Volume 116, Issue C11
Free Access

Penetration of UV irradiance into the global ocean

T. J. Smyth

T. J. Smyth

Plymouth Marine Laboratory, Plymouth, UK

Search for more papers by this author
First published: 15 November 2011
Citations: 42

Abstract

[1] A new global ocean-atmosphere model has been developed to determine the penetration of ultraviolet (UV) radiation through the water column. This is accomplished by combining an atmospheric UV irradiance model, taking into consideration the effects of aerosols, clouds, and the air-sea interface, with empirical in-water diffuse attenuation coefficient (Kd(λUV)) relationships. These empirical relationships are derived from simultaneous in situ profiles of visible wavelength inherent optical properties and downwelling UV irradiances. The combined model is applied to global data sets using a look-up table approach to speed up calculation time. The atmospheric model compared against ∼3000 data points gave a root-mean-square error (RMSE) of between 10% and 15% at wavelengths of 305, 325, 340, and 380 nm; the coupled global model compared against 30 independent in-water irradiance profiles gave a logarithmic RMSE of between 0.15 and 0.35 at these wavelengths. On the global scale the 10% irradiance levels were found to be deepest in the oceanic gyres (∼18, 32, 44, and 70 m at 305, 325, 340 and 380 nm, respectively) and shallowest in the optically complex continental shelf regions. The calculated UV doses were shown to be spectrally and seasonally variable, with the highest values being encountered in the eastern Mediterranean during July, with values of ∼0.5, 4, 7, and 10 kJ m−2 d−1 nm−1 at 305, 325, 340, and 380 nm, respectively.

Key Points

  • Distribution and dosage of UV radiation in the global pelagic ocean
  • Novel Kd UV relationships with visible wavelength IOPs over ocean provinces
  • Global quantification of UV impacts on biogeochemistry

1. Introduction

[2] Ultraviolet (UV) radiation, with a wavelength between 280 and 400 nm, plays an important role in regulating biogeochemical cycles within the global ocean. Previous research has focused on concerns raised by the well-publicized reduction in stratospheric ozone [Farman et al., 1985; Stolarski et al., 1991; Gleason et al., 1993] and the consequential predicted rise in surface levels of UV radiation. This has included the possible deleterious consequences for the health of marine ecosystems [de Mora et al., 2000], the photolysis of volatile organoiodine compounds of marine origin [Martino et al., 2006], the photo degradation of colored dissolved organic matter (CDOM) [Mopper and Kieber, 2002], the production of UV radiation–absorbing mycosporine-like amino acids (MAAs) by phytoplankton [Karsten et al., 1999; Llewellyn and Harbour, 2003], and damage and alterations to DNA structure [Häder and Sinha, 2005].

[3] The current generation of ocean color satellites (e.g., Sea-viewing Wide Field-of-view Sensor (SeaWiFS), Moderate resolution Imaging Spectroradiometer (MODIS), and Medium-Resolution Imaging Spectrometer (MERIS)) estimate water leaving radiances within the visible part of the electromagnetic spectrum. The shortest wavelengths are in the far blue (∼410 nm). Therefore to determine the in-water propagation of UV radiation on a global scale, empirical [Fichot et al., 2008; Hirata, 2008] or analytical Vasilkov et al. [2001, 2005] modeling approaches are required to relate the visible wavelength measurands to useful UV radiation parameters such as the diffuse attenuation coefficient, Kd(λUV).

[4] The value of the diffuse attenuation coefficient, Kd(λUV), determines the rate of decay with depth of the in-water spectral UV radiation field. Once this parameter is known the magnitude of the spectral UV irradiance within the water column can be determined if an atmospheric model [Sabziparvar et al., 1999; Vasilkov et al., 2001; Mayer and Kylling, 2005] is used to predict the surface UV irradiance. Further complexity can be introduced if the mixed layer depth [de Boyer Montégut et al., 2004] over which the UV irradiance effectively acts is known. The depth dimension is critical if the UV dose that a parcel of water encounters as it circulates over the mixed layer depth is to be determined. This will have importance in studies of photophysiological and photochemical effects of UV radiation.

[5] In this paper robust empirical relationships between visible wavelength inherent optical properties (IOPs) and multispectral values of Kd(λUV) are derived from in situ data collected over a wide spread of biogeographical locations. These are then used to derive global maps of Kd(λUV) spanning the entire SeaWiFS mission (1998–2009). Surface irradiance values are provided by the libRadTran (v1.4) atmospheric UV model [Mayer and Kylling, 2005] to allow the quantification of the amount of UV irradiance through the water column. The model developed here is essentially a hybrid ocean-atmosphere model and some error may be incurred in the simplification of the air-sea interface boundary term. Differences between various coupled ocean-atmosphere models, including differences in the way radiation is propagated between the atmosphere and the ocean, have been previously investigated by Mobley et al. [1993].

2. Method

[6] The model developed in this paper comprises two components. The first is an empirical in-water model, developed using in situ measurements of downwelling UV spectral irradiance, Ed(λ), and coincidental visible wavelength inherent optical properties (IOPs). Satellite derived fields of IOPs [Smyth et al., 2006], specifically the total absorption at 443 nm, atot(443), were then used to produce global maps of Kd(λ) at four wavelengths in the UV domain. The second is the libRadTran (v1.4) [Mayer and Kylling, 2005] atmospheric model, which was used to provide the above surface UV irradiance, Ed0+(λ). The models were coupled across the air-sea interface, taking account of the direct and diffuse parts of the downwelling irradiance. A schematic diagram (Figure 1) shows the individual components of the coupled atmospheric in-water model.

Details are in the caption following the image
Schematic diagram showing the atmospheric and in-water models and their inputs, outputs, and coupling across the air-sea interface.

2.1. In-Water Model

[7] In-water optical measurements were taken on four separate cruises in contrasting bio-optical regions. These were: close to the Cape Verde Islands (Surface Ocean Lower Atmosphere Study (SOLAS) Investigation of Near-Surface Production of Iodocarbons–Rates and Exchange (INSPIRE), November 2007); in the Mauritanian upwelling region (SOLAS impact of coastal upwelling on air sea exchange of climatically important gases (ICON), April 2009); and two north–south transects of the Atlantic Ocean (Atlantic Meridional Transect (AMT)-18, October 2008; AMT-19, October 2009).

[8] Ed(λ,z) at four UV wavelengths (305, 325, 340 and 380 nm) was measured using a Satlantic UV-507 radiometer attached to an optical profiling rig. A Wetlabs ac9+ was simultaneously deployed on the same optical rig to determine two IOPs in the visible spectral domain: absorption (a(λ)) and attenuation (c(λ)) coefficients. At the start of each deployment, the optical rig was lowered to a depth of about 5 m for 2–3 min to allow the instruments to temperature stabilize and to purge the pumped systems of bubbles. The rig was then raised to just below the surface and immediately lowered to a depth of approximately 200 m and then raised slowly through the water column using the ship's crane, extended as far from the superstructure as possible, at a rate of approximately 0.2 m s−1. No tilt and roll sensor was available, so the wire angle was closely monitored and profiling ceased if this became greater than 10°, only resuming when the necessary maneuvering of the ship had rectified this. The combination of slow profiling and high data rates together with strict monitoring of the wire angle ensured sufficient data volumes to correct for wave focusing and angular deviations. All casts were carried around within ±2 h of solar noon to achieve the minimum zenith angle for a given location; this was generally between 10 and 40° in the tropical SOLAS cruises, but necessarily greater on the midlatitude autumn portions of the AMT cruises (up to 60°). The rig was deployed on the sunward side of the ship to avoid shading issues and care was taken to avoid rapidly changing skies, to the extent that profiling was ceased if the solar disk became temporarily covered during a cast. Surface hyperspectral (Δλ = 1.5 nm) irradiance was monitored every 5 min throughout the day using a Trios Rameses-ACC UV (280–500 nm) sensor during three of the cruises (INSPIRE, ICON, and AMT-19). Both the Satlantic and Trios UV sensors were returned to their respective manufacturers for an annual calibration.

[9] Kd(λ), was determined by the slope of ln(Ed(λ)) against depth between the surface and the spectrally varying 1% irradiance level using the upcast part of the vertical profile. This was done by median binning the Ed(λ) profile into 2 m depth intervals from a depth of 1 m and below and then carrying out the regression. Only profiles where R2 > 0.98 for the Kd(λ) regression were selected for further analysis. This strict criterion was set to automatically and objectively remove vertical profiles where there were strong vertical inhomogeneities or remaining surface irradiance changes.

[10] The ac9+ was routinely returned to the manufacturer for recalibration; pure water field calibrations carried out during the period of a cruise; and corrections for scattering, salinity and temperature applied following the manufacturer's protocols. The mean and associated standard deviation of the total absorption, excluding the contribution from pure water, at 443 nm, atot(443), was extracted from the individual profiles from the surface down to the 1% irradiance level depth (determined using the Ed(380) measurements). The total absorption, atot(443), here implies the absorption contributions at 443 nm from phytoplankton, detrital matter and CDOM combined.

[11] Monthly averaged global (18 km) satellite remote sensing reflectances (Rrs) derived from SeaWiFS (G. C. Feldman and C. R. McClain, Ocean Color Web, SeaWiFS reprocessing 2009.1, http://oceancolor.gsfc.nasa.gov) were used as input to the Smyth et al. [2006] IOP model for the period between 1998–2009. The atot(443) global maps [Smyth and Artioli, 2010] were then used to derive spectral Kd(λ) on the basis of relationships described in section 3, between atot(443) and Kd(380), Kd(380) and Kd(340), Kd(380) and Kd(325), and Kd(380) and Kd(305).

2.2. Atmospheric Model

[12] The above surface measurements of hyperspectral irradiance using the Trios Rameses-ACC UV sensor were used to configure and validate the libRadran (v1.4) atmospheric model [Mayer and Kylling, 2005]; the settings and data sets used to run the model are shown in Table 1. The libRadtran model was run in two different modes. Firstly a validation mode to compare with in situ measurements and secondly a global mode designed to construct a look-up table to speed up calculations when large input data sets are used.

Table 1. Input Details for the libRadTran Atmospheric Model
Parameter Data Set or Value Notes
Atmospheric profile Anderson et al. [1986] function of latitude and date
Solar flux ATLAS-3 Woods et al. [1996]
Earth-Sun distance function of date
Ozone TOMS monthly zonal average 1997–2004
Aerosol type Maritime Shettle [1989]
Aerosol season summer/winter function of latitude and date
Background aerosol above 2 km
Aerosol SSA 0.96 Single Scattering Albedo
τa SeaWiFS aerosol optical thickness, 550 nm
ɛa SeaWiFS aerosol spectral slope, 865–765 nm
Surface albedo 0.06 Jin et al. [2002]
Solar zenith Calculated function of time and position
RTE solver disort2 radiative transfer equation, Stamnes et al. [2000]
Number of streams 16
λ range 299–600 nm Δλ = 1 nm
Slit function Triangular Δλ = 10 nm

[13] Common to both modes was the use of atmospheric constituent profiles for arctic, midlatitude, and tropical atmospheres during summer and winter [Anderson et al., 1986]. The profile used was selected on grounds of season and position. Tropical atmospheres were selected for locations between 30°S and 30°N, midlatitude locations were 30°N–60°N or 30°S–60°S, and arctic locations were 60°N–90°N or 60°S–90°S. The season was selected using the equinoxes as the boundary between summer or winter depending on latitude. This was also true for the aerosol seasonal profile. The extraterrestrial solar spectrum was adjusted according to day of year to account for the elliptical nature of the Earth's orbit. For ozone, monthly zonally averaged (10° increments) climatological Total Ozone Mapping Spectrometer (TOMS) data were utilized (ftp://toms.gsfc.nasa.gov/pub/eptoms/data/zonal_means/ozone); the value was selected on the basis of position and month of the year. The aerosol type was fixed as maritime [Shettle, 1989], as was the aerosol single scattering albedo (SSA) at 0.96. The surface albedo was set at 0.06 [Jin et al., 2002]. A triangular slit function was used with a bandwidth of 10 nm.

2.2.1. Validation Mode

[14] For this mode, the model was run with aerosol optical thickness (τa) at 550 nm measurements, made using a handheld Microtops sunphotometer [Smirnov et al., 2009]. The daily median value was used as being representative of the whole day. The Ångstrom coefficient, which is the logarithmic slope of the aerosol optical thickness between two or more wavelengths (865 and 765 for the SeaWiFS aerosol correction parameter, ɛa) was also calculated. This is indicative of the aerosol size distribution and the relative radiative contribution of smaller and larger aerosols.

[15] Daily surface UV irradiance time profiles were visually inspected for clouds, and clear sky days or fraction of days selected for validation against modeled data. The data were modeled using 1 nm spectral resolution. Coincidental instantaneous spectral comparisons were made between modeled and measured. In addition, for the purposes of this work, modeled and measured Ed0+(λ) were extracted from the hyperspectral data at 305, 325, 340 and 380 nm, these being the operational wavelengths of the in-water Satlantic UV sensor.

2.2.2. Global Mode

[16] To allow the atmospheric model to be tractably run repeatedly over large global data sets, a set of four dimensional look-up tables were generated as a function of (1) day of year, (2) latitude, (3) ozone concentration, and (4) aerosol optical thickness, τa(550). The parameters used and the increments employed are shown in Table 2; this resulted in 50,616 element look-up tables. These were separated into the different wavelengths (305, 325, 340 and 380 nm) and Ed(λ) into direct, Edir(λ), and diffuse, Ediff(λ) components to facilitate ease of use when the effect of clouds and the air-sea interface are considered. A weighted 4-D interpolation scheme was implemented when using the look-up tables on the satellite data. In all eight look-up tables, each of 50,616 elements were generated (four wavelengths, direct and diffuse).

Table 2. Parameters Used in Generating the Spectral Edir(λ) and Ediff(λ) Look-up Tables From the libRadtran Model
Parameter Range Increment N
Day of year 1–361 10 37
Latitude −90–90 10 19
Ozone (DU) 100–500 50 9
τa (550) 0.01–3.2 0.01, 0.05, 0.1, 0.2, 0.4, 0.8, 1.6, 3.2 8

[17] Daily images of global subsurface irradiance, Ed0−(λ), were created using the look-up tables for the period 1998–2009 at a resolution of 1.25° × 1.0°. The atmospheric model was run at each grid point for local noon, effectively creating the maximum value encountered at each wavelength during the day. Daily ozone data was obtained from the Earth Probe TOMS (EPTOMS, 1998–2005) and Total Ozone Analysis using SBUV/2 and TOVS (TOAST, 2006–2009) data sets (http://oceandata.sci.gsfc.nasa.gov/Ancillary/Meterological) and daily cloud amounts obtained from the European Centre for Medium Range Weather Forecasting (ECMWF) operational analysis data set. Global fields of τa(550) were calculated using a combination of the SeaWiFS τa(865) and ɛa aerosol products on a seasonal climatological basis.

[18] To calculate the effect of clouds on Edir(λ) and Ediff(λ), the algorithm of Kasten and Czeplak [1980] was used, where C is the cloud cover as a decimal fraction; if C > 0.25,
equation image
These cloud modified (CM) components were then propagated through the sea surface, assuming a wind speed of 5.0 m s−1 using equations from the HYDROLIGHT (V5) radiative transfer code [Mobley and Sundman, 2008] and described by Mobley [1994].

2.2.3. Sensitivity Study

[19] The atmospheric model was tested for the sensitivity to the following, while keeping the all other parameters at their default (shown in parenthesis): solar zenith angle (45°), ozone (300 DU), surface albedo (0.06), atmospheric profile (midlatitude), season (summer), aerosol type (marine), τa(0.1), and aerosol single scattering albedo (0.96). The solar zenith angle, θz, was initially varied over the range 25°–75°, but then over smaller ranges: 10° ± 5°, 45° ± 5°, 75° ± 5°. Ozone amount was varied between 200 and 450 DU, surface albedo varied between 0.04 and 0.08, the atmospheric profile was varied for all scenarios (tropical, midlatitude, and arctic) for both seasons (summer and winter), the aerosol type was varied for all types (rural, urban, maritime, and tropospheric), τa varied between 0.01 and 3.2, and aerosol single scattering albedo varied between 0.80 and 0.99, which covers the range reported by Krotkov et al. [2005b].

2.3. Combined Model Validation

[20] The combined model (Figure 1) was tested in its entirety for the global implementation against in situ profiles of Ed(λ) from two independent data sets. These data were from (1) the Biogeochemistry and Optics South Pacific Experiment (BIOSOPE) cruise (October–December 2004) as described by Tedetti et al. [2007] and provided by M. Tedetti and R. Sempéré and (2) the Aerosol Characterization Experiment (ACE)-Asia experiment (March–April 2001) as described by Vasilkov et al. [2005] and provided by A. Vasilkov, M. Kahru, and G. Mitchell. On the BIOSOPE cruise, downward irradiance was measured at 305, 325, 340, and 380 nm using a Satlantic radiometer whereas the ACE-Asia experiment utilized a MER PRR-800 measuring downward irradiance at 313, 320, 340, 380, and 395 nm. Direct comparison with the model output then was possible for all wavelengths on the BIOSOPE cruise but only at 320 (deemed close enough to 325), 340 and 380 nm for the ACE-Asia cruise. Profiles were selected on the basis of the following criteria: (1) if profiles were taken within quick succession in the same location (i.e., within 1 h), then only one profile was selected, and (2) the stability of the coincidental above-water irradiance during the period of the optical cast or largely clear skies recorded within the metadata was considered. This resulted in 11 profiles being selected from the BIOSOPE cruise and 19 from the ACE-Asia experiment.

[21] The comparison used the daily files of Ed0−(λ), generated from the daily cloud and ozone data and seasonal climatologies of τa(550), and the global monthly composites of Kd(λ). Any time offset of the in-water profile and solar noon was accounted for by assuming a sinusoidal variation in Ed0−(λ) as a function of time and day length. In an attempt to disentangle errors from the atmospheric and in-water components, the comparison was also run in normalized mode. This involved normalizing the surface modeled value by the coincidental measured value.

2.4. Global Model Implementation

[22] Using the monthly global fields of Kd(λ), the 10% irradiance depth was calculated using
equation image
and climatologies for the period 1998–2009 were produced for each month of the year.
[23] The mean daily UV dose received within the mixed layer depth (MLD) may be expressed mathematically [Boelen et al., 2000; Miles et al., 2009; Wambeke et al., 2009] as
equation image
where DL is day length in seconds and Ed0−(λ, t) is calculated each hour assuming a sinusoidal variation in Ed0−(λ) given the noon maximum value and the day length. Hm(λ) was calculated at each pixel at an hourly time step between sunrise and sunset. Daily files of global Ed0−(λ) were used together with monthly composites of Kd(λ) to ensure the maximum number of valid pixels. The model used climatological MLD fields on a 2° × 2° grid [de Boyer Montégut et al., 2004]. The daily time integrated values are normalized by the MLD giving a spectral dose in units of kJ m−2 d−1 nm−1 [Sliney, 2007]. Climatologies for the period 1998–2009 were produced for each month of the year. Comparison with literature values of UV dose was carried out by using the monthly global climatologies and extracting the nearest pixel to the station locations reported by Boelen et al. [2000] and Wambeke et al. [2009].

3. Results

[24] Figure 2 and the associated regression statistics (Table 3) show good correlations (R2 > 0.786) between the various multispectral UV wave bands, measured using the four channel Satlantic UV-507 radiometer. The error bars shown in Figures 2b2d were determined from the standard error in regression in determining Kd from individual irradiance profiles. There is a weaker relationship between atot(443) and Kd(380) compared with the longer UV wavelength regressions (i.e., excluding 305 nm), this is possibly due to the following reasons. First, the total absorption at 443 nm is made up of contributions from phytoplankton, CDOM and suspended particulates. Once into the UV part of the spectrum it is likely that the IOPs are dominated by CDOM, excluding the wavelength-dependent contribution from pure water [Quickenden and Irvin, 1980; Pope and Fry, 1997; Morel et al., 2007]. As a possible exception to this assumption, in some locations, there may be absorption due to MAAs around 320–330 nm [Klisch and Häder, 2002]; indeed, Whitehead and Vernet [2000] report variable absorption due to MAAs and other unidentified absorbing compounds throughout the UV, and this could explain the nonlinear bio-optical relationships in Figure 2. Second, Kd is dependent upon the directional structure of the ambient irradiance field (i.e., an apparent optical property) whereas IOPs are not. Nevertheless, Figure 2 and Table 3 give confidence in extrapolating visible channel determined measurands from satellite remote sensing to model the in-water irradiance field characteristics in the UV domain. Indeed data from other published sources [Tedetti et al., 2007; Vasilkov et al., 2005] shows good agreement in Figure 2, although the range over which their data were collected is smaller than for those presented here and their data fall within the near-linear portions of the polynomial fits.

Details are in the caption following the image
In situ derived relationships between (a) atot(443) and Kd(380), (b) Kd(380) and Kd(340), (c) Kd(380) and Kd(325), and (d) Kd(380) and Kd(305). Error bars on atot(443) are the standard deviation over the surface layer down to the Ed(380) 1% irradiance level, and Kd(λ) error bars are derived from the standard error in the log linear regression to the 1% irradiance level. Statistics for the regression (solid lines) are shown in Table 3. Open squares on Figures 2b–2d are taken from Tedetti et al. [2007], and open diamonds on Figures 2b and 2c use data from Vasilkov et al. [2005]; these two data sets are not used to derive the empirical relationships shown in Table 3.
Table 3. Second-Order Polynomial Statistics for Regressions Shown in Figure 2 With Associated Standard Error, R2, and Number of Data Points (N)a
Parameters a b c R2 N Figure
atot(443) versus Kd(380) 3.313 ± 0.986 0.887 ± 0.184 0.03 ± 0.006 0.871 106 2a
Kd(380) versus Kd(340) −1.965 ± 0.188 1.979 ± 0.064 −0.009 ± 0.004 0.991 72 2b
Kd(380) versus Kd(325) −3.486 ± 0.349 2.546 ± 0.121 −0.007 ± 0.008 0.975 62 2c
Kd(380) versus Kd(305) −7.070 ± 1.739 4.424 ± 0.616 −0.008 ± 0.042 0.786 52 2d
  • a The polynomial is of the form ax2 + bx + c.

[25] Figure 3 shows a direct comparison between the modeled and measured hyperspectral Ed0+(λ). The case shown is for a single acquisition of the Trios Rameses-ACC on 13 November 2007 at 12:40 UTC at position 27.265°N, 17.054°W on the INSPIRE cruise. Experimentation showed that using a 10 nm bandwidth with a triangular slit function (Figure 3b) within the libRadtran model gave more comparable results with the measurements. This is because the atmospheric signal, when using finer spectral resolution and a rectangular slit (Figure 3a), gives large variability over small wavelength increments, especially for the range between 350 and 400 nm. The large variability in the finer spectral resolution in Figure 3a) between 370 and 400 nm is consistent with the results shown in Figures 2 and 8 of Mayer and Kylling [2005] and are attributable to Fraunhofer lines [Kirchhoff, 1860]. The modeled accuracy of Ed0+(λ) at 305, 325, 340, and 380 nm was +8%, +13%, +6%, and −2%, respectively. It is also worthwhile to note that maintaining accurate calibration for UV radiation instrumentation is a necessary requirement, especially when operating in high irradiance areas such as the tropics where UV filters will degrade more quickly. A start and end of cruise calibration is recommended with daily trends calculated to correct for any degradation of the filters. This was not possible for these data as the calibration coefficients were not made available at the time by the manufacturer; instead, the calibration was applied internally. The calibration drift was noticed by the modeled data being in good agreement with the data at the start of a cruise (i.e., just after the laboratory calibration) but progressively getting poorer as the cruise went on. This drift was synthesized by taking single clear sky hyperspectral acquisitions at the beginning and end of the cruise, calculating the offset from the modeled data at both of these fixed points, and then calculating this offset drift for each day of the cruise and correcting the measurements on the basis of this.

Details are in the caption following the image
Comparison between Trios Rameses-ACC UV sensor hyperspectral surface measured (solid line) and modeled (dashed) irradiance (Ed0+(λ)) using libRadtran: (a) 2.5 nm bandwidth and rectangular slit function and (b) 10 nm bandwidth and triangular slit function. Measurement taken at 12:40 UTC on 13 November 2007 at position 27.265°N, 17.054°W.

[26] Figure 4 shows the intercomparison between the in situ Trios Rameses-ACC measured (corrected using the method described above, and extracted at the wavelengths of interest) and modeled multispectral Ed0+(λ). Only clear sky cases have been included in this comparison, although it cannot be ruled out that some cloud contamination may exist from sources such as thin cirriform cloud. Table 4 shows the regression statistics to be good, with the root-mean-square error (RMSE) expressed as a percentage being less than 14% at all wavelengths.

Details are in the caption following the image
Intercomparison between modeled and measured surface irradiance (Ed0+(λ)) at 305, 325, 340, and 380 nm. Multispectral values are extracted from hyperspectral measurements taken with a Trios Rameses-ACC UV sensor.
Table 4. Regression Statistics for a Multispectral Intercomparison Between Measured and Modeled Ed0+(λ)a
305 325 340 380
Slope 1.02 0.94 0.94 0.92
Intercept −1.74 27.33 39.08 55.9
R2 0.94 0.92 0.92 0.90
Percent RMSE 13.9% 10.2% 10.3% 11.9%
N 2925 2925 2925 2925

[27] Table 5 shows that the atmospheric model is most sensitive to changes in solar zenith angle (θz), aerosol optical thickness (τa) and ozone concentration. The sensitivity to atmospheric profile, season, aerosol type, single scattering albedo and ground (sea surface) albedo are relatively low and can be considered second-order effects. The effect of O3 concentration on Ed0+(λ) is obviously spectrally dependent with the greatest sensitivity at the shorter wavelengths of 305 and 325 nm. The sensitivity to θz, shown here for 45° ± 5°, shows the importance of accurate calculation based on known position, time, and date information. Indeed for higher solar zenith angles (∼70°) the sensitivity is much higher, with a ±85% variation at 305 nm when θz is varied between 65° and 75°. The model is also sensitive to aerosol loadings with a ±22% variation in Ed0+(λ) when τa(550) is varied over a realistic range.

Table 5. Sensitivity of the Atmospheric Model to Various Input Parametersa
λ θz O3 GA τa(550) Atmosphere Season Type SSA
305 ±30.3% ±64.0% ±0.5% ±22.6% ±2% ±0.2% ±0.5% ±2.0%
325 ±13.3% ±4.1% ±0.6% ±22.1% ±0.3% ±0.2% ±0.3% ±2.1%
340 ±12.1% ±0.4% ±0.5% ±22.2% ±0.1% ±0.2% ±0.3% ±2.0%
380 ±11.2% ±0% ±0.4% ±22.3% ±0.1% ±0.1% ±0.3% ±1.7%
  • a Values are expressed as a percentage variation in Ed0+(λ). GA is the ground albedo, and SSA is the single scattering albedo.

[28] Figure 5 shows an intercomparison between a single in-water irradiance profile and the global implementation of the model which used the daily global Ed0−(λ) data for that day together with that month's (October 2004) global atot(443) map to calculate Kd(λ). There is good agreement between the measured and modeled data, with the logarithmic root-mean-square (RMS) down to the spectrally varying 10% irradiance level being 0.164, 0.0629, 0.042, and 0.041 at 305, 325, 340, and 380 nm, respectively. Table 6 shows the log RMS difference between 30 in-water irradiance profiles and global implementation modeled values of Ed(λ, z). There are a larger number of data points (1 m depth bins) for longer wavelengths as the statistics are calculated down to the spectrally varying 10% irradiance depth. By normalizing the modeled profile at the surface to the measured Ed0−(λ) values (effectively shifting the surface intercept and keeping the slope fixed), closer agreement is found between the in situ and modeled data. This allows the errors in the model due to the atmospheric component and in-water component to be separated. The log RMS is then reduced to around 0.17 at all wavelengths, this being a metric of how accurately the global 18 km satellite derived monthly averaged fields of Kd(λUV) reproduce the in situ point measurements.

Details are in the caption following the image
Comparison between modeled (solid line) and measured (asterisks) Ed(λ, z) for 305, 325, 340, and 380 nm for 23:11 UTC on 29 October 2004 at 8.319°S, 141.23°W.
Table 6. Logarithmic RMS Difference Between Measured and Modeled Spectral Ed(λ,z) Down to the 10% Irradiance Levela
Cruise 305 nm 325 nm 340 nm 380 nm
BIOSOPE 0.222(137) 0.149(237) 0.154(312) 0.189(521)
BIOSOPE, normalized 0.121(137) 0.122(237) 0.136(312) 0.167(521)
ACE-Asia - 0.297(174) 0.344(210) 0.354(337)
ACE-Asia, normalized - 0.155(174) 0.162(210) 0.177(337)
  • a The number of data points is the total included within the RMS statistics (1 m bin intervals). BIOSOPE data [Tedetti et al., 2007] comprise 11 profiles; ACE-Asia [Vasilkov et al., 2005] comprises 19 profiles.

[29] Figure 6 clearly shows spatial variability in the amount of UV radiation that penetrates the pelagic zone. This variability is spectral as well as spatial, with the shorter (and as a consequence most harmful) UV irradiance being more strongly attenuated over relatively shallow depths. The 10% irradiance level at all wavelengths is at a maximum depth in the oceanic gyres (∼18, 32, 44, and 70 m at 305, 325, 340 and 380 nm, respectively) [Tedetti et al., 2007], whereas on the continental shelf, upwelling and coastal regions, irradiance at all wavelengths in the UV domain is extinguished within the top 0–5 m of the water column. The increased biological activity of the equatorial upwelling regions in the Atlantic, and particularly the Pacific, reduces the 10% irradiance level in these regions to around 30–40 m (at 380 nm) in comparison with 60–70 m in the gyre regions to the north and south. April marks the start of increased biological activity in the North Atlantic, and this shoals the penetration depth of UV irradiance in this region to less than 20 m at all wavelengths. The southern Atlantic and Indian Oceans both have lower UV irradiance penetration depths (10–15 m at 325 nm) and this could be caused by a combination of higher biological activity and the input of terrigenous CDOM from the Patagonian Shelf which is then entrained and flows unimpeded around the oceanic regions below 40°S.

Details are in the caption following the image
Climatologically averaged (over the period 1998–2009) 10% irradiance level depth for April at (a) 305 nm, (b) 325 nm, (c) 340 nm, and (d) 380 nm. Grid lines represent 30° increments in latitude and 60° increments in longitude centered on 0°N, 0°W.

[30] Figures 7 and 8 show the combined effect of introducing the modeled atmospheric UV irradiance forcing together with a measure of the MLD to the in-water optical signature determined by Kd(λ). The strongest atmospheric UV irradiances are encountered in the clear sky tropical and equatorial regions. When this is combined with a relatively shallow MLD, such as off the Californian coast, the Sargasso and the Mediterranean Seas in July and the northern Indian Ocean in April, the highest surface mixed layer UV doses occur. Figure 8 shows that the doses at 305 nm are in the range 0.3–0.5 kJ m−2 d−1 nm−1 in these areas; Figure 7 shows that at the longer wavelengths of 325, 340, and 380 nm the doses are more than an order of magnitude larger with values in the range 8–10 kJ m−2 d−1 nm−1. The dose at 305 nm is an order of magnitude less than the other wavelengths as a result of lower surface irradiances and higher values of Kd(λ). This is shown clearly in Figures 3, 4, and 5. Figure 8 shows that appreciable amounts of the most harmful UVB radiation (280–320 nm) are restricted to the tropical regions. Outside of this zone, the low incident surface irradiances and, to a second-order, relatively deep MLD, restrict short wavelength UV irradiance intensity. The effect of a deep mixing can clearly be seen in the Southern Ocean [Dong et al., 2008] and the North Atlantic where the MLD doses are in the range 0–2 kJ m−2 d−1 nm−1. In these regions the effect of increased biological activity during the spring and summer months will also serve to restrict the UV doses by increasing Kd.

Details are in the caption following the image
Climatologically averaged (over the period 1998–2009) mixed layer depth UV doses for July at (a) 305 nm, (b) 325 nm, (c) 340 nm, and (d) 380 nm. Units are kJ m−2 d−1 nm−1, and grid lines represent 30° increments in latitude and 60° increments in longitude centered on 0°N, 0°W.
Details are in the caption following the image
Climatologically averaged (over the period 1998–2009) mixed layer depth UV doses at 305 nm for (a) January, (b) April, (c) July, and (d) October. Units are kJ m−2 d−1 nm−1, and grid lines represent 30° increments in latitude and 60° increments in longitude centered on 0°N, 0°W.

[31] The climatologies presented here compare well with literature values of UV dose. Boelen et al. [2000] calculated mean daily biological effective doses for a series of stations in the Tropical north Atlantic in August between 0.12 and 0.25 kJ m−2 d−1. The global climatological data for Hm(305) extracted at the station locations yielded a range of 0.17–0.27 kJ m−2 d−1 nm−1. The coastal stations around the Island of Curaçao reported by Boelen et al. [2000] ranged between 0.06 and 0.57 kJ m−2 d−1, whereas the global climatological figures ranged between 0.01 and 0.09 kJ m−2 d−1 nm−1. This is probably due to the global model poorly representing optical water types in coastal areas (18 km resolution) and the MLD being at a coarse (2° × 2°) resolution. Wambeke et al. [2009] reported Hm(305) and Hm(380) for the BIOSOPE cruise during October and November 2004: Hm(305) varied between 0.11 and 0.30 kJ m−2 d−1 nm−1 which compares with 0.06–0.24 kJ m−2 d−1 nm−1 calculated for the global climatologies; Hm(380) varied between 4.2 and 11 kJ m−2 d−1 nm−1, which is higher than the global climatology value of between 2.3 and 6.6 kJ m−2 d−1 nm−1.

4. Discussion and Conclusions

[32] The combined ocean-atmosphere model presented in this paper depends upon empirical IOP to Kd(380) and then Kd(380) to Kd(λ) pair relationships to propagate the spectral irradiance field through the water column. These relationships are quadratic in form and therefore this raises the possibility of larger errors being introduced at higher values of Kd, especially at the shorter UV wavelengths. The quadratic form is statistically justifiable in terms R2: the linear regression between Kd(305) and Kd(380) yields a value of 0.716 in comparison with 0.786 for the higher-order polynomial. Using a monthly mean Kd(λ) derived from the SeaWiFS IOP data set is also likely to lead to errors, especially in highly dynamic regions of the global oceans such as the upwelling zones or those affected by intense, short-duration phytoplankton blooms. The errors here are likely to be greater in comparison with the center of the oligotrophic gyres where changes in biological activity are low and operate over longer time scales.

[33] The handling of aerosols within the atmospheric model are necessarily simplistic and rely upon the aerosol optical thickness at 550 nm, which is derived from SeaWiFS data when the model is run in global mode. The single scattering albedo has been assumed to be spectrally flat and fixed to a value of 0.96; Krotkov et al. [2005b] show that this may be lower than this at short UV wavelengths (0.92), but the UV radiation model is not particularly sensitive to this parameter (±2%). Possible errors will occur when there are dust outbreaks from, e.g., the Sahara desert or episodes of industrial pollution from urban areas. The radiative characteristics of the aerosols will be spectrally changed in these instances, reducing the values of the surface UV irradiance. As the UV irradiance model is particularly sensitive to the effects of aerosol optical thickness (±22%) the vexed issue of extrapolating aerosol optical properties from the near infrared to UV wavelengths will need addressing [Krotkov et al., 2005a]. The sensitivity of the atmospheric model to the solar zenith angle is a source of error when considering monthly mean values, especially at the middle to high latitudes during the periods of greatest change (i.e., around the equinoxes).

[34] The sensitivity of the atmospheric model to using different parameterizations for the effects of clouds was not carried out. The libRadTran model [Mayer and Kylling, 2005] contains several parameterizations for different cloud types (ice or water), however the disconnect between this and the global modeling developed here, is the ability to realistically predict cloud level and type in a tractable manner from readily available data sets. The Kasten and Czeplak [1980] model is similar in form to other empirically derived cloud cover correction functions. Kim and Hofmann [2005], in a comparison between such algorithms, highlight that these simplistic, two dimensional approaches are unable to capture the three dimensional morphology of clouds. Indeed using even daily averaged cloud amounts fails to reproduce subhourly variability in the radiative characteristics of the sky. Cloud cover can also alter the spectral properties of the wavelength-dependent downwelling irradiance [Siegel et al., 1999]. The complexity of the issue of clouds was therefore simplified by using a readily available algorithm that is implemented as standard within the HYDROLIGHT [Mobley and Sundman, 2008] hydrological optics model.

[35] The inclusion of the MLD within the combined ocean-atmosphere model together with the diurnal variation in the incident irradiance field at all depths is an attempt at determining the daily dose of UV irradiance through the water column. This necessarily gives a bulk daily integrated dose, whereas the more likely impact on individual biological organisms or on water parcels and their attendant photochemical properties will be more subtle. This will depend on where in the water column the individuals are at any given moment during the day and on the vertical mixing rates. If photoinhibition effects are under investigation, then time spent in the layer where the irradiance levels are of sufficient magnitude will be required.

[36] Photochemical rates are dependent upon the spectral irradiance penetrating to a particular depth [Martino et al., 2006], which is readily available from this model, together with values of the wavelength-dependent absolute [Martino et al., 2006] or apparent [Fichot and Miller, 2010] quantum yield. To calculate fluxes on a global or regional scale, as has already been demonstrated by Fichot and Miller [2010] then assumptions need to be made about the concentrations of the photoreactive substances. Fichot and Miller [2010] showed some success in doing this for carbon monoxide (CO) on the global scale, by invoking relationships with CDOM the bio-optical properties of which may be determined from satellite data. However, only sparse data sets exist for other species such as diiodomethane, methyl iodide and dimethyl sulphate which are thought to play an important role in the chemistry of the marine boundary layer. Also, no known link with satellite derived products has yet been forthcoming, unlike the case for CDOM and CO.

[37] The results from this model have served to highlight the regions of the highest potential photochemical or photoinhibitory impacts based on bio-optical and physical factors. These are in the Sargasso and Mediterranean Seas, the eastern Equatorial Pacific, the northern Patagonian Shelf, the northern Indian Ocean, and a latitude band in the southern hemisphere between 20°S and 35°S, with seasonal variability superimposed upon this. Not only has this been qualitatively described in terms of geography but quantitative values have been attached to the bulk daily integrated UV doses. The next step is to couple this approach to photochemical and photobiological models to determine rates of production or destruction of climatically important compounds.

Acknowledgments

[38] This work was funded by U.K. SOLAS INSPIRE (reference NE/D006511/1), U.K. SOLAS ICON (NE/C517176/1), and the NERC Oceans 2025 Atlantic Meridional Transect and is AMT paper 205. The ECMWF data were provided by the British Atmospheric Data Centre. T.S. particularly thanks Marc Tedetti and Richard Sempéré [Tedetti et al., 2007] and Alexander Vasilkov, Greg Mitchell, and Mati Kahru [Vasilkov et al., 2005] for providing UV irradiance profiles from their respective cruises. T.S. is partly funded by the National Centre for Earth Observation (NCEO).