.3).
The increase in clay δ 18 O, by itself, is not compelling evidence for a change in the seasonality of precipitation. Many other factors could drive the same isotopic trend, including an increase in δ 18 O p driven by warmer temperatures (Dansgaard, 1964), an increase in clay-water isotopic fractionation driven by cooler temperatures (Sheppard & Gilg, 1996), greater soil evaporation or evaporation of falling raindrops (Barnes & Allison, 1983;Lee & Fung, 2008;Zimmermann et al., 1967), and more upwind recycling of moisture back into the atmosphere Kukla et al., 2019;Mix et al., 2013;Salati et al., 1979;Winnick et al., 2014). However, global climate both warmed and cooled across the OHT (Westerhold et al., 2020;Zachos et al., 2001), so temperature is unlikely to explain the unidirectional δ 18 O increase. Soil and post-condensation evaporation can increase with drying, but the isotopic effect would likely be stronger in soil carbonates than clays because evaporation is highest in warmer months when soil carbonates tend to form. This is inconsistent with our finding that the main δ 18 O increase occurs in clays, while carbonate δ 18 O stays the same or decreases. Finally, upwind moisture recycling has little effect on δ 18 O p at sites close to the coast where there is minimal upwind land area for recycling to occur, such as in the western domain (Chamberlain et al., 2012;Kukla et al., 2019Kukla et al., , 2021. Thus, it is unlikely that upwind recycling could drive a δ 18 O increase in all three domains. We also note that there is no change in the mineralogy of clay samples across the δ 18 O shift. Importantly, the mechanism for the increase in clay δ 18 O must also be consistent with the carbonate δ 18 O data, which either stay the same or decrease across the OHT. If clays and carbonates are generally recording the same information through time then we would not expect diverging δ 18 O trends. The lack of a corresponding increase in soil carbonate δ 18 O supports the hypothesis that the clay δ 18 O shift is driven by a decrease in winter precipitation. Mid-latitude soil carbonates should generally be less sensitive to changes in winter precipitation due to their strong bias to forming in warmer months, as supported by clumped isotope studies (e.g., Kelson et al., 2020). This delay between winter precipitation and warm season soil carbonate formation implies that soil carbonates either (a) form in contact with precipitation that falls in warmer months (higher δ 18 O) that is not influenced by changes in winter precipitation, or (b) carbona form in contact with a combination of warm season precipitation and evaporating groundwater (which is often sourced from winter moisture; Jasechko et al., 2014), such that summer precipitation dampens any winter signal (Quade et al., 1989). Indeed, late Quaternary soil carbonates in the western U.S. suggest that winter moisture can be incorporated in summer carbonate formation (Kelson et al., 2020), but the influence of winter moisture appears small except at the most winter-wet sites ( Figure S9 in Supporting Information S1; Kelson et al., 2020). This result is consistent with clays being more sensitive to winter drying than carbonates, explaining why carbonate δ 18 O does not show the same increase observed in the clay data.
The decrease in eastern domain carbonate δ 18 O can also be explained by winter drying. Studies of modern (or latest Quaternary) soil carbonates have found that carbonates tend to form in the summer in more winter-wet conditions, and spring or fall in summer-wet conditions (e.g., Gallagher & Sheldon, 2016;Kelson et al., 2020;Peters et al., 2013). Wetter winter conditions prior to the OHT could have restricted soil carbonate formation to the warmest months of the year when δ 18 O p is highest. Subsequent winter drying would lead to drier springtime soils, likely increasing carbonate formation in the spring when soil moisture δ 18 O is lower due to a combination of lower precipitation δ 18 O and less evaporatively enriched groundwater recently recharged by winter (low-δ 18 O) precipitation and snowmelt (e.g., Jasechko et al., 2014). We would not expect to see the same shift in the timing of carbonate formation in the central and western domains because these regions are more winter-wet today, indicating that drier winters across the OHT would not have established such a summer-wet climate as exists in the east. Indeed, the greater influence of summer moisture in the eastern domain can also explain the relatively muted increase in clay δ 18 O compared to the central and western records (Table S1 in Supporting Information S1). If summer precipitation was already high in the east, drier winters would have a diminished effect on precipitation-weighted annual (and therefore clay) δ 18 O.

Precipitation Seasonality Before and After the OHT
The difference between clay and carbonate δ 18 O (Δδ 18 O clay−carb ) before and after the OHT provides additional insight to the role of precipitation seasonality. If carbonates are generally less sensitive to winter drying than clays, then Δδ 18 O clay−carb will differ from the co-equilibrium value of ∼−3.5‰, especially in more winter-wet regions. Specifically, because winter is the low-δ 18 O season in WCNA today, Δδ 18 O clay−carb will become more negative with a more winter-wet climate due to an amplified winter bias in clay δ 18 O compared to carbonate. Alternatively, we expect Δδ 18 O clay−carb to increase in a less-winter wet (more summer-wet) climate. If summer is wet enough to restrict soil carbonate formation (and its oxygen isotope bias) to spring and fall conditions, as argued above, then Δδ 18 O clay−carb will increase and may exceed −3.5‰ due to a summer (high-δ 18 O) bias in clay δ 18 O compared to carbonate. Ultimately, because the relative biases in clay and carbonate mineral formation and δ 18 O depend on precipitation seasonality, changes in Δδ 18 O clay−carb can encode changes in the seasonal balance of precipitation in the past.
An implicit assumption in our analysis is that, like today, winter has been the low-δ 18 O p season across WCNA since the Paleogene (see Figure 1b). Isotope-enabled climate models and oxygen isotope data from seasonal bivalves confirm that winter has been the low-δ 18 O season since at least the early Eocene (∼50 Ma; Feng et al., 2013;Morrill & Koch, 2002;Norris et al., 1996). Moreover, the atmospheric circulation patterns that set WCNA δ 18 O seasonality-high-δ 18 O convective precipitation during warm months and synoptic-scale, low-δ 18 O precipitation during cold months (Feng et al., 2013;Z. Liu et al., 2010)-are robust in models, even in hot Eocene climates (Feng et al., 2013;Sewall & Sloan, 2006) and with much lower WCNA topography (Feng et al., 2013). Strong correspondence between west-east trends in paleo-proxy and modern δ 18 O provides additional support that the basic circulation patterns are unchanged ( Figure S8 in Supporting Information S1). The seasonal δ 18 O amplitude and baseline very likely changed in the past, but this does not violate our assumption.
By assuming winter was the low-δ 18 O season throughout our record, we can infer precipitation seasonality from Δδ 18 O clay−carb . As stated above, we interpret higher Δδ 18 O clay−carb values as consistent with a more summer-wet climate and lower Δδ 18 O clay−carb as more winter-wet (illustrated in Figure 4a). In theory, the Δδ 18 O clay−carb co-equilibrium value of −3.5‰ represents the threshold between winter-wet and summer-wet climates in WCNA, but this has yet to be verified with data from modern (or late Quaternary) soils. When Δδ 18 O clay−carb is equal to the co-equilibrium value we can infer that (a) no season is wet enough to inhibit carbonate formation so clays and carbonates form at the same time of year; or (b) clays and carbonates form at different times of year but with the same mean δ 18 O p and temperature conditions. The latter is possible if, for example, clays form year-round while carbonate formation is restricted to the spring or fall with δ 18 O p and temperature approximating annual mean conditions. We argue that the OHT coincides with a shift to drier winters that may mark the establishment of modern precipitation seasonality across WCNA. Increasing Δδ 18 O clay−carb after the OHT points to drier winters that increased the summertime (high-δ 18 O p ) fraction of annual precipitation (increasing clay δ 18 O) with a negligible effect on summer (and soil carbonate) δ 18 O in the west and central domains and a counteracting effect on the timing of carbonate formation in the east. The increase in Δδ 18 O clay−carb occurs in all domains and, if −3.5‰ represents a winter-wet versus summer-wet threshold, would likely mark the onset of summer-wet conditions in the east. This spatial pattern of precipitation seasonality is similar to modern (although the central domain Δδ 18 O clay−carb is slightly higher than expected), suggesting the modern gradient was established around the time that open, grassy habitats expanded.
Other mechanisms that can alter Δδ 18 O clay−carb are unlikely to explain our results. For example, an increase in winter δ 18 O could explain a shift toward higher clay δ 18 O in the west and central domains, but global cooling and Cascades uplift since the OHT (Bershaw et al., 2019;Kohn & Fremd, 2007;McLean & Bershaw, 2021;Pesek et al., 2020;Reiners et al., 2002;Takeuchi et al., 2010;Westerhold et al., 2020;Zachos et al., 2001) should, all else being equal, decrease winter δ 18 O. Colder clay formation temperatures are also consistent with increasing clay δ 18 O and Δδ 18 O clay−carb , but any cooling would likely occur in the winter and summer, causing the same signal in soil carbonate δ 18 O. The OHT itself might decrease soil CO 2 by decreasing productivity, which can affect carbonate δ 18 O by changing the seasonal timing of carbonate formation. However, if carbonate formation tracks the timing of peak photosynthesis and root water uptake (Meyer et al., 2014), we expect the OHT to shift soil carbonate to the summer, likely increasing carbonate δ 18 O and decreasing, not increasing, Δδ 18 O clay−carb , in conflict with our findings. Additionally, lower soil productivity should increase soil carbonate δ 13 C by decreasing soil respiration (Caves et al., 2016;Licht et al., 2020;Rugenstein & Chamberlain, 2018), and this is not observed ( Figure S10 in Supporting Information S1).
Our results are also difficult to reconcile with the idea that carbonates and clays, although forming at different times of the year, are recording essentially the same wet season moisture (e.g., Torres-Ruíz et al., 1994;B. Liu et al., 1996;Stern et al., 1997). If this moisture is unevaporated, Δδ 18 O clay−carb should stay close to the co-equilibrium value of −3.5‰, although it is more likely that carbonates would record some evaporative offset ( 18 O enrichment) of this moisture (Quade & Cerling, 1995;Quade et al., 1989;Stern et al., 1997). If this evaporative offset is constant over space relative to precipitation-weighted δ 18 O p , then most Δδ 18 O clay−carb variations will be due to differences in mineral formation temperature (since clays and carbonates would effectively record the same source water), in which case Δδ 18 O clay−carb should be higher in winter-wet regions due to a relative cold bias in clay formation that increases clay δ 18 O. This would lead to Δδ 18 O clay−carb decreasing from west-to-east, which is not observed either before or after the OHT. Alternatively, a larger evaporative offset in winter-wet climates due to drier summers could reverse this effect by increasing carbonate δ 18 O relative to clay and decreasing Δδ 18 O clay−carb . However, this means the evaporative offset would have to decrease in each domain across the OHT in order to increase Δδ 18 O clay−carb . Since carbonates would be forming in contact with this less-evaporated moisture, a decrease in summer evaporation would primarily affect carbonate, rather than clay δ 18 O, making this scenario is unlikely because the clay δ 18 O data drive most of the WCNA Δδ 18 O clay−carb increase. Additionally, a decrease in the evaporative offset of δ 18 O would make clay and carbonate formation waters more isotopically similar, likely shifting Δδ 18 O clay−carb closer to −3.5‰, yet Δδ 18 O clay−carb increases away from −3.5‰ in the eastern domain (the only site where carbonate δ 18 O decreases). Thus, it is difficult to reconcile the spatial pattern of Δδ 18 O clay−carb and its increase across the OHT with a scenario where clays and carbonates are recording the same moisture source, even if there is some evaporative offset between the two minerals. We therefore conclude that our Δδ 18 O clay−carb results require relative seasonal source moisture biases between clays and carbonates. Figure 4 presents a conceptual model for our results, linking Δδ 18 O clayclay−carb to precipitation seasonality ( Figure 4a) and mapping these seasonal trends on our WCNA domain (Figure 4b). Based on the relationship between the average post-OHT Δδ 18 O clay−carb and modern precipitation seasonality (DJF/(JJA + DJF)) in each domain, the increase in Δδ 18 O clay−carb values is consistent with about a 25% decrease in the winter fraction of precipitation across the OHT (Figure S11 in Supporting Information S1; Figure 4b; this analysis effectively assumes modern precipitation seasonality was established post-OHT). This approximation is crude, but it captures the winter precipitation trend and illustrates two useful points. First, the Cascades Range, which marks the boundary between very winter-wet and somewhat winter-wet climates (Figure 4b) was likely a less effective barrier to the wintertime westerlies before the OHT (Figure 4b). Second, summer precipitation accounted for a smaller fraction of annual precipitation before the OHT, precluding a summer-wet climate in Oregon prior to grassland expansion (Retallack, 2004b).
Our results are not compatible with the demise of a summer-wet climate in the western domain at the OHT (Retallack, 2004b), even when we ignore our assumption that winter has remained the low-δ 18 O p season. For example, in summer-wet "monsoon" climates summer is often the low-δ 18 O p season. If the OHT marks the demise of summer-wet conditions and low-δ 18 O p summers, then summer drying would shift the wet season from a low-δ 18 O p summer to a low-δ 18 O p winter. This effect would likely lead to no change or a decrease in clay δ 18 O and possibly an increase in carbonate δ 18 O (see Text S3 in Supporting Information S1), in conflict with our results. Changes in mineral formation temperatures might counteract these effects, but are unlikely to erase them because Δδ 18 O clay−carb is nearly 3x more sensitive to precipitation seasonality than to temperature seasonality ( Figure S4 in Supporting Information S1) and soil temperature seasonality is generally dampened relative to surface temperatures (Gallagher et al., 2019;Hillel, 1982). While we cannot unequivocally rule out temperature and summer precipitation change across the OHT, our results indicate that the largest climatic signal comes from a decrease in winter precipitation.

A Mechanism for Winter Drying
While our analysis cannot confirm a cause for winter drying, it allows us to propose a testable hypothesis based on three observations. First, the isotope trends appear unidirectional (like the OHT itself), so the forcing may have been unidirectional, too. Second, topography is the main driver of the spatial pattern of modern precipitation seasonality today. The Cascades ridgeline divides very winter-wet from somewhat winter-wet climates, and the continental divide separates somewhat winter-wet from neutral or summer-wet climates (Lora & Ibarra, 2019; Figure 1). Third, the clay and carbonate δ 18 O trends are not the same in each domain, and the eastern clay δ 18 O increase lags behind the western domain. Based on these observations, we propose winters became drier due to tectonics-most likely the uplift of the Cascades Range.
Cascades uplift is a plausible driver of drier winters because it was (mostly) unidirectional and the Cascades mark a significant boundary in precipitation seasonality today (Figure 4b). The Cascades block more winter precipitation than summer precipitation as they directly intercept winter westerly moisture (Rutz et al., 2014;Siler & Durran, 2016;Siler et al., 2013), and this seasonal difference in rainshadow strength is evident in oxygen isotopes of precipitation-δ 18 O p decreases more over the Cascades in the winter and less in the summer (Z. Liu et al., 2010). The seasonality of rainshadow strength means that, east of the Cascades, uplift will have two effects on annual mean δ 18 O p that act in opposite directions: (a) more rainout with uplift decreases δ 18 O p and (b) greater blocking of winter precipitation means that summer (high δ 18 O p ) accounts for a larger fraction of total annual precipitation, thereby increasing δ 18 O p . These two effects are necessarily linked-lower δ 18 O implies more windward rainout (effect 1), and more windward rainout implies less precipitation inland (effect 2). The clay mineral response in the lee (to the east) of uplift depends on which effect is greater. Results from a simple two end-member mixing model show that it is plausible for the seasonality effect to outpace the uplift effect, causing an increase in δ 18 O p with Cascades uplift that is consistent with the magnitude of the clay δ 18 O increase (Text S4 and Figure S12 in Supporting Information S1).
The uplift of the Cascades is generally consistent with the relative changes in δ 18 O across the OHT. Because the Cascades form a stronger winter rainshadow, we expect their uplift to have the largest effect on precipitation seasonality in the most winter-wet region. This is consistent with the largest clay δ 18 O increase occurring in the west and the smallest in the east. Still, when analyzing the full data set (rather than site-averaged δ 18 O) the increase in Δδ 18 O clay−carb in the west is dampened compared to the other domains due to an increase in carbonate δ 18 O (statistically significant in the full data set, but not site averaged δ 18 O). This increase in carbonate δ 18 O, if confirmed with more data, could be driven by a decrease in the prevalence of groundwater (lower-δ 18 O) carbonates that appear common near the OHT ( Figure S7 in Supporting Information S1), or a greater sensitivity of carbonates to winter moisture in the western domain due to a combination of a winter-dominant groundwater supply and low summer precipitation rates. The latter scenario is consistent with the apparent influence of winter-δ 18 O in late Quaternary carbonates that form in the most winter-wet soils ( Figure S9 in Supporting Information S1). Additionally, the onset of soil carbonate formation just prior to the OHT and the apparent shift from groundwater to vadose zone carbonates across the OHT are consistent with the onset and strengthening of the dry Cascades rainshadow.
Despite its distance from the Cascades, the eastern domain is still hydrologically connected to the windward and leeward sides of the Cascades today ( Figure S13 in Supporting Information S1) and climate model simulations show that lower topography in this region would increase the winter precipitation fraction in the Great Plains in the past (Feng et al., 2013). Still, the eastern domain clay δ 18 O increase appears delayed relative to the west (Figure 2a). While more data will help determine the precise timing of these transitions, it is possible that greater summer moisture in the eastern domain prior to Cascades uplift dampened the increase in δ 18 O from the seasonality effect, perhaps failing to reverse the opposing isotopic effect of uplift. In this case, the delayed increase in clay δ 18 O may be attributable to the subsequent extension of the Basin and Range (e.g., Dickinson, 2002;Loughney et al., 2021) which would likely further restrict the supply of winter moisture by increasing orographic blocking of westerly and Arctic air-masses as well as the distance moisture must travel from the Pacific coast.
However, this hypothesis of Cascades uplift as the OHT driver remains speculative. The uplift history of the Cascades is debated (Bershaw et al., 2019;Kohn & Fremd, 2007;McLean & Bershaw, 2021;Pesek et al., 2020;Reiners et al., 2002;Takeuchi et al., 2010), and while tectonic forcing can explain many aspects of our results, more isotope data (especially clay data) with robust age constraints are needed to rigorously test any links. Improved constraints on the spatial extent of the OHT would also help address the possibility that it was driven by Cascades uplift. Other open-habitat shifts occur globally in the Cenozoic, but generally asynchronously with one another and the timing presented herein (e.g., Strömberg et al. (2013); Karp et al. (2018); Andrae et al. (2018); Barbolini et al. (2020)). In the U.S., phytolith records outside of WCNA are sparse and limited to the last ∼17 million years. These data indicate grassy, open habitats in southern California by ∼17 Ma (Loughney et al., 2020;Smiley et al., 2018) and Kansas by at least 8 Ma , but it is unclear if the onset of grassy conditions correlates with the WCNA signal. In general, phytolith preservation is poor south of WCNA. Nevertheless, whether tectonics is the underlying cause, the shift to drier winters provides new context to understand the drivers of grassland and open habitat expansion across the OHT.

Implications for OHT Vegetation Dynamics and Biogeography
Our results support the hypothesis that the OHT was a consequence of limited water availability (Harris et al., 2017;Webb & Opdyke, 1995;Wing, 1998;Wolfe, 1985) and, specifically, that drier winters decreased the annual supply of water by decreasing precipitation and spring snowmelt. However, changes in global temperature have also been proposed to trigger the OHT and while our results are unlikely to be driven by temperature (e.g., Figure S4 in Supporting Information S1), we cannot rule out the possibility that temperature contributed to the OHT. A key line of evidence for temperature change is the expansion of pooid grasses, which tend to be adapted to cool or cold climates (Edwards & Smith, 2010;Schubert et al., 2019;Strömberg, 2005Strömberg, , 2011. However, many pooids also specialize in extreme aridity, and the lack of additional evidence for cooling suggests that drying can account for the pooid expansion (Edwards & Smith, 2010;Harris et al., 2017). Further, the survival of frost-intolerant palms across the OHT indicates that if cooling occurred, it must have been minor (Reichgelt et al., 2018;Strömberg, 2005Strömberg, , 2011. For now, any link between global temperature and the OHT remains ambiguous (Harris et al., 2017;Strömberg, 2011), but we suggest that neither warming nor cooling are required to explain grassland expansion across the OHT.
What do our results mean for the sensitivity of WCNA tree cover to climate? Much like WCNA tree cover today, forests spanning WCNA were likely reliant on winter precipitation before the OHT (Berkelhammer et al., 2020;Hu et al., 2010;Knowles et al., 2018). With the westerlies more easily traversing the Cascades (and Basin and Range) before the OHT, providing more winter moisture further inland, winter precipitation likely supported closed, wooded habitats as far east as the Great Plains. Still, it is not clear whether the expansion of open habitats occurred synchronously across WCNA, making it difficult to determine the precise conditions supporting greater forest coverage beyond wetter winters.
Even if the shift to drier winters occurred asynchronously, our finding that winter precipitation decreased in all domains may help explain why grass communities, unlike other aspects of WCNA floras, became more uniform from west to east after the OHT. Prior to the OHT bambusoid grasses were common in the understory of eastern domain forests but rare elsewhere, while after the OHT similar pooid-dominated, open-habitat communities expanded in at least the eastern and central domains (Miller et al., 2012;Strömberg, 2005). In contrast, woody taxa maintained their biogeographic affinities through and long after the OHT (Leopold & Denton, 1987;Strömberg, 2005Strömberg, , 2011. We suggest that greater water stress with winter drying across WCNA favored the expansion of open-habitat, primarily pooid-dominant grass communities during the OHT. Meanwhile, other aspects of plant communities, like woody taxa, likely persisted in places where conditions remained favorable. A key implication of drier winters prompting grassland expansion is that grasses did not expand everywhere. Today, forests still prevail on mountain slopes that intercept westerly (winter) moisture and near perennial rivers that recharge groundwater (see Figure 1e; Schimel et al., 2002). A shift to drier winters probably decreased precipitation in the intermontane valleys that cover much of the WCNA or on east-facing slopes that do not intercept westerly moisture. Orographic precipitation and perennial rivers are reliable moisture sources in places that do intercept the westerlies, thus dampening the decrease in winter precipitation and supporting tree cover in these regions. In addition to orographic precipitation, colder temperatures at higher elevation help preserve snowpack and limit evaporation in the warm, growing season, further maintaining the water supply (Clow, 2010;Hu et al., 2010). We hypothesize that the OHT was predominantly a low-relief (and east-facing slope) phenomenon and vegetation fed by orographic precipitation or groundwater from perennial rivers was not as severely impacted.
The aridification associated with the OHT reorganized WCNA floral and faunal communities and may have increased mammal diversity. Phytolith indicators of forests, closed habitat grasses, and moisture-dependent gingers and palms are all present after the OHT, just in much lower abundances (Strömberg, 2005). The survival of these floras while open, grassy habitats expanded likely dampened the loss of mammals adapted to forests while promoting the niche-filling diversification of mammals adapted to new, open habitats (Samuels & Hopkins, 2017). The number of mesodont and hypsodont taxa (adapted to eating silica-rich vegetation in dusty environments like grasslands) increased in both small and large mammalian herbivores near the start and end of the OHT, respectively (Janis et al., 2000;Jardine et al., 2012;Samuels & Hopkins, 2017). This increase in taxonomic richness, however, appears short-lived-each niche-filling pulse is followed by a subsequent decline in taxonomic richness driven by the loss of taxa adapted to feeding on trees and shrubs, less likely to be covered in dust (Janis et al., 2000;Jardine et al., 2012;Samuels & Hopkins, 2017).
Overall, we suggest that winter drying triggered the expansion of open, grassy habitats by decreasing winter precipitation and spring and summer snowmelt, and increasing forest water stress. Forests relied on winter precipitation before the OHT, as they do today, and drier winters would have inhibited forest survival, allowing open habitat grasslands to expand. Other factors like fire frequency and intensity may have also promoted grassland expansion by reducing summer soil moisture, but charcoal data and organic biomarker data of fire are sparse and more data are needed to test this hypothesis. Drier winters across the OHT mark a shift to more mosaic landscapes with the expansion of open, grassy ecosystems representing an important step from the closed-forest vegetation of the Paleogene to the grasslands, scrublands, and deserts that span WCNA today.

Concluding Remarks
Our results refute the hypothesis that drier summers caused grassland expansion across the OHT (Retallack, 2004b) and instead implicate drier winters due to a decrease in the contribution of westerly precipitation. The cause of winter drying, however, remains unknown. Topography sets the step-wise spatial gradient of WCNA precipitation seasonality today and we tentatively suggest that tectonic change, namely the uplift of the Cascades, caused drier winters across the OHT. Still, more data and rigorous model analysis are needed to test if this is compatible with the isotope record.
The expansion of open-habitat communities with drier winters across the OHT demonstrates that the modern, positive relationship between winter precipitation and western U.S. biomass, productivity, and tree cover (Berkelhammer et al., 2020;Hu et al., 2010;Knowles et al., 2017Knowles et al., , 2018 has existed since the Eocene. Our findings suggests that the link between winter precipitation and vegetation is robust over time and across timescales. Drier winters can decrease WCNA biomass on short (<10 2 yr) and long (>10 6 yr) timescales despite differences in the time available for plant communities to adapt. As winter moisture declines with ongoing warming, our study emphasizes the fundamental challenges that WCNA forests face when winter water is limited.

Data Availability Statement
Isotope and phytolith data in this manuscript can be found in the Supporting Information S1 and are also available on Github (https://github.com/tykukla/Kukla_etal_2022-Drier_Winters).