INT. J. REMOTE SENSING,
2003,
VOL.
24,
NO.
13, 2683–2702
Underwater light characterisation for correction of remotely sensed images EVANTHIA KARPOUZLI{, TIM MALTHUS{, CHRIS PLACE{, ANTHONY MITCHELL CHUI{, MARTHA INES GARCIA{ and JAMES MAIR§ {Department of Geography, University of Edinburgh, Drummond St, Edinburgh EH8 9XP, Scotland, UK; ?e-mail:
[email protected] {Corporation for the Sustainable Development of the Archipelago of San Andres, Old Providence and Santa Catalina-CORALINA, San Andres Island, Colombia §Department of Civil and Offshore Engineering, Heriot-Watt University, Edinburgh EH14 4AS, Scotland, UK
Abstract. Objective measurement of habitat change using remote sensing requires processing of the images to enhance the bottom reflectance signal. This process typically uses correction techniques to remove the influence of the water column on bottom reflectance, and to enable the accurate correction of the imagery for varying bathymetry. Such correction measures depend on reliable estimates of water column light attenuation. An investigation into the spatial variation in attenuation in a typical tropical region was undertaken. Measurements of gross spatial variations in downwelling attenuation around San Andres and Old Providence islands in the western Caribbean were made using a PAR sensor. Measurements of specific attenuation were also made for blue, green and red light using filters fitted to the sensor. High spectral resolution attenuation measurements were also made using a spectroradiometer. Results showed a four-fold variation in light attenuation in shallow littoral regions alone. Spectral attenuation measurements suggested that this variation was largely the result of scattering by particulate matter rather than varying concentrations of dissolved yellow substances. These findings suggest that the results of studies where single measurements of ‘average’ attenuation have been used to depth-correct remotely sensed imagery should be interpreted with a high degree of caution. The paper goes on to show that simple models can be empirically obtained where attenuation can be spatially predicted with confidence, based on the variables of water depth, distance to and size of mangrove beds, and distance to and size of towns. The models obtained showed high statistical significance, with 89% and 78% of the spatial variation in attenuation explained for San Andres and Old Providence, respectively. It is postulated that the use of such approaches for the estimation of attenuation will lead to more accurate depth correction and hence improved interpretation of remotely sensed imagery for littoral regions.
Presented at the Remote Sensing of the Marine Littoral Environment Meeting, Linnean Society, London, 15–16 December 1999. International Journal of Remote Sensing ISSN 0143-1161 print/ISSN 1366-5901 online # 2003 Taylor & Francis Ltd http://www.tandf.co.uk/journals DOI: 10.1080/0143116031000066972
2684
E. Karpouzli et al.
1.
Introduction Optical remote sensing offers a non-invasive technique with which to rapidly monitor changes in the cover and health of submerged habitats. However, its full potential is still to be exploited in littoral environments, where the strong attenuating influence of the water column has been a limiting factor. A combination of optical properties in the water column (absorption, scattering) results in a significantly reduced and spectrally altered bottom reflectance. This impact gets greater with increasing depths and with waters of greater turbidity (Spitzer and Dirks 1987, Gould and Arnone 1998). To extend the potential of optical remote sensing in littoral applications, such as for monitoring coral reef health and for qualitative and quantitative monitoring of seagrass habitats, the influence of the water column must be removed from the remotely sensed image. Such a process also significantly improves the accuracy of classification of such habitats (Mumby et al. 1998). Despite the existence for some time of a number of algorithms for correcting for water column depth and turbidity effects (e.g. Lyzenga 1978, 1981, Moussa et al. 1989, Bierwirth et al. 1993), few studies of tropical marine habitats attempt such correction techniques before classification. Mumby et al. (1998) reported only four studies out of forty-five (9%) that attempted such pre-processing methods and concluded that authors were generally unaware of such correction methods. In order to address the confounding influence of the water column in images of the littoral zone, a knowledge of water depth and the local light attenuation (Kd ) is necessary. Few studies have used independently acquired estimates of attenuation, with most extracting Kd values for relevant bands directly from their imagery in areas of uniform bottom type (e.g. sand) and known depth (e.g. Lyzenga 1981, Bierwirth et al. 1993). Nearly all of these studies assume that Kd values extracted for one area can be applied to other regions. However, it is likely that attenuation in the tropical littoral zone spatially may be highly variable. Green et al. (2000) acknowledge this fact with their recommendation to segment images into regions with different water quality for separate water column correction. However, this idea has not yet been tested. This paper reports the first attempt to investigate the nature of the spatial variation in water column attenuation for a tropical littoral region and infers its potential influence on calculated bottom reflectance. 2.
Study site description This study focuses on the littoral habitats of the Archipelago of San Andres, Old Providence and Santa Catalina, Colombia, an expanse of 300 000 km2 isolated within the western waters of the Caribbean Sea, approximately 180 km east of the Nicaraguan coast and 800 km north-east of the Colombian coast (figure 1). Its two main islands are: San Andres (12‡ 34’ N; 81‡ 43’ W), and Old Providence (13‡ 22’ N; 81‡ 23’ W). The archipelago is surrounded by trenches and faults of very deep water of the order of 2–3 km. Both islands are located on the tectonic fracture of the Lower Nicaraguan Rise and have a similar volcanic history but San Andres does not exhibit the later volcanic reactivation observable in Old Providence (Geister and Diaz 1996). Significant increases in the human population migrating to San Andres from the 1950s through to the 1980s saw a dramatic rise in population from 5675 inhabitants in 1952 to around 80 000 by 1992, making it the most densely populated island in the whole of the Caribbean (Vollmer 1997). Still today San Andres is by far the
Remote Sensing of the Coastal Marine Environment
Figure 1.
2685
Map showing location of the islands of San Andres and Old Providence in the south-western Caribbean Sea.
most developed island of the archipelago and a major Colombian national tourist destination. San Andres has a smooth topography and an area of approximately 27 km2 . A series of hills runs the length of the island not exceeding 90 m in elevation. The main extent of its platform is to the east and north-east of the island bordered by a barrier reef with depths ranging between 1 m and 15 m before dropping rapidly to over 1000 m at the edge of the platform. The lagoonal area behind the reef receives some input of diffuse organic pollution since most urban development is concentrated in the north
2686
E. Karpouzli et al.
end of the island in the town of San Andres, along with the main harbour, the airport and a number of hotel resorts. Census data from 1990 indicate that over 70% of the population lives in this northern sector of the island (Diaz et al. 1995). The lagoon enclosed by the barrier reef to the north-east has a limited influence from the open sea and relatively calm waters. The municipal sewage system currently only covers 8% of the island, and is diverted to the west coast where the platform is much narrower and plunges steeply to deep waters. The remainder is handled by septic tanks — most of which do not meet technical specifications and leach effluent rapidly into the sea through the porous limestone substrate, and direct discharge into the sea and gullies (M. W. Howard, 2000, personal communication). In contrast to San Andres the more recent volcanic history of Old Providence has given rise to a slightly smaller island (18 km2 ) but with a mountainous landscape (peak at 330 m) and a number of small and often ephemeral fresh-water streams that flow down to the coastline. Flat land is scarce and is where the main settlements can be found, its population today is still under 4500 inhabitants. The platform surrounding Old Providence is more extensive at between 5 m and 10 m depth. To the north, it extends for 60 km with the second longest barrier reef in the Caribbean bordering its eastern side. The substrate is finer darker silty sand comparing with the white coarse sediment of coraline origin of the platform around San Andres. The typical submerged habitats found around both islands are seagrass (mainly Thalassia and Syringodium genera) and algal beds in different proportions, soft and hard coral habitats, as well as sandy and rocky substrates. A number of mangrove habitats of different sizes are also scattered around the coastlines of both islands, which represent an extra source of natural eutrophic waters. The effects of increased tourism, population growth and uncontrolled development to both islands in recent years but mostly to San Andres have had a significant effect on the clarity of the surrounding waters as they have been receiving increasing inputs of organic pollution and increased boat traffic. Old Providence has also been subject to recent deforestation of the hillsides, particularly from cattle raising which may have caused increased sedimentation through the stream outflows. 3. Methods 3.1. Broad-band irradiance measurements Measurements of gross spatial variations in downwelling PAR attenuation at stations around both San Andres and Old Providence islands were made using a submersible PAR cosine quantum irradiance meter (Macam model Q203 PAR, sensor 5638). In addition to broad-band PAR measurements, profiles of specific attenuation were also made for red, green, and blue light using filters fitted to the sensor closely matching the three visible bandwidths of the Landsat Thematic Mapper2 sensor. The sensor was fixed facing upwards on a weighted lowering frame and lowered below the water surface. Measurements were made at 1 m, 0.5 m or 0.25 m intervals to a maximum of 10 m depending on the depth and degree of attenuation of the water column. In that way a vertical profile comprising typically ten measurements was produced for each station. Measurements were referenced to above-surface incident irradiance using an identical continuously logging PAR cosine sensor on deck to correct for fluctuations in surface incident flux due to drifting clouds. An electronic damping circuit, part of the submersible meter, was also used to temporally average the readings, to smooth out rapid fluctuations in
Remote Sensing of the Coastal Marine Environment
2687
irradiance intensity produced by wave action. All measurements were made between 10:00 and 15:00 hours local time each day, to minimise solar angle effects. The measurements were made at selected sites chosen around San Andres, and Old Providence after dividing up the coastal zone into arbitrary ‘optical zones’ and ensuring that at least one sampling site was contained in each. The division was made from estimates of the optical properties of the water column from Secchi disc depths that are routinely carried out by the collaborating institution CORALINA, the natural resource management agency that represents the Colombian national environment system (SINA) in the Archipelago. Measuring site positions were located to an accuracy of less than 4 m using a differential global positioning satellite (GPS) system. The first set of PAR measurements (AS1–AS25 and AP1–AP18) took place over a one-month period (first collection period-A) from 16 April 1999 to 14 May 1999. During that period (A) a total of 25 profiles (AS1–AS25) of downwelling PAR irradiance were measured around San Andres and 18 around Old Providence (AP1– AP18). A second set of 21 PAR measurements were carried out in San Andres a few weeks later (BS1–BS24) during the period 6 June 1999 to 23 June 1999 (second collection period – B). The second set of PAR measurements around San Andres (B) included a number of sites that were also sampled during the first collection period (A) which allowed for comparisons of temporal changes in attenuation. Pairs of stations within 300 m distance from each other were considered to be the same site, and subsequently 11 pairs of measurements were compared. These stations, in close proximity, were used to test the hypothesis that temporal factors did not influence the Kd (PAR) measurements, at least within the time frame that this study took place. To test this a paired-sample t-test was carried out for the 11 paired stations around San Andres. A set of band-specific measurements using the red, green and blue filters were made around San Andres during the second collection period (B) at most stations at which PAR measurements were made. A set of band-specific downwelling measurements was also made around Old Providence and Santa Catalina Islands between 25 May 1999 and 2 June 1999. For all downwelling measurements, the diffuse vertical attenuation coefficients (Kd ) were calculated as the linear slope coefficient of the logarithm of downwelling irradiance (of PAR, red (R), green (G), or blue (B) bands) with respect to depth (Kirk 1994). The majority of regressions between ln-irradiance and depth gave R2 values of 0.95 or above. Poor linear relationships were omitted. 3.2. Spectral measurements In conjunction with the first set of PAR measurements, high spectral resolution attenuation measurements were made at selected sites around San Andres and Old Providence during the first collection period. All spectral measurements were made using a GER 1500 spectroradiometer system comprising of two radiometers linked to a notebook computer. The GER 1500 is a rapid scanning radiometer, using a linear photodiode array to measure radiance over the visible to near-infrared wavelength range (300–1100 nm) with a nominal dispersion of 1.5 nm and resolution of 3 nm. One radiometer sensor head was fitted with a 4 m fibre optic probe and cosinecorrected sensor and used to make measurements of incident downwelling irradiance. The cosine-corrected sensor was fixed on a lowering frame and, while kept horizontal,
2688
E. Karpouzli et al.
was lowered and triggered at 0.5 m intervals down to 2.5 m below the water surface. The second sensor head was fitted with a 15‡ field-of-view receptor and was used to reference the downwelling measurements to above-surface incident irradiance reflectance from a calibrated Spectralon2 reflectance panel. Both instruments were triggered simultaneously in dual field-of-view mode using the notebook computer, thus obtaining pairs of target and reference measurements. 3.3. Laboratory analysis Samples of 5 l of water from ten sites in San Andres and two of 4 l from Old Providence were collected and filtered though Whatman GF/C glass fibre filters under low vacuum and stored until analysis for determination of absorption by aquatic humus (dissolved colour). After transportation to the UK these samples were further filtered through 0.2 mm membrane filters and the absorption of the filtrate determined at 380 nm in a Perkin-Elmer Lambda 40 spectrometer using 10 cm pathlength cuvettes and a reference of distilled water. In addition to aquatic humus absorption measurements, the samples were also analysed for chlorophyll concentration (performed on site). Particulate matter was collected on Whatman GF/C filters, the pigments were extracted using boiling ethanol and absorption of the extract measured in a spectrophotometer, following the method of Moed and Hallegraeff (1978).
3.4. Predictive models for spatial estimation of attenuation The construction of simple models to predict spatial variation in attenuation was attempted for the separate San Andres and Old Providence attenuation datasets. A number of predictor terms were considered and their significance to Kd tested using stepwise multiple regression, performed using the Systat2 software package. Terms were validated by being added and removed from the analysis to lead to an optimal subset of variables that gave the best possible regression equation. The predictor terms that were considered are listed below. . Water depth (D or 1/D) – the depth of the water column at each sampling
station, measured during sampling using a hand-held echosounder. . Influence of mangroves (MA /LM or MA /LM 2 ) – a cumulative predictor
hypothesized to influence Kd as a factor of mangrove bed surface area (MA ) and their distance from the sampling stations (LM ). In the absence of detailed information on current directions around the islands it was assumed that all mangroves on the coast facing the particular station would have some influence on it. Thus, sampling stations were divided into east or west groups and only the east coast mangroves were considered to have an effect on the east-lying stations and the contrary for the west-lying stations. . Influence of major towns (TA /LT or TA /LT 2 ) – In the absence of detailed population estimates for the different towns or of waste water volumes from different sources, the area of the settlements (TA ) and their distance from the sampling stations (LT ) was used to form this cumulative predictor. As for the mangroves, only those coastal towns on the east or west coasts were considered, depending on the particular location of the sampling station. . Influence of river and sewage outlets (1/LR or 1/LR 2 ) – a cumulative predictor
Remote Sensing of the Coastal Marine Environment
2689
estimated assuming that their influence on Kd is inversely related to their distance (LR ) from a particular sampling station. The predictor was then the sum of the inverse distances from the outlets to the given sampling point or the sum of the distances squared. Again, only east or west outlets were considered, depending on sampling station location. 4. Results 4.1. Analysis of water samples The water sample analysis revealed very low aquatic humus concentrations for both islands (table 1). The highest measurement (0.1 m{1 ) was found to be Station AS8 at the outflow of the largest mangrove bed of San Andres island (mangrove Bahia Hooker), followed by station AS7 off the main port (figure 2). Absorption by aquatic humus in these islands is high compared to values for other Caribbean and tropical regions reported by Kirk (1994), although most of those reported are for more open oceanic waters. Similarly, chlorophyll concentrations showed very low pigment concentrations ranging from 0.13 mg m{3 , at station AS5 north of the port of San Andres, to 0.98 mg m{3 at station AP5 off Santa Catalina harbour on Providence (figures 2 and 3). In general, the highest values for San Andres were found at stations AS11, AS23, AS24 and AS4 which were located near mangrove Bahia Hooker, off a sewage outlet and by coastal tourist developments, respectively. Surprisingly, comparatively lower concentrations of chlorophyll were found at station AS7 off the main port of San Andres while the same station showed high absorption by aquatic humus (figure 2). From a total of 11 samples taken in May and September 1992, Diaz et al. (1995) found chlorophyll concentrations around San Andres ranging from 0.013–0.492 m2698g m{3 . Although the chlorophyll measurements of both Diaz et al. (1995) and this study are limited in number, they may indicate a slight increase in chlorophyll concentration around the island over the seven-year period. Both Providence stations showed overall higher pigment concentrations but lower concentrations of aquatic humus when compared to San Andres. Overall, the Table 1.
Concentrations of aquatic humus and chlorophyll from a number of sites around San Andres and Old Providence islands.
Station San Andres AS20 AS21 AS22 AS23 AS24 AS1 AS4 AS5 AS7 AS8 Old Providence AP5 AP11
Total pigment (mg m{3 )
Aquatic humus absorption at 380 nm (m{1 )
1999 1999 1999 1999 1999 1999 1999 1999 1999 1999
0.21 0.29 0.30 0.40 0.36 0.30 0.36 0.13 0.18 0.90
0.069 0.091 0.075 0.035 0.043 0.046 0.063 0.056 0.100 0.160
28 April 1999 28 April 1999
0.98 0.68
0.080 0.079
Collection date 16 16 16 16 16 19 19 19 19 19
April April April April April April April April April April
2690
E. Karpouzli et al.
Figure 2. Variation in measured PAR attenuation (Kd ) in coastal waters around the island of San Andres. Sampling stations marked A denote first sampling period, B second sampling period.
Remote Sensing of the Coastal Marine Environment
Figure 3.
2691
Variation in measured PAR attenuation (Kd ) in coastal waters around the island of Old Providence during the first sampling period.
measurements for both islands suggest waters of high transparency but indicate localised eutrophication effects. 4.2. Downwelling PAR measurements – temporal variations Table 1 compares the Kd values for the first collection period stations (A) that were re-visited during the second collection period (B). Their locations are shown in figure 2.
2692 Table 2.
E. Karpouzli et al. Comparison of Kd (PAR) values for revisited stations around San Andres island corresponding to those shown in figure 2.
Date
Station
13 May 1999 16 April 1999 16 April 1999 16 April 1999 22 April 1999 22 April 1999 14 May 1999 15 June 1999 14 May 1999 7 June 1999 14 May 1999
AS2 AS23 AS22 AS21 AS19 AS19 AS17 BS13 AS14 BS3 AS9
Kd (m{1 ) 0.139 0.081 0.193 0.161 0.088 0.088 0.299 0.249 0.262 0.157 0.272
Station BS4 BS22 BS21 BS19 BS18 AS20 BS12 BS12 BS9 BS2 AS8
Kd (m{1 ) 0.110 0.130 0.139 0.108 0.113 0.045 0.287 0.287 0.304 0.153 0.292
No consistent pattern in variation in attenuation is evident, although differences do exist at some stations. The results of a paired-sample t-test on the data indicated no significant consistent differences due to sampling period (n~11, t~20.511, pw0.05). The power of the test for detecting a true difference of at least 0.05 m{1 in Kd at the revisited stations was 0.98; thus 98% of the time such a difference would be detectable using a sample of 11 stations at a significance level of 0.05. Therefore, differences between stations were considered to be more due to spatial as opposed to temporal variation in light attenuation. Subsequent analysis of the dataset treated all the data combined, irrespective of sampling period. 4.3. Downwelling PAR measurements – spatial variations Tables 3 and 4 present a comparison of the calculated downwelling diffused attenuation coefficients for PAR (Kd ) at the different sampling stations around the two islands, along with the depth of the euphotic zone (Zeu ) for each site, the latter estimated as the depth of penetration of 1% of sub-surface irradiance light level. Spatial variation in Kd around the islands is also shown in diagrammatic form in figures 2 and 3. Attenuation around San Andres ranged from 0.05–0.57 m{1 , corresponding to a range of depth in the euphotic zone of 102 m down to 8.1 m. The highest attenuation was measured at locations closest to the port, the mangrove areas and nearest to the developed coastline by the major towns. The lowest attenuation values were found off the north-west coast and at the southernmost point of the island (figure 2). Similarly, for Old Providence attenuation ranged from 0.06–0.38 m{1 , corresponding to variation in euphotic zone depths of 74.2 m down to 12.1 m. The greatest attenuation was measured near the harbour by the town of Santa Isabel and along the west coast of the island where most of the tourist development, made up of small inns and cabins, is found (figure 3). 4.4. Downwelling waveband-specific measurements Along with the PAR profiles for San Andres collected during the second data collection period (B), 22 band-specific profiles (BSF1–BSF25) were obtained using
Remote Sensing of the Coastal Marine Environment Table 3.
2693
Values of Kd and depth of euphotic zone for stations around San Andres island during the first and second collection periods (A and B).
Station East coast, north to south BS4 AS2 AS1 AS3 BS5 AS4 BS6 AS5 AS6 AS7 AS8 AS9 AS10 AS11 AS12 AS13 AS14 BS9 BS10 AS15 AS16 BS11 BS12 BS13 AS17 AS18 BS14 BS15 BS16 West coast, south to north BS18 AS19 AS20 BS19 AS21 BS20 BS21 AS22 BS22 AS23 BS23 AS24 BS24 BS1 AS25 BS2 BS3
Kd (PAR) (m{1 )
Zeu (m)
0.11 0.14 0.11 0.13 0.30 0.25 0.29 0.17 0.17 0.12 0.29 0.27 0.25 0.57 0.21 0.30 0.26 0.30 0.17 0.10 0.22 0.13 0.29 0.25 0.30 0.26 0.18 0.12 0.11
41.82 33.20 41.40 36.60 15.59 18.10 16.03 27.80 26.80 39.70 15.70 16.90 18.20 8.10 21.90 15.20 17.60 15.13 26.90 48.40 20.90 35.94 16.03 18.47 15.40 17.40 26.29 39.66 41.07
0.11 0.09 0.05 0.11 0.16 0.27 0.14 0.19 0.13 0.08 0.08 0.12 0.07 0.16 0.07 0.15 0.16
40.71 52.00 102.2 42.59 28.60 16.85 33.09 23.80 35.38 56.80 56.79 38.90 66.67 28.75 70.00 30.07 29.30
Station locations correspond to those in figure 2.
2694 Table 4.
E. Karpouzli et al. Values of Kd and depth of euphotic zone for stations around Old Providence island during the first collection period (A).
Station West coast, north to south AP1 AP2 AP3 AP4 AP5 AP6 AP7 AP8 AP9 AP10 AP11 AP12 East coast, south to north AP13 AP14 AP15 AP16 AP17 AP18
Kd (PAR) (m{1 )
Zeu (m)
0.06 0.17 0.14 0.25 0.31 0.18 0.18 0.22 0.19 0.23 0.38 0.09
74.2 27.9 33.6 18.8 14.9 26.0 25.4 21.3 23.8 19.8 12.1 54.1
0.19 0.12 0.21 0.16 0.22 0.18
24.2 37.4 22.1 28.9 20.9 25.4
Station locations correspond to those in figure 3.
red, green and blue colour filters. A further 15 profiles were also obtained for Old Providence (PF1–PF15). The calculated Kd values for red, green and blue light are given in tables 5 and 6 and graphically illustrated for blue light in figures 4 and 5. As can be seen from figures 4 and 5, the spatial variation of attenuation in an individual band closely followed the pattern of the PAR measurements, being highest at the same stations that PAR attenuation was highest. Lowest attenuation was almost always observed at the blue region, suggesting that blue light is the most penetrative (Tables 6 and 7). Red light was most attenuated, attributed to the influence of the greater absorption of the light in this region by water itself. In the few cases that green band attenuation approached or was less than that for blue light, these were generally in the most turbid areas or where it might be expected localised eutrophication effects were operating (e.g. near ports and mangroves). 4.5. Spectral measurements Twelve attenuation spectra from San Andres and 14 from Old Providence were calculated from measurements made using the spectroradiometer fitted with the fibre optic probe (figure 6). Station numbers correspond to the PAR profiles from figures 2 and 3. Little variation in spectral attenuation was observed between the different stations, and between the islands, where the shape of attenuation was largely influenced by the absorption properties of water itself across the visible spectrum. This suggests that differences between stations were largely the result of differences in scattering caused by varying concentrations of particulate matter. Some stations (AS8, AP10, AP11) showed increased blue light attenuation in areas
Remote Sensing of the Coastal Marine Environment Table 5.
2695
Values of band-specific Kd measured using colour filters for stations around San Andres island.
Station
Kd Blue (m{1 )
Kd Green (m{1 )
Kd Red (m{1 )
BSF1 BSF2 BSF3 BSF4 BSF5 BSF6 BSF7 BSF8 BSF9 BSF10 BSF11 BSF13 BSF14 BSF15 BSF16 BSF17 BSF18 BSF19 BSF20 BSF21 BSF22 BSF23 BSF24 BSF25
0.13 0.09 0.10 0.10 0.22 0.15 0.20 0.54 0.31 0.15 0.16 0.33 0.10 0.17 0.12 – 0.04 0.10 0.36 0.05 0.12 0.07 0.05 0.08
0.12 0.18 0.14 0.15 0.21 0.16 0.23 0.52 0.29 – 0.16 – 0.16 0.16 0.12 – 0.12 0.12 0.55 0.15 0.22 0.13 0.13 –
0.43 0.43 0.42 0.42 0.43 0.49 0.47 0.66 0.58 – 0.43 – 0.54 0.39 – 0.39 0.41 0.41 0.62 0.45 0.46 0.41 0.50 –
Station numbers correspond to those in figure 4. Table 6. Station PF1 PF2 PF3 PF4 PF4 PF5 PF6 PF7 PF8 PF9 PF11 PF12 PF13 PF14 PF15
Values of band-specific Kd measured using colour filters for stations around Old Providence island. Kd Blue (m{1 )
Kd Green (m{1 )
Kd Red (m{1 )
0.05 0.13 0.21 0.05 0.41 0.28 0.11 0.34 0.16 0.13 0.13 0.16 0.20 0.32 0.20
0.18 0.17 0.21 0.28 0.38 0.34 0.19 0.38 0.13 0.21 0.25 0.18 0.27 0.40 0.16
0.44 0.49 0.98 0.36 0.59 0.53 0.52 0.66 0.49 0.42 0.44 0.56 0.50 0.50 0.50
Station numbers correspond to those in figure 5.
where increased turbidity of water was encountered either as a result of proximity to mangroves or probably due to effects of wind-induced mixing and resuspension of particulate matter emanating from mangrove beds.
2696
Figure 4.
E. Karpouzli et al.
Variation in measured blue light attenuation (Kd , B) in coastal waters around the island of San Andres.
4.6. Predictive models results To process a remotely sensed image to bottom reflectance in order to investigate underwater habitats and to enable the accurate correction of the imagery for varying bathymetry, an accurate measure of the water column attenuation at each
Remote Sensing of the Coastal Marine Environment
Figure 5.
2697
Variation in measured blue light attenuation (Kd , B) in coastal waters around the island of Old Providence.
pixel is required. However, the results of the spatial survey indicate a high degree of variation in light attenuation in the waters around both islands under study. Thus, if a single measure of ‘average’ attenuation is used for water column correction of imagery, the accuracy of the result may be questionable. However, it is possible to generalise some factors about the behaviour of light attenuation around the islands. First, Kd values generally decrease with greater
2698
Figure 6.
E. Karpouzli et al.
Spectral attenuation at some San Andres and Old Providence sampling stations.
Remote Sensing of the Coastal Marine Environment
2699
distance from the shore of each island. Second, Kd values are generally lower over deeper waters than shallower ones. These trends would suggest that it may be possible to develop simple relationships which may be used to explain and map the distribution in attenuation in such waters. Thus, to better understand the sources of variation in water attenuation around the coastal zones of San Andres and Old Providence, multiple regression analysis was employed to develop the simple models to estimate PAR attenuation. A predictive model to estimate Kd in areas where such measurements were not made would also give rise to better estimation of attenuation as opposed to direct interpolation between the measured Kd points. For San Andres the measurements of PAR attenuation for both the first and second collection periods were used (table 2) whereas for Old Providence the values in table 5 were used. Deep water sites were not included in the analysis since the precise water depth for these sites was not known. For San Andres the strongest correlation between attenuation and a single variable was with water depth: ð1Þ Kd ~0:061z0:673:1=D, n~44, r2 ~0:752, pv0:001, indicating the major influence of depth in determining relative water clarity around the island. With the addition of further variables, it was found that the regression was improved to Kd ~0:0778z0:4781:1=Dz0:0074MA =L2 z0:0028TA =L2 , n~44, r2 ~0:890, M
pv0:001,
r
ð2Þ
With the elimination of Station AP11 (Southwest Bay) from the dataset for Old Providence island, which was indicated as an outlier, depth was also a strongly correlating variable with Kd : ð3Þ Kd ~0:069z0:646:1=D, n~16, r2 ~0:633, pv0:01, The relationship between Kd and depth for Old Providence is very similar to that obtained for San Andres above. With the addition of more variables into stepwise regression analysis the model was improved to: ð4Þ Kd ~0:0757z0:4145:1=Dz0:0006TA =Lr , n~16, r2 ~0:78, pv0:05 The addition of further variables did not improve the models of both islands. Correlations between individual variable terms included in the regressions were all non-significant (Pearson correlation coefficient rv0.5 in all cases) indicating their relative independence. 5.
Discussion This investigation has shown that there was a three- to four-fold spatial variation in the attenuation of tropical ‘clear’ littoral waters around the islands of San Andres and Old Providence in the south-western Caribbean Sea, corresponding to extremely wide variations in the depth of light penetration around both islands. The general similarity in the shape of high spectral resolution attenuation spectra made using a spectroradiometer suggested that this variation was largely the result of attenuation due to scattering by particulate matter in the water column, rather than varying concentrations of dissolved yellow substances. Broad-band spectral measurements indicate that blue light was the most penetrating. Although there is some evidence for slightly increasing chlorophyll concentrations around San
2700
E. Karpouzli et al.
Andres, light is still of sufficient quality to support a wide variety of plant and coral habitats. The limited sampling over time in this study (April to June) indicated that there was little influence of a temporal component to variation in Kd (table 1). However, this would need greater investigation in order to ascertain the degree of variation over the different seasons. These results therefore suggest that it may be highly inappropriate to use a single or few measures of ‘typical’ attenuation obtained in deep water or shallow areas for water column correction of remotely sensed images acquired over tropical regions where reflectances are influenced by both variations in water depth and bottom type. It further suggests that the results of such corrections should be interpreted with a high degree of caution. Deep water attenuation values tend to be much lower than those over related littoral zones and, as a result of scattering by varying concentrations of particulate matter, attenuation values in shallower waters are highly variable. Clearly, field data from deep water sites, as well as in the proximity of more turbid or organic-rich water sources, are required to objectively correct remotely sensed images for water column attenuation. In this study, shallower waters generally showed greater attenuation compared to more open and deeper sites. However, there was considerable variation in attenuation in shallow regions alone, with up to three-fold variations encountered. It is thought this variation was a function of: . . . .
depth of the water column; proximity of the sampling stations to the coastline; proximity to centres of population or mangrove and lagoon environments; local currents.
That attenuation was lowest further away from shore tends to indicate the decreased influence of land processes and deeper waters which both encourages settlement of sediments and discourages their re-suspension. From the empirical modelling studies, water depth had the largest influence on Kd . Deeper water would be expected to be both further away from the shoreline and clearer, due to the larger volume of water over which turbid particles may diffuse, and also due to lessened influence of sandy substrates where sediment may become re-suspended. The results from the multivariate regression analysis gave very similar models for both islands for the prediction of attenuation from water depth alone, suggesting that a general model for both islands would be obtainable. For both islands, predictions of Kd were significantly improved by including terms for the proximity and hence influence of, mangrove beds (known to be significant sources of suspended sediments) and towns. The overall strength of the equations developed for San Andres was high, with nearly 90% of the variation in Kd explained by variations in depth, distance to and size of mangroves and distance to and size of towns. Relationships for Old Providence were less strong, with only 78% of the variation in Kd explained by depth and distance to and size of towns. This may be partially the result of the lower number of sampling stations measured around this island but may also suggest that other factors that were not included in the analysis may have been responsible in part for the variation encountered in Kd . One such variable not included is the influence of current strengths and directions. These may have had more influence around Old Providence island — where the
Remote Sensing of the Coastal Marine Environment
2701
platform is more exposed to the predominantly north-eastern wind-induced currents as the barrier reef is discontinuous around where stations were located (pinnacle belt) — than for San Andres (Diaz et al. 1996, 1997, Geister and Diaz 1996). Current hydrologies around both islands may thus be markedly different. The results of this study suggest that, using the equations developed, Kd may be predicted spatially with a high degree of confidence for all locations on the platforms around both these tropical islands. This would then enable remotely sensed images to be more accurately corrected for water column effects than if based on ‘average’ attenuation values alone. It further suggests that similar relationships with driving variables may be obtained for other littoral areas. The models reported here were developed for broad PAR attenuation. For the correction of multi-spectral remotely sensed images over littoral regions, estimates of band-specific Kd would be required. However, the similarities in behaviour between the broad-band blue, green and red attenuation measurements made during this study and the broader PAR attenuation values suggests that models specific for individual sensor bands should be easily obtainable. More accurate measurements of distance to sources of turbid water may lead to improvements in the accuracy of the models. For example, populations of towns may be a better variable to describe the influence of towns as opposed to their size as was used in this study. Although rivers were ephemeral in this study region, the influence of river flows and sewage outfalls may also be more quantitatively related to their discharge volumes rather than simply distance from their locations. Furthermore, more detailed information on hydrology may allow greater understanding of the diffusion of pollutants in the water column and hence lead to better estimates of Kd . Nevertheless, the results from this study are particularly encouraging, given the limited information available for the study region. Acknowledgments The project was financially supported by the Darwin Initiative, the Onassis Foundation, the Carnegie Trust, the Small Project grant scheme of Edinburgh University, and the Moray Fund. It was conducted in collaboration with CORALINA in San Andres. The assistance of a number of staff during data collection is gratefully acknowledged. The GER 1500 spectroradiometer was obtained on loan from the Natural Environment Research Council Equipment Pool for Field Spectroscopy, UK. Special thanks are extended to Phil Lovell and Callan Duck for their valuable help during fieldwork, and Phil Lovell and Lex Hibby for helpful discussions and for comments which have greatly improved this paper. References BIERWIRTH, P. N., LEE, T. J., and BURNE, R. V., 1993, Shallow sea-floor reflectance and water depth derived by unmixing multispectral imagery. Photogrammetric Engineering and Remote Sensing, 59, 331–338. DIAZ, J. M., DIAZ-PULIDO, G., GARZON-FERREIRA, J., GEISTER, J., SANCHEZ, J. A., and ZEA, S., 1996, Atlas de los arrecifes coralinos del Caribe Colombiano. Vol 1. Complejos arrecifales oceanicos Serie Publicaciones Especiales, No. 2 (Santa Marta, Colombia: Instituto de Investigaciones Marinas y Costeras (INVEMAR)), 1–83. DIAZ, J. M., GARZON-FERREIRA, J., and ZEA, S., 1995, Los arrecifes coralinos de la Isla de San Andres Colombia: Estado actual y perspectivas para su conservacion (Bogota, Colombia: Academia Colombiana de Ciencias Exactas, Fisicasy Naturales). DIAZ, J. M., SANCHEZ, J. A., and GEISTER, J., 1997, Development of lagoonal reefs in oceanic
2702
Remote Sensing of the Coastal Marine Environment
reef complexes of the southwestern Caribbean: Geomorphology, structure and distribution. Proceedings of the 8th International Coral Reef Symposium, Panama City, Panama, Volume 1, pp. 779–784. GEISTER, J., and DIAZ, J. M., 1996, A field guide to the atolls and reefs of San Andres and Providencia (Colombia). In Proceedings of the 8th International Coral Reef Symposium, Panama City, Panama, Volume 1, pp. 235–236. GOULD, R. W., and ARNONE, R. A., 1998, Three-dimensional modelling of inherent optical properties in a coastal environment: coupling ocean colour imagery and in situ measurements. International Journal of Remote Sensing, 19, 2141–2159. GREEN, E. P., MUMBY, P. J., EDWARDS, A. J., and CLARK, C. D., 2000, Remote Sensing Handbook for Tropical Coastal Management Coastal Management Sourcebooks 3, edited by A. J. Edwards (Paris: UNESCO). KIRK, J. T. O., 1994, Light and Photosynthesis in Aquatic Ecosystems, 2nd edn (Cambridge: Cambridge University Press). LYZENGA, D. R., 1978, Passive remote sensing techniques for mapping water depth and bottom features. Applied Optics, 17, 379–383. LYZENGA, D. R., 1981, Remote sensing of bottom reflectance and water attenuation parameters in shallow water using aircraft and Landsat data. International Journal of Remote Sensing, 2, 71–82. MOED, J. R., and HALLEGRAEFF, G. M., 1978, Some problems in the estimation of chlorophyll a and phaeopigments from pre- and post-acidification spectrophotometric measurements. Internationale Revue gesamten Hydrobiologie, 63, 787–800. MOUSSA, H. B., VIOLLIER, M., and BELSHER, T., 1989, Remote-sensing of macrophytic algae in the Molene Archipelago: ground-radiometry and application to SPOT satellitedata. International Journal of Remote Sensing, 10, 53–69. MUMBY, P. J., CLARK, C. D., GREEN, E. P., and EDWARDS, A. J., 1998, Benefits of water column correction and contextual editing for mapping coral reefs. International Journal of Remote Sensing, 19, 203–210. SPITZER, D., and DIRKS, R. W. J., 1987, Bottom Influence On the Reflectance of the Sea. International Journal of Remote Sensing, 8, 279–290. VOLLMER, L., 1997, The History of the settling process of the Archipelago of San Andres, Old Providence and St. Catherine, 1st edn (San Andres, Isla: Ediciones Archipelago).