Carbon isotope ratios of bulk organic matter in sedimentary rocks (δ13CDOM) are a potential source of paleoenvironmental information in terrestrial stratigraphic sequences. However, insufficient understanding of the range of depositional and post-depositional controls on δ13CDOM values makes interpretations of these data difficult. Here we evaluate the effects of organic matter (OM) provenance and preservation on δ13CDOM using records spanning the Paleocene Eocene Thermal Maximum (PETM) in the Bighorn Basin (Wyoming, USA) as a case study. We sampled sedimentary rocks spanning the PETM in two well-studied locations—Polecat Bench (PB) and Highway 16 (HW16)—in the Bighorn Basin. Independent carbon isotope records from biomarkers and pedogenic carbonates at these sites suggest that local shifts in plant and soil δ13C values associated with the PETM CIE were broadly similar and were characterized by an abrupt ∼5‰ decrease followed by a plateau and a eventual return to pre-PETM δ13C values. The δ13CDOM records from both sites differ significantly from these reference curves in both amplitude of change and in preserving high-frequency isotopic fluctuations and large isotopic anomalies superimposed on the general pattern of isotopic change through the CIE. For each location, we separated organo-mineral fractions (MOM), concentrated macerals from 20 stratigraphic levels and analyzed the carbon isotope ratio (δ13C) of each fraction. At both sites the δ13C of the fine and coarse MOM differ significantly from each other and from δ13CDOM. Concentration-weighted mixing of these isotopically distinct OM fractions explains high resolution δ13CDOM fluctuations but does not explain the large isotopic anomalies observed at both sites. At HW16, we identified two thermally and isotopically distinct populations of macerals interpreted as being indigenous and recycled OM. At this site, one over total organic carbon (1/TOC) values correlate with δ13CDOM for pre-PETM and PETM strata and both relationships converge toward the δ13C of recycled OM for low TOC. At PB, macerals display homogeneous thermal maturity, but the proportion of isotopically distinct vitrinite and liptinite varies between facies. Relationships between 1/TOC and δ13CDOM are also present within specific stratigraphic intervals at PB, but values do not converge on a single isotopic value across the sampled interval. These observations are consistent with variable mixing of OM fractions having different provenance—mixing of exotic recycled OM at HW16 and locally reworked OM at PB with indigenous “fresh” OM at both sites—and explain the large anomalies observed in the δ13CDOM records at both sites. Our findings raise questions about the assumption that OM in ancient sediments is indigenous and dominantly records δ13C variations of local plants.
 Carbon isotope ratio (δ13C) variations of plant-derived organic matter preserved in sediments have been used to constrain carbon cycle variations throughout geological time [e.g., Bird et al., 1994; Jahren et al., 2001; Beerling and Royer, 2002], to reconstruct taxonomical [e.g., Nordt et al., 1994; Arens and Jahren, 2000; Wang et al., 2008], and physiological response of plants to past environments [Farquhar et al., 1989; Beerling and Royer, 2002] and as a tool in chemostratigraphy [Grocke, 1998; Arens and Jahren, 2002]. Well-preserved deposits rich in organic matter (OM) such as coals or shales provide ideal sources of material for δ13C analysis in paleoenvironmental studies [Grocke, 1998, 2002], but the scarcity of these deposits in the geological record limits spatial and temporal resolution of the records. Consequently, interest is increasing in δ13C measurements of dispersed OM (δ13CDOM) preserved in ancient floodplain sediments, which are widely available and can potentially provide a high temporal resolution substrate for paleoenvironmental reconstruction. In particular, dispersed OM (DOM) from fossil soils (paleosols) is often assumed to derive primarily from plants growing in situ during pedogenesis and thus record information on the paleoenvironmental conditions experienced by those plants. Many records of δ13CDOM from floodplain sediments have been produced in the past decade for chemostratigraphy purposes [Jahren et al., 2001; Magioncalda et al., 2004; Wing et al., 2005; Yans et al., 2006; Carvajal-Ortiz et al., 2009; Domingo et al., 2009; Foreman et al., 2012; Baczynski et al., 2013], but difficulties persist in interpreting δ13CDOM records for more detailed paleoenvironmental work given that δ13CDOM records often exhibit: (1) large isotopic variability over short stratigraphic distances and (2) inconsistency with carbon isotopic trends recorded by other substrates at the same location.
 The use of DOM as a substrate for paleoenvironmental reconstruction revolves on the assumption that δ13CDOM reflects the δ13C value of ancient plants (δ13Cplant) living at or close to the location and time of sediment deposition. One obvious drawback of using DOM as a substrate for paleoenvironmental reconstruction is that only a very small fraction of the original soil OM is preserved in paleosols, and this fraction might not preserve the characteristics of the initial plant remains [Retallack, 1991]. During soil formation, oxidation and/or microbial respiration of soil OM and selective preservation/addition of OM fractions with different δ13C [Melillo et al., 1989; Ehleringer et al., 2000; Sollins et al., 2006] can lead to divergence between δ13Cplant and δ13C of soil OM. Deposition and mixing of isotopically distinct, recalcitrant, recycled OM from ancient sediments with fresh OM in contemporary soils can also complicate the interpretation of δ13C of soil OM [Blair et al., 2010; Clark et al., 2013]. For buried sediments, thermal maturation and fluid circulation can further alter OM during burial history of ancient floodplain sediments [Retallack, 1991]. Although the potential influence of many of these factors on δ13CDOM variations has been previously recognized, there are relatively few examples examining the degree to which they affect and compromise the interpretation of δ13CDOM for paleoenvironmental reconstruction.
2. Case Study
2.1. The Paleocene Eocene Thermal Maximum in the Bighorn Basin
 The Paleocene Eocene Thermal Maximum (PETM) is a massive global perturbation of the carbon cycle identified by a prominent negative carbon isotope excursion (CIE) in stratigraphic records. Variations of δ13C of atmospheric CO2 (δ13Catmo) during the PETM are well constrained by terrestrial and oceanic records and give a solid reference for temporal changes in δ13C that should propagate to sedimentary substrates preserving atmosphere-derived carbon across the event [McInerney and Wing, 2011]. Paleogene alluvial sediments deposited in the Bighorn Basin (BHB) yielded the first terrestrial chemostratigraphic record of the PETM [Koch et al., 1992]. Through >20 years of subsequent research, the PETM has been extensively studied and the CIE has been identified in several substrates and exposures throughout the BHB [Bowen et al., 2001; Bains et al., 2003; Magioncalda et al., 2004; Wing et al., 2005; Yans et al., 2006; Smith et al., 2007; Baczynski et al., 2013]. In the BHB, the PETM coincides with significant warming [e.g., Fricke et al., 1998] and large fluctuations in precipitation including a transient drying period during the PETM and wetter conditions before and after the event. Radical faunal and floral turnovers are associated with those climatic changes, with immigration of plant species formerly found only at lower latitudes [Wing et al., 2005], temporary rise of angiosperms and decrease of gymnosperms [Smith et al., 2007], and appearance of new mammal species [Gingerich, 2003]. The multiple records of the CIE of the PETM in the BHB and the associated paleoenvironmental research offer a unique opportunity to investigate the preservation of the well-documented global CIE in δ13CDOM.
2.2. δ13C Records of the PETM in the BHB
 Our work focuses on two intensively studied BHB localities, Polecat Bench (PB) and Highway 16 (HW16) (Figure 1). The HW16 (43.96°N, 107.65°W) and PB (44.77°N, 108.89°W) sites are located at opposite ends of the BHB and are separated by 133 km (Figure 1). Late Paleocene and early Eocene rocks at both sites are part of the Paleogene Fort Union and/or Willwood Formations (Fm) and were deposited by fluvial systems near the margin of the subsiding intermontane BHB. Both sections preserve sediments deposited in a suite of dynamic fluvial environments including small channels, crevasse splays, avulsion belts, abandoned channels, and stable floodplains [Kraus and Riggins, 2007; Adams et al., 2011]. Records of the CIE have been produced using long-chain, plant-derived n-alkanes at HW16 [Smith et al., 2007] and pedogenic carbonate nodules at PB [Bowen et al., 2001; Bains et al., 2003; Koch et al., 2003]. Those records broadly resemble each other and CIE records from other well-resolved terrestrial sections [McInerney and Wing, 2011]. They are characterized by three distinct periods: a rapid 4–5‰ initial shift (“onset”), a following ∼100 kyr period where δ13C values remain quasi-constant (“body”), and a return to initial δ13C values (“recovery”) [Bowen et al., 2006]. δ13CDOM records have also been generated for both sites, and display substantial differences relative to the biomarker and carbonate-derived records, including: (1) CIEs with lower magnitude; (2) stratigraphically lagged CIE onset and/or CIE recovery; and (3) large, high resolution δ13C fluctuations. The δ13CDOM record at HW16 displays lower CIE magnitude and lower absolute δ13CDOM values than at PB. We selected HW16 and PB (Figure 1) as case studies to investigate the potential influence of organic matter provenance and preservation on δ13CDOM variations because (1) independent δ13C records of the PETM from other substrates were available at these sites for comparison, (2) strong differences exist between the δ13CDOM records and the independent records, and (3) estimates of mean annual precipitation (MAP) and soil moisture at these sites provide further constraints relevant to the interpretation of δ13CDOM variations.
3. Material and Methods
3.1. Sampling and Field Methods
 Sections were measured with a Jacob's staff. At each site, we collected 45 samples including strata deposited before the PETM, strata from the onset and body of the PETM, and strata deposited during the recovery. We collected material from a trench dug to a depth at which fresh rock was exposed and no modern plant roots were visible. A variety of floodplain strata with different degrees of pedogenesis were collected (supporting information Tables A1 and A2).11 In each section, we also sampled two OM-rich beds (carbonaceous shales, DOM > 0.5%), one situated stratigraphically below and a second within the PETM (CS1 and CS2 in Figure 2a and CS3 and CS4 in Figure 3a).
3.2. Bulk DOM Stable Isotopes Analysis
 We applied several physical and chemical pretreatment steps to process the rock samples before analyzing δ13CDOM. For all steps, laboratory equipment was carefully cleaned to limit contamination from modern sources. Before grinding, rock pieces were carefully examined to remove any visible surface oxidation. Subsequently, we powdered the samples in a ball mill.
 We removed carbonates by treating samples with hydrochloric acid (HCl). Recent work has demonstrated that different acid treatments to remove carbonates could produce different δ13C results [Brodie et al., 2011]. We tested several acid treatment methods including different variants of “fumigation,” “capsule,” and “rinse” methods following published protocols (supporting information Table A3). For each treatment, we used and analyzed 10 replicates of one of our field-collected samples (HW-10-001; supporting information Table A3). For our samples, treatment by “fumigation” resulted in poor precision (1 standard deviation = 1.2‰) and the “capsule” and “rinse” methods gave comparable results with a standard deviation around 0.3‰ between replicates. Therefore, we selected a “rinse” method to be used for our samples. Some potential uncertainties associated with this method and discussed by Brodie et al.  include: (1) the absence of total dissolution of resistant carbonate phases (e.g., dolomite, siderite), (2) the loss of a soluble OM fraction during the rinsing steps, and (3) contamination by volatile organic compounds during the drying phase in plastic tubes (S. Dworkin, personal communication, 2013).
 The powdered samples (0.5 g) were treated using a 2N HCl solution at 50°C for 24 h in 50 mL plastic centrifuge tubes. We centrifuged the treated samples at 5000 rotation per minute (rpm) for 30 min, removed the acid, and rinsed the sample with distilled water until the pH of the supernatant became neutral. We dried the samples in an oven at 50°C, powdered them with a mortar and pestle and packed between 1 and 40 mg of powdered sample into tin capsules. δ13C, defined as δ13C = ((13C/12Csample)/(13C/12CPDB) − 1) × 1000 with PDB = Pee Dee Belemnite (standard), and TOC were measured with an Elemental Analyzer coupled to a Thermo Delta-Plus Advantage Isotope Ratio Mass Spectrometer (EA-IRMS) at the SIRFER lab (University of Utah). We used internal lab reference materials calibrated to the V-PDB scale, UU-CN-1, UU-CN-2, and NIST 2711 Montana Soil, for data correction. Analytical precision and standard deviations for reference materials and most samples was ±0.3‰ (1 σ). However, duplicate samples with carbon content lower than 0.1% had a precision of ±0.5‰ (supporting information Tables A1 and A2).
3.3. Physical Fractionation
 Physical fractionation is widely used in modern soil studies to refine δ13CSOM interpretation by isolating OM fractions that commonly have different macromolecular composition, age, and origin. The procedure separates soil particles based on density and allows isolating functional OM fractions because OM is increasingly bound to minerals as decay progresses [Sollins et al., 2009]. Fractions separated through this processes include particulate OM (POM) not bound to minerals, and presumably less processed, and one or more organo-mineral fractions (MOM). We selected 20 samples from each section that represented a variety of lithologies and carbon contents and used a physical fractionation protocol to separate the DOM in three OM fractions [Christensen, 2001]. Slight modifications were made to the published method to facilitate work with our samples, which were OM poor and required large amount of material to extract significant quantity of OM.
 Depending on the OM content, we used between 0.5 and 20 g of untreated and finely powdered sample and placed this powder in one or more 50 mL centrifuge tubes (maximum 3 g of sample per tube). We added 40 mL of sodium iodide (NaI) solution with a density of 1.7 g cm−3 in each tube. The samples were shaken and mildly sonicated to disperse the sample particles and break weak organo-mineral bounds. Subsequently, they were centrifuged at 5000 RPM for 30 min, and we isolated the floated POM by filtering the supernatant on a precombusted 2 µm glass fiber filter. We treated the POM on the filters with 2N HCl for 1 h. The filters and POM were thoroughly rinsed with 500 mL of distilled water (DI) and cut in two parts: one half was used for analyzing the δ13C of the POM (δ13CPOM) and the other half for organic petrography. We further separated the residual organo-mineral fractions (MOM) between fine (FMOM) and coarse (CMOM) by centrifuging at 750 RPM for 3 min in DI water [Christensen, 2001]. After drying the fine and coarse fractions in an oven at 50°C for 48 h, the dry weight of each mineral fraction was measured and aliquots were analyzed for carbon content and carbon isotopic ratios of the fine and coarse MOM (δ13CFMOM and δ13CCMOM, respectively).
 For the four OM-rich beds (section 3.1), we repeated the method above to obtain two replicates of the POM for each sample. The first replicate was used to run bulk isotopic (as described in the section above) and petrographic analysis (section 3.4). The second replicate was used to further isolate maceral types based on density [Rimmer et al., 2006; Mastalerz et al., 2012]. The method applied sequential density fractionation in NaI at 1.2, 1.35, and 1.5 g cm−3 to separate liptinite and vitrinite macerals. We analyzed the organic petrography and δ13C of each separated maceral fraction.
3.4. Organic Petrography
 The commonly used whole rock pellet preparation method for petrographic analysis [Taylor and Glick, 1998] was not applicable to our samples because lithified floodplain sediments are soft, hard to polish, and lean in OM. We developed a method in which the petrographic analysis could be performed directly on the filters containing the POM. The filters were cast horizontally with Lucite®, and the Lucite blocks were ground in a two step process (using 60 and 600 grit paper) to expose the filters and OM at the surface. We polished the block surface in a three step process (0.3, 0.05, and 0.01 micron). Organic petrography analyses were made on these polished blocks at 500× magnification in oil immersion. Vitrinite reflectance (Ro)—a method used to determine thermal history of OM—was determined using plane-polarized incident white light, whereas for maceral analysis both white and blue light was used. For sample containing sufficiently high quality vitrinite at least 25 Ro measurements were obtained and averaged (supporting information Tables A1 and A2). We also characterized maceral types by differentiating on the basis of morphology, Ro, fluorescence, size, and polishing relief [Taylor and Glick, 1998]. For most samples, only small amounts (∼100 µg) of POM could be extracted which limited the resolution of petrographic analysis. Total vitrinite (macerals derived from cell-wall material and woody tissue), total liptinite (macerals derived from lipid rich plant material such as spore or cutine), and inertinite (fossilized charcoals) contents were semiquantified by visually estimating maceral proportions within 20 fields of view on each sample. Uncertainty related to this method is calculated and described in Appendix 1 and supporting information Tables A1 and A2.
4. Estimation of δ13C Reference Curves for Ancient Plants
4.1. Method and Uncertainty Associated With Developing δ13Cmodel-plant Curve
 To identify and characterize δ13CDOM variations that are driven by factors other than changes in the carbon isotope ratios of plants living at or near the sites of sediment deposition we first attempt to estimate the average δ13C of ancient plant tissues (δ13Cmodel-plant) through the PETM interval at each site. This is not a trivial undertaking because records that are unambiguously related to the δ13C values of ancient biomass are generally not available. Our approach combines information from two types of records in order to characterize the likely stratigraphic pattern and magnitude of δ13Cplant change through the PETM CIE. First, we use data from stratigraphically resolved proxy records (n-alkanes at HW16, pedogenic carbonate at PB) that are closely related to ancient plant values but may have δ13C values that are variably offset from δ13Cplant due to climatically driven changes in ecosystem composition [Diefendorf et al., 2010] or soil properties [Bowen et al., 2004]. Because independent data suggests the largest shifts in these properties occurred at the beginning and end of the PETM [Wing et al., 2005; Kraus and Riggins, 2007], with relative stability before and after these changes, we suggest that these records likely provide a reasonably accurate record of the stratigraphic pattern of δ13Cplant change even if the magnitude of change during the CIE is biased by environmental shifts.
 To independently check and constrain the magnitude of δ13Cplant change during the PETM and estimate the absolute values of plant δ13C we use data from OM-rich carbonaceous shales. These samples have much higher OM content and several lines of evidence suggest that they provide a relatively accurate record of ancient plant δ13C values, including: (1) the observation that the organic-rich layers (CS1–4) display exceptional preservation of plant material and appear to be less affected by δ13C effects associated with decay and/or microbial processing of soil OM [Wing et al., 2005], (2) other δ13CDOM values [Baczynski et al., 2013] from OM-rich samples before (−25.8‰ ± 0.6, n = 11) and during the PETM (−29.8‰ ± 1.2, n = 13) are close to δ13Cplant values estimated from δ13C of n-alkanes, assuming a fixed, empirically derived isotopic offset between n-alkane and bulk biomass δ13C [Smith et al., 2007]. Adopting the assumption that δ13Cplant equals the δ13C of the OM-rich samples at the stratigraphic levels where we obtained these samples, we shift and linearly rescale (with respect to δ13C, where necessary) the stratigraphically resolved proxy records to derive site-specific δ13Cmodel-plant curves that pass through the OM-rich “anchor points.”
4.2. δ13Cmodel-plant at HW16
 At HW16, Smith et al.  estimated δ13Cplant from long-chain n-alkanes by adding 4.9‰ to the δ13C values measured for C31 n-alkanes extracted from sediments. This n-alkane based proxy record almost perfectly matches the δ13CDOM values from our two OM-rich samples at HW16 (Figure 2a), and as a result we make no further modifications to this curve and adopt it as our δ13Cmodel-plant curve for HW16. One limitation associated with using this δ13Cmodel-plant record to isolate depositional and post-depositional factors influencing δ13CDOM at HW16 is its relatively low sampling resolution in comparison with the δ13CDOM record. Although we consider it unlikely given the relatively smooth δ13C variation in the n-alkane record, we cannot rule out the possibility that plant δ13C values might have exhibited higher-frequency changes, not captured in the δ13Cmodel-plant record but affecting the δ13CDOM record, associated with taxonomic and/or physiological responses to PETM environmental fluctuations. Uncertainty and consistency of the δ13Cmodel-plant record at HW16 are further discussed in supporting information (Appendix 1).
4.3. δ13Cmodel-plant at PB
 At PB, δ13C values from n-alkanes have not been obtained yet, making estimation of δ13Cplant values more challenging. We derived a δ13Cmodel-plant curve from δ13C measurements of pedogenic carbonate nodules (δ13Ccarb) [Bowen et al., 2001]. Carbon in carbonate nodules originates primarily from soil-respired CO2, and thus from the decay of plant-derived OM [Cerling et al., 1989]. δ13Cplant variations should thus be recorded in δ13Ccarb, but because a variable portion of soil CO2 also originates from isotopically-distinct atmospheric CO2 the offset between δ13Cplant and δ13C of carbonates is sensitive to changes in factors that control this mixing ratio, such as soil permeability, atmospheric pCO2, and rates of soil respiration [Cerling et al., 1989; Breecker et al., 2009]. The magnitude of the CIE in the δ13Ccarb record at PB is amplified (∼5.5‰) in comparison with either the δ13Cmodel-plant at HW16 or the δ13CDOM values of OM-rich samples from PB, an observation that has previously been attributed to changes in soil properties during the PETM [Bowen et al., 2004]. In order to account for this effect we linearly rescaled the range of δ13C values from the carbonate record by a factor of 0.85 and shifted the curve, aligning it with the δ13CDOM values of our 2 OM-rich samples at PB (Figure 3a). The rescaling process implicitly assumes that changes in environmental factors controlling the offset between plant and carbonate δ13C values occurred in parallel with changes in plant δ13C values (largely driven by the global PETM CIE), which is consistent, to first order, with lithological and geochemical proxy data [Figure 3; Kraus and Riggins, 2007]. Ultimately, the δ13Cmodel-plant curves at HW16 and PB are similar in most respects and only differ in absolute δ13C values and sampling resolution (Figures 2a and 3a). Uncertainty and consistency analysis of the δ13Cmodel-plant estimates at PB are available in supporting information (Appendix 1).
5.1. Comparison Between δ13Cmodel-plant and δ13CDOM Records From OM-Poor Samples
 Our δ13CDOM records from OM-poor samples at PB and HW16 (Figures 2a and 3a) are relatively consistent with previous δ13CDOM records at the same locations [Magioncalda et al., 2004; Baczynski et al., 2013]. Differences between our records and existing δ13CDOM at the same site likely reflect differences in the strata sampled and differences in sampling resolution (Figures 2a, and 3a). Although δ13CDOM records at both sites display a negative shift of about 3.5‰ at the onset of the PETM, both records display several large isotopic fluctuations toward more 13C-enriched values which do not match δ13Cmodel-plant (Figures 2a and 3a). At HW16, δ13CDOM shifts from −24.3‰ to −27.6‰ at the CIE onset, recording a maximum CIE magnitude of 3.3‰. However, δ13CDOM values averaged through the body interval of the CIE, as defined by δ13Cmodel-plant, are only ∼1.6‰ lower than pre-CIE baseline values, and in several stratigraphic intervals within the CIE body δ13CDOM values are essentially indistinguishable from baseline values. The body and recovery of the PETM are not easily identified in the δ13CDOM record. The δ13CDOM values for pre-PETM and PETM samples at HW16 are on average 5‰ higher than the calculated δ13Cmodel-plant values (Figures 2a). At PB, δ13CDOM values transition from −24.9‰ to −27.9‰ at the CIE onset, recording a maximum CIE magnitude of 3.0‰. Average δ13CDOM values during the CIE body are ∼2.2‰ lower than pre-event values. The CIE is overprinted by rapid isotopic fluctuations, in particular a substantial increase in values between the 31 and 41 meter levels of our section, but is more clearly defined than at HW16. The absolute δ13CDOM values are also much closer to δ13Cmodel-plant (Figure 3a).
5.2. δ13C of POM and MOM
 At HW16, δ13CFMOM is on average 0.7‰ higher than δ13CDOM whereas δ13CCMOM and δ13CPOM are on average 1.1‰ and 1.2‰ lower than δ13CDOM, respectively. At HW16, patterns of variation in the δ13C of the POM and MOM fractions do not resemble δ13Cmodel-plant, similar to what was observed for the δ13CDOM record (Figure 2a). At PB, δ13CFMOM is on average 0.3‰ higher than δ13CDOM, whereas δ13CCMOM and δ13CPOM are on average 0.3‰ and 0.2‰ lower than δ13CDOM. At PB, the δ13CPOM record resembles δ13Cmodel-plant more closely and appears to record the CIE more faithfully than does δ13CDOM (Figure 3a).
5.3. Correlation Between δ13CDOM and TOC
 A relationship between one over total organic carbon (1/TOC) and δ13CDOM is observed at HW16 for PETM strata (R2 = 0.53) (Figure 4a). At HW16, both non-PETM and PETM strata converge toward similar values of ∼−24‰ for low-TOC samples. Similar relationship for PETM and non-PETM strata from high resolution δ13CDOM records had been previously reported at this site and other sites of the Southern Bighorn Basin [Wing et al., 2005; Baczynski et al., 2013]. Relationships between 1/TOC and δ13CDOM are also observed at PB for PETM strata (Figure 4b). Strata deposited during the body and recovery of the PETM show distinct relationships which do not converge toward a single unique isotopic value. Strata deposited before the PETM and during the onset do not display significant correlation between 1/TOC and δ13CDOM.
5.4. Organic Petrography of POM
 POM makes up only a small part (<1%) of the DOM in our samples, but understanding factors controlling δ13CPOM variation, particularly identifying changes in provenance and preservation, might help guide interpretations of δ13CDOM. We first analyzed petrographic characteristics of the two OM-rich samples from each site, and then used the results to inform our petrographic analysis of OM-poor samples.
5.4.1. OM-Rich Samples
 In POM samples at both sites we identified terrigenous macerals (vitrinite and inertinite) but did not find fluorescent macerals of lacustrine and/or marine origin. At HW16, both OM-rich samples show a variety of macerals including vitrinite, liptinite, and inertinite (Figures 5a, 5b, 5e, and 5f). Both samples (pre and during PETM) contain around 70% vitrinite and 30% liptinite. We observed minor amounts of inertinite (mostly semifusinite). The vitrinite fraction is dominated (95%) by a homogeneous, nonoxidized vitrinite (Figures 5a and 5b). We identified minor amounts (5%) of an oxidized vitrinite characterized by Ro ranging from 0.84% to 1.12%, weathering cracks, and oxidation rims (Figures 5a and 5b). The liptinite fraction is dominated by sporinite, with some minor amount of cutinite and resinite, and has a green-yellow fluorescence. Ro of the dominant vitrinite fraction for both samples is around 0.45% and this level of thermal maturity is consistent with green-yellow fluorescence of liptinite. At PB, OM-rich samples (pre and during PETM) show less variety in maceral types than at HW16. Both samples are dominated by vitrinite and only minor amounts of liptinite and inertinite are present (Figures 5e and 5f). The liptinite displays an orange-red fluorescence. The vitrinite fraction is homogeneous with no oxidation rims and with a Ro around 0.72% to 0.91%, consistent with an orange-red fluorescence of liptinite.
 For both sites, the sequential density fractionation (section 2.4) of these OM-rich samples allowed a partial separation of vitrinite and liptinite macerals. POM floated at lower density (1.2 and 1.35 g cm−3) contains a higher proportion of litpinite (fliptinite) and lower proportion of vitrinite (fvitrinite) in comparison with the initial POM (Table 1). POM samples extracted at low density and with high fliptinite have lower δ13C values than POM samples containing more vitrinite (Table 1).
Table 1. Isotope Composition and Proportion of Counted Macerals in OM-Rich Samples From HW16 and PB, respectively
CS1, CS2, CS3, and CS4 are OM-rich samples from HW16 and PB, respectively (section 2.2).
5.4.2. OM-Poor Samples
 At both sites, the POM fraction of OM-poor strata displays a variable maceral composition, but vitrinite and/or liptinite are the dominant macerals. Minor amounts of inertinite (mostly semifusinite) are present in some strata. For most OM-poor strata, the vitrinite fraction is more fragmented and harder to polish than in OM-rich samples. This likely reflects higher degree of in-soil OM oxidation and/or respiration within OM-poor in comparison with OM-rich strata. As in OM-rich strata, we identified two populations of vitrinite in OM-poor strata at HW16: a vitrinite population with Ro ranging from 0.48 to 0.64%, and an oxidized vitrinite fraction with Ro ranging from 0.76 to 1.12% (Figures 5c and 5d). The oxidized fraction is characterized by a more altered aspect with broken edges, weathering cracks and presence of oxidation rims (Figures 5c and 5d). In most of the OM-poor strata deposited during the PETM at HW16, the oxidized vitrinite is the dominant maceral population. The liptinite fraction in OM-poor strata at HW16 displays green-yellow fluorescence consistent with a Ro around 0.5%, supporting a common origin for the nonoxidized vitrinite and liptinite that is distinct from that of the oxidized vitrinite population. At PB, the proportion of vitrinite and liptinite in OM-poor strata is highly variable. Nevertheless, the characteristics and thermal maturity of the vitrinite fraction are consistent throughout the section (Figures 5e, 5f, 5g, and 5h). Only one population of vitrinite can be identified with a Ro ranging from 0.72 to 0.91%. The liptinite fluorescence is orange-red, consistent with a Ro around 0.8%.
6. Interpretation and Discussion
6.1. Effect of Soil Functional Fractions on δ13CDOM
 For each site, δ13C values of the different OM fractions are significantly different within and between strata (Figures 2a and 3a). For HW16, the observed average offsets between isotopic values of the different extracted fractions follow the relationship δ13CPOM < δ13CCMOM < δ13CFMOM (Figure 2a). At PB, this relationship is not as clearly defined but the different fractions remain significantly different (Figure 3a).
 We hypothesize that variations in the proportion of these isotopically distinct OM fractions could explain some of the δ13CDOM fluctuations observed at both sites. We test this hypothesis by developing a simple model predicting δ13CDOM as a function of the mixing of OM fractions:
where fPOC, fCMOC, and fFMOC are the fractions of the total organic carbon contained in the POM, coarse and fine MOM, respectively (Figures 2c and 3c). We assume fPOC ≈ 0 and that fFMOC = 1 − fCMOC because the quantity of POM extracted through floatation was negligible (<1% of TOC) in comparison with the carbon contained in the MOM. We calculate the percentage of carbon in each of the other fractions as:
where CCMOM and CFMOM are the weight percent of carbon present in the coarse and fine fractions obtained from the elemental analyzer, and fcoarse and ffine are the proportion of each grain size fraction in the strata obtained by weighing each dried fraction (Figures 2b and 3b). Details on the uncertainty calculation associated with δ13CDOM-model are given in supporting information Tables A1 and A2.
 At both sites values of δ13CDOM-model correlate better with δ13CDOM (Figures 6c and 6f) than do values of δ13CCMOM (Figures 6a and 6d) or δ13CFMOM (Figures 6b and 6e) individually. This result largely reflects the fact that the model accounting for mixing ratios of the two fractions better reproduces the high-frequency δ13CDOM variations than does the δ13C of individual MOM fractions (Figures 2e and 3e). The correlation between δ13Cmodel-plant and δ13C of individual MOM fractions is weaker at HW16 (Figure 2a) in comparison with PB (Figure 3a), suggesting that other factor(s) control δ13C variations in individual MOM fractions at HW16.
6.2. Effect of Isotopically Distinct Maceral Types and Provenance on δ13CPOM
 Results of sequential density fractionation on OM-rich samples showed that POM extracted at different densities is characterized by different maceral composition and δ13C values (Table 1). For both sites, the liptinite-dominated POM fractions have lower δ13C than vitrinite-dominated POM fractions. We calculated the average δ13C values of pure indigenous liptinite (δ13Cliptinite) and vitrinite (δ13Cvitrinite) fractions at HW16 and PB for pre-PETM and PETM samples, which we assume represent a pure, indigenous source of these macerals, by solving equations of maceral mixing using the proportion of liptinite and vitrinite recovered at each of the three densities (Table 1). Across all samples, δ13Cliptinite is on average 1.6‰ lower than δ13Cvitrinite. This value is similar to the 2‰ isotopic difference between liptinite and vitrinite fractions in shales from Rimmer et al. . The isotopic difference between vitrinite and liptinite is likely related to their chemical composition: vitrinite is essentially derived from 13C-enriched cellulose-rich woody material whereas liptinite is rich in 13C-depleted aliphatic compounds [Kruge et al., 1991]. For each OM-rich sample, we also calculated the isotopic difference between δ13Cliptinite and δ13CDOM and between δ13Cvitrinite and δ13CDOM. δ13Cliptinite is on average 1.25‰ ± 0.12 lower than δ13CDOM and δ13Cvitrinite is on average 0.35‰ ± 0.25 higher than δ13CDOM (Table 1).
 Our organic petrography results show that at HW16 two populations of vitrinite coexist in both the OM-rich and OM-poor strata. We interpret these two populations as consisting of indigenous OM derived from vegetation living on the paleo-soil surface and recycled material derived from erosion of rocks and sediments at the Basin's margins. The indigenous population is characterized by low thermal maturity, with a Ro around 0.5%, whereas the recycled population has a higher thermal maturity with Ro from 0.84% to 1.12%. The recycled vitrinite is also characterized by a more altered morphology, with presence of oxidation rims and weathering cracks, in comparison with the more homogeneous indigenous vitrinite that commonly preserves wood structure. The higher thermal maturity suggests that the recycled vitrinite had already matured prior to its deposition at HW16. The higher degree of alteration of this vitrinite population suggests that the recycled vitrinite was likely altered during erosion, transport, and prior to burial in the HW16 sediments.
 At PB, we only identified one population of vitrinite, which we characterize as likely to be indigenous based on the lack of oxidation and alteration features. All vitrinite at this site is characterized by higher thermal maturity than at HW16, however, which may reduce our ability to distinguish a recycled population based on Ro.
 We developed a mixing model accounting for effects of maceral type and maceral provenance to predict δ13CPOM. This model has three end-members with distinct δ13C values: (1) indigenous vitrinite, (2) indigenous liptinite, and (3) recycled vitrinite:
where fvitrinite, frecycled, and fliptinite are the proportion of indigenous vitrinite, recycled vitrinite, and liptinite (Figures 2d and 3d) and δ13Cvitrinite, δ13Crecycled, and δ13Cliptinite are the isotopic compositions of pure fractions of indigenous vitrinite, recycled vitrinite, and indigenous litpinite. Based on the results above we assumed constant offsets between δ13Cvitrinite, δ13Cliptinite, and δ13Cmodel-plant. We calculated δ13Cvitrinite and δ13Cliptinite for each stratum by adding 0.35‰ or subtracting 1.25‰ from δ13Cmodel-plant, respectively (Table 1). For HW16, we back-calculated δ13Crecycled as −24.1‰ (using equation (3)) based on δ13CPOM values for an OM-poor stratum, HW-10–033, where recycled vitrinite makes up 90% of the POM. This value is also supported by the relationship between 1/TOC and δ13CDOM in our HW16 data set, which converges on this value for low TOC values, and by Baczynski et al.  who report δ13CDOM values of potential parent rocks of the Willwood Fm in this region (−24.3‰, n = 10 formations). Details on uncertainty associated with δ13CPOM-model are given in supporting information Tables A1 and A2.
 While values of individual maceral fractions are only weakly correlated with δ13CPOM at both sites (Figures 7a and 7c), δ13CPOM-model accounting for the mixing of those maceral fractions is strongly correlated with δ13CPOM (Figures 7b and 7d). This result indicates that most of the variability in δ13CPOM can be predicted by considering both changes in paleo-vegetation δ13C values and variation in the mixing ratios of isotopically distinct vitrinite and liptinite fractions (Figures 2f and 3f). At PB, the δ13CPOM curve is similar to δ13Cmodel-plant in both amplitude and absolute values and appears to record the CIE more faithfully than does δ13CDOM (Figure 3a). δ13CPOM values diverge from δ13Cmodel-plant values during the CIE recovery. This anomaly is reproduced by the δ13CPOM-model and can be attributed to higher contents of 13C-depleted liptinite in these strata (Figure 3d). At HW16, δ13CPOM shows highly variable values, a low CIE magnitude, and outlier values, similar to what is observed for δ13CDOM (Figure 2a). In addition, δ13CPOM values are on average 4‰ higher than δ13Cmodel-plant values. The POM model incorporating mixing of different maceral types and recycled material, however, produces δ13CPOM-model values that are strongly correlated with and concordant with δ13CPOM (Figure 7b), suggesting that the mixing of isotopically distinct recycled and indigenous fractions explains the majority of the observed mismatch between δ13CPOM and δ13Cmodel-plant (Figure 2a).
6.3. Interpreting δ13Cmodel-plant and δ13CDOM Mismatch
 In the previous sections, we demonstrated that mixing of isotopically distinct MOM influences δ13CDOM and that mixing of isotopically distinct maceral types with different provenance controls δ13CPOM variations. Although we cannot directly measure the presence or abundance of recycled OM in the mineral fractions, the identification of recycled POM at HW16 suggests the possibility that a fraction of the more abundant mineral-associated organic carbon may also be derived from exotic sources. If we assume that the provenance of macerals in the POM fraction is representative of the provenance of macerals in the bulk DOM then we can evaluate whether variation in provenance contributes to the observed variability of δ13CDOM values at our study sites. We adopt this assumption for the following discussion but it should be considered tentative because the extracted POM represents only ∼1% of the total DOM. In this section, we integrate these ideas to develop a comprehensive model for observed δ13CDOM variations at the two sample sites and show that the model incorporating variation in provenance and preservation resolves apparent discrepancies between the organic fraction records and the reference plant curves (Figures 2a and 3a).
 Two striking differences between δ13CDOM and δ13Cmodel-plant records at HW16 are: (1) the large isotopic offset between δ13CDOM values and δ13Cmodel-plant estimates throughout section (Figure 2a) and (2) the 13C-enriched δ13CDOM values during the middle of the PETM body. Strata deposited during the PETM body at HW16 are characterized by higher content of recycled POM in comparison with other strata in the section (Figure 2d). These strata are also associated with inferred changes in soil moisture content and mean annual precipitation (MAP) at this site [Figure 2g and Adams et al., 2011; Kraus et al., 2013]. The soil moisture/MAP index applied by these authors suggests wetter conditions outside the PETM and during the onset and recovery of the PETM and drier soil conditions for the middle of the PETM body (Figure 2g). The transient period of improved drainage conditions correlates well with several changes in DOM proprieties: strata from the middle of the PETM body have on average lower carbon content (Figure 2c), higher proportion of recycled OM (Figure 2d) and lower content of liptinite (Figure 2d) than other strata in the section. The early and later portions of the PETM body more closely resemble the non-PETM intervals and display higher OM content (Figure 2c), lower proportion of recycled OM (Figure 2d) and higher liptinite content (Figure 2d).
 We suggest that these observations are consistent with more extensive in-soil oxidation and/or respiration of the OM in these strata, which would be expected under conditions of higher temperature, transient drying and improved drainage in the middle of the PETM body. Based on our organic petrography results and on the δ13CDOM/TOC relationships observed at HW16 (Figure 4a), we further hypothesize that in strata with low soil moisture where the degree of OM oxidation and/or respiration is high, the proportion of exogenous recycled OM (δ13CDOM ∼ −24‰) increases relative to the proportion of indigenous OM (δ13CDOM ∼−31‰ and ∼−26.5‰ for PETM and non-PETM strata, respectively) and leads to higher δ13CDOM. This mechanism would explain the relationships between 1/TOC and δ13CDOM for PETM strata and the convergence of δ13CDOM toward δ13CDOM ∼−24‰ for strata with low TOC for both PETM and non-PETM samples. The lack of correlation between 1/TOC and δ13CDOM for non-PETM strata in comparison with PETM strata (Figure 4a) is likely due to both the smaller isotopic difference between non-PETM indigenous OM and recycled OM and the lower abundance of recycled OM in those strata (Figure 2d). Assuming a constant flux of recycled material to the section, these observations might be expected if the recycled OM fraction is more recalcitrant than the indigenous OM fraction. The recycled vitrinite has higher Ro values than the indigenous vitrinite, which indicates that the recycled fraction was thermally mature prior to deposition. Thermally mature vitrinite is much more recalcitrant than fresh OM and is preferentially preserved in soils [Elmquist et al., 2006]. Thus, the strong divergence between CIE body δ13CDOM and δ13Cmodel-plant values and poor preservation of the CIE in the HW16 section can largely be attributed to “contamination” of the record by a large fraction of recycled OM varying with soil moisture content.
 Two striking differences between δ13CDOM and δ13Cmodel-plant records at PB (Figure 3a) are: (1) the lagged onset of the CIE in the δ13CDOM record and lag or apparent absence of the end-PETM carbon isotope recovery in δ13CDOM and (2) the relatively high δ13CDOM values within the middle of the PETM body in strata where the reference curve values remain near the CIE minimum values. While no exogenous recycled OM was identified in the POM fraction at PB, relationships between 1/TOC and δ13CDOM are apparent within some intervals of the PB record. In contrast to HW16, these relationships do not converge toward a single isotopic value for strata with low TOC (Figure 4b). Instead samples from within the recovery and the body of the PETM each display different trends, converging toward distinct isotopic values for strata with low TOC. We suggest that this observation is consistent with local reworking of OM, leading to a degree of time averaging in the δ13CDOM signal. Strata deposited during the recovery could contain a mixture of indigenous, post-PETM OM with relatively high δ13C values (∼−24‰) and a proportion of isotopically lighter (∼−27.5‰), locally reworked DOM derived from eroding upland soils formed within the PETM body. Similarly, samples from the body of the PETM could include a mixture of indigenous OM with low δ13CDOM values (∼−27.5‰) and reworked OM from older pre-PETM strata (∼−23‰). This interpretation requires that locally reworked OM is more recalcitrant than indigenous OM because the proportion of reworked OM appears to increase for strata with lower TOC. Reworked OM likely originates from erosion of upland soils during river channel migration and downcutting. Eroded sediments from those strata were likely pedogenically modified before erosion, leading to the loss of less stable OM and enhancing the chemical and/or physical resilience of the eroded OM.
 The interval of anomalously high δ13CDOM values during the body of the CIE at PB is associated with changes in soil moisture content and MAP at this site [Figure 3g and Kraus and Riggins, 2007; Smith et al., 2008]. The soil moisture index applied by these authors suggests similar results as at HW16, with wetter soil moisture conditions outside the PETM interval and drier conditions for the middle of the PETM body (Figure 3g). The transient period of drier conditions correlates well with changes in DOM proprieties: strata from the middle of the PETM body have on average lower TOC (Figure 3c) and lower liptinite content (Figure 3d) than other strata in the section. We suggest that in well drained and/or drier soils of the PETM body the degree of OM oxidation and/or respiration is higher than in more saturated soils forming before and after this time, leading to higher proportions of recalcitrant, locally reworked pre-PETM OM preserved in these strata. Within more poorly drained soils forming in other parts of the PETM OM was less extensively altered, leading to preservation of a larger pool of OM including a higher content of indigenous OM. As a result δ13C values from these beds more faithfully record δ13Cplant.
6.4. Recycled OM Provenance
 Our organic petrography analysis cannot specifically pinpoint the provenance of the recycled OM at HW16, but allows us to constrain possible sources. The recycled material observable in our POM fractions is of terrigenous origin (vitrinite, woody material, only), suggesting that this material may have been derived predominantly from the erosion of continental sedimentary rocks. Although algal OM was not identified at HW16, our results cannot entirely rule out marine rocks as a source of recycled OM because OM in marginal marine deposits often consists of a mixture of marine and terrestrial OM. It is possible, then, that algal material could have been selectively removed during thermal maturation of the parent rock or erosion and transport, leaving recalcitrant terrestrial OM as the dominant recycled material preserved in our samples. The presence of recycled dinoflagellate cysts and shark teeth in the Willwood Fm [Baczynski et al., 2013; Wing and Harrington, 2001] indicates that Willwood Fm. parent material was derived in part from marine rocks. The altered morphology of the recycled vitrinite (broken edges, oxidation rims) could reflect alteration during fluvial transport of OM eroded from basin margin uplands. Given this context, the presence of recycled vitrinite at HW16 and absence of recycled vitrinite at PB is consistent with the idea that the HW16 recycled vitrinite originated from OM-rich Cretaceous rocks that were not a significant source of parent material at PB.
 Recycled OM sources contributing to HW16 were likely located along the southeast margin of the BHB, with material being transported to the northwest following the dominant paleoflow direction during the Paleocene and Eocene in the southern BHB [Neasham and Vondra, 1972]. Many OM-rich Upper Cretaceous rocks deposited at the continental margin, including those of the Lance and Meeteetse Fm, currently outcrop along the southern and eastern margins of the BHB and were presumably being uplifted and eroded along the flanks of the Basin during the Paleocene [Roberts, 1998]. A simulation of the thermal maturity history of the BHB using Petromod1D ® shows that those Upper Cretaceous Fms had a Ro ranging from 0.6% to 1% throughout the basin during the Late Paleocene [Roberts et al., 2008], broadly consistent with values inferred for the recycled vitrinite fraction in the HW16 samples. We hypothesize that the recycled vitrinite found at HW16 originated primarily from the uplift and erosion of these Cretaceous rocks at the basin margins during the time of Fort Union and Willwood Fm deposition, and that the lack of recycled OM at PB largely reflects the absence or low contribution of those parent rock sources to that site.
7. Implications and Conclusions
 We demonstrated in this case study that OM provenance and preservation are primary and interactive controls on δ13CDOM variations for Paleogene floodplain strata of the BHB. Isotopic variations related to these factors can blur isotopic signals even as large as the CIE of the PETM. This finding calls into question the assumption that DOM in floodplain paleosols originates exclusively from indigenous plant material, complicating interpretations of δ13CDOM data that do not account for OM provenance and preservation. In the BHB, δ13CDOM is often the primary tool available to identify the stratigraphic position of the PETM or other Paleogene hyperthermal events and develop high-resolution age constraints for a given exposure. Errors in the correlation of these events due to provenance and preservation effects on isotopic records could propagate in the interpretation of biostratigraphy, geochemistry, or sedimentology data sets [e.g., Baczynski et al., 2013]. Other paleoenvironmental interpretations of δ13CDOM variations to calculate proportions of C3 and C4 plants [e.g., Wang et al., 2008], to reconstruct PCO2 in ancient geological times [e.g., Cerling and Quade, 1989; Ekart et al., 1999; Myers et al., 2012], or to investigate physiological response of plants to climate [e.g., Beerling and Royer, 2002] should be carefully evaluated with consideration of the factors identified here. Caution is particularly necessary when parent rocks of the studied sedimentary deposits are OM-rich, thermally mature (i.e., likely to contain highly resilient OM), and deposited in different times and/or depositional environment than the studied rocks, and/or when of the total content of OM in the studied materials is low. Extending this case study to other ancient and modern sedimentary basins could help constrain the degree to which the factors identified here are a recurrent and significant control on δ13CDOM variations and help to establish a framework for more robust interpretation of δ13CDOM in paleoenvironmental studies.
 This work was supported by a Geological Society of America Graduate Student Research grant awarded to C. P. Bataille, US National Science Foundation grants EAR-0628302 and OCE-0902882 to G. J. Bowen, and the donors of the American Chemical Society Petroleum Research Fund. We thank S. Wing and M. Kraus for assistance with sample collection and for providing background information at HW16. We thank Brad Errkila for assistance in analyzing δ13C at the SIRFER laboratory. We also thank Chris Brodie, Steeve Dworkin, and Brady Foreman for their reviews which significantly improved the science and clarity of this manuscript.
 Additional supporting information may be found in the online version of this article.