Coral Skeletal Luminescence Records Changes in Terrestrial Chromophoric Dissolved Organic Matter in Tropical Coastal Waters

Terrigenous dissolved organic matter (tDOM) carried by rivers represents an important carbon flux to the coastal ocean, which is thought to be increasing globally. Because tDOM is rich in light‐absorbent chromophoric dissolved organic matter (CDOM), it may also reduce the amount of sunlight available in coastal ecosystems. Despite its biogeochemical and ecological significance, there are few long‐term records of tDOM, hindering our understanding of its drivers and dynamics. Corals incorporate terrestrial humic acids, an important constituent of CDOM, resulting in luminescent bands that have been previously linked to rainfall and run‐off. We show that luminescence green‐to‐blue (G/B) ratios in a coral core growing in waters affected by peatland run‐off correlate strongly with remote sensing‐derived CDOM absorption. The 24‐year monthly resolution reconstructed record shows that rainfall controls land‐to‐ocean tDOM flux from this protected peatland catchment, and suggests an additional impact by solar radiation, which degrades tDOM at sea.

that absorbs light. CDOM plays an important role in aquatic ecosystems by absorbing sunlight and reducing its transmission through the water column (Kowalczuk et al., 2005;Mascarenhas et al., 2017). This is partly beneficial for aquatic life: photodegradation can release nutrients from refractory organic compounds (Bushaw et al., 1996), while absorption of ultraviolet radiation protects biota (Dunne & Brown, 1996;Pienitz & Vincent, 2000). However, light absorption by CDOM also reduces the availability of light for primary producers, which particularly affects benthic ecosystems such as seagrass meadows, kelp forests, and coral reefs (Gattuso et al., 2006;Jones, 1998).
Long-running measurements have shown that tDOC fluxes in North America and Europe have increased in recent decades. While this is largely attributed to reductions in atmospheric pollution, changes in climate and land-use have also contributed to these increases (Evans et al., 2005;Monteith et al., 2007;Noacco et al., 2017;Wauthy et al., 2018). Land-use change, in particular, is thought to have accelerated land-ocean organic carbon fluxes world-wide (Butman et al., 2015), and may have led to particularly dramatic increases in tDOC fluxes from disturbed peatlands in Southeast Asia (Moore et al., 2013;Yupi et al., 2016). Given the ecological and biogeochemical impacts of tDOC on coastal ecosystems, the implications of such increases in tDOC fluxes are potentially profound. Indeed, indirect evidence from parts of northern Europe suggests that increases in terrestrial CDOM have reduced light availability sufficiently to have noticeable ecological impacts (Aksnes et al., 2009;Frigstad et al., 2013). However, even in the best-studied parts of the world (i.e., Europe and North America) we still largely lack long time-series of tDOC dynamics in coastal waters. Across the tropics, there are no such observational time-series data, even though tropical rivers contribute around half of the global land-ocean tDOC flux (Dai et al., 2012).
Currently, this paucity of time series can only be overcome using optical remote sensing, which allows CDOM to be calculated from the water-leaving reflectance spectrum (Odermatt et al., 2012). This offers high spatial and temporal coverage, and in many coastal locations, CDOM or the CDOM spectral characteristics can then be used to infer DOC or even tDOC concentrations (Fichot & Benner, 2012;Liu et al., 2019;Shanmugam, 2011). However, to measure tDOC from satellite remote sensing in optically complex coastal waters still requires regional algorithms to be developed, which first necessitates a substantial effort to collect optical and biogeochemical data in the field. Consequently, such remote sensing analyses only exist for a limited number of locations (ChunHock et al., 2020;D'Sa et al., 2006;Mannino et al., 2008;Siswanto et al., 2011). Moreover, remote sensing can only extend records of coastal CDOM back over the past few decades, and cannot measure through cloud cover or during the dark, which limits the utility of this technique, especially in equatorial and polar regions. It is therefore highly attractive to look to possible natural archives for longer-term time series of CDOM and tDOC.
Humic-like substances are an integral component of tDOC, and an important constituent of the CDOM pool (Coble, 2007). The term encompasses a mixture of high-molecular-weight substances formed during the partial decomposition of terrestrial organic matter in soils (Piccolo et al., 2018). They are strongly fluorescent (Coble, 2007), and corals incorporate such humic-like substances into their skeletons during growth, where they can form sub-annual growth layers that luminesce under ultraviolet (UV) light (Matthews et al., 1996;Susic et al., 1991;Tanzil et al., 2016). Such molecules most likely diffuse passively to the coral calcification site via paracellular pathways through the coral tissue (Tambutté et al., 2012). Corals such as Porites spp. also offer exceptional chronological constraints, allowing monthly resolution proxy reconstructions over hundreds of years. Coral core luminescence has been used to reconstruct past rainfall and river discharge from the Burdekin River on the Great Barrier Reef (Isdale, 1984;Isdale et al., 1998;Lough, 2007;Lough et al., 2002;Rodriguez-Ramirez et al., 2014), in Madagascar (Grove et al., , 2013Maina et al., 2012) and correlates with salinity in Southeast Asia (Tanzil et al., 2016).
Using fluorescence excitation-emission matrix measurements, we recently demonstrated that coral luminescence is quantitatively related to the fluorescence intensity of terrestrial humic substances and that terrestrial humic substances, but not plankton-derived marine humic substances, are preferentially incorporated into aragonite (Kaushal et al., 2020). Because humic substances are such a key component of terrestrial CDOM and therefore of tDOC, this implies that coral luminescence is more directly a proxy for terrestrial CDOM and tDOC. Coral luminescence is more indirectly a proxy for rainfall or salinity to the extent that terrestrial CDOM and tDOC are typically closely related to salinity in shelf seas (Carr et al., 2019;Fichot et al., 2014;Martin et al., 2018). Here, we show that coral skeletal luminescence is indeed an accurate proxy for terrestrial CDOM, by calibrating coral core luminescence with a 12-year time series of satellite-derived CDOM from north-western Borneo. We then reconstruct a 24-year record of terrestrial CDOM with this core and investigate monthly-to annual-scale drivers of tDOM variability in this region.

Study Site
The coral core was collected from the western coast of Pulau Talang Besar (1.918°N, 109.774°E), which is part of a National Park referred to here as the Talang Islands. These small islands are situated in the southern South China Sea, 20 km off the western coast of Sarawak, Malaysian Borneo ( Figure 1a). Martin et al. (2018) and Zhou et al. (2019) reported characteristics of chromophoric and fluorescent DOM of rivers and coastal waters in this region. The Talang Islands are chiefly affected by the Samunsam River, which is a peatland-draining blackwater river that carries very high concentrations of DOC (1,200-1,800 µmol L −1 ) and CDOM (reported as absorption coefficient at 350 nm, a CDOM (350), of 71-98 m −1 ). The Sematan and Lundu rivers drain catchments with limited peatland cover and carry much less DOC (240-500 µmol L −1 ) and CDOM (a CDOM (350): <15 m −1 ). Despite seasonal changes in the DOC and CDOM concentrations, the relationship between DOC and CDOM across all the coastal waters and rivers sampled was very strong and showed no seasonality, demonstrating that CDOM is an effective proxy for tDOC in this region (Martin KAUSHAL ET AL.  Sarawak has an equatorial climate with rainfall throughout the year, but rainfall approximately doubles during the Northeast Monsoon (December-February) relative to other months ( Figure 1b). The cloud cover exceeds 70% year-round but is lower during April to August. Sea surface temperature (SST) decreases around 2°C from December to March, but the air temperature is close to 27 °C in all months, dropping by only 0.5°C during the Northeast Monsoon.

Coral Collection, Age Model, and Luminescence Measurements
The coral core was collected in May 2015 from the main growth axis of a Porites sp. colony at 2-3 m depth using a pneumatic drill with a 5-cm diameter, 50-cm length diamond bit core barrel. A slice of ∼0.7 cm was cut and cleaned for 48 h with a 1:4 mix of household bleach (NaOCl, 3%-7% reactive chlorine), and sonicated in deionized water for 30 min (water changed every 10 min). Bleaching removes contaminants and increases the luminescence intensity (Nagtegaal et al., 2012). The cleaned section was photographed under UV light (excitation wavelength 365 nm) using spectral line scanning (SLS) on an Avaatech XRF scanner (Grove et al., 2010(Grove et al., , 2015Tanzil et al., 2016). The light source and camera are moved down the core as one unit scanning multiple lines that are stitched together to produce a continuous image. The light entering the camera lens is split into three wavelength bands (red, green, blue) by a dichroic RGB beam splitter prism and is recorded by separate sensors. The luminescence green-to-blue ratio (G/B ratio) for the coral section was obtained along the main growth axis for a 2-mm wide track at a resolution of 143 pixels cm −1 . The G/B ratio was shown to be a more accurate tracer of terrestrial input than total luminescence, because the ratio normalizes for luminescence variation caused by other factors, such as variation in skeletal microstructure (Grove et al., 2010). The age model was constructed by considering each peak luminescence G/B layer as the Northeast Monsoon peak of each year with the upper-most layer as the Northeast Monsoon of 2014. We assumed that the coral had a linear growth rate between adjacent peaks, amounting to a growth rate of ∼1 cm/year. These assumptions are reliable in numerous Porites sp. cores collected across Southeast Asia (Tanzil et al., 2016(Tanzil et al., , 2019. For this study, a core section of ∼25 cm in length covering 24 years of growth from 1990 to 2014 was selected.

Satellite-Derived CDOM Measurements
We used the MODIS-Aqua Collection 6.1 Level 2 ocean color product, with 1-km resolution, from https:// oceancolor.gsfc.nasa.gov/for the period July 2002 to June 2014. MODIS was selected because of its high (daily) revisit frequency. The MODIS data were processed with a regionally parameterized inversion model, described in Cherukuru et al. (2021), based on the Generalized Inherent Optical Property model (Werdell et al., 2013). Briefly, this model uses a spectral optical library of in situ biogeochemical and optical measurements and estimates the concentrations of CDOM, DOC, and total suspended matter (TSM) based on selecting the most suitable set of specific inherent optical properties (SIOPs) from the optical library. This selection is made for each pixel in the satellite image by identifying which of the in situ measured remote sensing reflectance spectra in the optical library best matches the MODIS-measured remote sensing reflectance and then selecting the SIOPs of the stations where the matching reflectance spectrum was measured. The SIOPs are the DOC-specific CDOM absorption (i.e., absorption per unit DOC), the TSM-specific particulate absorption, and the TSM-specific backscattering. These SIOPs are then used in a forward model to simulate the best match between forward-modeled remote sensing reflectance and MODIS-measured remote sensing reflectance by iteratively varying the DOC and TSM concentrations and deriving the inherent optical properties according to the SIOP values. We used the model to retrieve a time series of CDOM, quantified as the CDOM absorption coefficient at 440 nm, a CDOM (440). This represents the absorption of light at 440 nm by CDOM over a unit of distance, and is reported in units of m −1 . To create a satellite-derived time series of CDOM absorption to compare to the coral core data, we calculated the average a CDOM (440) at daily resolution over a 3 × 3 km area immediately west of the Talang Islands, centered at 1.914°N 109.741°E, and further calculated the monthly averages of satellite-derived CDOM.

Peak Alignment
The time series of coral core G/B ratio and of monthly mean a CDOM (440) both showed well-defined seasonal peaks during the late Northeast Monsoon, with a CDOM (440) typically peaking in February. We peak-aligned the time series of coral G/B ratio with satellite-derived a CDOM (440) maxima and minima using Analyseries 2.0 (Paillard et al., 1996). This process rescales the luminescence G/B data and gives monthly averaged luminescence G/B.

Proxy Development
The Talang coral luminescence green-to-blue ratio (G/B ratio) ranged between 0.58 and 0.75 from January The satellite-derived CDOM absorption, a CDOM (440), ranged between 0.03 and 0.41 m −1 , with a single peak observed each year (Figure 2a). The peak was usually in February, that is, during the late Northeast Monsoon, with a few years showing a peak in the later months of March and April. Owing to cloud cover, the Northeast Monsoon months of December to February have the fewest measurements, with some months having only a single day of data. The months of April to August have the lowest cloud cover and therefore highest data counts of >3 days a month (Figure 2c). For 17 of the total 143 months, there are no cloud-free data.
The time series of satellite-derived a CDOM (440) and coral G/B ratio showed highly correlated seasonality (Figure 2a). We calibrated the G/B ratio as a proxy for a CDOM (440) (Figure 2b). We then applied this relationship to reconstruct CDOM from the G/B ratio through the full length of the coral core , providing a ∼24-year record of CDOM.
The reconstructed a CDOM (440) record ranged between 0 and 0.35 m −1 , with a single peak each year (Figure 2d). Two short sections in 1990 (March-August) and 1991 (May-September) had G/B ratios <0.6, yielding slightly negative reconstructed a CDOM (440) values that were set to 0. There was strong seasonality, with a CDOM (440) peaking between December to March, likely driven by seasonal changes in climatic drivers.
There is a small step-change in the record for the a CDOM (440) minima during the Southwest Monsoon, with very low values (<0.01 m −1 ) before 1995 giving way to higher minima thereafter (>0.03 m −1 ). Despite this increase in the Southwest Monsoon minima, there was only a small and statistically nonsignificant longterm trend in the annually averaged a CDOM (440) (linear regression, 0.0015 m −1 y −1 , p = 0.19). The Northeast Monsoon a CDOM (440) maxima do not show any systematic increasing trend, but there is a gradual shift in the timing of the a CDOM (440) peak, from January in the early 1990s to February after 2002. Overall, the time series shows large interannual variability in maximum and minimum a CDOM (440).

Seasonal and Interannual Drivers of CDOM Variability
The Northeast Monsoon peaks in a CDOM (440) are the most distinctive feature of the CDOM record (Figure 2d). Since this is the rainiest season, rainfall is the most obvious driver of the seasonal variation in CDOM. The month-to-month correlation (Spearman's rho) between reconstructed a CDOM (440) and monthly rainfall is 0.51 (Figure 3a), which increases to 0.53 if a CDOM (440) is lagged by one month (both p < 0.05). When the data are seasonally averaged, with the high-rainfall Northeast Monsoon months of December to February considered as one season and the remaining months as a second season, the correlation considerably increases to 0.68, since the seasonal averaging accounts for possible offsets between rainfall and a CDOM (440) accumulation at sea. Interestingly, we also found a month-to-month correlation between reconstructed a CDOM (440) and percentage cloud cover of lower, but still significant (p < 0.05) value of 0.40 (Figure 3b). When these data were seasonally averaged, with the low cloud cover months of April to August as one season and the remaining months as a second season, the correlation was 0.68, as high as the correlation of CDOM with seasonal rainfall. A surprisingly high correlation was also found between monthly a CDOM (440) and monthly SST, with r s = −0.59 (p < 0.05), with lower SSTs corresponding to higher a CDOM (440) (Figure 3c). This correlation also increased when we averaged the data seasonally (r s = −0.64) with the low SST months of December to March as one season and the remaining months as a second season. However, neither rainfall, cloud cover, nor SST were significantly correlated with a CDOM (440) when examined at annual resolution (all p > 0.05), indicating that none of these variables explained interannual variability in CDOM well. This is perhaps not surprising, since a CDOM (440) only varied between 0.07 to 0.15 m −1 interannually.

Discussion
Our results show that luminescence green-to-blue (G/B) ratios in coral cores are fundamentally driven by the concentration of terrigenous dissolved organic matter in the sea and that the variation in these ratios provides a palaeo-proxy for CDOM. The data gaps caused by cloud cover in our satellite record necessarily introduce uncertainties in our monthly mean CDOM estimates. Moreover, our satellite CDOM time series only represents snapshots at one time of day, which might be affected by short-term (e.g., tidal) variability that would be smoothed out in the coral record. These factors likely account for a large part of the scatter in our G/B versus CDOM relationship, and we anticipate that with better CDOM records, the calibration could prove to be considerably stronger. Nevertheless, the small RMSE of our calibration demonstrates that accurate CDOM reconstructions are possible.
This proxy is likely to reconstruct total CDOM accurately wherever the CDOM pool is dominated by terrestrially derived CDOM, such as in coastal waters of Southeast Asia that are influenced by peatland-draining rivers. Elsewhere, marine plankton is also a source of humic substances (Helms et al., 2013;Osburn et al., 2019). Such autochthonous marine CDOM usually differs spectrally from terrestrial CDOM (Astoreca et al., 2009;Painter et al., 2018;Stedmon & Markager, 2001). Although it has been speculated that such marine humic substances might also contribute to coral luminescent banding   compared to marine CDOM, yielding much higher fluorescence in the aragonite (Kaushal et al., 2020). This is consistent with previous studies showing that coral luminescent bands are a proxy for river discharge and salinity (Isdale, 1984;Isdale et al., 1998;Lough, 2007;Lough et al., 2002;Tanzil et al., 2016). However, if coral luminescence is in fact most directly a proxy for terrestrial CDOM, it becomes important to consider not only variation in CDOM input but also potential variation in CDOM removal, when interpreting luminescence records.
Our site is primarily exposed to discharge from the peatland-draining, DOC-and CDOM-rich Samunsam River (Figure 1). Seasonal changes in CDOM production in tropical peatlands are unlikely because of the small temperature range, and porewater concentrations of DOC, specific UV absorbance, and fluorescence of DOM appear to vary by ≤10% throughout the year (Gandois et al., 2013). The strong seasonal peak in CDOM during the Northeast Monsoon thus more likely points toward a rainfall-driven increase in land-tosea CDOM flux, yet the monthly level correlation between reconstructed CDOM and rainfall was relatively weak. This may partly be due to an expected lag between rainfall on land and accumulation of CDOM at sea, yet a lagged correlation was only slightly stronger. While fluvial fluxes of DOC (and hence of CDOM) do generally increase with freshwater discharge both in peat and nonpeat soils (Clark et al., 2007;Moore & Jackson, 1989) the complex hydrology of peatlands means that DOC concentrations in peatland-draining rivers typically decrease during periods of high rainfall due to acrotelm overflow (Clark et al., 2007(Clark et al., , 2008. This rainfall-driven dilution also occurs in Southeast Asian peatlands (Rixen et al., 2016), and is likely a reason for the higher DOC and CDOM concentrations measured in the Samunsam River in September compared to March (Martin et al., 2018). Consequently, the relationship between rainfall and the flux of DOC and CDOM to the sea is likely to be complex in this region, which may partly explain the only modest correlation of our coral record with monthly rainfall, and the complete lack of a correlation at annual resolution. This result reinforces that while rainfall is a driver of CDOM flux and hence of coral luminescence, the relationship is likely to be controlled by complex hydrological processes in peatland regions.
These complexities of peatland hydrology would likely dampen the seasonal variability in fluvial CDOM flux relative to the seasonal variability in rainfall. Consistent with this expectation, ChunHock et al. (2020) inferred that quarterly average fluxes of tDOC from rivers in Sarawak as estimated by satellite remote sensing varied by at most threefold from 2013 to 2018. In contrast, our reconstructed CDOM record shows high seasonal variability, with very low CDOM values during the Southwest Monsoon, which suggests that it might also be driven by seasonal changes in processes removing CDOM. Experiments suggest that Southeast Asian peatland DOM is photolabile over short time-scales (Martin et al., 2018;Rixen et al., 2008). Photodegradation breaks down organic matter and can cause a rapid loss of CDOM (Helms et al., 2014;Miller & Zepp, 1995;Moran & Hodson, 1990;Stutter & Cains, 2016). The percentage cloud cover, number of cloudy days in our satellite CDOM record, and SST data are indicative of seasonal changes in irradiance, and therefore of the potential for CDOM photodegradation. While the lower rainfall during the less cloudy months of April to August will already reduce the CDOM flux to sea, the higher solar irradiance most likely reduces the CDOM concentrations further still. In contrast, during the Northeast Monsoon months, the increase in rainfall-driven CDOM flux is probably amplified by a lower rate of photodegradation. While CDOM mostly mixes conservatively across estuaries in Sarawak, the absorption and fluorescence spectral characteristics measured in coastal waters of Sarawak are consistent with a contribution from partly photodegraded tDOC, and further photodegradation was observed when seawater from this region was exposed to sunlight (Martin et al., 2018;Zhou et al., 2019). Along similar lines, pCO 2 data indicate that a large proportion of peat-derived tDOC is remineralized in coastal waters off Sumatra, suggesting that this material is biogeochemically labile after mixing across estuaries (Wit et al., 2018).
In addition to rainfall and photodegradation, seasonal variation in physical dilution by currents might impact coastal CDOM concentrations. However, our site most likely experiences stronger mixing and dilution during the stormier Northeast Monsoon period than during other seasons, as shown by the higher average wind speeds during December to February ( Figure S1). Physical modeling further confirms that water residence times along the entire coastline of Sarawak (and across the entire Sunda Shelf) are lowest during the Northeast Monsoon (Mayer et al., 2015). Our data thus show that the CDOM concentration in coastal waters is highest at the time when the physical CDOM dilution rate is greatest but decreases to the lowest values as the dilution rate also decreases. Conversely, the potential for photodegradation of CDOM increases during the less cloudy months from April to August, coincident with the observed seasonal decrease in CDOM. While our data thus confirm that coral luminescence can be used to reconstruct hydrological variability, it may also be important to consider potential variability in processes removing terrestrial CDOM over the time-scale of the record.
On interannual time-scales, the coral CDOM record shows notable changes in seasonal variability, especially with the period 1996-2001. Our data show that years that have high Northeast Monsoon rainfall (i.e., ≥700 mm/month) have high CDOM maxima but subsequently low CDOM minima. On the other hand, years (particularly consecutive years such as between 1996 to 2001) that do not have such a large peak rainfall during the Northeast Monsoon have lower CDOM maxima but also higher CDOM minima. This would be consistent with the fact that rainfall acts both as a driver of tDOC flux but also acts to dilute or concentrate tDOC. During 1996-2001, the Northeast Monsoon rainfall is some of the lowest observed, which would necessarily limit the total quantity of tDOC flux. However, rainfall amount in the other seasons, that is, other than the Northeast Monsoon, is similar to that in other years. Because of the lower annual rainfall during this 1996-2001 period, the riverine tDOC concentration is likely increased because of overall lower rainfall-driven dilution, which would then result in a larger tDOC flux during the Southwest Monsoon, and a corresponding increase in the CDOM minima during these years. Analysis of more coral-based CDOM reconstructions from tropical peatland-influenced regions will help to shed light on the drivers of such interannual variability and on the hydrological drivers of peatland tDOC fluxes.
The 24-year long Talang record did not show a statistically significant long-term trend in CDOM. Most of the forest within the Samunsam catchment is protected and has remained relatively intact, including the peatlands, as shown by the Centre for International Forestry Atlas (accessible under https://atlas.cifor.org, Gaveau et al., 2016). This indicates that forest and peatland protection measures can be effective at stabilizing soil carbon pools. Our study demonstrates that coral luminescence G/B measurements can be used to reconstruct terrestrial CDOM, which provides a way to study land-ocean carbon fluxes and promises to reduce the severe lack of historical biogeochemical data in tropical regions.

Data Availability Statement
All data for this study can be found under https://doi.org/10.21979/N9/5CZV22