Coral Reefs

, Volume 23, Issue 1, pp 39–47

Use of SeaWiFS ocean color data to estimate neritic sediment mass transport from carbonate platforms for two hurricane-forced events

Authors

    • NASA Goddard Space Flight Center, Code 902
  • Alexander Vasilkov
    • Science Systems and Applications Inc.
  • Denis Nadeau
    • NASA Goddard Space Flight Center, Code 902
  • Norman Kuring
    • NASA Goddard Space Flight Center, SeaWiFS Project, Code 970.2
Report

DOI: 10.1007/s00338-003-0355-9

Cite this article as:
Acker, J.G., Vasilkov, A., Nadeau, D. et al. Coral Reefs (2004) 23: 39. doi:10.1007/s00338-003-0355-9

Abstract

Plumes of neritic sediment caused by the passage of Hurricane Gert near Bermuda in 1999, and by the passage of Hurricane Michelle over Cuba’s Gulf of Batabano in 2001, were observed by the Sea-viewing Wide Field-of-view Sensor (SeaWiFS). The mass of sediments in each of these plumes, which consist largely of neritic carbonate particles, was estimated using an algorithm for the calculation of suspended sediment concentrations. The Bermuda and Batabano plumes transported 0.22 and 1.2–1.35 million kg of sediment, respectively. The algorithm results were compared with the results from two other sediment mass algorithms and proved to be consistent. These results indicate the potential use of remote sensing to estimate carbonate flux from coral reefs and banks and atolls as an augmentation to in situ studies. In addition, the use of remote sensing data may improve estimates of the annual global carbonate sediment flux, a quantity important to models of global carbonate production and the global carbon cycle.

Keywords

HurricanesNeritic carbonatesRemote sensingSediment transport

Introduction

The two environments in the global ocean that have the largest production rates of biogenic calcium carbonate (CaCO3) are coral reefs and shallow carbonate platforms (Milliman 1993; Milliman and Droxler 1996). Quantitative estimates of the mass of CaCO3 that is transported to the deep ocean from these environments remain uncertain by factors of up to 100%. Due to the increased solubility (in seawater) of the high magnesian (high-Mg) calcite produced in these environments, these fine-grained carbonate sands and muds may play an important role in the global carbonate cycle (Sabine and Mackenzie 1995).

The report of the 2001 US Joint Global Ocean Flux Study (JGOFS) Workshop on Marine Calcification (Iglesias-Rodriguez et al. 2001, 2002) indicated that uncertainties in carbonate production, flux, and accumulation are greatest for carbonate shelves and slopes. One of the key unknowns is the mass of carbonate sediment derived from neritic export. The report recommended measuring CaCO3 fluxes from shelf to open ocean, “combining satellite, profiling-float, and sediment-trap measurements.” The need for this measurement is the fact that this flux is deemed significant to both carbon burial and the overall alkalinity flux in the deep ocean.

Both production rates and off-bank flux estimates for shallow neritic carbonate sediments are highly uncertain. One reason for this degree of uncertainty is that the mechanisms that induce large sediment-mass transport events are variably episodic. The primary mechanisms likely to induce sediment transport are meteorological, either wind-driven transport (potentially augmented by tidal outflows) due to strong storms (Hine et al. 1981), or offshore density flows due to a density differential created by the passage of a winter cold front (Wilson and Roberts 1992).

Milliman (1993) provided a comprehensive evaluation of carbonate production and flux with a substantial discussion on reef and bank production and transport. He estimated reef production at 1,500 g CaCO3 m-2 year–1 for reefs and 500 g CaCO3 m-2 year–1 for banks and lagoons. These values represent a total annual global production rate of 0.9 gigatons (Gt). Milliman indicates that a reasonable estimate of offshore transport is 10% of production, though Hubbard et al. (1990) estimated that 25% of reef neritic carbonate production was exported, and Wilber et al. (1990) indicate that up to 50% of the carbonate sediment produced on Grand Bahama Bank is transported to deeper waters. Thus, while banks are significantly less productive than reefs, more of their carbonate production may be transported offshore due to the mobility of fine-grained lagoon and bank sediments.

Using a model of carbonate production and a new map of the area and depth of isolated low-latitude banks and reefs, Vecsei (2001) estimated a total annual production from these environments of 0.2 Gt CaCO3 per year. He indicated that this production may contribute to a major carbonate flux to the pelagic ocean.

Several studies have measured mass fluxes from reefs and bank margins and have formulated carbonate production and export budgets for these areas (Hubbard et al. 1990; Delesalle et al. 1998; Schrimm et al. 2002; Ouillon et al. 2003). These studies may be for only a portion of the reef environment, such as a reef pass, or for a single reef/atoll system. Even Sabine and Mackenzie (1995), who estimated carbonate sediment transport rate for the Hawaiian Archipelago, derived their results from sampling at a few selected sites within the entire archipelago. Due partly to the difficulty of sampling, and also the problem of performing synoptic observations on remote coral archipelagos and large reef and bank systems, large-scale quantification will be difficult to achieve. Remote sensing provides a potential additional source of data for large-scale sediment transport estimates.

However, the use of visible remote sensing data to quantify suspended sediment mass is a difficult research goal. The problem is complicated by the high amount of variability in the optical properties and mineralogical composition of near-shore oceanic sediments. Most sediment sources are riverine, and sediments from these sources are frequently mixed with other substances, such as colored dissolved organic matter (CDOM), which complicate analysis of the data.

Neritic carbonate sediments derived from carbonate banks, however, provide a somewhat simpler optical situation. These sediments are usually highly reflective, and because they are generated in island environments, they are minimally influenced by terrigenous materials, either erosion-derived sediments or organic matter (particulate or dissolved). Also, because these sediments are quite homogenous, an estimate of the sediment mass is approximately equivalent to the mass of CaCO3 that is being exported. Thus, quantification of “large” neritic carbonate sediment transport events from carbonate banks is feasible, and this data would substantially assist the estimation of total carbonate flux from these environments.

The primary drawback to analysis and quantification of these events is their episodic nature, which severely limits attempts to validate (“sea-truth”) the results of analytical algorithms applied to remote sensing data. Sediment plumes generated by storms consist of rapidly settling particles, and because the storm effect on a given region is usually short-lived, the wind forcing that creates the features is also short-lived. Therefore, observation of these ephemeral features by visible range remote sensing is fortuitous and dependent on favorable satellite overpass timing and rapid cessation of cloud cover in the vicinity. However, smaller plumes that are recurrent may provide a validation pathway if remote-sensing data with resolutions greater than 1 km can be obtained (see the Discussion).

This paper presents the results of applying an algorithm for sediment mass quantification to two events: the passage of Hurricane Gert near Bermuda, in 1999, and the passage of Hurricane Michelle over the Gulf of Batabano adjacent to the island of Cuba, in 2001. The Bermuda event provided one of the first indications that satellite remote sensing could be used to observe the transport of neritic sediments from carbonate platforms (Acker et al. 2002). The Gulf of Batabano event produced a remarkably large sediment plume that was observed for several days after the passage of Hurricane Michelle. In fact, the Gulf of Batabano plume is likely the largest neritic carbonate sediment transport event ever observed by remote sensing.

Materials and methods

Study areas

The island of Bermuda (32 N, 64 W) and the surrounding shallow carbonate bank are considered the northernmost coral reef/carbonate environment in the Atlantic Ocean due to the warm waters of the Gulf Stream and Sargasso Sea. Bermuda is commonly affected by the passage of hurricanes during the Atlantic hurricane season from July to November. The depth of Bermuda’s large North Lagoon ranges from 10–20 m. Although the land area of the island of Bermuda is only about 60 km2, the lagoon and reef complex is approximately 750 km2 in size.

The Gulf of Batabano (Golfo de Batabano) lies south of the western end of the island of Cuba, and is partly bounded on the southwest by the Island of Youth (Isla de la Juventud). Shelf islands form a broken chain eastward from the Island of Youth to the main island of Cuba. The Gulf of Batabano is a shallow carbonate bank similar to the Bahamas Banks, but subject to more terrigenous influence due to its nearness to the main island of Cuba. The Gulf of Batabano is approximately 18,000 km2 in size. Due to its central location in the Caribbean Sea, the island of Cuba is also frequently affected by hurricanes, which form in both the equatorial Atlantic Ocean and in the Caribbean Sea.

Ocean color data processing

SeaWiFS images of the two sediment transport events analyzed in this paper were acquired by the High Resolution Picture Transmission (HRPT) stations located at the Bermuda Biological Station for Research (station HBBS) and the University of South Florida (station HUSF). The Orbview-2 satellite, which carries SeaWiFS, broadcasts direct downlink data at 1-km resolution to all HRPT stations. These data files are then sent to the SeaWiFS Project at NASA Goddard Space Flight Center (NASA GSFC) and subsequently to the Goddard Earth Sciences Distributed Active Archive Center (GES DAAC) for archive and distribution to SeaWiFS Authorized Research Users. The data files for the two events were acquired from the GES DAAC. All HRPT station data files are level 1A, consisting of geolocated raw radiance data for the eight SeaWiFS visible bands.

The level 1A data files were processed to level 2 geophysical products, which include normalized water-leaving radiances, using the SeaWiFS Data Analysis System (SeaDAS) version 4.2. Default atmospheric correction parameters were employed for this processing. Subsequent to this processing, the data were analyzed by a SeaDAS extension created to generate suspended sediment concentration according to the algorithm of Vasilkov (1997). The algorithm is described below.

Suspended sediment concentration algorithm

All algorithms for retrieval of seawater constituents from spectral reflectance can be divided into two groups. The first are referred to as empirical algorithms. They are based on empirical correlation between reflectance or radiance band ratios and water constituent concentrations. However, these simple empirical algorithms are not reliable for case 2 coastal waters. In such cases, the second group, so-called analytical algorithms, are more promising. In analytical algorithms, inherent optical properties (IOPs) of seawater are derived from inversion of the reflectance model. The non-linear function minimization of the spectral difference between the modeled reflectance and the in situ measured reflectance has been employed in several studies (Burenkov et al. 1995; Lee et al. 1994; Roesler and Perry 1995; Garver and Siegel 1997). The minimization of a non-linear function of several variables may be computationally expensive and, therefore, cannot be used for operational purposes. An alternative approach is based on a direct inversion technique. The radiance model is transformed into a set of equations with unknowns related to water constituent concentrations. Non-linear equations are solved in a manner similar to Carder et al. (1991).

A relationship between the normalized water-leaving radiance and diffuse reflectance is given in Hoge and Lyon (1996). The water-leaving radiance is expressed through the reflectance (neglecting a small term of 0.47R in comparison with unity):
$$ L_{w} {\left( \lambda \right)} = F_{0} {\left( \lambda \right)}t_{\lambda } {\left( {\theta _{0} } \right)}\cos _{0} MR{\left( \lambda \right)}/Q{\left( \lambda \right)} $$
(1)
where λ is the wavelength; F0 is the extraterrestrial solar irradiance; t is the diffuse transmittance of the atmosphere; M=(1–rE) (1–rL) / m2, rE and rL are the Fresnel reflection coefficients of the sea surface for downward irradiance and upward radiance, respectively; m is the index of refraction of seawater; and θ0 is solar zenith angle (SZA). From this equation, it follows [Lw]N=F0 M (R/Q). We assume (1–rL)=0.98, i.e., we neglect the very slight dependence on zenith viewing angle. The water-surface transmittance (1–rE), depends on wavelength, SZA, aerosol loading, and wind speed. However, this dependence is weak (Baker and Smith 1990), so we assume the water-surface transmittance is constant for our conditions (1–rE)=0.96.

The least-squares technique to solve an over-determined system of linear equations was used by both Sugihara et al. (1985) and Vasilkov (1997). The algorithm utilized in this paper is based on the least-squares technique. An important feature of this algorithm is the option to vary the parameters describing the spectral backscatter of suspended particulate matter (SPM) and the spectral absorption of dissolved organic matter (DOM) when inverting the reflectance model.

Many approaches to obtain an approximate solution to the radiative transfer equation for seawater provide roughly similar dependence of the reflectance on the IOPs, notably the absorption coefficient, a, and the backscattering coefficient, bb. This dependence can be expressed in a general form R=f(X), where R is the diffuse reflectance just beneath the sea surface, and
$$ X = b_{b} /{\left( {a + b_{b} } \right)} $$
(2)
A least-squares technique to retrieve the IOPs from the reflectance model (Vasilkov, 1997) requires that the equation R=f(X) should be analytically solvable with respect to X. Following Hoge and Lyon (1996), the reflectance model of Gordon et al. (1988) was chosen. According to this model the ocean reflectance R can be directly related to X by
$$ R/Q = l_{1} X + l_{2} X^{2} $$
(3)
where the Q factor is the ratio of the upwelling radiance to the upwelling irradiance toward zenith, l1=0.0949 and l2=0.0794 are constants.
The total IOPs are the sums of the IOPs of the pure seawater and the three major scattering and absorbing water substances:
$$ b_{b} {\left( \lambda \right)} = b_{{bw}} {\left( \lambda \right)} + b_{{bp}} {\left( \lambda \right)}; \quad a{\left( \lambda \right)} = a_{w} {\left( \lambda \right)} + a_{{ph}} {\left( \lambda \right)} + a_{{dom}} {\left( \lambda \right)} $$
(4)
where subscripts w, p, ph, and dom denote the pure seawater, the particulate matter, the phytoplankton pigments, and the dissolved organic matter (DOM), respectively. The detritus absorption is included in the DOM absorption because of its approximately identical spectral dependence (Carder et al. 1991). The pure seawater IOPs were obtained from Pope and Fry (1997). Phytoplankton backscatter is included in the particulate matter backscattering coefficient, and is much lower than particulate matter backscatter because the index of refraction of living phytoplankton is close to that of seawater.
The SPM backscattering coefficient and the DOM absorption coefficient are accepted in the conventional form:
$$ b_{{bp}} {\left( \lambda \right)} = b_{0} {\left( {\lambda _{0} /\lambda } \right)}^{n} ; \quad a_{{dom}} {\left( \lambda \right)} = a_{0} \exp {\left[ { - S{\left( {\lambda /\lambda _{0} } \right)}} \right]} $$
(5)
where n is the backscatter wavelength ratio exponent and S is the DOM spectral slope. The parameters S and n are allowed to vary when inverting the reflectance model.
The phytoplankton absorption coefficient is expressed through the normalized absorption spectrum depending on chlorophyll concentration:
$$ a_{{ph}} {\left( \lambda \right)} = a_{{ph0}} a^{*} _{{ph}} {\left( {\lambda ,C_{{chl}} } \right)} $$
(6)

Dependence of the normalized absorption spectrum on chlorophyll concentration accounts for an increasing package effect from oligotrophic to eutrophic waters and a possible effect of changing the pigment composition.

Equation (2) can then be rewritten as a linear expression with respect to the three unknowns: b0, aph0, and a0. Using this expression for each of the instrument band wavelengths λi, i=1,2,...,N obtains an over-determined system of linear equations for the unknown IOPs: Ax=B, where A is the N×3 matrix, x is a 3-dimensional vector {b0, aph0, a0}T, and B is an N-dimensional vector.

A linear least-squares technique is used to solve the set of equations [Eq. (4)] for the predefined pair of the parameters S and n. The parameters are allowed to vary by step increments in the range determined from available data specific for the coastal area concerned. The set of equations is solved for each pair of S and n. The pair of parameters providing the minimum standard error of the least-squares inversion, together with the respective IOPs, are taken as the final solution. One of the main advantages of this algorithm is the capability to vary S and n, rather than relying on a single set of predefined values for these variables.

In the case of the normalized phytoplankton absorption coefficient dependent on the chlorophyll concentration, an iteration technique is used. The normalized phytoplankton absorption coefficient is calculated for an initial value of chlorophyll concentration. Then, by solving the system of equations, the new value of the chlorophyll concentration is found. The iteration convergence is normally fast, usually requiring only a few iteration steps.

Sediment concentration P is then calculated using the generated value of the backscattering coefficient at 555 nm, bb(555), using the regression equation
$$ {\text{b}}_{{\text{b}}} {\left( {555} \right)}{\left[ {{\text{m}}^{{ - 1}} } \right]} = 0.015{\text{ P}}{\left[ {{\text{g/m}}^{3} } \right]} $$
(7)
which was determined by analysis of spectral reflectance data for North Sea coastal waters (Althuis et al. 1996). Note that the value of the coefficient (0.015) is dependent on particle size and can be varied according to sediment characteristics.

Note also that although Eq. (7) was obtained for riverine sediments, this semi-analytical algorithm should be broadly applicable because it uses data from all of the SeaWiFS bands to determine the IOPs. For case 1 waters, it was found that a simple relationship b(550)=P, where b(550) is the sediment scattering coefficient at 550 nm, is a good approximation (Gordon and Morel 1983). Morel (1988) suggested a parameterization of the sediment-backscattering coefficient as a function of phytoplankton pigment concentration in which the coefficient of proportionality varies from 0.02 m2/g for eutrophic waters to 0.002 m2/g for oligotrophic waters. For typical pigment concentration of 0.1 mg/m3 in case 1 waters, the proportionality coefficient is 0.017 m2/g. For case 2 waters, the value b*=0.015 m2/g is recommended by both Tassan (1994) and Althuis et al. (1996). This value is very close to the above value for typical case 1 waters. We chose the value b*=0.015 m2/g for our study as reasonably applicable for both open ocean and coastal waters. We believe this choice minimizes uncertainty in our estimates of sediment transport.

Event observation chronologies

Hurricane Gert/Bermuda

Hurricane Gert formed as a tropical depression west of Africa on 11 September 1999 and reached hurricane strength on 13 September. The storm moved westward and then curved northward. On 21 September, the center of Gert came within 300 km of Bermuda, subjecting the island to 110-knot winds and causing extensive beach erosion.

On 22 September, SeaWiFS viewed Bermuda from a near-nadir position. The true-color image showed a high reflectance feature originating on the platform and extending westward into a small eddy. This feature is clearly visible in an image of normalized water-leaving radiance at 555 nm [nLw(555)] (Fig. 1). The feature is located at approximately 32.2°N, and extends from 65.4°W to 65.8°W. The coincidence of Gert’s passage with the appearance of this feature, combined with reports of extensive beach erosion, strongly suggested that this feature was a plume of beach- and bank-derived sediments from the Bermuda Platform, caused by the increased turbulence and mixing induced by Gert’s high winds.
Fig. 1

SeaWiFS nLw(555) grayscale image of Bermuda and the adjacent Atlantic Ocean acquired on 22 September 1999, showing a plume of bank-derived neritic sediments extending southeastward from the Bermuda platform. The island of Bermuda and the shallow waters of the lagoon and reef system are covered by the high-radiance mask created by SeaDAS processing of the SeaWiFS level 1A file to level 2

Hurricane Michelle/Cuba

Hurricane Michelle originated as a tropical depression near Nicaragua on 29 October 2001. It intensified to tropical storm strength on 1 November, and began a slow northward drift. Michelle became a hurricane on 2 November, and wind speed intensified to 115 knots on 3 November. As Michelle moved northeastward on 4 November, wind speed increased to 120 knots, making Michelle a Category 4 hurricane (Saffir–Simpson scale) as it crossed the Gulf of Batabano and the island of Cuba. Michelle weakened considerably during the destructive passage over Cuba and moved northeastward through the Bahamian archipelago.

SeaWiFS provided observations of Cuba and the Gulf of Batabano during the passage of Michelle and on subsequent days. A SeaWiFS image acquired on 4 November 1999 shows the eye of Michelle situated directly over the Gulf of Batabano. On the following day, 5 November, a very bright plume of sediment was observed south of the Gulf, and the shallow waters within the Gulf were a blue-white color, indicating very high turbidity. Observations of the Gulf of Batabano region from 6–9 November followed the evolution of the sediment plume as it extended southwestward, eventually reaching its maximum extent on 9 November. Figure 2a, b shows the nLw(555) images of the Gulf of Batabano region and the Caribbean Sea on 5 and 9 November, respectively.
Fig. 2 a

SeaWiFS nLw(555) image of western Cuba, the Gulf of Batabano, and the Caribbean Sea acquired on 5 November 2001. The green arrows indicate the leading edges of the bilobate sediment plume caused by the passage of Hurricane Michelle. White areas in the Gulf of Batabano (to the north and west of the Island of Youth, the large comma-shaped island) indicate turbidity and shallow water. Some of these areas were covered by the high-radiance mask during data processing, because nLw(555) radiances for the turbid waters of the Gulf of Batabano and the incipient plume were both very high on 5 November. b SeaWiFS nLw(555) image of the same region, acquired on 9 November 2001. The plume of sediments has now extended a considerable distance to the southwest. In both images, light gray and white areas on the coast of Mexico and Central America also indicate either shallow water or turbidity

Results

The SeaWiFS data file for 22 September 1999 from station HBBS was processed to level 2 using SeaDAS 4.2 and then analyzed by the suspended sediment mass algorithm, generating an image of suspended sediment concentration. The area of the sediment plume was then approximated using a polygonal blotch over the pixels in which sediments were present. The mean suspended sediment concentration in the plume was multiplied by the volume of water contained in the top meter of the area of the plume (a critical assumption) to obtain an initial total sediment mass estimate. The concentration of sediment in the plume ranged from 0.42 g/m3 near the shore to minimum values of 0.24 g/m3 offshore. This calculation yields an estimate of 2.2×108 g (0.22 million kg) of sediment in the plume (Fig. 3).
Fig. 3

Suspended sediment concentration image of Bermuda and the adjacent Atlantic Ocean. Sediment concentrations are expressed in mg/l, which is equivalent to g/m3

The same procedure was applied to SeaWiFS data files from station HUSF acquired on 5 and 9 November 2001. For the 5 November sediment plume, observed the day after the passage of Hurricane Michelle, the mean sediment concentration was 0.71 g/m3, and the estimated total sediment mass was 1.35×109 g (1.35 million kg) of sediment (Fig. 4a). For the sediment plume observed on 9 November, four separate areas were analyzed, with mean sediment concentrations ranging from 0.50 to 0.37 g/m3. The estimated total mass of sediment in the 9 November plume was 1.2×109 g (1.2 million kg) of sediment (Fig. 4b). The similarity of the estimated sediment mass in the plume for 5 and 9 November has significant implications for the dynamics of sediment transport during this event.
Fig. 4 a

Suspended sediment concentration image of western Cuba, the Gulf of Batabano, and the Caribbean Sea on 5 November 2001. Note that the color palette is slightly different than for Fig. 3. b Suspended sediment concentration image for the same region on 9 November 2001

Discussion

Although the waters of the Bermuda reef/lagoon complex are shallow, the adjacent waters are quite deep, which minimizes any potential bottom reflectance effects. Acker et al. (1997) illustrated the coastal proximity of the 18-, 180-, and 900-m isobaths on the Bermuda Platform, particularly on the southwest side of the island where the plume originates. The bathymetry south of the Gulf of Batabano is similar; the 200-m isobath is less than 10 km from the shelf edge and, on 9 November, much of the plume was situated over waters in excess of 4,000-m depth. Thus, bottom reflectance is likely to influence only the closest pixels to the shelf edge for either the Bermuda Platform or Gulf of Batabano, and will not significantly affect the sediment mass algorithm.

A more significant effect may be due to the uncertainty in atmospheric correction over highly turbid waters because the reflectance in the 765- and 865-nm bands used for atmospheric correction of SeaWiFS data may not be zero in the pixels over the sediment plume. Although it is beyond the scope of this initial observational study, an evaluation of how changing the atmospheric correction parameters affects the results of the algorithm would be a useful extension of this research.

Because it is impossible to acquire actual sediment concentration data for these events that can be compared with the algorithm output, the data files processed by the algorithm used in this paper were provided to two other research groups for processing with different suspended-sediment concentration algorithms. The sediment concentration algorithms of Ahn et al. (2002) and Gould et al. (2002) were searched for comparison. (The Gould et al. algorithm could only be used to analyze the Gulf of Batabano event because the necessary mapping projection for the Bermuda region had not been created.)

Comparison of the results indicated a significant level of agreement between the Vasilkov algorithm and the Gould et al. algorithm. For the 5 November Gulf of Batabano plume, where the mean value for the plume was 0.71 g/m3, the Gould et al. algorithm indicated concentrations ranging between 0.58 and 1.40 g/m3.In the highest concentration regions of the feature, the Gould et al. algorithm indicated concentrations of 1.1–1.4 g/m3, and the Vasilkov algorithm indicated concentrations of 1.0–1.2 g/m3 . For the 9 November plume, the Gould et al. algorithm indicated concentrations ranging from 0.51 to 0.59 g/m3, whereas the Vasilkov algorithm indicated concentrations of 0.48 to 0.60 g/m3. In the absence of actual sea-truth data, this agreement is quite remarkable.

Figures 5 and 6 show a comparison of the results from the Vasilkov and Gould et al. algorithms. Processing the data with these algorithms created files in which the original latitude/longitude information was lost, so pixels with the same xy coordinates were compared. For the 5 November sediment plume (Fig. 5), five pixels were selected on a “transect” extending southeast from the Island of Youth to the edge of the feature. Thus, pixel 1 is closest to the island, and pixel 5 is farthest from the island. For the 9 November plume (Fig. 6), ten pixels were selected starting at the plume terminus and moving northeastward toward the Gulf of Batabano. In this figure, pixel 1 is the farthest from the island and pixel 10 the closest.
Fig. 5

Comparison of results for the Vasilkov suspended sediment algorithm (circles) and Gould et al. suspended sediment algorithm (squares) for five selected pixels in the 5 November 2001 image of the Gulf of Batabano sediment plume. The transect extended southeastward from the island; pixel 1 was closest to the island, pixel 5 at the edge of the plume

Fig. 6

Comparison of results for the Vasilkov suspended sediment algorithm (circles) and Gould et al. suspended sediment algorithm (squares) for ten selected pixels in the 9 November 2001 image of the Gulf of Batabano sediment plume. The transect commenced in the plume terminus and extended northeastward toward the Gulf of Batabano; pixel 1 is the farthest offshore, pixel 10 the closest to the Island of Youth

This initial comparison indicates that there is a greater degree of variability between the two algorithms when sediment concentrations are higher and the radiances are correspondingly greater. The close agreement and roughly similar pattern of variation for the 9 November data indicates that the agreement is improved when the radiance values are moderate. This initial data comparison suggests that both approaches are empirically valid, but validation of the algorithms with in situ data is still required.

For the 5 November plume, the algorithm of Ahn et al. indicated considerably higher concentrations (10–20 g/m3) compared with the results of either the Vasilkov or Gould et al. algorithms. For the 9 November Gulf of Batabano plume, the Ahn et al. algorithm indicated concentrations of 5–8 g/m3. There are several related reasons for the higher estimates of sediment concentration returned by this algorithm. This algorithm was developed in case 2 waters for sediments in the Yellow Sea, and these sediments are considerably less reflective than the neritic carbonate sediments exported from either Bermuda or the Gulf of Batabano. Ahn et al. note directly that the results of their algorithm will be sensitive to the mineralogy of the suspended sediment. Furthermore, the clarity of the waters around Bermuda and the Gulf of Batabano may present quite different optical properties than the case 2 waters which were analyzed in the development of the Ahn et al. algorithm. The Yellow Sea waters have considerably higher chlorophyll and suspended sediment concentrations than are found in the tropical, open-ocean waters of the Sargasso Sea and Caribbean Sea. In fact, much of the Yellow Sea has higher suspended sediment concentrations than are present in the sediment plumes observed here. Because the Ahn et al. algorithm relies exclusively on nLw(555) values, it may be that the results of the algorithm are more sediment specific than the results of the semi-analytical Vasilkov algorithm or the Gould et al. algorithm. It should also be noted that the suspended sediment concentration images generated by the Ahn et al. algorithm delineated details in the sediment plumes, particularly on 9 November, which are not as clearly resolved by the other two algorithms. This observation also indicates that the use of the 555-nm band data by the Ahn et al. algorithm is quite good at detecting sediment features because the 555-nm band is quite sensitive to sediment reflectance. For Bermuda, the Ahn et al. algorithm also indicates higher sediment concentrations than the Vasilkov algorithm.

For several reasons, the mass estimates calculated for these events are very conservative, minimum estimates of the carbonate mass exported offshore. The primary reason that these are minimum estimates is that a sediment plume is not a shallow surface feature: it possesses a definite three-dimensional character due to the rapid settling rates of CaCO3 sediments. This observation means that the mass of sediment in the terminal eddy of the feature has been underestimated by the algorithm that uses surface water-leaving radiances because the sediment in the terminal eddy has very likely settled several meters below the water surface. So the lower radiances that are observed at the far extensions of the plumes (as compared with the higher radiances observed closer to shore) are not necessarily due to a reduced mass of sediments; they may be because the water-leaving radiances are attenuated by scattering in the intervening water column.

The volume occupied by the sediment plume is also a critical factor. The assumption that the sediment occupies a 1-m depth at the surface is made because sediment concentrations, similar to chlorophyll concentrations, are reported in terms of concentration per cubic meter, as if all of the material was contained in the top meter of the water column. It is obvious, however, that sediment plumes and phytoplankton blooms will have a three-dimensional character and will be distributed in the water column. Sediment plumes will occupy a depth distribution that is defined by the minimum and maximum settling rates of the sediment particles. Because the observed sediment plumes appear quickly and disappear in a few days, and also because light penetration in the water column is limited to a maximum of approximately 30 m, the volume that the sediments occupy is likely to be fairly well defined. Studies of settling rates using particle-size distributions for these sediments would help to clarify this question quite easily and would allow improved estimation of the total mass of sediment exported by these events. The observed plumes are somewhat different from sediment plumes originating in rivers because the river water carrying the sediment would be buoyant compared with saline ocean waters, constraining the sediments near the surface.

The study of Hubbard (1992) provides one datum for comparison with the estimated results from this study. Using geological measurements, Hubbard estimated that Hurricane Hugo removed ~2 million kg of sediment from the Salt River canyon on the north coast of the island of St. Croix. The amount of sediment estimated in the Bermuda sediment plume generated by Hurricane Gert, 0.22 million kg, is an order of magnitude smaller than Hubbard’s value for St. Croix. Yet, if the calculation was modified using the reasonable assumption that the sediments were constrained to a zone 10 m wide in the water column rather than 1 m wide, the sediment mass estimate would then become 2.2 million kg, comparable to Hubbard’s value. Similarly, the value estimated for the Gulf of Batabano plume, 1.35 million kg, is the same order of magnitude as Hubbard’s value, but the immense size of the Gulf of Batabano event intuitively indicates that a larger volume of sediment was moved than would have been transported from the island of St. Croix. Therefore, the estimates reported in this paper are clearly conservative minimum estimates. Analytical studies of particle settling and possible sea-truth studies using artificially generated plumes, described in Acker et al. (2002), both offer the potential for significant improvement to these initial estimates.

The use of higher-resolution data than the maximum 1-km resolution of SeaWiFS may also provide a method for validating the results of the algorithm. Some reef/lagoon complexes may produce a persistent or regularly recurring plume of neritic sediments that could be observed by the higher resolution bands of other instruments while in situ sampling was performed. The Moderate Resolution Imaging Spectroradiometer (MODIS) has a 500-m resolution band centered at 555 nm that could be used for this purpose, and the 300-m nadir resolution data of the Medium Resolution Imaging Spectrometer (MERIS) on Envisat could also provide useful observations. With some good fortune, Landsat 15-m resolution data might even be obtained for a coordinated sampling program.

Another factor is that this method for estimating IOP and deriving sediment concentrations allows for variability of the parameters S (DOM spectral slope) and n (backscatter wavelength ratio exponent). Vasilkov (1997) indicates that knowledge of the correct values of S and n allows considerable improvement in the retrieval accuracy of sediment backscatter and DOM absorption. Determination of these values would require field observations, which could also be accomplished in situ for recurrent sediment plumes.

Perhaps the most interesting result of this study is a comparison of the total suspended sediment mass estimates for the sediment plume emanating from the Gulf of Batabano on 5 and 9 November. Our estimates indicate that the amount of sediment in the plume on both days was nearly the same. Given the likelihood that some settling of particles has occurred in the water column over the 4-day interval, this is an unexpected result. It would be expected that the radiance would decrease as the sediments settled deeper in the water column, due both to the more rapid loss of heavier particles and the attenuation of the radiance from particles that are suspended at a depth below the water surface. Because this does not appear to have happened, it can be inferred that the primary mass of sediments observed in the images consists of very fine-grained particles in the mud/silt size range, and that these particles do not settle very much over the course of 4 days. These fine-grained particles act as a visible “tracer” of plume movement. It is also possible that the sediments were mixed within a mass of low-salinity water that was created by the heavy rainfall from Hurricane Michelle, and this mass of sediment remained near the surface for several days due to its lower density. The reduction in suspended sediment concentration near the periphery of the Gulf of Batabano observed in the 9 November data indicates that there was minimal additional contribution of sediments from the turbid shallow waters of the Gulf following the initial pulse of sediment export that occurred immediately after the passage of Hurricane Michelle.

In conclusion, this study presents, for the first time, quantification of neritic sediment mass transport from two carbonate environments after the passage of a hurricane, demonstrating the potential of remote-sensing observations for this type of measurement, which is difficult to assess in situ. We hope this study will generate research interest in order to better calibrate the results for future catastrophic or recurrent events, and to allow an assessment of how our stated assumptions affect the mass transport estimates.

Acknowledgements

We are grateful to Dr. Richard Gould of the Naval Research Laboratory, Stennis Space Center, Mississippi and Paul Martinolich of Neptune Science Inc. for processing the data files with their algorithm and providing the results for our examination. We are also grateful to Dr. Yu-Hwan Ahn for processing the data files with his algorithm and providing these results to us. SeaWiFS data are the property of Orbimage, Inc. and are used for this research in accordance with the SeaWiFS Research Data Terms and Conditions Agreement of the NASA SeaWiFS Project. We would also like to acknowledge the operators of the HRPT stations at the Bermuda Biological Station for Research and the University of South Florida Institute for Marine Remote Sensing (ImaRS) for acquiring the SeaWiFS data used for this research.

Copyright information

© Springer-Verlag 2003