The studied outcrop section, Lodo Gulch, is part of the late Paleocene–early Eocene Lodo Formation located in the Panoche Hills of central California (Fig. 1). During the late Paleocene, the section was situated at a paleolatitude of ∼ 42° N, roughly at the boundary between the dry subtropical highs and mid-latitude low-pressure systems. The Lodo Formation is comprised primarily of siltstone with a relatively low abundance of calcareous microfossils truncated by thin glauconitic sand layers (Brabb, 1983). Depositional facies are consistent with a neritic–bathyal setting along the outer shelf (John et al., 2008).

Particle size was measured by laser diffraction using a Beckman Coulter device with polarization intensity differential scatter (PIDS) housed at UCSC (see the Supplement). For each sample, 2 to 5 mg of bulk sediment was powdered and put through a 2 mm sieve following the protocols in Blott et al. (2004). A total of 39 samples was measured, each in duplicate or triplicate to ensure reproducibility.

Sample preparation followed a slightly modified version of Kemp et al. (2016). Roughly 5 to 10 g of sediment was powdered with a mortar and pestle and then placed in a Calgon (sodium hexametaphosphate) solution on a shaker table for 72 h. Samples were sorted through a 63 µm sieve while collecting the fluid with the < 63 µm fraction. The collected fluid and suspended fine fraction (< 63 µm) were allowed to settle for a period determined by Stokes' law to keep < 2 µm size clay particles remaining in suspension. The fluid was then decanted and dried in the oven at 40 °C. Approximately 150 mg clay from each sample were used to prepared oriented mounts for X-ray diffraction (XRD) analysis. A total of 38 clay samples were prepared from the Lodo Formation. The sample residues were measured on a Philips 3040/60 X'Pert Pro X-ray diffraction instrument at UCSC. Clay species (i.e., smectite, illite, kaolinite, and chlorite) were identified based on the peak positions and intensities representing each clay mineral.

Sediment extraction, compound isolation, and compound-specific isotope measurements were conducted following Tipple et al. (2011). Briefly, the sediments were freeze-dried, powdered (∼ 500 g), and extracted with a dichloromethane (DCM) and methanol ratio (2:1; v/v) using a Soxhlet extractor. Total lipid extracts were concentrated and then separated by column chromatography using silica gel. N-alkanes were further purified from cyclic and branched alkanes using urea adduction following Wakeham and Pease (2004). N-alkane abundances were determined using a gas chromatograph (GC) with a flame ionization detector (FID). Isotope analyses were then performed using a GC coupled to an iRMS interfaced with a GC-C III combustion system or a high temperature conversion system for δ13C and δ2H analyses, respectively. A total of 59 samples was processed with a fused silica, DB-5 phase column (30 m × 0.25 mm internal diameter, 0.25 µm film thickness) with helium as the carrier at a flow of 1.5 mL min-1. The GC oven temperature program was 60–320 °C at 5 °C min-1 and isothermal for 30 min. A Thermo Trace mass spectrometer (MS) was used for detection with the mass spectrometry scanning from 50–800 m/z or exclusively m/z of 191, 217, 218, 370, 372, 386, and 400 for single-ion monitoring. Biomarkers were identified by the elution time and mass spectra of in-house petroleum standards with published biomarker distributions (Peters et al., 2005).

δ13C and δ2H values are expressed relative to Vienna Pee Dee Belemnite (VPDB) and Vienna Standard Mean Ocean Water (VSMOW). Individual n-alkane isotope ratios were corrected to n-alkane reference materials (for δ13C, C20, C25, C27, C30, and C38 of known isotopic ratio and for δ2H; “Mix A” from Arndt Schimmelmann, Indiana University) analyzed daily at several concentrations. In addition, H2 reference gas of known isotopic composition was pulsed between sample n-alkane peaks to confirm if the normalizations were appropriate. Standard deviations (SDs) of n-alkane reference materials were ±0.6 ‰ for δ13C and ±6 ‰ for δ2H.

Climate simulations from two models were used in this paper for (1) comparison with leaf wax proxy data and (2) extreme events analyses. (1) The water-isotope-enabled Community Earth System Model version 1.2 (iCESM1.2) simulates changes in climate and water isotopic composition during the PETM (Zhu et al., 2020) with a horizontal resolution of 1.9 × 2.5° in atmosphere and land and a nominal 1° in the ocean and sea ice components. Water isotope capabilities have been incorporated into all the components of the Community Earth System Model (CESM) 1.2 (Brady et al., 2019) which include the Community Atmosphere Model, version 5 (CAM5), for the atmosphere; the Parallel Ocean Program, version 2 (POP2), for the ocean; the Community Land Model, version 4 (CLM4), for the land; the River Transport Model (RTM) for river flow; and Community Ice Code, version 4, for sea ice. All simulations were run with the identical boundary conditions (including early Eocene paleogeography, land–sea mask, vegetation distribution, and pre-industrial (PI) non-CO2 greenhouse gas concentrations, soil properties, natural aerosol emissions, solar constant, and orbital parameters), following the DeepMIP protocol (Lunt et al., 2017), and differ only in atmospheric CO2 concentration. Crucially, the models with reduced latitudinal temperature gradients (e.g., Geophysical Fluid Dynamics Laboratory (GFDL) and CESM) more closely reproduce proxy-derived precipitation estimates and other key climate metrics (Cramwinckel et al., 2023). Increased climate sensitivity with warming and cloud feedback in CESM1.2 over earlier models improved water vapor sensitivity. (2) Using the same CESM1.2 framework, high-resolution (0.25°) simulations were conducted with forced sea surface temperatures (SSTs) and active atmosphere and land components (CAM5 and CLM4). RTM was run at 1° resolution, and forced SSTs were calculated from consistent 2° fully coupled PETM simulations (see details in Rush et al., 2021, and references therein). The much higher horizontal resolution in the atmosphere enables improved the simulation of the extreme events. Hourly, daily (CAM5), and monthly (iCESM1.2) temporal resolution precipitation outputs from both sets of climate simulations were utilized in this paper, with 100 years taken from the equilibrated iCESM1.2 simulations and 15 years from the forced SST high-resolution CAM5 simulations.

A carbon isotope excursion is present in both bulk organic (Fig. 2a) and carbonate-based δ13C records (John et al., 2008) across the P-E boundary, marking the PETM onset of the Lodo section. The terrestrial leaf wax n-alkane records all capture the carbon isotope excursion (CIE) with a pattern that roughly parallels the other published records (i.e., planktonic foraminifera) (John et al., 2008), though is much less noisy than the bulk δ13Corg record, which is not unexpected given the potentially variable composition of the bulk organic matter. The magnitude of the Δδ13Cn-alkane is roughly -4 ‰ (average of n-C27, n-C29, and n-C31) at the onset of the CIE, followed by a gradual recovery that is truncated at the disconformity between 20.3 and 23.5 m (coincides with nannofossil biozone boundary NP10 and NP11), thus marking the top of the PETM body (Fig. 2a). The disconformity coincides with a global sea level regression (John et al., 2008). Following the recovery, above the disconformity, the mean δ13Cn-alkane is depleted relative to the pre-PETM baseline, as observed in some other PETM sections (Cui et al., 2021; Garel et al., 2013; Hasegawa et al., 2006).

The leaf wax δ2Hn-alkane values range from -150 to -213 ‰ over the entire sampled section that has an initial decrease of 25 ‰ (from -150 to -175 ‰ in C29) just prior to the CIE onset and then followed by a slight rise (∼ 6 ‰) right after the onset. The relatively invariable δ2Hn-alkane through the PETM is punctuated with two brief intervals of more negative values (-202 ‰ at 6.26 m and -213 ‰ at 22 m). The second larger anomaly coincides with the disconformity (related to local sea level regression). The post-PETM δ2Hn-alkane values are on average lower than for the upper Paleocene/PETM, although analytical errors may bias the values. Given the limited number of samples to establish a baseline for the upper Paleocene, the significance of the pre-CIE (and post-CIE) shifts/anomalies in δ2Hn-alkane should be considered with some caution. Several other sections do show pre-CIE shifts, both positive and negative, and typically an enrichment with the CIE (Handley et al., 2008, 2011; Jaramillo et al., 2010; Tipple et al., 2011). Such minor changes likely reflect unconstrained orbital influences on regional precipitation (Rush et al., 2021; Campbell et al., 2024), especially considering the variable direction of change from location to location.

Clay assemblages and particle grain size should to some extent be influenced by regional hydroclimate. At Lodo, the clay assemblages (Fig. 3) are dominated by smectite throughout. The minor clay components of illite and chlorite show several spikes relative to smectite within the lower (8 to 10 m) and upper CIE (∼ 19 m), whereas the ratio of kaolinite gradually increases (0.5 to 1.5) only over the upper portion of the CIE (10 to 20 m). A delayed rise in kaolinite has also been observed in a few other PETM sections, whereas some show an immediate rise (Tateo, 2020; Gibson et al., 2000). The smectite concentration and kaolinite/smectite ratio remain high in the post-PETM interval. The late Paleocene, with a relatively coarse sand size, shows slight spikes of kaolinite associated with other minerals. Grain size, largely for silt and clay, shows a distinct shift toward finer fractions (i.e., clay) with the onset of the CIE (Fig. 3e).

We obtained and processed temperature and precipitation outputs from two community Earth system/climate models: the isotope-enabled iCESM1.2 and high-resolution CAM5 models (daily precipitation over 15 years) forced by a range of greenhouse conditions (1×, 3×, 6×, and 9× pCO2 pre-industrial), both with Eocene paleogeography. For comparisons with observations, we used the output from the 3× to 6× pCO2 simulations which best replicated the observed SSTs (ΔSST) for the pre-PETM and PETM (Zhu et al., 2020). Overall, monthly winter precipitation for the study region decreases (∼ 30 %) during the PETM in both simulations, with a slight increase in the summer (Figs. 4, 5). CAM5 output shows a modest decrease in the mean annual precipitation, with significant seasonal shifts during the PETM (Fig. 5a). Seasonal changes in the monthly averaged δ18O and δ2H from the mean monthly precipitation in iCESM1.2 of central California are consistent with CAM5. On average, the δ2Hprecip increases by ca. 5 ‰–10 ‰ from pre-PETM to PETM, especially in the winter/spring, with a smaller shift in summer/fall (1 ‰–2 ‰) (Fig. 4a, b, c). The extreme value index (ξ), a representation of the distribution of the exceedance right tail (see the Supplement), shows a small but statistically robust increase in the wet extremes of winter (December–February, DJF), with a significant increase in summer (June–August, JJA) during the PETM (Fig. 5b).

In all model simulations of the PETM forced with higher pCO2 (e.g., 3× to 6× pre-industrial), the hydrological cycle intensifies, as manifested by increases in global mean precipitation and meridional vapor transport (Kiehl and Shields, 2013; Kiehl et al., 2018; Carmichael et al., 2016; Zhu et al., 2020). Regionally however, the magnitude and even the sign of precipitation change can differ considerably from global means (Carmichael et al., 2016, 2017). This is most evident in the latest low- and high-resolution model simulations of the PETM (Zhu et al., 2020; Shields et al., 2021). For central California, the simulations yield an overall decline in the mean annual precipitation that is mainly due to a notable decline in winter precipitation with only a slight increase in summer (Fig. 4). This pattern is produced by both the water-isotope-enabled iCESM1.2 and the higher-resolution CAM5 with an overall shift to lower-amplitude seasonal cycles (i.e., drier winter/spring and a slightly wetter summer) (Figs. 4, 5). This seasonal wet–dry shift appears to be driven in part by a pronounced northward shift in atmospheric rivers (ARs) in winter along the North American Pacific coastline (Shields et al., 2021). As ARs deliver most of the winter precipitation to the mid-latitude Pacific coast, less frequent occurrences result in drier winters during the PETM. Moreover, the extreme value index (ξ) shows a small but statistically robust increase in winter (DJF) wet extremes, with a significant increase in the probability of summer (JJA) wet exceedance during PETM (Fig. 5b). The latter might be due to elevated tropical storm activity along the Pacific coast during PETM (Fig. S7; Kiehl et al., 2021).

Assuming leaf wax δ2H is primarily influenced by local meteoric water, how does the seasonal distribution of precipitation influence the bulk signal? In iCESM1.2 simulations with increasing pCO2 (i.e., 3× to 6× pre-industrial) and SSTs, the seasonal shifts in δ2H of mean monthly precipitation from pre-PETM to PETM are significant. During the winter, as precipitation amounts decline, δ2Hprecip increases by 10 ‰ while decreasing by ∼ 1 ‰ to 5 ‰ during late summer/fall in central California (Fig. 4). To estimate how this seasonal change in δ2Hprecip and precipitation amount influences leaf water δ2H, we applied a leaf wax proxy model (see the Supplement) which computes the combined effects of changes in seasonal precipitation and growing season length. The model shows leaf water δ2H is enriched ca. 4 ‰ to 7 ‰ from pre-PETM to PETM. Arguably, this would be consistent with the minor ∼ 6 ‰ enrichment observed in the Lodo record at the onset of the PETM. We also examined other sites for comparison of the predicted leaf water differences from pre-PETM to PETM with the fossil leaf wax Δδ2H (see the Supplement) and find a similar pattern in other mid-latitude sites. Other factors to consider include precipitation source waters and a shift in a mixing endmember between proximal and distal sources of water in the coast. For example, with a summer shift in the source water from the Pacific to subtropics (i.e., summer monsoons), the effect of increasing distance and distillation would isotopically deplete vapor (Hu and Dominguez, 2015), thus offsetting the temperature-related enrichment of local δ2Hprecip. In addition, infrequent but high-intensity tropical cyclones during the PETM (Kiehl et al., 2021) would tend to deliver relatively depleted precipitation (i.e., a more negative δ2H) during summer months.

Finally, a related record that might indirectly reflect on precipitation amount (i.e., atmospheric humidity) is the magnitude of the CIE recorded by leaf wax δ13Cn-alkane. Recalcitrant higher plant leaf wax n-alkane carbon isotope ratios (n>25 with an odd-over-even preference) mainly reflect the carbon source (Diefendorf et al., 2010). However, photosynthetic carbon isotope fractionation (Δp) is sensitive to atmospheric pCO2 variations, generally increasing with rising concentrations assuming a constant photosynthetic fractionation factor and humidity (Diefendorf et al., 2010). The δ13Cn-alkane of Lodo section displays a sharp negative shift of ca. 4 ‰ (average of n-C27, n-C29, and n-C31) across the onset of CIE (Fig. 2b), which is consistent with global mean atmospheric CIE (Sluijs and Dickens, 2012) but generally smaller than observed in other leaf wax records (Handley et al., 2008, 2012; Jaramillo et al., 2010; Pagani et al., 2006; Tipple et al., 2011). The smaller δ13Cn-alkane CIE recorded in Lodo would be consistent with a reduction in local humidity which should reduce the magnitude of Δp during photosynthetic carbon fixation.