Survival of graphitized petrogenic organic carbon through multiple erosional cycles supplementary material

Abstract Graphite forms the endpoint for organic carbon metamorphism; it is extremely resilient to physical, biological and chemical degradation. Carbonaceous materials (CM) contained within sediments, collected across Taiwan and from the Gaoping submarine canyon, were analyzed using Raman spectroscopy to determine the crystallinity. This allowed the erosional and orogenic movements of petrogenic organic carbon (OCpetro) during the Taiwanese orogeny to be deduced. After automatically fitting and classifying spectra, the distribution of four groups of CM within the sediments provides evidence that many forms of OCpetro have survived at least one previous cycle of erosion, transport and burial before forming rocks in the Western Foothills of the island. There is extensive detrital graphite present in rocks that have not experienced high-grade metamorphism, and graphite flakes are also found in recently deposited marine sediments off Taiwan. The tectonic and geological history of the island shows that these graphite flakes must have survived at least three episodes of recycling. Therefore, transformation to graphite during burial and orogeny is a mechanism for stabilizing organic carbon over geological time, removing biospheric carbon from the active carbon cycle and protecting it from oxidation during future erosion events.


Proportion of lithologies in each catchment
shows the relative proportion of each lithological group contained in the studied catchments. These proportions were calculated using ArcMap (ESRI). 2-D area percentage was calculated by extracting catchments from a digital elevation model, and generating shapefiles of the area upstream of the sample locations. These were then combined with a geological map shapefile (Chen et al., 2000) and the intersection of catchment and lithology generated new shapefiles. The metadata for these were exported to a table. Two-dimensional areas were extracted from the table and summed for the various lithology groups.

Sample collection
We chose to sample fluvial and marine sediment rather than directly sampling Taiwanese bedrock, for a number of reasons. Firstly, Beyssac et al. (2007) undertook a thorough investigation of the metamorphic rocks of the Hueshuan Range and Central Range, and we do not feel that repeating this work is worthwhile. They collected three transects of bedrock samples, and analysed the CM contained within these using Raman. Secondly, rivers integrate erosional signals from an entire catchment. Sediment samples are more representative of the relative volume of sediment erosion in a catchment compared to a bedrock survey. If a bedrock sampling campaign found a graphiterich formation, but its contribution to the sediment load in the catchment was negligible, this would bias the characterisation of that river's output. Thirdly, and conversely, if the bedrock survey missed a small but rapidly eroding graphite-rich layer, there may be unexpected graphite exported by the river.
Since Raman spectroscopy analyses individual particles of OC petro , the catchment wide averaging of sediment erosion does not lead to averaging of Raman measurements. Erosively mixed Highly Graphitised and Disordered OC petro will be identified as a bimodal distribution of these two spectral groups. Samples were collected during by different teams during multiple campaigns. Therefore there is a wide range of sampling techniques and locations used in the study, including sediment coring, manual bedload collection and automated suspended load sampling. We believe that this does not compromise our ability to observe and characterise all available forms of carbonaceous material eroded by the various rivers. Usually, the entire range of gransizes have been collected in one way or another. Laonung samples (LN) include silt through to coarse sand, due to collection from both a fluvial sand bar and fine-grained riverbank material. Gaoping Canyon (CY) cores contained sand and mud sized material, which were both investigated. Gaoping Shelf (SH) sediments only contained mudsized particles, but we observed the full range of carbonaceous material within these samples. Chenyoulan (CY) material was collected from the suspended load using an autosampler, but since SH samples contained the whole range of carbonaceous material grades, we do not expect systematic loss of any particular type here. Grass Lake Creek was sampled by walking up the semi-dried river bed during the dry season, and sampling cm-scale cobbles of sedimentary rock. These were crushed and ground, which should release and homogenise the samples, allowing all carbonaceous material within to be observed.

Raman spectra collection and fitting
Raman spectra were collected using Renishaw InVia and Ramascope-1000 Raman spectrometers. Dry sediment samples were ground for 12 minutes at 250 rpm in a Retsch PM-400 agate ball-mill grinder. Short-period grinding does not affect Raman charateristics of OC petro (Sparkes et al., 2013). One spatula (˜0.25 g) of material was pressed between glass slides to produce a flattened sample area with 2 cm diameter. The process of flattening between slides tends to align graphite flakes with the sample surface, meaning that the laser beam is incident perpendicular to the basal planes. Within this area, 10-20 flakes of POC were usually found using a 50 x magnification objective lens. The field of view as rastered across the sample to ensure that no POC flakes were missed.
Measurements were taken using a 514 nm Ar-ion laser, set to between 0.75 and 1.8 mW for 30 seconds to avoid damaging the target. Raman-shift was measured from 800 -2000 cm −1 with an 1800 l mm −1 grating. Spectra were fitted using the method described by Sparkes et al. (2013), in which peaks representing the G, D1, D2, D3 and D4 Raman bands were automatically fitted by a computer algorithm. This process allows rapid and objective analysis of a large number of spectra from complex samples. Spectrum metadata (location, height, width and area for each peak, plus peak height and area ratios) was used to classify each sample (Sparkes et al., 2013(Sparkes et al., , 2018. Samples were categorised into "Disordered", "Intermediate", "Mildly Graphitised" and "Highly Graphitised" using a combination of the metamorphic temperature predicted by the "R2" and "RA2" geothermometers of (Beyssac et al., 2002;Lahfid et al., 2010), and the increased width of spectroscopic peaks seen in disordered CM (Sparkes et al., 2013). These geothermometers have been carefully calibrated against metamorphic temperature, using field and laboratory studies (Beyssac et al., 2002(Beyssac et al., , 2003 but show no systematic relationship with pressure. Increased pressures lead to faster graphitization, but not the final degree of graphitization (Beyssac et al., 2003).
Whilst highly graphitised CM can be successfully differentiated using just R2 and the width of the D1 peak, these measurements do not permit separating intermediate and disordered CM. The RA2 calibration of Lahfid et al. (2010), coupled with the combined width of the D1, D2 and G peaks, allows these two groups to be distinguished. The metadata from each spectrum are included as Supplementary Dataset 1, and displayed graphically by plotting the sum of peak widths against the calculated temperature from either the "R2" or "RA2" geothermometers ( Figure  2, main paper).
A script implementing the automatic fitting procedure is maintained at https://github.com/robertsparkes/ raman-fitting