Evaluating ultraviolet (UV) based photochemistry in optically complex coastal waters using the Hyperspectral Imager for the Coastal Ocean (HICO)

Estuarine, Coastal and Shelf Science 215 (2018) 199–206

Contents lists available at ScienceDirect

Estuarine, Coastal and Shelf Science journal homepage: www.elsevier.com/locate/ecss

Evaluating ultraviolet (UV) based photochemistry in optically complex coastal waters using the Hyperspectral Imager for the Coastal Ocean (HICO)

T

Fang Caoa, Deepak R. Mishrab, John F. Schallesc, William L. Millera,∗ a

Department of Marine Sciences, University of Georgia, Athens, GA, 30602, USA Department of Geography, University of Georgia, Athens, GA, 30602, USA c Department of Biology, Creighton University, Omaha, NE, 68102, USA b

ARTICLE INFO

ABSTRACT

Keywords: Ultraviolet radiation (UV) Photochemistry Remote sensing ocean color HICO CDOM Absorption coefficient Diffuse attenuation coefficient (Kd)

Knowledge of light partitioning into different optically active constituents, particularly chromophoric dissolved organic matter (CDOM) in the ultraviolet (UV) is indispensable for understanding UV dependent biogeochemical issues including photochemical processes in optically complex waters. Herein a new approach is presented to investigate photochemistry by blending two ocean color algorithms, namely the composite SeaUV (Cao et al., 2014) and the SeaCDOM (Cao and Miller, 2015) algorithms, and applying them to visible remote sensing reflectance (Rrs) measured using the Hyperspectral Imager for the Coastal Ocean (HICO). As illustrated using photochemical carbon monoxide (CO) production from CDOM, this model approach allows high resolution examination of UV optical details with estimates of both depth-specific and depth-integrated photoproduction rates in a dynamic estuarine/coastal environment. Decoupled retrievals of inherent and apparent optical properties (i.e. diffuse attenuation coefficient (Kd) and CDOM absorption coefficient (ag)) using two distinct ocean color algorithms over the entire UV spectrum allow a synoptically dynamic view of CDOM's contribution to light attenuation (ag/Kd). This provides new potential to probe UV processes in complex coastal waters on regional as well as global scales using remote sensing of ocean color.

1. Introduction Estuarine and coastal waters function as critical interfaces between terrestrial and marine systems. Despite covering a small area compared to the global ocean, coastal waters play an active role in regulating carbon (C) fluxes and transformations (Bauer et al., 2013). Dissolved organic matter (DOM) delivered by rivers is estimated to be ∼0.25 Pg C per year and a large fraction of this riverine DOM is remineralized during transport to the open ocean via multiple physical and biogeochemical processes (Cauwet, 2002). Among these, ultraviolet (UV) photochemistry is a crucial sink for photo-labile chromophoric dissolved organic matter (CDOM), producing new bioavailable organic materials and environmentally important inorganic and organic carbon species (e.g. carbon monoxide (CO), carbon dioxide, carbonyl sulfide, etc.) (Blough and Del Vecchio, 2002; Mopper and Kieber, 2002). On the other hand, photochemical breakdown of CDOM, by influencing underwater UV irradiance, could have ecological consequences by altering phytoplankton assemblages (Domingues et al., 2014) or inducing photoacclimation and UV adaption (Neale et al., 1998). Since UV light is responsible for CDOM photochemistry, understanding the coastal

∗

carbon cycle requires detailed knowledge of UV processes and related biogeochemical modifications of CDOM in coastal waters. Modeling photochemistry driven by UV–CDOM interactions in highly dynamic coastal waters is challenging yet potentially achievable using ocean color. Uncertainty due to atmospheric correction errors in coastal environments, the relatively coarse spatial resolution of most existing satellite sensors that are unsuited for use in nearshore regions, as well as errors associated with many existing ocean color algorithms that would retrieve CDOM absorption coefficients in ultraviolet wavelengths, all remain a challenge for obtaining better estimates of photochemistry. In this study, we introduce a new approach to estimate photochemical fluxes using CO photoproduction as an example. It employs two distinct ocean color algorithms to retrieve UV downwelling diffuse attenuation coefficient (SeaUV; (Cao et al., 2014)) and CDOM absorption coefficient (SeaCDOM; (Cao and Miller, 2015)) from a single Hyperspectral Imager for the Coastal Ocean (HICO) image centered on Sapelo Island (latitude 31° 24′ 50" N and longitude 81° 11′ 0" W, January 8th, 2013), Georgia, USA. The enhanced spatial resolution (∼90 meters at nadir) of HICO makes it well suited for providing spatial details of coastal features and creates opportunities to

Corresponding author. E-mail address: [email protected] (W.L. Miller).

https://doi.org/10.1016/j.ecss.2018.10.013 Received 30 October 2017; Received in revised form 9 October 2018; Accepted 16 October 2018 Available online 23 October 2018 0272-7714/ © 2018 Elsevier Ltd. All rights reserved.

Estuarine, Coastal and Shelf Science 215 (2018) 199–206

F. Cao et al.

Fig. 1. (a) Map of coastal Georgia, USA; (b) True color HICO image over coastal Georgia on January 18th, 2013.

understand fine-scale processes with explicit descriptions of spatial complexity. With this blended approach, we model depth-specific, as well as depth-integrated CO photoproduction for the high resolution image and demonstrate the utility of blending two independent ocean color algorithms to address the relative optical contributions of photoactive materials in a highly dynamic and spatially heterogeneous estuarine system. This approach allows a more robust examination of biogeochemical processes and mechanisms generated by UV-dependent photochemistry.

column. There are, however, several significant limitations when applying the FM approach to coastal waters. First, the Fichot et al. (2008) SeaUV algorithms, developed using few high CDOM samples, can significantly underestimate Kd(UV) for darker inshore waters (see Fig. 4 in Cao et al. (2014)). Second, the FM approach retrieves CDOM absorption coefficient at 320 nm (ag(320)) from modeled Kd using a fixed ag(320)/ Kd(320) ratio of 0.68 (i.e., ag (320) = 0.68 × K d (320) ). This ratio was derived from limited studies that measured Kd and ag concurrently at mostly offshore and blue water stations, consequently this ratio could be very different for near-shore waters (Johannessen et al., 2003). Furthermore, modeling ag at wavelengths other than 320 nm requires use of a single pre-defined CDOM spectral slope coefficient (S) in the FM approach. S, however, is known to vary significantly, especially in terrestrial-influenced coastal waters (Twardowski et al., 2004). These assumptions could lead to erroneous results in estimating the UV optical properties required for photochemical calculations, particularly in coastal and estuarine systems. With new ocean color algorithms for retrieval of UV optical properties, the FM approach to estimate photochemistry can now be improved upon. Cao et al. (2014) extended the original SeaUV algorithms to better characterize complex coastal/inshore waters, and presented a composite set of SeaUV algorithms more suited for Kd retrieval in the UV for all water types with errors less than ± 15%. Cao and Miller (2015) further developed algorithms to allow direct estimates of spectrally-resolved CDOM absorption spectra (275–450 nm) from ocean color. Our modeling approach for the HICO image here is similar to the FM approach, but with the unique capability to independently retrieve both the apparent and inherent UV optical properties needed to estimate in situ photochemical processes from ocean color using improved algorithms appropriate for dynamic and complex coastal areas.

2. Background Quantitative modeling of depth-specific CO photoproduction rates (hereafter ΨCO(z), where z denotes depth in meters) from remote sensing ocean color, requires knowledge of four spectral parameters (Fichot and Miller, 2010): (1) the incident radiant energy as downwelling scalar irradiance right below the sea surface (E0d(λ,0–), W m−2), (2) the diffuse attenuation coefficient of downwelling irradiance (Kd, m−1),(3) the absorption coefficient of CDOM (ag, m−1), and (4) the photochemical efficiency of the reaction in question. As in Fichot and Miller (2010), the reaction considered here is CO production via CDOM photolysis. Efficiency is described as spectral apparent quantum yield (AQY, hereafter ϕCO, mol (CO) mol (photons) −1), calculated as moles of CO formed per mole of photons absorbed by CDOM. CO photoproduction rates can therefore be calculated over the entire photoreactive spectrum as in Eq. (1), max

co (z )

=

E0d ( , 0 ) × exp( K d ( ) × z ) × ag ( ,z ) ×

CO (

, z)d

min

(1) Even though photochemical reactions are acknowledged to be important in many marine biogeochemical cycles, few studies have made synoptic estimates of marine photochemistry using remote sensing (e.g., Bélanger et al., 2008; Xie et al., 2012). Notably, Fichot and Miller (2010) proposed a “practical” model (using Eq. (1)) to quantify depthspecific photochemical fluxes with the SeaUV algorithms (Fichot et al., 2008) and ocean color climatology to give a global estimate of CO photoproduction. This approach, hereafter referred to as the “FM” approach”, is highly idealized, but nonetheless serves as a good starting point for quantifying photochemical processes occurring in the water

3. Methods 3.1. Study area and HICO data This study was focused on the U.S. Georgia coastal zone, spanning roughly from −81.7°E to −80.6°E and from 30.8°N to 31.9°N (Fig. 1(a)) (Schalles and Miller, 2012). The scene encompassed three distinct estuarine systems (Altamaha, Doboy, and Sapelo Sounds). A detailed description of hydrological patterns in these three adjacent 200

Estuarine, Coastal and Shelf Science 215 (2018) 199–206

F. Cao et al.

estuaries can be found in Cao et al. (2014). Overall, this coastal estuarine system is turbid and heavily influenced by substantial CDOM input from intertidal marsh habitats and the Altamaha River (Cai, 2011). The HICO image (scene ID: 12459, Fig. 1(b)) used here was acquired during a single pass of the International Space Station on January 18th, 2013, under mostly clear sky conditions. The data was obtained from Oregon State University (http://hico.coas.oregonstate.edu/). Three atmospheric correction schemes were evaluated to find the optimal application for such a turbid, inshore environment. These were the ATmosphere REMoval algorithm (ATREM) developed by the Naval Research Laboratory (Gao and Davis, 1997), the Atmospheric Correction Algorithm for the Land (i.e., Tafkaa_6s) processed by the HICO research team, and the method of near infrared iterations implemented in the SeaWIFS Data Analysis System (SeaDAS). Rrs obtained via these three atmospheric correction schemes were also compared with Level-2 Rrs from a MODIS-Aqua image (spatial resolution of 1.1 km) (Supporting Information). Since the optical algorithms (SeaUV and SeaCDOM) applied here were initially developed with the multispectral Sea-Viewing Wide Field-of View Sensor (SeaWiFS) platform, we obtained Rrs at SeaWiFS wavebands centered at λ = 412, 443, 490, 510, 555, and 670 nm by interpolating Rrs for corresponding atmospherically corrected HICO wavelengths.

Fresnel

t)

i i

+

t)

2

+

tan( tan(

t)

i i

+

t)

2

(3)

3.2.3. Determination of AQY spectra for CO (ϕCO(λ)) Previous seasonal studies have shown the photochemical efficiency of CO production in coastal Georgia estuarine waters to be fairly well constrained (Reader and Miller, 2012). Here we added four more CO AQY spectra to the Reader and Miller (2012) data set from samples collected within the study area in May 2013. All laboratory irradiations, optical measurements, and modeling of CO AQY were performed as described in Reader and Miller (2012). Our new CO AQY values compared well with prior measurements (Fig. 2 (d)) and demonstrated that coastal Georgia waters can be reasonably modeled with a single CO AQY spectrum, representative of our small-scale study area. The single CO AQY spectrum used to model CO photochemistry obtained by refitting all 41 available AQY experiments is displayed in Eq. (4):

3.2.1. Spectral solar downwelling scalar irradiance (E0d(λ,0–)) E0d(λ,0–) was modeled from spectral solar downwelling irradiance above the sea surface (Ed(λ,0+)) for a cloudless day near solar noon to match the HICO overpass of our study area (at 11:08 Eastern Time), using the Simple Model of the Atmospheric Radiative Transfer of Sunshine model (SMARTS, version 2.9.2; http://www.nrel.gov/rredc/ smarts/). SMARTS, considered to be a robust prediction model for solar spectral irradiance, especially in the UV wavelengths (Stubbins et al., 2006), calculates both direct and diffuse components of global solar spectral irradiance with a 1.0 nm resolution between 280 and 490 nm. We limited our use of SMARTS output to the photoreactive wavelength range of 290–450 nm in this study. For modeling inputs to the SMARTS model, we obtained aerosol optical depth and ozone concentration over the study area from the NASA Goddard Earth Sciences Data and Information Services Center (https://giovanni.gsfc.nasa.gov/giovanni/). Total column aerosol optical depth at 550 nm (AOD550 nm) from the Moderate-Resolution Imaging Spectroradiometer and ozone concentrations from the Aura Ozone Monitoring Instrument were used in the SMARTS model. E0d(λ,0–) was calculated from the direct and diffuse irradiance above the water surface obtained from SMARTS (Fig. 2 (a)) using Eq. (2) following Galí et al. (2016):

CO

( ) = exp ( (8. 216 + 0. 034 × (

269. 54))

(4)

To compare results from this study with previous research, modeled CO photoproduction rates ( µ mol (CO) m−3 s−1) were scaled to hourly production ( µ mol (CO) m−3 h−1), using the simplifying assumptions that irradiance intensity and optical properties remained constant within a one-hour time window over the entire study area. 3.2.4. Computation of depth-integrated CO photoproduction rate (PCO) and differences in PCO using two different approaches Depth-integrated CO photoproduction rates using the “blended” blending , µ mol (CO) m−2 approach described in this study (hereafter PCO −1 h ) were calculated with a trapezoidal integral method that assumed a homogeneous distribution of CDOM throughout the photic zone for each pixel. The PCO estimates from blending the composite SeaUV and SeaCDOM algorithms were directly compared to those determined FM using the FM approach (hereafter PCO , µ mol (CO) m−2 h−1). To obtain FM PCO , the SeaUV algorithms were implemented exactly as described in Fichot et al. (2008) to the same HICO scene and derived Kd and ag(320) were calculated using the relation of ag (320) = 0.68 × K d (320) . To obtain the entire CDOM spectra for the FM approach, ag at other waveag ( ) = ag (320) × lengths was estimated using exp( 0.0175 × ( 320)) , where 0.0175 is S, calculated from CDOM samples collected in the study area (Cao and Miller, 2015). The CO AQY spectrum in Eq. (4) was used for PCO calculations in both approaches and the differences between models were calculated with Eq. (5) as follows:

Fresnel )

(2) +

sin( sin(

3.2.2. Derivation of optical properties We used the ATREM scheme for our HICO scene since it was the only atmospheric correction that produced Rrs values most closely matching those generated for MODIS-Aqua (Fig. S1). Downwelling diffuse attenuation coefficients, Kd(λ), at discrete UV and visible wavelengths (i.e., λ = 320, 340, 380, 412, 443, and 490 nm, e.g. Fig. 2(b)) were estimated from visible Rrs using the composite SeaUV algorithms as detailed in Cao et al. (2014). Kd(λ) from 320 to 450 nm was subsequently derived at 5 nm intervals with a cubic interpolation and extrapolated into the UV-B (λ = 290–320 nm) assuming an exponential increase of Kd over decreasing wavelengths and a spectral slope value calculated using Kd(320) and Kd(340). Spectrally resolved CDOM absorption coefficients ag (290–450 nm; 5 nm resolution) were modeled using the SeaCDOM algorithm (Cao and Miller, 2015) (e.g., Fig. 2 (c)).

Our model incorporated several assumptions from the FM approach, including: (1) UV wavelengths from 290 to 450 nm dominate photochemical reactions involving CDOM, (2) total solar scalar irradiance is approximated by solar downwelling scalar irradiance (E0d(λ,0–)) and upwelling irradiance is negligible, and (3) attenuation of downwelling scalar irradiance can be approximated with Kd(λ). Derivations of input parameters and analysis of the resulting uncertainty are discussed below.

0.066) + Ed,DIR ( ,0+) × (1

1 × 2

where is the incident angle (solar zenith angle) and t is the transmitted angle. The solar zenith angle was modeled with SMARTS. t was obtained using Snell's Law with refractive index values of 1 and 1.34 for air and seawater, respectively.

3.2. Modeling procedures

E0d ( , 0 ) = Ed,DIFF ( ,0+) × (1

=

+

where Ed,DIFF(λ,0 ) and Ed,DIR(λ,0 ) are the diffuse and direct beam in the global horizontal irradiance above the sea surface, respectively. We assumed a constant reflection of 0.066 for the diffuse irradiance. Fresnel corrects the direct irradiance being reflected and was calculated according to the Fresnel's equation (Eq. (3)):

BL Difference = (PCO

201

FM BL FM PCO )/[(PCO + PCO )/2]

(5)

Estuarine, Coastal and Shelf Science 215 (2018) 199–206

F. Cao et al.

Fig. 2. Input parameters for the calculation of CO photoproduction rates with (a) E0d(λ,0–), with simulation condition as: latitude = 31.5°N, longitude = 81.76 °W, clear sky with ozone concentration of 265 DU and AOD550 nm of 0.048 on January 18th, 2013; (b) Kd(340) derived from the composite SeaUV algorithms; (c) ag(340) derived from the SeaCDOM algorithms; and (d) Apparent quantum yield spectra of CO, where grey background lines denote data published in Reader and Miller (2012) (N = 37), black lines represent our data set sampled during 2013 (N = 4), and the blue line represents the spectrum used in this study obtained with a total of 41 measurements, with the spectral equation AQY(λ) = exp (–(8.216 + 0.034× (λ-269.54))) used for modeling. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

4. Results

inshore water z = 1 m was generally one order of magnitude lower than at the sea surface (z = 0−), ranging from < 0.1 to 2 µ mol m−3 h−1, with a mean of 1.4 µ mol m−3 h−1 (Fig. 3 (b)). Spatial comparison of Ψ at these two different depths shows a striking difference for pixels obtained behind the barrier islands, showing the lowest rates for z = 1 m while Ψ remaining highest for z = 0−. This is understandable, given the high abundance of optically active constituents in estuaries including CDOM (Fig. 2 (c)) and particles, both contributing to the high Kd observations in the first optical depth (Fig. 2 (b)) (Schalles, 2006). Rapid attenuation of UV irradiation resulted in little solar energy available for photochemistry at z = 1 m, creating low photoproduction rates within estuaries at the 1-meter isobaths. ΨCO at z = 1 m was higher for nearshore areas due to reduced UV attenuation, even with lower CDOM

4.1. Depth-specific distribution of CO photoproduction The spatial distribution of depth-specific CO photoproduction rates (ΨCO, µ mol (CO) m−3 h−1) varied partially due to the different depths considered. The rates calculated at the sea surface (z = 0–) demonstrated remarkable spatial variability, ranging from > 40 µ mol m−3 h−1 in interior inshore waters to 0.2 µ mol m−3 h−1 further offshore, with a mean of 4.2 µ mol m−3 h−1 (Fig. 3 (a)). These values compare well with coastal waters reported elsewhere, for example in the Gulf of Maine (Ziolkowski and Miller, 2007) and the Canada Basin (Song et al., 2013). ΨCO at the depth right above the first optical depth in the interior

Fig. 3. Model output of depth-specific CO photoproduction rates (ΨCO(z), µ mol (CO) m−3 h−1) at (a) z = 0– m; and (b) z = 1 m isodepths. 202

Estuarine, Coastal and Shelf Science 215 (2018) 199–206

F. Cao et al.

blending Fig. 4. Model output of depth-integrated CO photoproduction rates (PCO , µ mol (CO) m−2 h−1).

values compared to estuaries and offshore. These spatial trends in UV attenuation and CDOM resulted in the highest values of Ψ at 1 m in the inner shelf (Ψ of > 1.8 µ mol m−3 h−1), with lower values (Ψ of < 1 µ mol m−3 h−1) inshore and further offshore (Fig. 3 (b)). 4.2. Depth-integrated CO photoproduction The spatial detail of depth-integrated photochemistry obtained with blending our model (PCO ) varied by nearly two orders of magnitude within the scene (0.1–16 µ mol m−2 h−1), with a mean of 4.4 µ mol m−2 h−1 blending (Fig. 4). An interesting result was that the lowest PCO values occurred in the innermost inshore estuarine waters, exactly where CDOM and surface rates (ΨCO(z = 0–)) were the highest. In other words, the most photochemically productive waters, in terms of PCO, were decoupled from their photochemical source material (CDOM) in these BL estuaries. Seaward transects for PCO showed decreasing values and were similar to ΨCO at the surface and at 1-meter depth. Distributions of depth-integrated rates using the FM approach original blending , Fig. 5 (a)), however, bear little resemblance to PCO (Fig. 4). (PCO blending original Compared to PCO , PCO exhibited far less spatial variability and original blending the difference between PCO and PCO was quite striking, especially for inshore waters and downstream of the Altamaha River outflow, which differed by as much as ± 100% (Fig. 5 (b)). A detailed discussion of these discrepancies is provided in Section 5.2.

Fig. 5. (a) Modeled depth-integrated CO photoproduction rates using the aporiginal ) ; and (b) differences between proach in Fichot and Miller (2010) (PCO original blending PCO and PCO .

5. Discussion

characterization of spectral photochemical efficiency introduce uncertainties into the calculation of P or Ψ. For example, with the optimistic assumption that Rrs is 100% accurate, even the best UV optical models inherently introduce an error of 15% and 25% respectively, using the composite SeaUV and SeaCDOM algorithms when deriving Kd(340) and ag(340). A second source of uncertainty stems from the variability of laboratory derived photochemical quantum yield spectra. Numerous studies have shown that AQY for several photoproducts critically depends on CDOM characteristics (Stubbins et al., 2011) and therefore AQY spatial distributions can vary appreciably (e.g., Bélanger et al., 2008; Bélanger et al., 2006; Johannessen and Miller, 2001). CO AQY spectra, however, remain relatively constant in oligotrophic open ocean

Based on the optical and photochemical parameters, approaches, and assumptions involved in producing the results above, the following discussion explores three issues: (1) uncertainties in estimating photoproduction from ocean color; (2) potential causes of differences between the depth-integrated CO photoproduction results in this study blending (i.e., PCO ) and values calculated using the FM approach (i.e., original PCO ); and (3) implications for UV-dependent aqutic processes derived from the use of two independent ocean color algorithms to retrieve Kd and ag. 5.1. Uncertainties in CO photoproduction estimation from ocean color Both

the

derivation

of

UV

optical

parameters

and

the 203

Estuarine, Coastal and Shelf Science 215 (2018) 199–206

F. Cao et al.

waters (Zafiriou et al., 2003) and so reasonably allow the usage of one average AQYCO spectrum in global studies such as Fichot and Miller (2010). Estuarine and coastal areas, on the other hand, are subjected to variable terrestrial influence and AQY variations can span several orders of magnitude (Powers and Miller, 2015b), almost certainly due to the increased complexity of CDOM sources and photochemical mechanisms by which CDOM produces CO (Gao and Zepp, 1998). This natural variability of CDOM complicates efforts to properly capture photochemical rates in coastal waters. This final difficulty is partly overcome in our coastal study area in two ways. First, the area within our HICO image is relatively small and represents only a subset of possible CDOM sources that might create variability in AQY. Second, AQY spectra for this area have been determined with a field sampling program covering three years of seasonal study, allowing us to capture the AQY variability and define a single AQYCO spectra that adequately represents our specific study area. We caution, therefore, that, photochemical modeling results from this study cannot simply be generalized to other estuaries or the entire coastal ocean with the same confidence. Although significant improvements in the accuracy of retrieving Kd and ag in a wide variety of coastal waters have been achieved (Cao et al., 2014; Cao and Miller, 2015), they may be largely overshadowed by the large uncertainties resulting from variable AQY spectra over larger geographic areas with diverse sources and fates of organic carbon. In light of this, to apply the approach proposed in this study elsewhere, rigorous modeling and precise knowledge of regional AQY spectra are essential to obtain optimal estimates of photoproduction rates from satellite data. In addition, ocean color algorithms in coastal waters that rely on satellite platforms are often subjected to errors resulting from the imperfect atmospheric correction of blue wavebands, in particular the 412 nm waveband (Goyens et al., 2013). It is worth noting that both the composite SeaUV and SeaCDOM algorithms were developed from optical buoy data that included Rrs(412) and the inclusion of uncertainties due to atmospheric correction can lead to biased optical property estimates when applied to satellite data. Unfortunately, there were no robust ground-truth data acquired in this study, preventing direct determination of the exact uncertainty associated with this recognized issue. Nevertheless, this potential source of error should not be neglected and future ocean color algorithms for coastal waters would benefit from approaches that do not require Rrs(412) for their implementation.

Fig. 6. Distribution of spectral contribution of CDOM to total light attenuation at 340 nm (ag(340)/Kd(340)).

To help with the interpretation, we examined the spatial distribution of the fraction of UV radiation absorbed by CDOM relative to light attenuation at λ = 340 nm (i.e., ag(340)/Kd(340)) estimated from the HICO image. As shown in Fig. 6, ag(340)/Kd(340) varied remarkably, with inshore waters generally having lower ag(340)/Kd(340) values than coastal and offshore waters, suggesting a dynamic partitioning of UV light into different optically active components that can include CDOM, mineral particles (Song et al., 2013), plankton, especially those containing various UV-absorbing pigments such as mycosporine-like amino acids (Ayoub et al., 2012). Waters from the Altamaha River have large particle loadings which absorb and scatter light and show the lowest ag(340)/Kd(340) (∼0.1–0.2) in the study area. The attenuation of irradiance by particles declines during estuarine mixing as particles are removed from the water column due to flocculation and sinking, resulting in higher values (∼0.9–1) of ag(340)/Kd(340) in nearshore coastal areas where CDOM remained high in January 2013. This value is similar to field observations made by Bittar et al. (2016), reporting DOC concentrations 10 times higher than particulate organic matter in a nearby Georgia coastal estuary during a high tide in January 2013. This was the same month that our HICO image was obtained, and therefore, we assume there was little scattering in the UV contributed by organic particles. Further offshore, the ratio decreased, likely representing dilution and/or photobleaching of CDOM with increasing distance offshore. This dynamic range of ag(340)/Kd(340) observed from ocean color agrees reasonably well with prior field observations within our study area that ranged from 0.33 to 1 (Cao et al., 2014; Cao and Miller, 2015). While direct field verification is warranted, these results confirm the potential of this blended approach to quantify the contribution of CDOM absorption coefficient to the downwelling attenuation of photons in surface waters using remote sensing. Our modeling results provide insight into the underlying UV optical drivers that govern spatial distribution of depth-integrated photochemical and photobiological processes. There is a close coupling of blending patterns of ag/Kd (Fig. 6) and PCO (Fig. 4), with lower PCO associated with lower ag/Kd values. These values very likely track particle concentrations within estuarine waters and show increased PCO in coastal waters where we calculated higher ag/Kd. Similar correlations have been observed in other optically complex waters (Bélanger et al., 2008; Xie et al., 2012). Unlike modeling depth-specific photoproduction rates that requires explicit knowledge of both Kd and ag, estimation of

5.2. Interpretation of depth-integrated CO photoproduction blending The large PCO differences in this study between (PCO ) and the FM original approach (PCO ), can be explained by differences in modeling procedures. In applying both models, we used the same ϕCO(λ) and irradiance data, and therefore the spectral contribution of the CDOM absorption coefficient to light attenuation, ag(λ)/Kd(λ), or the ratio of ag(λ)/atotal(λ) in Bélanger et al. (2006), becomes the single driver behind PCO variations between these two approaches. It is known that the value of ag(λ)/Kd(λ) not only varies with wavelength (Johannessen et al., 2003) but depends on water optical properties (Tedetti et al., 2007; Zepp et al., 2008). Based on a more consistent ratio developed from coastal and blue water field data, the FM approach used a constant of 0.68 for ag(320)/Kd(320) and thus cannot accomodate known spatial heterogeneity in complex coastal and inland waters. While appropriate for global models, this simplistic assumption leads to the homogeneous original distribution of PCO seen in Fig. 5 (a), and fails to capture spatially resolved PCO in the coastal environment where concentrations of various optically active materials occur independently. Comparatively, decoupled retrievals of Kd and ag from two separate algorithms applied independently to each pixel, captured this variance and seemed to recover more reasonable spatial details for the optically complex waters found on the Georgia coast.

204

Estuarine, Coastal and Shelf Science 215 (2018) 199–206

F. Cao et al.

CDOM-driven photochemical processes integrated over the entire water column is primarily a function of the proportion of the spectral radiation absorbed by CDOM to the total loss of photons to other mechanisms, namely ag/Kd. Whereas this partitioning of photons has been noted previously in the visible regime (Bélanger et al., 2006, 2008), to the best of our knowledge, our study provides the first synoptic estimate of this optical phenomenon in the UV. Revealing these spatial patterns from satellite-derived ocean color underscores the value of this novel ability to blend two independently derived ocean color algorithms to probe UV processes in the marine environment. Using this approach, it may be possible to further examine the long-term and spatial extent of seasonal and episodic variability in CDOM outflow, coastal wetland exchange, riverine particle loading, photochemical bleaching of CDOM, and re-suspension of sediments in optically complex coastal environments.

synoptically assess and monitor optically complex inland, estuarine, and coastal waters. In addition, evaluation of UV light partitioning into different optically active constituents can, at present, only be made for the dissolved phase. Nevertheless, this determination is informative. Finally, while our “snapshot” assessment of UV-driven photochemistry from HICO ocean color appears robust, in should again be noted that the AQY spectra for CO (or for any other marine photochemical reaction quantified relative to CDOM absorbance) in coastal and inshore systems are likely regional and exert a dominant control on photochemical calculations. Upscaling CO production values obtained in this study to other coastal waters could create unrealistic results unless the temporal and spatial variability of AQY is recognized. The challenge of constraining AQY spectra is common to many coastal study locations, particularly considering the natural variability of AQY. Consequently, our study is less an evaluation of the absolute accuracy of modeled CO photoproduction rates than it is as a demonstration of the use of remote sensing to address UV driven biogeochemical processes at a fine scale in optically dynamic inshore systems. Accurate seasonal and yearly quantification of photochemical rates at coastal spatial scales will likely require long term, systematic ocean color climatology as well as rigorous modeling of regional AQY for any photoproduct in question. Photochemical rates dependent on UV CDOM photolysis can easily be pursued for photoproducts other than CO. For instance, the photoproduction of hydrogen peroxide has been mapped on a global scale using remote sensing (Powers and Miller, 2015b). In turn, hydrogen peroxide mapping may potentially serve as a proxy to allow estimates of the photoproduction of carbon dioxide (Powers and Miller, 2015a,b), a reaction of great interest from climate change and carbon cycling perspectives. In this respect, future work to better quantify the significance of photochemistry in controlling marine carbon inventories and cycling, as well as investigation of UV-affected ocean ecology should benefit from our two-algorithm, blended approach.

5.3. Implications of decoupled retrievals of optical properties In addition to the substantial benefits inherent in the independent retrievals of two optical parameters by different ocean color algorithms for estimating CDOM photochemistry, new insights should be obtainable for estimating contributions from photochemistry occurring in the particulate phase. Recent studies note significant photochemical reactions that involve suspended particles (Xie and Zafiriou, 2009 and references therein). Notably, Song et al. (2013) demonstrated that particulate organic matter, regardless of its origins, could be more susceptible to photo-alteration than CDOM in terms of CO photoproduction. More generally, Estapa and Mayer (2010) concluded that quantitative knowledge of light absorption partitioning into different photochemically active materials is indispensable to clarify the photoredox mechanisms of particulate organic matter. Hence, spectral partitioning of the CDOM absorption in the UV relative to the total attenuated photon budget, as described in this study, could help describe variability in the UV light allocation among materials in different phases (dissolved vs. particulate) and further aid in disentangling metal redox (e.g. Fe, Cu, Mn) effects on the photochemistry of particulate organic matter in the ocean.

Acknowledgements HICO image data were provided through a data grant from the Office of Naval Research to J.F.S and W. L.M. We are grateful to Bo-Cai Gao in the Naval Research Laboratory and to Curtiss O. Davis and Jasmine Nahorniak at the Oregon State University for helping with the atmospheric correction on the HICO image. Some materials in this manuscript are based on work supported while WLM was serving at the National Science Foundation. We thank the NASA Ocean Biology Processing Group for providing access to the ocean color satellite data and geolocation data. We also thank the Editor and the two reviewers for their detailed and careful reviews on this manuscript.

6. Concluding remarks and outlook This work has shown the potential of blending two different ocean color algorithms to address UV dependent processes in a coastal environment using high spatial resolution satellite observations. Two notable contributions are (1) the successful demonstration of modeled, high resolution depth–specific and depth–integrated photochemical production of CO in an estuarine setting, and (2) the value of decoupled retrievals of Kd and ag for the appraisal of optical dynamics that estimate the percentage of UV light attenuated by CDOM at a high spatial resolution using remote sensing. This later contribution appears to overcome a persistent challenge in remote sensing applications for predicting marine photochemistry and could provide opportunities to quantify diverse UV-dependent processes in the ocean. With good estimates of the contribution of CDOM absorption to the total UV attenuation, the impact of photochemical reactions important to biogeochemical issues, including carbon cycling, redox chemistry, and marine ecosystem responses to changing UV radiation, can be more accurately addressed. With regard to remotely sensed optical variables (i.e., Kd and ag), uncertainties stemming from the inclusion of Rrs in the blue wavelengths, namely Rrs(412), in satellite ocean color algorithms are clearly recognized as problematic due to current errors in atmospheric corrections. Because of this, further refinements of the Cao et al. (2014) and Cao and Miller (2015) algorithms for Kd and ag that omit Rrs(412) may reduce errors by eliminating the atmospheric correction uncertainties in this spectral region. Such refinements should improve optical retrievals from ocean color and improve our abilities to

Appendix A. Supplementary data Supplementary data to this article can be found online at https:// doi.org/10.1016/j.ecss.2018.10.013. References Ayoub, L.M., Hallock, P., Coble, P.G., Bell, S.S., 2012. MAA-like absorbing substances in Florida Keys phytoplankton vary with distance from shore and CDOM: implications for coral reefs. J. Exp. Mar. Biol. Ecol. 420, 91–98. Bauer, J.E., Cai, W.-J., Raymond, P.A., Bianchi, T.S., Hopkinson, C.S., Regnier, P.A., 2013. The changing carbon cycle of the coastal ocean. Nature 504, 61–70. Bélanger, S., Babin, M., Larouche, P., 2008. An empirical ocean color algorithm for estimating the contribution of chromophoric dissolved organic matter to total light absorption in optically complex waters. J. Geophys. Res.: Oceans 113 (1978–2012). Bélanger, S., Xie, H., Krotkov, N., Larouche, P., Vincent, W.F., Babin, M., 2006. Photomineralization of terrigenous dissolved organic matter in Arctic coastal waters from 1979 to 2003: interannual variability and implications of climate change. Global Biogeochem. Cycles 20. Bittar, T.B., Berger, S.A., Birsa, L.M., Walters, T.L., Thompson, M.E., Spencer, R.G.M., Mann, E.L., Stubbins, A., Frischer, M.E., Brandes, J.A., 2016. Seasonal dynamics of dissolved, particulate and microbial components of a tidal saltmarsh-dominated estuary under contrasting levels of freshwater discharge. Estuar. Coast Shelf Sci. 182,

205

Estuarine, Coastal and Shelf Science 215 (2018) 199–206

F. Cao et al. 72–85. Blough, N.V., Del Vecchio, R., 2002. Chromophoric DOM in the coastal environment. In: Biogeochemistry of marine dissolved organic matter, pp. 509–546. Cai, W.J., 2011. Estuarine and coastal ocean carbon paradox: CO2 sinks or sites of terrestrial carbon incineration? Annu. Rev. Mar. Sci. 3, 123–145. Cao, F., Fichot, C.G., Hooker, S.B., Miller, W.L., 2014. Improved algorithms for accurate retrieval of UV/visible diffuse attenuation coefficients in optically complex, inshore waters. Remote Sens. Environ. 144, 11–27. Cao, F., Miller, W.L., 2015. A new algorithm to retrieve chromophoric dissolved organic matter (CDOM) absorption spectra in the UV from ocean color. J. Geophys. Res.: Oceans 120, 496–516. Cauwet, G., 2002. DOM in the coastal zone. In: Biogeochemistry of marine organic matter, pp. 570–611. Domingues, R.B., Guerra, C.C., Barbosa, A.B., Brotas, V., Galvão, H.M., 2014. Effects of ultraviolet radiation and CO2 increase on winter phytoplankton assemblages in a temperate coastal lagoon. J. Plankton Res. 672–684. Estapa, M.L., Mayer, L.M., 2010. Photooxidation of particulate organic matter, carbon/ oxygen stoichiometry, and related photoreactions. Mar. Chem. 122, 138–147. Fichot, C.G., Miller, W.L., 2010. An approach to quantify depth-resolved marine photochemical fluxes using remote sensing: application to carbon monoxide (CO) photoproduction. Remote Sens. Environ. 114, 1363–1377. Fichot, C.G., Sathyendranath, S., Miller, W.L., 2008. Sea UV and Sea UVC: algorithms for the retrieval of UV/Visible diffuse attenuation coefficients from ocean color. Remote Sens. Environ. 112, 1584–1602. Galí, M., Kieber, D.J., Romera-Castillo, C., Kinsey, J.D., Devred, E., Pérez, G.L., Westby, G.R., Marrasé, C., Babin, M., Levasseur, M., 2016. CDOM sources and photobleaching control quantum yields for oceanic DMS photolysis. Environ. Sci. Technol. 50, 13361–13370. Gao, B.C., Davis, C.O., 1997. Development of a Line-by-line-based Atmosphere Removal Algorithm for Airborne and Spaceborne Imaging Spectrometers, Optical Science, Engineering and Instrumentation, vol 97. International Society for Optics and Photonics, pp. 132–141. Gao, H., Zepp, R.G., 1998. Factors influencing photoreactions of dissolved organic matter in a coastal river of the southeastern United States. Environ. Sci. Technol. 32, 2940–2946. Goyens, C., Jamet, C., Schroeder, T., 2013. Evaluation of four atmospheric correction algorithms for MODIS-Aqua images over contrasted coastal waters. Remote Sens. Environ. 131, 63–75. Johannessen, S.C., Miller, W.L., 2001. Quantum yield for the photochemical production of dissolved inorganic carbon in seawater. Mar. Chem. 76, 271–283. Johannessen, S.C., Miller, W.L., Cullen, J.J., 2003. Calculation of UV attenuation and colored dissolved organic matter absorption spectra from measurements of ocean color. J. Geophys. Res.: Oceans 108 (1978–2012). Mopper, K., Kieber, D.J., 2002. Photochemistry and the Cycling of Carbon, Sulfur, Nitrogen and Phosphorus. Academic Press, New York. Neale, P.J., Cullen, J.J., Davis, R.F., 1998. Inhibition of marine photosynthesis by

ultraviolet radiation: variable sensitivity of phytoplankton in the Weddell-Scotia Confluence during the austral spring. Limnol. Oceanogr. 43, 433–448. Powers, L.C., Miller, W.L., 2015a. Hydrogen peroxide and superoxide photoproduction in diverse marine waters: a simple proxy for estimating direct CO2 photochemical fluxes. Geophys. Res. Lett. 42, 7696–7704. Powers, L.C., Miller, W.L., 2015b. Photochemical production of CO and CO 2 in the Northern Gulf of Mexico: estimates and challenges for quantifying the impact of photochemistry on carbon cycles. Mar. Chem. 171, 21–35. Reader, H.E., Miller, W.L., 2012. Variability of carbon monoxide and carbon dioxide apparent quantum yield spectra in three coastal estuaries of the South Atlantic Bight. Biogeosciences 9, 4279–4294. Schalles, J.F., Miller, W.L., 2012. HICO Imagery for Geospatial Analysis of Organic Carbon, Phytoplankton, and Tidal Wetlands of the Georgia Coast & SAB. Schalles, J.F., 2006. Optical remote sensing techniques to estimate phytoplankton chlorophyll a concentrations in coastal. In: Remote sensing of aquatic coastal ecosystem processes, pp. 27–79. Song, G., Xie, H., Bélanger, S., Leymarie, E., Babin, M., 2013. Spectrally resolved efficiencies of carbon monoxide (CO) photoproduction in the western Canadian Arctic: particles versus solutes. Biogeosciences 10, 3731–3748. Stubbins, A., Law, C.S., Uher, G., Upstill-Goddard, R.C., 2011. Carbon monoxide apparent quantum yields and photoproduction in the Tyne estuary. Biogeosciences 8, 703–713. Stubbins, A., Uher, G., Law, C.S., Mopper, K., Robinson, C., Upstill-Goddard, R.C., 2006. Open-ocean carbon monoxide photoproduction. Deep Sea Res. Part II Top. Stud. Oceanogr. 53, 1695–1705. Tedetti, M., Sempéré, R., Vasilkov, A., Charriere, B., Nérini, D., Miller, W.L., Kawamura, K., Raimbault, P., 2007. High penetration of ultraviolet radiation in the south east Pacific waters. Geophys. Res. Lett. 34. Twardowski, M.S., Boss, E., Sullivan, J.M., Donaghay, P.L., 2004. Modeling the spectral shape of absorption by chromophoric dissolved organic matter. Mar. Chem. 89, 69–88. Xie, H., Bélanger, S., Song, G., Benner, R., Taalba, A., Blais, M., Tremblay, J.-É., Babin, M., 2012. Photoproduction of ammonium in the southeastern Beaufort Sea and its biogeochemical implications. Biogeosciences 9, 3047–3061. Xie, H., Zafiriou, O.C., 2009. Evidence for significant photochemical production of carbon monoxide by particles in coastal and oligotrophic marine waters. Geophys. Res. Lett. 36. Zafiriou, O.C., Andrews, S.S., Wang, W., 2003. Concordant estimates of oceanic carbon monoxide source and sink processes in the Pacific yield a balanced global “blue‐water” CO budget. Global Biogeochem. Cycles 17. Zepp, R.G., Shank, G.C., Stabenau, E., Patterson, K.W., Cyterski, M., Fisher, W., Bar-Tels, E., Anderson, S.L., 2008. Spatial and temporal variability of solar ultraviolet exposure of coral assemblages in the Florida Keys: importance of colored dissolved organic matter. Limnol. Oceanogr. 53, 1909. Ziolkowski, L.A., Miller, W.L., 2007. Variability of the apparent quantum efficiency of CO photoproduction in the Gulf of Maine and Northwest Atlantic. Mar. Chem. 105, 258–270.

206