Spectral Signatures of Coral Reefs

www.elsevier.com/locate/sna

Spectral Signatures of Coral Reefs:Features from Space

Dan Lubin,* Wei Li,† Phillip Dustan,‡ Charles H. Mazel,§and Knut Stamnes†

The spectral signatures of coral reefs and related scenes, be feasible using satellite remote sensing, but that detailedreef mapping (e.g., species identification) may be moreas they would be measured above the Earth’s atmosphere,difficult. Elsevier Science Inc., 2001are calculated using a coupled atmosphere-ocean discrete

ordinates radiative transfer model. Actual measured reflec-tance spectra from field work are used as input data. Four

INTRODUCTIONcoral species are considered, to survey the natural range ofcoral reflectance: Montastrea cavernosa, Acropora palmata, The world’s coral reefs are highly susceptible to damageDichocoenia stokesii, and Siderastrea siderea. Four non- by a variety of human activities, and by global climatecoral objects associated with reefs are also considered: sand, change. Studies of individual coral reef ecosystems havecoralline algae, green macroalgae, and algal turf. The reflec- reported catastrophic damage by either anthropogenic pol-tance spectra as would be measured at the top of the atmo- lution or the water column or by physical impacts such assphere are substantially different from the in situ spectra, ship groundings (Dustan and Halas, 1987; Hughes, 1994;due to differential attenuation by the water column and, Hatziolos et al., 1998). It is not an exaggeration to suggestmost importantly, by atmospheric Rayleigh scattering. The that the world’s coral ecosystems could, through neglect,result is that many of the spectral features that can be used suffer severe degradation leading to near-extinction andto distinguish coral species from their surroundings or from ecosystem collapse within a few decades. To date thereone another, which have been used successfully with sur- has been no successful program to produce a completeface or aircraft data, would be obscured in spectral mea- global map of the world’s coral reefs. Of the known reefsurements from a spacecraft. However, above the atmo- ecosystems, some are very well studied by ongoing fieldsphere, the radiance contrasts between most coral species programs while the health of others remains unmonitoredand most brighter noncoral objects remain noticeable for due to their remoteness.water column depths up to 20 m. Over many spectral inter- To map the global distribution of coral reef ecosys-vals, the reflectance from dark coral under shallow water tems, and to monitor the growth or deterioration of coralis smaller than that of deep water. The maximum top-of- reefs worldwide, satellite remote sensing will be requiredatmosphere radiances, and maximum contrasts between in some capacity. The Landsat and Systeme Probatoire descene types, occur between 400 nm and 600 nm. This study l’Observations de la Terre (SPOT) multispectral shortwavesupports the conclusions of recent satellite reef mapping imagers offer some potential for coral reef identificationexercises, suggesting that coral reef identification should (Bour and Pichon, 1997; Holden and Ledrew, 1999; Miller

and Cruise, 1995; Mumby et al., 1998). For example, Mumbyet al. (1998) tested a retrieval algorithm using the Landsat

* Scripps Institution of Oceanography University of California, San Thematic Mapper (TM), comparing the satellite-based es-Diego, La Jolla, California timates of coral reef extent against similar 1-m-resolution† Geophysical Institute, University of Alaska, Fairbanks, Alaska

‡ College of Charleston, Charleston, South Carolina retrievals from an airborne imaging spectrometer; the satel-§ Physical Sciences, Inc., Andover, Massachusetts lite-based method yielded an overall accuracy of 75%.Address correspondence to D. Lubin, Scripps Inst. of Oceanogra- In this study, we use a coupled ocean-atmosphere radi-

phy, Univ. of California, San Diego, 9500 Gilman Dr. La Jolla, CA 92093- ative transfer model to further investigate the potential of0221. E-mail: [email protected] 27 May 1999; revised 11 January 2000. satellite remote sensing, by examining how the spectral and

REMOTE SENS. ENVIRON. 75:127–137 (2001)Elsevier Science Inc., 2001 0034-4257/00/$–see front matter655 Avenue of the Americas, New York, NY 10010 PII S0034-4257(00)00161-9

128 Lubin et al.

radiometric signatures of coral reef objects should appearabove from low earth orbit after being modified by watercolumn and atmospheric attenuation. For example, it iswell known that sand has a considerably higher reflectanceat all visible wavelengths than any living coral. Does thissignificant reflectance difference that we can measure insitu manifest itself in suitable radiance contrast at the topof the atmosphere? If so, is this also true for other noncoralobjects (e.g., algal turf) that are more reflective than livingcoral, but not as reflective as sand?

We can also investigate the spectral reflectances ofcoral reef objects as they would appear above the atmo-sphere. Holden and Ledrew (1999) have shown i) that thereare robust differences between in situ reflectance spectraof living corals and those of related noncoral objects, ii)that these differences are independent of geographic sam-pling and of coral morphology, and iii) that these spectralcontrasts can be used to construct a scene identificationalgorithm based on differential reflectance. Does this typeof differential reflectance spectroscopy remain applicableto scenes observed from above the atmosphere? It is knownthat differential absorption by the water column will modifyconsiderably the spectral reflectance of an object at depth(Lyzenga, 1981; Maritorena et al., 1994), but what is thecombined effect of the water column and the atmosphere?

Finally, we can estimate both the spectral radiancesand reflectances that would be measured by a satellite instru-ment. This gives us the ability to i) make a first-order esti-mate of what spatial resolution might be available usingstandard CCD technology with spectral bandwidths opti-mized for coral reef studies and ii) discuss the suitability of

Figure 1. Spectral reflectances of the eight scene typesexisting sensors such as Landsat TM, which were originallyconsidered in this study, measured by an underwaterdesigned for terrestial applications, for measuring the ra-spectrometer. Reflectance shown is relative to spectralon, A) fordiometric signatures of coral reef objects.the noncoral objects and B) for the living corals.

RADIATIVE TRANSFER MODEL ANDof Iguana Cay, a small island approximately 3 km north ofINPUT DATALee Stocking Island (238N, 768W), Exumas, Bahamas. The

We base this study on the measured reflectance spectra of coral community structure and distribution has been de-eight objects or scene types: sand, coralline (red crustose) scribed by King (1995). Each reflectance spectrum shownalgae, green macroalgae, algal turf, and four different coral in Figure 1 is an average of a small set of sample measure-species. It is obviously necessary to consider sand, if we ments. We note that our sample spectrum of green mac-are discussing the potential for distinguishing coral reefs roalgae has a reflectance that is similar to that of darkfrom their surroundings. The coralline algae, green mac- brown macroalgae reported by Maritorena et al. (1994).roalgae, and algal turf reflectance spectra are considered The conclusions of this study are related to both the widebecause they represent components of the shallow water variability and continuous gradient in spectral reflectancesreef community. Montastrea cavernosa and Acropora pal- of objects related to coral reefs, and do not require consid-mata are chosen as two coral species having very different eration of the standard deviations of each sample set inspectral reflectances (brighter and darker than average our radiative transfer modeling.corals, respectively). Dichocoenia stokesii and Siderastrea To examine how these scene types might appear fromsiderea are chosen because they are intermediate in bright- space, after light reflected from them has been furtherness at all visible wavelenghts between M. cavernosa and attenuated by the water column and scattered by the atmo-A. palmata. These reflectances, shown in Figure 1, were sphere, we utilize a discrete ordinates radiative transfermeasured relative to spectralon using an underwater spec- algorithm (Stamnes et al., 1988). Lyzenga (1981) and Mari-trometer (Mazel, 1997). The study site was a small Baha- torena et al. (1994) have shown how the water column will

modify the apparent reflectance spectrum of an object onmian reef patch, Rainbow Gardens, located on the lee side

Spectral Signatures of Coral Reefs: Features from Space 129

ance calculation within the water column. Jin and Stamnes(1994) have derived a method that adds the appropriatenumber of quadrature points to the water column, givena desired number of quadrature points in the atmosphere.This method includes both scattering and absorption by bothmedia, and allows the vertical structure of both media to bespecified by any number of horizontally homogeneous layers.

As input data to this radiative transfer model, we usethe measured reflectance spectra of Figure 2 as the lowerboundary condition, specifying a Lambertian surface al-bedo. The Lambertian surface albedo is a modeling as-sumption, as the scene bidirectional reflectance distribu-tion functions were not measured by the underwaterspectrometer. The water column is treated as a single layerwhose thickness (depth) we can vary. Spectral attenuationcoefficients in the water column are taken from Smith andBaker (1981) for clear ocean waters. The radiative transfermodel requires that we assume a flat air-water interface,and neglect effects of ocean surface roughness at present.The vertical structure and composition of the atmosphereis specified by the tropical model atmosphere in LOW-Figure 2. Diagram of the important features of the coupled

ocean-atmosphere discrete-ordinates radiative transfer TRAN 7 (Kneiszys et al., 1988). Tropospheric aerosols aremodel. The atmosphere, modeled using 35 layers each of included, as specified by the maritime background modeluniform optical depth, has total optical depth sa; sa is the in LOWTRAN 7. The ozone column abundance (relevantsum of Rayleigh scattering, aerosol scattering/absorption, and

to visible wavelengths via the Chappuis absorption bands)ozone absorption optical depths, and is a function ofis set at 250 Dobson units, which is appropriate for tropicalwavelength. The water column is modeled as a single layer

having total optical depth sw, which is also a function of latitudes year round. The radiance calculations are carriedwavelength as specified by the Smith and Baker (1981) out with 16 (atmospheric) computational streams for 4pattenuation coefficients. The extraterrestrial solar flux is steradians, with the radiances solved for each of the 16l0F0, where l0 is the cosine of the solar zenith angle. At the

Gaussian quadrature angles (Chandrasekhar, 1960). Forair-water interface, the direct solar beam is refracted intothe angle l0w. As light from all downward directions (2p clarity, we discuss radiances at the Gaussian quadraturesteradians) crosses the air-water interface, it is refracted angle closest to the downward-looking direction (11.48 offinto a cone of less than 2p steradians (region II). In the water nadir), as this angle is the most relevant to a potentialcolumn, light scattered outside this cone (region I) cannot

satellite instrument operating with high spatial resolution.reach the atmosphere. The reflectance of the reef floor isapproximated as being Lambertian (isotropic).

SPECTRAL SIGNATURES OF CORAL REEFS

We first examine the upwelling spectral radiance at thethe bottom, and how this effect can be treated in remoteair-water interface, when the water column depth is fivesensing retrievals. Here we are interested in both thismeters. Figure 3 shows this upwelling radiance, at viewingspectral transformation and in estimates of the actual radi-angle 11.48, for all eight objects. We see that the usefulance at the top of the atmosphere. Therefore, for this studysignal for remote sensing lies between 400 nm and 600 nm.we use a discrete ordinates radiative transfer formulationThe maxima of in situ scene reflectances (Fig. 1) for longerby Jin and Stamnes (1994), to solve for the top-of-atmo-wavelengths are offset by the strong water column attenua-sphere radiances directly (Fig. 2). This formulation allowstion, such that the overall radiances for wavelengths longerfor numerical solution of the radiative transfer equation inthan 600 nm are very small (,0.01 W m22 sr21 nm), anda medium where the index of refraction changes (i.e., atthe radiance contrasts between scene types are poor. Ifthe air-water interface). As light propagating through thewe examine the upwelling radiance at the top of the atmo-atmosphere in all downward directions (2p steradians)sphere (Fig. 4), we see how the water-leaving spectral radi-crosses the air-water interface, it is refracted into a coneance is modulated by the atmospheric Rayleigh scattering.of less than 2p steradians (angular region II in Fig. 2). InFor wavelengths between 400 nm and 500 nm, the top-of-the water column, light from within this cone can be scat-atmosphere radiance is typically a factor of 2 or more largertered to and from the region outside it, but light scatteredthan the water-leaving radiance, due to Rayleigh scatteringoutside this cone (angular region I in Fig. 2) cannot reachbeing the strongest at the shorter wavelengths. Betweenthe atmosphere. Therefore, a complete radiative transfer400 nm and 600 nm, some scene types remain distinct fromsolution for a coupled atmosphere-ocean system requires

that additional computational streams be added for the radi- one another. The sand and coralline algae are the brightest

130 Lubin et al.

Figure 4. The spectral radiances at the top of the Earth’sFigure 3. Upwelling spectral radiance at the air-water atmosphere, for a solar zenith angle of 308, A) noncoralinterface, in the near-nadir direction (polar angle 11.48), objects, B) living corals. The water column depth is 5 m.for A) the noncoral objects and B) the coral species. The watercolumn depth is 5 m. The solar zenith angle is 308.

Figure 5 shows the top-of-atmosphere spectral reflec-tance, defined as Eq. (1):

scenes, with maximum spectral radiances of 0.145 W m22 Rk(h0, h, u)5100pIk(h0, h, u)/Fk(h0), (1)sr21 nm and 0.100 W m22 sr21 nm, respectively; while all

where Fk(h0) is the extraterrestrial solar spectral irradiance.four coral species considered here are noticeably darkerThe reflectance defined this way is not the true reflectancethan coralline algae and yield top-of-atmosphere radiances(albedo) of the ocean-atmosphere system (which would be

nearly a factor of 2 smaller than that of sand. However, the ratio of upwelling flux to extraterrestrial solar flux),the two other noncoral objects yield top-of-atmosphere but is instead a convenient and often used convention forradiances that are much closer to those from living corals. scaling radiances when displaying satellite imagery. For aSpectral radiances from the algal turf are typically only nadir or near-nadir view, this scaled radiance is actually20% larger than those from the brightest coral (M. cav- smaller than the true albedo, due to the angular redistribu-ernosa), and spectral radiances from the green macroalgae tion of radiation that favors the direction of the solar beamlie squarely within the range of those from living coral. (specular reflection, e.g., Lubin and Weber, 1995). WhenThus, we see that the natural variability in spectral reflec- the calculations are examined with this scaling, it is appar-tance of coral reef objects yields a continuous gradient in ent that the reflectance contrasts between sand, corallineradiance as measured from above the atmosphere, and that algae, and the corals are large enough to be practical forliving corals and noncoral objects are not always distinct remote sensing; however, this is not true for green macroal-from one another. Although sand should be readily distin- gae, algal turf, and the corals. Furthermore, the effect ofguishable from other objects, there may always be some atmospheric Rayleigh scattering is to render the slopes ofambiguity in resolving some types of algae from living coral the top-of-atmosphere reflectance spectra between 400 nm

and 600 nm very similar for most of the objects (corallinewith a satellite instrument.

Spectral Signatures of Coral Reefs: Features from Space 131

spectral radiances at the ocean surface. Between 500 nmand 600 nm for all objects, the upwelling radiance decreasesnoticeably with increasing water depth as the water columnattenuates the bottom-reflected radiance component. Forthe four coral species, the radiance changes very little withwater depth for wavelengths shorter than 450 nm, and atthese shorter wavelengths the largest water-leaving radi-ances occur for water column depth 20 m. At these shorterwavelengths, the corals have a reflectance that is less thanthe reflectance of the model’s optically deep water. At adepth of 20 m, there is very little difference in radiancebetween the four scene types around 550 nm (top-of-atmo-sphere radiances generally less than 0.04 W m22 sr21 nm),signifying that most of the bottom-reflected radiance isattenuated by the water column at this depth. For sandand coralline algae, there is substantial variability in radi-ance with water column depth for wavelengths between500 nm and 600 nm, which we would expect as thesescene types have higher intrinsic reflectances which wouldcontribute to the total space-measured radiance over agreater range in water column depth. The intrinsic reflec-tances of the M. cavernosa and algal turf are large enoughthat the top-of-atmosphere radiances, between 500 nm and600 nm, over water column depths 1–3 m are twice as largeas the radiances over water column depth 20 m (essentiallyinfinite ocean depth). A. palmata and green macroalgae,on the other hand, provide such a small bottom reflectancecomponent that the only appreciable variability in top-of-atmosphere radiance with water column depth occursaround 560–580 nm. Coral objects this dark will be theeasiest to distinguish from surrounding sand, but the ambi-

Figure 5. The spectral reflecance, or scaled radiance defined guity between green macroalgae and living corals remainsas 100pIk(h0, h, u)/Fk(h0), at the top of the Earth’s

apparent at all depths.atmosphere, for a solar zenith angle of 308, for A) noncoralobjects, B) living corals. The water column depth is 5 m.

UTILITY OF THE LANDSATTHEMATIC MAPPER

algae, green macroalgae, algal turf, A. palmata, D. stokesii,Given the limited success some investigators have shownS. siderea). The exceptions are sand, with its higher overallwith Landsat TM images of reef ecosystems (e.g., Mumbyreflectance, and M. cavernosa between 450 nm and 480et al., 1998), it is worth integrating our modeled spectralnm. Thus, many of the objects’ unique spectral featuresradiances over the relevant TM bandwidths to examinethat are evident in situ or at the air-water interface (Figs.how well suited the TM instrument’s radiometry is for this1 and 3) are obscured by the atmospheric column.purpose. This type of analysis needs to be done in theFollowing principles used in ocean color remote sens-context of the expected dynamic range. When designing aing, it should be possible to use a radiative transfer modelsatellite sensor for a specific remote sensing task, one needsto remove atmospheric effects from a satellite measure-to know the expected radiance range from the scenes ofment and retrieve the ocean surface reflectance. However,interest so that the detector sensitivity and analog-digitalfor this to be successful, the satellite sensor must haveconverter can be chosen accordingly. For example, theenough sensitivity and precision to detect the spectral dif-Landsat TM was designed to show as much distinction asferences between coral reef objects after they have propa-possible between various cloud-free land surface types,gated through the entire atmosphere column. If these top-and the dynamic ranges of the TM1 (450–520 nm) andof-atmosphere differences are smaller than the resolutionTM2 (520–600 nm) bands were set at approximately (pre-or precision of the sensor, then removing the atmosphericlaunch values) 0–11.1 and 0–24.7 W m22 sr21, respectivelyeffects will yield no information.(Barker, 1984). Clouds usually give rise to larger backscat-In Figure 6, the upwelling spectral radiances at thetered radiances, but cloud mapping for weather or climatetop of the atmosphere are shown as a function of water

column depth. Figure 7 shows the corresponding upwelling study was not part of the Landsat mission and clouds satu-

132 Lubin et al.

Figure 6. The spectral radiance at the top of the Earth’s atmosphere, in the near-nadir direction (polar angle 11.48), as a functionof water column depth, for A) sand, B) coralline algae, ) green macroalgae, D) algal turf, E) M. cavernosa, F) A. Palmata, G) D.stokesii, and H) S. siderea.

rate the TM1 and TM2 bands. When using Landsat for the radiance from this scene type decreases by 95 counts(approximately 1/3 of the sensor’s dynamic range) as watercoral reef mapping, we are restricted to a sensor whose dy-

namic range has been optimized for general use over all column depth increases from 1 m to 20 m. The radiancefrom the coralline algae scene spans a range of 121–98tropical and temperate-latitude land surface scene types. It

is worth investigating how much of the TM1 and TM2 dy- counts as water column depth increases from 1 m to 20m. The radiance from algal turf spans a similar range, butnamic range we are actually using when viewing coral reefs.

We integrated the top-of-atmosphere spectral radi- is slightly less sensitive to water column depth. The fourliving corals, and the green macroalgae, yield radiancesances of Figure 6 over the Landsat TM1 and TM2 bands,

weighting the spectral radiance by the instrument response that encompass the lower third of the TM1 band’s dynamicrange. The radiance from A. palmata actually increasesfunctions of these bands (Barker, 1984). For the eight

objects considered in this study, these integrated radiances slightly with water column depth. A. palmata has such alow intrinsic reflectance that, at the shallower depths, moreare shown as a function of water column depth for the

TM1 band (Figs. 8A,B) and the TM2 band (Figs. 8C,D). photons are absorbed than in a semiinfinite water column.From Figures 8A,B we conclude that the Landsat TM1The radiances are given in W m22 sr21 on one vertical axis,

and in 8-bit counts as used by the TM digitization (scale band should be well suited for distinguishing living coralreefs from most surrounding objects, at least from a radio-of 0 to 255 over the full dynamic range of each sensor

band) on the other vertical axis. We see that the TM1 band metric perspective, because the range of backscatteredradiances from the objects of interest makes full use ofis already nearly ideally optimized for coral reef identifica-

tion. The brightest scene of interest—sand at the shallowest this band’s dynamic range.Figures 8C,D show that the Landsat TM2 band is lessdepth—yields a radiance corresponding to 211 counts, and

Spectral Signatures of Coral Reefs: Features from Space 133

Figure 6. Continued.

well optimized for coral reef identification. The radiance the top-of-atmosphere radiances to see what might be real-ized in practice. In Table 1, we show the reflectance differ-from sand corresponds to only 98 counts at water column

depth 1 m, decreasing rapidly to 31 counts at water column ences at the ocean surface between the TM1 and TM2bands (as they might be measured by a TM “simulator”depth 10 m. There is a useful contrast (tens of counts)

between the other three scene types for shallower water aboard a low-altitude aircraft), as a function of water col-umn depth and for each coral reef object. Over the shallow-column depths, but by depth 10 m the radiances from all

eight objects occupy only the lower 15% of the TM2 band’s est depths, this reflectance difference is negative, and forall objects the reflectance difference eventually becomesdynamic range.

For this exercise, we extended our calculations for all positive with increasing depth. At the shallowest depths,there does not appear to be much obviously useful informa-eight objects to a depth of 100 m, to verify convergence of

radiances over deep water. In the TM1 band, the radiances tion for distinguishing between coral species, or for distin-guishing coral from coralline algae or algal turf. However,converge to a mean value of 3.052 W m22 sr21, with a dif-

ference of no more than 3% between bands. Most of this there may be useful information about water column depthin the reflectance difference over sand. The reflectancedifference is due to the radiance over sand, which is slightly

larger than that of all other objects. In the TM2 band, the difference for sand increases noticeably with intermediatewater column depth (3–7 m), toward a maximum at 10 m.radiances converge to a mean value of 1.717 W m22 sr21,

with a difference of no more than 0.8% between bands. If the reflectance difference over sand can be used to makea first-order estimate of water column depth (after theIt is also worth investigating if the differential reflec-

tances between the TM1 and TM2 bands contain informa- absolute scene reflectance in either band has been used toidentify sand), then perhaps other reflectance differencestion about object discrimination or about water column

depth. First, we examine the water-leaving radiances to see could help distinguish corals from coralline algae or algalturf. The success of such an attempt would depend on theif any information is present in principle; then we examine

134 Lubin et al.

Figure 7. Upwelling spectral radiance at the air-water interface, in the near-nadir direction (polar angle 11.48), as a function of watercolumn depth: A) sand, B) coralline algae, C) green macroalgae, D) algal turf, E) M. cavernosa, F) A. Palmata, G) D. stokesii,and H) S. siderea.

existence of homogeneous scene types on the scale of a TM throughout various reef components, this result is trueeven for a sensor such as TM which proves to be radiomet-pixel (of order 30 m). In Table 2, we show the reflectance

differences as they would be measured from space. For rically very well suited for imaging coral reefs. Due towater column attenuation at wavelengths longer than 600most objects and depths, the reflectance differences be-

tween objects and depths are preserved above the atmo- nm and obscuration by atmospheric Rayleigh scattering atwavelengths shorter than 500 nm, the intrinsic spectralsphere, although they are shifted positive by about 3%.

The noise-equivalent reflectance in the TM instrument is signatures of various reef components are somewhat chal-lenging to use from the vantage point of a satellite. Radia-of order 0.2% (Barker, 1984). Therefore, if the information

in Tables 1 and 2 is useful for coral reef mapping, most tive transfer calculations of the type presented here canillustrate the required dynamic range and sensitivity re-of this information should be detectable by TM.quired for a satellite instrument to detect the spectralsignatures of coral reef objects.DISCUSSION Additionally, the radiative transfer calculations pre-sented here are for idealized conditions: a maritime back-The radiative transfer modeling studies presented here sup-

port the results of previous case studies using satellite data ground tropospheric aerosol burden, and clear water columnattenuation as specified by Smith and Baker (1981). For(e.g., Mumby et al., 1998); previous studies have shown

that sand is relatively easy to distinguish from coral reef those coral reefs found in turbid waters (e.g., Miller andCruise, 1995), the obscuring effects of higher water columnobjects, while individual reef objects are usually difficult

to distinguish from one another. Due to the continuous attenuation might be more severe than those discussedhere. If the oceanographic community were to undertakegradation in both spectral and broadband reflectance

Spectral Signatures of Coral Reefs: Features from Space 135

Figure 7. Continued.

the daunting but worthwhile task of mapping and monitor- areas. A thorough survey of all of the world’s reefs mightnot be cost-effective using these commercial sensors. Theing all of the world’s reefs, radiative transfer studies of the

type presented here would have to be carried out in much recent deployment of Landsat 7 may partially remedy thecost-effectiveness problem; as part of this spacecraft’s Longgreater detail. In addition to higher water turbidity, other

radiative effects and sources of error include varying aero- Term Acquisition Plan, some well-known coral reefs arein the ongoing data collection queue and can thus be moni-sol opacity, effects of ocean surface roughness, and the ef-

fects of errors in the knowledge of mean ocean depth for tored effectively. However, Landsat 7’s onboard recorderscannot be programmed to collect data everywhere; therea given scene. During rough water conditions, the reefs

will become obscured with breaking waves and whitecaps. will always be a scarcity of Landsat 7 data over remotetropical regions where many coral reefs have yet to beWhile this would seriously reduce our ability to extract useful

biooptical data, the information might be used to “mark” or identified, let alone monitored.A polar orbit enables an instrument to overfly the sameconfirm the position of shallow water shoals which might

help to confirm the positions of reefs in remote regions. point on the Earth’s surface every few days. This mightbe an advantage for monitoring specific scenes. However,Another fundamental design criterion concerns the

orbital deployment. Commercial remote sensors such as because present spacecraft are Sun-synchronous, at anygiven low latitude location there will be only one or twoLandsat TM or SPOT are on polar-orbiting spacecraft.

Because the sensors’ flight times over tropical regions are potentially useful images per day, and many of them canbe expected to suffer cloud contamination at the time ofonly a small fraction of the total orbital period, a large

number of orbits would have to be considered and a large overflight. Orographically or thermally induced low cloudformation often occurs when the satellite is on the scene.number of images acquired in order to provide adequate

composites for coral reef monitoring over large geographic This makes the use of commercial sensors such as Landsat

136 Lubin et al.

Figure 8. The top-of-atmosphere near-nadir radiance for all four scene types, as a function of water column depth: A) noncoralobjects integrated over the Landsat TM1 band; B) corals integrated over the Landsat TM1 band; (C) noncoral objectsintegrated over the Landsat TM2 band; D) corals integrated over the Landsat TM2 band.

TM very costly and cumbersome for a global coral reef ent daylight hours (thus partially circumventing the prob-lem of cloud contamination). This might be an advantageproject, but may not be a limitation with a dedicated instru-

ment where the image extraction and processing proce- for a mission intended primarily to identify reefs and tomake the first complete world map of their distribution.dures are set up solely for coral reef studies. A low inclina-

tion orbit would enable the instrument to view the tropics One possible limitation with a low inclination orbit is thatthe ground track would rarely repeat itself. To providethroughout most of its operation, and would allow the in-

strument to view a given geographic region at many differ- complete geographic coverage, the instrument would need

Table 1. Difference in Ocean Surface Reflectance (Scaled Radiance in Percent) between Landsat TM1 and TM2 Bands

Coralline Green Algal Montastrea Acropora Dichocoenia SiderastreaDepth (m) Sand Algae Macroalgae Turf cavernosa palmata stokesii siderea

1 21.55 22.19 21.03 22.73 23.36 22.08 22.43 22.412 2.32 20.37 20.60 21.21 21.86 21.53 21.57 21.653 4.77 0.91 20.27 20.14 20.79 21.11 20.93 21.084 6.40 1.82 20.01 0.64 0.00 20.77 20.46 20.655 7.56 2.52 0.21 1.25 0.62 20.49 20.06 20.297 8.75 3.34 0.51 1.98 1.38 20.09 0.43 0.18

10 9.32 3.92 0.81 2.55 2.00 0.31 0.88 0.6320 7.86 3.82 1.21 2.74 2.37 0.91 1.36 1.18

Spectral Signatures of Coral Reefs: Features from Space 137

Table 2. Difference in Top-of-Atmosphere Reflectance (Scaled Radiance in Percent) between Landsat TM1 and TM2 Bands

Coralline Green Algal Montastrea Acropora Dichocoenia SiderastreaDepth (m) Sand Algae Macroalgae Turf cavernosa palmata stokesii siderea

1 0.93 0.82 2.53 1.08 20.14 1.09 1.38 1.402 4.34 2.45 2.88 2.33 1.29 1.60 2.09 2.033 6.35 3.51 3.16 3.21 2.23 1.96 2.62 2.504 7.62 4.23 3.37 3.86 2.88 2.23 3.01 2.855 8.41 4.75 3.59 4.40 3.36 2.46 3.36 3.197 9.26 5.31 3.80 4.96 3.91 2.74 3.73 3.53

10 9.53 5.65 4.04 5.43 4.31 3.02 4.10 3.8920 8.20 5.42 4.36 5.58 4.45 3.42 4.48 4.33

Hatziolos, M., Hooten, A. J., and Fodor, M., Eds. (1998), Corala large cross-track swath. This would most likely requireReefs: Challenges and Opportunities for Sustainable Manage-the added cost and complexity of a mechanically scanningment, The World Bank, Washington, DC.instrument, as opposed to a simpler “push-broom” device,

Holden, H., and Ledrew, E. (1999), Hyperspectral identificationor a more advanced CCD with a much wider field of view.of coral reef features. Int. J. Remote Sens. 20(13):2545–2563.At present, one excellent platform for a coral reef

Hughes, T. P. (1994), Catastrophic phase shifts and large-scalemission might be the International Space Station (ISS), degradation of a Carribbean coral reef. Science 265:deployed at an orbital inclination of 51.68. The ISS would 1547–1551.offer an instrument a relatively large duty cycle over tropi- Jin, Z., and Stamnes, E. (1994), Radiative transfer in nonuniformlycal latitudes, and the cost of the spacecraft itself would be refracting media: atmosphere–ocean system. Appl. Opt. 33:eliminated (except for the costs of space qualification for 431–442.

King, A. K. (1995), Digital analysis of a patch reef community:the instrument). The potential for decades of operation,Rainbow Gardens Reef, Lee Stocking Island, Exuma Cays,routine maintenance by ISS crew, and even interactionBahamas, M.S. thesis, College of Charleston, 92 pp.with the crew for data collection, compares very favorably

Kneiszys, F. X., Shettle, E. P., Abreu, L. W., et al. (1988), Userswith a single small spacecraft whose low cost might limitGuide to LOWTRAN 7, Air Force Geophysics Laboratory,the design lifetime to less than 5 years.Report AFGL-TR-88–0177, Hanscom AFB, MA, 137 pp.This research was supported by the National Oceanic

Lubin, D., and Weber, P. G. (1995), The use of cloud reflectanceand Atmospheric Administration Climate and Global Change functions with satellite data for surface radiation budget esti-Marine Ecosystem Response Program, Award NA36GP0420. mation. J. Appl. Meteorol. 34:1333–1347.We thank the Caribbean Marine Research Laboratory for Lyzenga, D. R. (1981), Remote sensing of bottom reflectanceassistance with laboratory facilities and sampling. and water attenuation parameters in shallow water using air-

craft and Landsat data. Int. J. Remote Sens. 2:71–82.Mazel, C. H. (1997), Diver-operated instrument for in situ mea-REFERENCES surement for spectral fluorescence and reflectance of benthic

marine organisms and substrates. Opt. Eng. 36:2612–2617.Barker, J., Ed. (1984), LANDSAT-4 Science Investigations Sum- Maritorena, S., Morel, A., and Gentili, B. (1994), Diffuse reflec-

mary, Including December 1983 Workshop Results, NASA tance of oceanic shallow waters: influence of water depth andConference Publication 2326, National Aeronautics and Space bottom albedo. Limnol. Oceanogr. 39:1689–1703.Administration, Scientific and Technical Information Branch, Miller, R. L., and Cruise, J. F. (1995), Effects of suspendedWashington, DC. sediments on coral growth: evidence from remote sensing and

Bour, W., and Pichon, M. (1997), Discrimination of scleractinian- hydrologic modeling. Remote Sens. Environ. 55:177–187.dominated from other reef communities using SPOT satellite Mumby, P. J., Green, E. P., Clark, C. D., and Edwards, A. J.imagery. In 8th Int. Coral Reef Symp. (H. A. Lessios and I. (1998). Digital analysis of multispectral airborne imagery ofG. Macintyre, Eds.), Smithsonian Tropical Research Institute, coral reefs. Coral Reefs 17:59–69.Balboa, Panama, pp. 1487–1490. Smith, R. C., and Baker, K. S. (1981). Optical properties of the

Chandrasekhar, S. (1960), Radiative Transfer. Dover, New York, clearest natural waters. Appl .Opt. 20:177–184.393 pp. Stamnes, K., Tsay, S.-C., Wiscombe, W., and Jayaweera, K.

Dustan, P., and Halas, J. (1987), Changes in the reef-coral popula- (1988). Numerically stable algorithm for discrete-ordinate-tion of Carysfort Reef, Key Largo, Florida, 1975–1982. Coral method radiative transfer in multiple scattering and emitting

layered media. Appl. Opt. 27:2502–2509.Reefs 6:91–106.