Distribution and seasonal biomass of drift macroalgae
Transcription
Distribution and seasonal biomass of drift macroalgae
Journal of Experimental Marine Biology and Ecology 326 (2005) 89 – 104 www.elsevier.com/locate/jembe Distribution and seasonal biomass of drift macroalgae in the Indian River Lagoon (Florida, USA) estimated with acoustic seafloor classification (QTCView, Echoplus) Bernhard M. Riegl a,*, Ryan P. Moyer c, Lori J. Morris b, Robert W. Virnstein b, Samuel J. Purkis a a National Coral Reef Institute, Nova Southeastern University Oceanographic Center, 8000 North Ocean Drive, Dania Beach, FL 33004, USA b St. John’s River Water Management District, 4049 Reid Street, Palatka, FL 32177, USA c Department of Earth and Environmental Sciences, University of Pennsylvania, Philadelphia PA 19104 (formerly NCRI, Nova Southeastern University), USA Received 13 September 2004; received in revised form 20 May 2005; accepted 20 May 2005 Abstract Three areas of the Indian River Lagoon, Florida (USA) were surveyed to show seasonal changes in the distribution and biomass of macroalgae and seagrass. Acoustic seafloor discrimination based on first and second echo returns of a 50 kHz and 200 kHz signal, and two different survey systems (QTCView and ECHOplus) were used. System verification in both the field and a controlled environment showed it was possible to distinguish acoustically between seagrass, sparse algae, and dense algae. Accuracy of distinction of three classes (algae, seagrass, bare substratum) was around 60%. Maps were produced by regridding the survey area to a regular grid and using a nearest-neighbor interpolation to provide filled polygons. Biomass was calculated by counting pixels assigned to substratum classes with known wet-weight biomass values (sparse algae 250 g m 2, dense algae 2000 g m 2, seagrass 100 g m 2) that were measured in the field. In three study areas (Melbourne, Sebastian Inlet, and Cocoa Beach), a dependence of algal biomass on depth and season was observed. Seagrass most frequently occurred in water less than 1 m deep, and in November, seagrass beds tended to be covered by dense algae that also extended up- and downstream of shoals in the Lagoon. In March, the pattern was similar, with the exception that some areas of previously dense algae had started thinning into sparse algae. Macrophyte biomass was lowest in May in the Melbourne and Cocoa Beach study areas, with the opposite situation in the Sebastian Inlet study area. In May, seagrass areas were largely devoid of dense algae and most algae accumulations were sparse. In August, dense algae covered large areas of the deep Lagoon floor while shoals were largely free of algae or had only sparse cover. We suggest this summer pattern to reflect moribund algae being washed from the shallows to deeper channels and from there being removed from the lagoonal ecosystem either through tidal passages into the open ocean or by degradation and breakdown in situ. The differences * Corresponding author. Tel.: +1 954 262 3671; fax: +1 954 262 4027. E-mail address: [email protected] (B.M. Riegl). 0022-0981/$ - see front matter D 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.jembe.2005.05.009 90 B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 between the study areas indicate high spatial and temporal variability in biomass and distribution of macrophyte biomass in the Indian River Lagoon. D 2005 Elsevier B.V. All rights reserved. Keywords: Acoustic seafloor discrimination; Algae; ECHOplus; Florida; Indian River Lagoon; Macroalgal biomass; QTCView; Seagrass 1. Introduction Seagrass and macroalgae beds are key ecosystems in the Indian River Lagoon (Florida, USA), which is a shallow, largely enclosed, coastal waterbody separated from the Atlantic Ocean by a series of barrier islands (Fig. 1). The Lagoon’s salinity is variable with tidal movements, but generally lower than normal marine (35 ppt) and water is typically turbid. There are seven species of seagrass present. The three most abundant, canopy forming species are Halodule wrightii, Syringodium filiforme, and Thalassia testudinum (Morris et al., 2000). These species of seagrass often occur in dense beds together with or adjacent to drift macroalgae where biomass has been estimated at 3 times, in extreme cases 100 times, that of seagrass (Morris and Hall, 2001). Due to their great abundance, drift macroalgae are considered to have a habitat value comparable to that of seagrass and since the densities of animals on the two vegetation-types are similar and they share about 75% of the same species (Virnstein and Howard, 1987), the drift macroalgae are generally considered an extension of the seagrass habitat. The estimation of drift algae biomass is important for the determination of overall nutrient release and/or uptake in the Indian River Lagoon system (Virnstein and Carbonara, 1985). Hence, maps of seasonal distribution are needed. An easy and accurate way of mapping shallow benthic habitats is by passive optical remote-sensing using airphotos or satellite imagery (multi- or hyperspectral, Green et al., 2000; Dierssen et al., 2003; Fig. 1. Study areas in the Indian River Lagoon, central Florida, USA. B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 Zimmermann, 2003). However, the typically turbid conditions in the Indian River Lagoon, largely driven by nutrient input and sediment re-suspension, usually render optical methods inappropriate. Acoustic seafloor discrimination provides an alternative that is not dependent on water clarity. The method is well established (Chivers et al., 1990; Preston et al., 2000; Lawrence and Bates, 2001; Ellingsen et al., 2002; Riegl and Purkis, 2005) and several commercially available systems exist (among others RoxAnn, Biosonics, Echoplus, QTCView; for reviews see Hamilton et al., 1999; Lawrence and Bates, 2001; Bates and Whitehead, 2001; Sabol et al., 2002; Kenny et al., 2003) that are capable of detecting differences in sediment types (Hamilton et al., 1999; Preston et al., 2000; Freitas et al., 2003a,b) and can differentiate artifacts from sediments (Lawrence and Bates, 2001). It has also been used to detect biotopes by jointly using biological and acoustic sampling (Freitas et al., 2003a,b). In this present study we examined the suitability of the QTCView Series V (Quester Tangent Co., Sidney, B.C.) and Echoplus (SEA Ltd, Bath, UK) systems to detect boundaries of areas covered by different densities of seagrass and drift algae in shallow waters of mostly less than 2 m depth. It was our goal to estimate seasonal biomass and map distribution in three test areas of the Indian River Lagoon (Melbourne, Sebastian Inlet, Cocoa Beach). 2. Materials and methods Three areas within the Indian River Lagoon, Florida, USA (Fig. 1) were surveyed in November 2002, March 2003, May 2003 and September 2003. The winter (November 2002) and summer (May 2002) surveys covered the biggest area in order to account for maximum seasonal variability in seagrass and algal standing stock. The Melbourne survey area was situated immediately to the south of the US 192 causeway (Fig. 1) and was characterized by a navigation channel (4 m) in the middle, shallowing to depths of less than 1 m on the Lagoon’s sides. Also, two headlands existed on the eastern shoreline, from which shoals extended towards the Lagoon’s center. The Sebastian Inlet survey area (Fig. 1) was situated immediately to the south of Sebastian Inlet. Maximum depth in the area was 3.6 m 91 while the shallowest areas were at 0.9 m. No deep navigation channel was found within the Sebastian Inlet study area. The Cocoa Beach survey area was situated between the Port Canaveral and Cocoa Beach causeways and was characterized by shallow sides and a deeper navigation channel in the middle, where water depths reached over 5 m. It had the deepest and widest central trough of all three study areas. Also, two headlands existed on the eastern shoreline, from which shoals extended towards the Lagoon’s center. Of these, the northern shoal was bigger. Both shoals were, at least over the survey period, permanent features. The surveys were conducted from a 7-m survey vessel equipped with a Trimble Ag132 global positioning system (dGPS) that used coast-guard beacon differential corrections to obtain real-time horizontal positioning accuracies of mostly less than 1.2 m horizontal dilution of precision. Data were logged as NMEA-GGA string, which encodes the horizontal accuracy of each position, to allow quality control of positioning during data processing. Geo-rectified aerial photographs of the survey areas were loaded into the softwares Fugawi and Hypack, that were interfaced with the dGPS unit to allow real-time monitoring of vessel position with respect to the imagery and planned survey lines. Sonar signals were obtained using a 50 kHz signal from a Suzuki TGN60-50H12L transducer and a 200 kHz signal from a Suzuki TGW50-200-10L omnidirectional transducer, both with 0.4-ms pulse width and a 5-Hz sampling frequency with a beam angle of 428 (50 kHz) and 128 (200 kHz), respectively, provided by a Suzuki 5200/ 6dB depth sounder. Signal saturation was avoided by an autogain control feature in both survey systems. Depth was determined by using the QTCView bottom picking algorithm, which was accurate to at least 10 cm (tested with bar checks; Brinker and Minnick, 1994) and by direct cross-checks of displayed bathymetry and measured depth with a weighted line at several sites. Survey lines were spaced variably, but mostly not more than 100 m apart. The principles of acoustic ground-discrimination based on single-beam echosounders employed by the systems are well-reviewed elsewhere (Chivers et al., 1990; Hamilton et al., 1999; Preston et al., 2000; Lawrence and Bates, 2001; Bates and Whitehead, 2001; Freitas et al., 2003a,b; Riegl and Purkis, 92 B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 2005) and will not be reiterated here. We used the QTCView Series 5 and Echoplus turn-key survey systems which perform acoustic ground discrimination based on the shape of sonar returns (Hamilton et al., 1999; Quester Tangent Corporation, 2002). The two systems are based on similar assumptions and record the characteristics of reflected waveforms to generate habitat classifications based on the acoustic diversity of collected echoes which encode scattering and penetration properties of different types of seafloor (Chivers et al., 1990; Preston et al., 1999; Hamilton et al., 1999). The typical process involves a hydrographic survey during which acoustic data are collected. In QTCView, these are recorded as timestamped, dGPS-geolocated digitized envelopes of the first echo. Data are then processed in the software QTC Impact and checked by the operator for correct time-stamps, correct depths, and correct signalstrengths. All signals that do not pass an appropriate level of quality control are discarded and not used for further processing. Data are also displayed on a bathymetry plot, where recorded depths are checked against the blanking (minimum recordable) and maximum depths set for the survey and any faulty depth picks are removed manually before further processing. In QTC Impact software, the echoes are digitized, subjected to a variety of analyses (cumulative amplitudes and ratios of cumulative amplitudes, amplitude quantiles, amplitude histograms, power spectra, wavelet packet transforms) by the acquisition software (Preston et al., 2001, 2004). After being normalized to a range between 0 and unity, they are subjected to Principal Components Analysis (PCA) for data reduction. The first three principle components of each echo are retained (called Q values), according to the assumption that these explain the majority of variability in the data set (Quester Tangent Corporation, 2002). Datapoints are then projected into pseudothree-dimensional space along these three components, and subjected to cluster analysis using a Bayesian approach (Quester Tangent Corporation, 2002). The user decides on the number of desirable clusters and also chooses which cluster is split how often. Clustering decisions are guided by three statistics that are offered by the program called bCPIQ (Cluster Performance Index), bChi2Q and bTotal ScoreQ. Total Score decreases to an inflection point which is da strong indication of best split levelT (Quester Tangent Corporation, 2002). CPI increases with more cluster splitting (Kirlin and Dizaji, 2000; Freitas et al., 2003b), while Chi2 decreases with more cluster splitting, reaching maximum/minimum values at optimal split level (Quester Tangent Corporation, 2002). In our case, the number of desired clusters was easily identified, since we knew from the surveys which seafloor classes were encountered. These were always 4 or fewer classes (dense algae, sparse algae, seagrass, and bare seafloor that was not further differentiated). Reviews of the functioning of the QTC system and critiques can be found in Hamilton et al. (1999), Hamilton (2001), Legendre et al. (2002), Preston and Kirlin (2002), von Szalay and McConnaughey (2002), Ellingsen et al. (2002), Freitas et al. (2003a,b), Riegl and Purkis (2005), Moyer et al. (in press). The QTC dataset was reduced to a three-column matrix consisting of a single x,y geo-referenced class category z that was obtained from the cluster analysis. Echoplus uses first and second echo. It timegates the first echo signal to ignore the first strong peak(s) and processes only the latter part plus the entire second echo. The Echoplus is similar to RoxAnn (Hamilton, 2001), which was extensively tested by Hamilton et al. (1999). Echoplus is entirely selfcontained and internally compensates for frequency, depth, power level and pulse length. Pulse amplitude and length are measured on every transmission, the outputs scaled accordingly, and absorption corrections factored in. The first echo is digitized and time-gated in a way that only its tail (backscatter component) is used for analysis along with the entire second echo. The measurements from first and second echo are collapsed into two indices, E1 and E2, for the first and second echo respectively. The user has no influence on the formation of these indices and collects a geo-referenced string of variables (latitude, longitude, E1, E2). All data above the 95th and below the 5th percentile were rejected as outliers and all data were normalized to the 95th percentile, resulting in a range between 0 and unity. First return (E1) was plotted against second return (E2) data in a scatterplot, which showed whether data clustered or not. In order to produce spatially continuous habitat maps and bfill in the blanksQ between survey lines, we resampled the irregular survey data (grid exclusively along the survey tracks) to a regular grid of 10- B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 m pixel size. Interpolation used a nearest-neighbor algorithm for categorical data, i.e. the classes produced by QTC cluster analyses (Davis, 2002; Riegl and Purkis, 2005) or geostatistics, in this case ordinary point kriging, based on the spatial autocorrelation inherent in landscape patterns, for the continuous variable output of the Echoplus (Matheron, 1971; Greenstreet et al., 1997; Middleton, 2000; Hamilton, 2001; Papritz and Stein, 2002; Walter et al., 2002). On categorical data, kriging is not the method of choice, since fractional classes, such as will be produced by kriging, appear non-sensical. To evaluate the accuracy of the acoustic ground discrimination, groundtruthing transects consisting of geo-referenced images obtained by video camera drops were collected. The habitat types observed in the videos were then compared to the extrapolated maps. Accuracy of these maps was then assessed using a confusion matrix approach (Ma and Redmond, 1995), which assesses how often classes are confused (i.e. mapped as something that groundtruthing proves them not to be). An Atlantis AUW-5600 color underwater camera was used to capture video images along groundtruthing survey lines. The video signal was time stamped and merged with positioning information. Incoming video with GPS and time-stamp information was recorded in Digital Hi-8 format using a Video Walkman. Total linear distances of 1.4 km and 5.5 km were surveyed in the Sebastian Inlet and Melbourne areas, respectively. In Cocoa Beach, groundtruthing was performed by a stratified-random arrangement of video-camera drops. Data were more equally spaced over the survey area, rather than along lines, such as in the other two study areas. Groundtruthing statistics were only calculated for the Melbourne and Sebastian River survey areas. 2.1. Verification In order to verify that the classes obtained by QTCView and Echoplus really reflected seagrass, sparse algae, dense algae and bare seafloor, two approaches were taken: (1) for verification in the study area where the survey vessel was positioned over a discrete habitat patch, the habitat was verified by means of video drop-camera or spot-dive, and then a small dataset, containing between 1000 and 1500 echo traces, was obtained over dense algae (approxi- 93 mately 2000 g wet weight m 2, which is equivalent to approximately 230 g dry weight, was spread in the transducer footprint to cover the entire insonified seafloor), sparse algae (approximately 250 g wet weight m 2 were concentrated as a single clump in the center of the transducer footprint), seagrass, and bare seafloor. (2) verification in the Nova SE University marina (comparable salinity and temperature to the field trial area), where collected drift algae were placed in various densities underneath a suspended transducer. Data files of 1500 echo traces were obtained with an empty basket large enough to encompass the entire area of the footprint. Then, 250 g of algae were added into the basket (bsparse algaeQ) and another file of 1500 echo traces was obtained, then, an additional 1750 g algae (bdense algaeQ) were added and another file of about 1500 traces was obtained. This procedure was repeated in 1.8 m and 1.3 m depth. Additionally, separate files with different settings of blanking depth and signal length were obtained to evaluate whether these factors had any influence. Since the transducer was in a stable position and not moving, signal stacking performed by QTC View did not affect footprint size, which could easily be determined by trigonometry. Data were then subjected to cluster analysis in order to evaluate whether discrimination was obtained. Because seagrass and drift algae differed between the study areas in species composition, length, density, and other factors important for echo formation, and also differences in composition of sediment (category bbare sedimentQ) between the study areas was unknown, we considered it unwise to use the verification files as calibration files, but rather as a reasonable approximation of expected conditions. If during the verification trials all seagrass, algae and sediment categories that could be encountered in the field had been tested and true calibration datasets produced, evaluation of survey data could have proceeded using discriminant functions (Davis, 2002) rather than cluster analysis. 2.2. Estimation of macroalgal biomass Biomass was estimated by counting the colorcoded pixels assigned to each substratum class in the extrapolated maps for each survey area. Values were calculated as total biomass in kg per sample 94 B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 Fig. 2. Illustration of the density of algae referred to in the text. (A) sparse algae (~ 250 g m 2 uncleaned wet weight) (B) dense algae (~ 2000 g m 2 uncleaned wet weight). Both within a footprint of 57 cm radius (outlined by a rope), as would be realistic with the 50 kHz transducer (428 opening angle) at 1.5 m depth. area. Uncleaned wet weight (including animals, water, dirt) biomass was calculated as number of 1 m2 pixels times 2 kg for dense algae, 0.25 kg for sparse algae. These values were the same as those used for the acoustic verification described above. When adequately cleaned and remeasured, the biomass of the clumps used for verification was compatible to that occurring naturally (Morris et al., 2000). Seagrass biomass was considered to be on average 0.1 kg per m2 (range between 0.4 and 0.04 kg m 2 wet weight, with more sparse than dense seagrass occurring in the study area. We therefore biased data towards a lower value). The sizes of the survey areas were calculated from the total number of pixels in the extrapolated maps times pixel-size. Biomass was only calculated for the November 2002 and May 2003 surveys since the total surveyed areas were comparable. March and August 2003 encompassed smaller areas. All data evaluation in this paper was done with code written in Matlab 6.1. 3. Results 3.1. Verfication of QTCView’s ability to detect algae In the Sebastian Inlet survey area, algae were successively added into the transducer’s footprint at depths of 1.2 and 1.5 m (Fig. 2). Data split into three classes and a new unique class only appeared when algae were added over the sand. This class represented the acoustic signal of the algae (Fig. 3). When seagrass was introduced under the transducer, a clear 4- Fig. 3. Differentiation of algae from seagrass from bare substratum using QTCView. Left figure shows PCA and class assignment by cluster analysis. Right figure shows depth and sequence of signals. Algae and substratum signals occur together when algae are put under the transducer. Trials with pure bare substratum provided a clear and unique acoustic signature. B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 class split was achieved, which again corresponded to the a priori known number of seafloor classes. Fig. 3 also shows that not all signals were uniform when algae were introduced under the transducer. After cluster-splitting, a relatively high percentage was grouped with the category bare substratum. The physical reasons for this phenomenon remain unclear. In the verification experiments at the university marina (Fig. 4), data from trials at 1.8-m depth allowed a three-class split, where at least one class could be assigned to dense algae, while results were not clear for the sparse algae. Clear separation between data taken with dense algae, sparse algae and empty basket were found. A distinct class (black in Fig. 5) represented dense algae, and another class (grey in Fig. 5) represented the sparser algae and a 95 third class (white in Fig. 5), which was closer to the sparse algae class than the dense algae class, represented the absence of algae. The depth-plot of Trial 1 in Fig. 5 clearly shows that dense and sparse algae created a signal by reflecting the acoustic pulse at the depth of the basket (0.7 m). When removed, the acoustic pulse was reflected at the depth of the substratum (1.6 m) and not at that of the now empty basket indicating that indeed the algae, and not the basket, were responsible for the scatter. Also, the depth plot of Trial 2 (Fig. 5) showed that dense algae scattered the acoustic signal at their surface, which was about 5cm above the substratum. Sparse algae formed clumps resulting in a different depthreading than the bare substratum. 3.2. Echoplus results Fig. 4. The experimental setup consisting of a wire basket suspended underneath the transducer used to determine whether algae and seagrass really produced an acoustic signature that could be differentiated from other substrata. The signal encountered either the empty basket (recording the seafloor underneath), or dense algae (biomass ~ 2000 g m 2 uncleaned wet weight) or sparse algae (~ 250 g m 2 wet weight). A plot of first against second echoes from the August 2003 survey in the Sebastian Inlet survey area (Fig. 6A) collected by the Echoplus, showed three distinct clusters: one group of low E1 and E2 values, one group of high E2 but low E1 values, and one group of low E2 and high E1 values (Fig. 6). Since E1 largely encodes surface roughness and E2 surface hardness, low values in both were believed to correspond to flat and soft, muddy substratum (i.e., the category bare substratum), low E1 (roughness) and high E2 (hardness) indicated a stronger substratum than algae signal and thus was believed to encode shelly, hard sand. High roughness (E1) and low hardness (E2) indicated a stronger surface scatter component and thus was believed to encode dense algae. Sparse algae formed a cloud of data in between the dense algae and hard substratum clusters. These assumptions were verified with groundtruthing video camera drops. The bdigital numbersQ (numeric codes for E1 and E2) were also plotted in sequence (Fig. 6B), allowing for further groundtruthing of the assumption that E1 encoded the density of algae. Since the time sequence of the digital numbers corresponded to their geographical position, it was possible to groundtruth the information contained in the signals with camera drops. High E1 values (between 0.75 and 1) corresponded to dense algae, medium E1 values (0.6 to 0.75) corresponded to sparse algae, and low E1 values (0.55–0.6) corresponded to bare substratum. 96 B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 Fig. 5. Differentiation of algae from bare substratum using QTCView. Upper two figures represent PCA and cluster assignment by cluster analysis, lower figures plot the depth of each signal. Trial 1 was performed in the marina using a basket suspended under the transducer into which algae were introduced (see methods). Trial 2 was in the field at Sebastian Inlet where different densities of algae were put on bare seafloor underneath the transducer. PCA shows three distinct groups of echoes (black=dense algae, gray=sparse algae, white=bare substratum). Depthplot of Trial 1 demonstrates that dense and sparse algae create a signal return—they reflect the acoustic signal at the depth of the basket (0.7 m). When removed, the signal is reflected at the depth of the substratum (1.6 m). Second depth plot (Trial 2) shows that dense algae scatter the acoustic signal near their surface (about 5 cm above the substratum). A similar relationship was found for E2 values. High E2 values (0.8 to 1) were found associated with bare, shelly substratum and sparse algae, medium E2 values (0.5–0.7) corresponded to dense algae, and low E2 values (0.3–0.5) corresponded to bare substratum (Fig. 6C). 3.3. Distribution of algal biomass in Melbourne survey area The QTCView survey in November 2002 (50 kHz) discriminated clearly all four classes of substratum, with bsparse algaeQ covering less area than bdense algaeQ and having a far lower cumulative biomass in the survey area (Table 3). Algae were concentrated along the lateral shoals of the study area, the biggest patches occurring on the northeastern and northwestern shoals (Fig. 7A). Much of the entire eastern fringe was covered by dense algae. Where seagrass was found, it was usually associated with algae. Dense algae were also found covering large patches of the central survey area (Fig. 7A). The QTCView survey in March 2003 (not illustrated in Fig. 7) encompassed the two northern shoals that had shown the densest coverage. The dense algae had remained but decreased in areal coverage on both shoals, and the western shoal showed some seagrass signal but remained algae-dominated. Previously, in November, the dense algae cover in the seagrass had drowned most seagrass signals. Also on the eastern shoal, the overall amount of dense algae appeared to have decreased in March, replaced by sparse algae. In May 2003, using a 200 kHz signal and QTCView, four substratum classes were observed (Fig. 7B) which coincided with classes observed in November. Dense algae covered more area than sparse B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 B 1 1 Bare substratum (shelly) 0.9 E1 = first echo, roughness parameter E2=second echo, hardness parameter A 97 0.8 0.7 sparse algae 0.6 0.5 bare substratum (muddy) 0.4 0.3 0.2 0.5 dense algae 0.95 dense algae 0.9 0.85 sparse algae 0.8 0.75 0.7 0.65 0.6 0.55 (shelly) (muddy) bare substratum 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0 2000 4000 6000 8000 10000 12000 14000 signal sequence E1 = first echo, roughness parameter E2=second echo, hardness parameter C 1 0.9 0.8 0.7 sparse algae bare substratum (shelly) 0.6 0.5 dense algae 0.4 0.3 0.2 (muddy) 0 2000 4000 6000 8000 10000 12000 14000 signal sequence Fig. 6. Differentiation of algae from bare substratum using Echoplus (results from the Sebastian Inlet August 2003) (A) Scatterplot of first versus second echo digital number; (B) The first echo digital numbers in time sequence; and (C) The second echo digital numbers in time sequence. High E1 (roughness) and low E2 (hardness) values correspond to drift algae, high E2 (hardness) and low E1 (roughness) values correspond to bare substratum. Sparse algae form a point cloud in between the two clusters. algae and seagrass and had a higher biomass than both other categories together (Table 3). The northern shoals were covered by dense algae and seagrass occupied similar areas as in November. The northwestern shoal showed some mixture of seagrass and sparse algae, but algae still dominated. The northwestern shoal had less algae. Only few areas of dense algae cover remained on both the northern and southern sides of the shoal and in the deep channel. Also the southeastern shoal showed seagrass cover with a fringe of sparse algae and patches of dense algae on the upstream and downstream sides (Fig. 7 B). In the August 2003 QTCView and Echoplus surveys, only three classes were found in sufficient density to be mapped, the class bsparse algaeQ was rare. Several areas of dense algae were found in the lee of the shoals and in the deep channel. The northwestern shoal now showed more seagrass than algae. Dense algae clumps appeared moribund, being flaccid with little pigmentation. We assume that they had previously lived in and around the seagrass and on the shoals. Total biomass of macrophytes decreased by over 50% between autumn 2002 and spring 2003 in the Melbourne study area (Table 3). 98 B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 Fig. 7. Spatial distribution of sand, seagrass, and algae in November 2002 (A,C,E) and in May 2003 (B, D, F). The Cocoa Beach survey was smaller in May 03 than in November 02. Dark gray=dense algae, medium gray=sparse algae, black=seagrass, white=bare substratum. Coordinates are in UTM, Zone 17R North. Striped areas in side the region of interest are outside the extrapolated areas and of unassigned bottom type. 3.4. Distribution of algal biomass in Sebastian inlet survey area The November 2002 (Fig. 7C) 50-kHz survey discriminated four classes of substratum. Seagrass occurred on the northern and southern shoals, and was surrounded by sparse algae forming a dense apron. Almost a quarter of the survey area was covered by sparse algae (Table 3), with smaller patches of dense algae concentrated near the southern shoal. In March 2003, two surveys (50 and 200 kHz) found more dense algae on the northern shoal than in November. Dense algae covered most of the seagrass and sparse algae surrounded these areas, which only showed in the 50 kHz survey. Most of the deeper survey area was bare substratum. The 50 kHz survey found a contiguous area of dense algae towards the southern shoal, while the 200 kHz survey suggested several patches of sparse algae in roughly the same area. In May 2003 (Fig. 7D), a 200-kHz survey discriminated the same four classes. Unlike in the Melbourne study area, overall macrophyte cover had not decreased and dense algae covered as much area as sparse algae and seagrass (Table 3). The northern shoals, near Sebastian Inlet, were characterized by seagrass signals with only small areas showing dense algae cover with surrounding sparse algae (Fig. 7D). B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 Another large seagrass area was on the southernmost shoal. Most dense algae were concentrated in the southern and western sections and surrounded by sparse algae. Seagrass and dense algae also existed in the deep and sheltered northeastern part of the basin, adjacent to mangroves. Most deeper area were bare. Between November 2002 and May 2003, the overall biomass of macrophytes had remained almost unchanged (Table 3). The August 2003 Echoplus surveys were conducted on 200 kHz, discriminating bare substratum, sparse algae, and dense algae. No seagrass areas were 99 surveyed. Different to Melbourne, algal cover in the Sebastian Inlet survey area increased in August. Most of the deeper basin was covered by dense algae or sparse algae (Fig. 7 D). Large clumps of algae were found along the waterline. The least algal cover was found in the western and southern sections. 3.5. Distribution of algal biomass in Cocoa beach survey area The November 2002, QTCView survey discriminated four classes of substratum (Fig. 7E). Algae Fig. 8. Survey results in the Cocoa Beach area using Echoplus (A,C) and QTCView (B,D). (A) and (C) show the location of individual pings assigned to the algae class, (C) and (D) show a nearest-neighbor interpolation of the algae class versus the non-algae class. Drift algae signals are indicated as black circled or filled black surfaces. Where survey lines appear incomplete, signals were removed during quality control due to poor quality. Both areas show a higher concentration of algae signal in the northern section. Coordinates are decimal degrees in WGS 84. 100 B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 formed an almost continuous fringe along the lateral shoals, but were also found in the deeper central part of the study area. Seagrass was almost always neighbored by accumulations of algae and concentrated on the northwestern shallow fringe of the study area, the northeastern shoal, and the southern central shoal. Dense algae covered more area than sparse algae and seagrass (Table 3). In May 2003 (Fig. 7F), the area covered by the survey was smaller than that of the November survey (Table 3, Fig. 7F). Seagrass was found on the southern central shoal and along the eastern and western shoals, while algae cover in the shallows had decreased. Cover of sparse algae in the deeper parts of the study area had increased, however. Sparse algae covered far more area than seagrass or dense algae (Table 3). Between November 2002 and May 2003, macrophyte cover had increased due to sparse algae in the deep Lagoon, but biomass had dropped dramatically (Table 3). In August 2003, Echoplus and QTCView surveys were conducted in the northern part of the survey area following the same survey lines (Fig. 8). In this relatively deep area, only algae and bare substratum occurred. Both surveys differentiated algae from bare substratum, and identified an area of dense algae in the northernmost part of the area. For map extrapolation of the Echoplus survey, digital E1 numbers were binned into algae- and non-algae classes according to Fig. 6B. The two systems differed in detection of algae in the southern, sparser, part. The Echoplus showed few algae signals in this area, while the QTCView showed a significant number of wellspaced algae signals. 3.6. Groundtruthing and accuracy assessment Video surveys confirmed the presence of the four bottom types observed by the acoustic surveys. In the Melbourne 50 kHz survey area, all four seafloor classes were also seen in the groundtruthing data. For this reason, dense and sparse algae were considered separate classes for accuracy assessment. Overall, accuracy was only 36% (Table 1A) and high confusion was evident among all classes, with highest confusion between dense and sparse algae. The confusion matrix was then recalculated with dense and sparse algae classes pooled (Table 1B) resulting in a three-class accuracy of 59%. The Sebastian Inlet groundtruthing did not encounter dense algae, resulting in only one dalgaeT class (Table 2). Two confusion matrices were produced which compared the extrapolated maps derived from the acoustic surveys. Overall, accuracy was about 60% (Table 2). The surveys and resulting extrapolated maps were very good at predicting areas of algae Table 1 (A) Error matrix (Ma and Redmond, 1995) for Melbourne study area with dense and sparse algae as separate classes; (B) with dense and sparse algae pooled as a single class A Melbourne Sand Seagrass Sparse algae Dense algae Sum User’s accuracy Sand Seagrass Sparse algae Dense algae Sum Producer’s accuracy 38 11 47 2 98 38.8% 17 0 14 43 74 0% 48 29 89 74 240 37.1% 17 1 57 75 150 50% 120 41 207 194 562 31.7% 0% 42.9% 38.6% 35.94% B Melbourne Sand Seagrass Algae Sum User’s accuracy Sand Seagrass Algae Sum Producer’s accuracy 38 11 47 98 38.8% 17 0 57 74 0% 65 30 295 390 75.6% 120 41 401 562 31.7% 0% 73.6% 59.3% Columns are classified categories on the maps, rows are accuracy assessment data obtained from video surveys. Bold number in italics is the total accuracy. B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 Table 2 Class-by-class error matrix for Sebastian Inlet A Sebastian 50 kHz Sand Seagrass Algae Sum User’s accuracy Sand Seagrass Algae Sum Producer’s accuracy 14 66 12 28 4 16 30 110 46.7% 25.4% 1 81 2 42 115 135 118 258 97.5% 17.2% 66.6% 85.1% B Sebastian 200 kHz Sand Seagrass Algae Sum User’s accuracy Sand Seagrass Algae Sum Producer’s accuracy 0 5 4 9 0% 17 41 0 58 71% 13 64 114 191 59.7% 30 110 118 258 0% 37.7% 96.6% 60.9% 60.1% Accuracy for individual classes as well as overall performance of the model. (A) 50 kHz survey (B) 200 kHz survey with sparse and dense algae pooled as a single class. (96 and 97%), however, some confusion did exist between areas of sand and seagrass. 4. Discussion The study shows that two different acoustic ground discrimination systems, QTCView and Echoplus, were capable, within limits, not only of differentiating sediment types, which is well published (Hamilton et al., 1999; Morrison et al., 2001; Freitas et al., 2003a,b) but were also able to detect algae and seagrass. A three-class confusion matrix based on QTCView surveys, suggested a total accuracy of nearly 60% for seagrass-algae-bare substratum. The two algae classes (sparse versus dense) showed high confusion, and clearly more work is needed to improve discrimination, either through improved signal processing or data analysis at the post-processing stage. But since ground-truthing information was collected after surveys for logistical reasons and drift algae are mobile, it is possible that the observed high confusion values are partly a sampling artifact caused by the algae having moved (aggregated, dispersed, etc.) in the time between the surveys and the groundtruthing. From both verification experiments and field survey data, it was apparent that seagrass and drift algae indeed produced unique echo classes. 101 Under controlled conditions as well as in the field, it was difficult to obtain files containing only echoes that could be clearly ascribed to either algae or seagrass with the QTCView instrument. Usually all files containing echoes from algae or seagrass also included a large proportion of echoes (up to half) from bare substratum. This high intermixed proportion of bottom signals was more pronounced with the algae than with the seagrass, but occurred with both vegetation types. This suggests that the bottom signal occurred as frequently as the vegetation signals themselves, even when the area under the transducer appeared to be entirely covered by vegetation. We cannot provide a satisfactory explanation why this is so, and several possibilities exist. The easiest is to assume that unattached drift algae rolled in and out of the transducer’s footprint, as was indeed sometimes observed. However, even when the algae were anchored, and when the attached seagrass was used as a target, a proportion of signals still had the acoustic qualities of signals derived from bare substratum (Figs. 3 and 5). The Echoplus appeared to give a somewhat clearer distinction (Figs. 6 and 8), which could, however, largely be due to the way classes were assigned as binned intervals of the digital number provided by the Echoplus, while in the QTCView system classes are assigned based on the position of signals in pseudo3-dimensional space after PCA. Depending on the width of the chosen bins, accuracy of discrimination could be influenced. The QTCView system uses at that last step at least three variables (the first three principal components characterizing each echo) for the binning into groups, in contrast, we used only the single digital number (E1) of Echoplus for obtaining Fig. 8. Reasons for confusion of algae and bare substratum echoes may either be a relatively weak scattering ability of the algae and/or the different signal properties and/or the processing steps. QTCView uses the entire signal envelope of the first echo (it ignores all multipath echoes, Collins and Lacroix, 1997) which Preston et al. (2000) and Freitas et al. (2003a,b) showed to provide good discrimination ability of sediment geotechnical variables. However, Chivers et al. (1990) reports that the first peak(s) of the echo is strongly influenced by subsurface reverberation, while the echo’s tail primarily encodes scatter, which is the reasoning followed in signal processing of the 102 B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 Echoplus. It is possible that much acoustic energy passes through the vegetation layer to primarily interact with the substratum and many signals may be more strongly influenced by the underlying substratum than the overlying vegetation. Thus, at least in theory, some benefit could be seen in emphasizing the trailing edge and the multipath echo when discrimination of macrophytes or other purely surficial structures is desired, since this should minimize the subsurface reverberation component. Sabol and Johnston (2001) and Sabol et al. (2002) used higher frequency echosounders and different signal processing in their acoustic evaluations of submerged aquatic vegetation. Interpretation of the seafloor classification is aided by regridding the surveys and benhancingQ them through extrapolation and spatial statistical methods (Guan et al., 1999; Middleton, 2000; Davis, 2002; Walter et al., 2002; Papritz and Stein, 2002), which allows the production of maps with closed surfaces. While some artifacts may be introduced, it is difficult to evaluate the validity and meaning of either the QTCView or the Echoplus acoustic ground discrimination by evaluating the survey lines only. An alternative would be to cover the entire seafloor with acoustic signals, but this is not practical since the shallow water depths result in a small footprint even of swath systems, like side-scanning or multibeam sonars. Bathymetry influenced the distribution of drift algae in all three study areas. The pattern was clearer in Melbourne and Cocoa Beach than in Sebastian Inlet. In Cocoa Beach and Melbourne, the areas of densest algae accumulation were always towards the Lagoon’s edges where the shallow areas (b1.5 m deep) showed, depending on season, a continuous fringe of either dense or sparse algae. This situation was also observed by other surveys in the same area (Nielsen et al., 2000). In the Sebastian Inlet area, dense algae were found as well on the shoals as in the deeper areas. Seagrass was restricted to the shallowest areas (b 1 m depth) and algae had a clear tendency to accumulate around and within these seagrass meadows. This made the unequivocal acoustic discrimination of seagrass difficult. Drift algae were encountered in the deeper parts of the Lagoon and also accumulated around and within seagrass areas, particularly in winter. In the Melbourne and Cocoa Beach survey areas, macrophyte biomass was low in spring, while in the Sebastian Inlet area it was high (Table 3). Algae density, in particular of freely drifting algae in the deeper parts of the Lagoon, in the Melbourne and Cocoa Beach areas increased again towards autumn, in August. It is assumed that the higher values in algal biomass in the Sebastian Inlet study area, which is situated near a major inlet with a very active hydrodynamic regime, could have been due to local, current-driven, accumulation of algae. We assume that the observed absence of a uniform seasonal pattern in biomass and distribution of algae and seagrass among the study areas is an expression of a generally high spatial and temporal variability of macrophyte, in particular algae, dynamics in the Indian River Lagoon. 5. Conclusion ! Acoustic ground discrimination using QTCView and Echoplus proved useful tools not only to map the distribution of seagrass and drift malcroalgae in the Indian River Lagoon. Table 3 Space cover (percentage value of total survey area above) and biomass (in tons wet weight biomass per survey area below) of acoustically detected bottom types in the Indian River Lagoon before and after summer (peak growth season of macrophytes) Melbourne Nov. 02 Melbourne May 03 Sebastian Nov.02 Sebastian May 03 Cocoa Nov. 02 Cocoa May 03 Seagrass Dense algae Sparse algae Total macrophyte cover/biomass Bare substratum Size of survey % area (tons) % area (tons) % area (tons) % area (tons) % area in km2 7 5 6 12 4 4 26 10 8 12 19 2 3 4 22 12 2 30 36 19 36 36 25 36 64 81 64 64 75 64 8.9 8.5 9.6 8.6 13.9 9.8 (60.04) (41.91) (54.69) (102.5) (52.14) (39.56) (4704.2) (1795) (1623.8) (2140.6) (5296) (343.2) (71.7) (78.9) (533.6) (267.6) (86.6) (736.4) (4835.9) (1915.8) (2212.1) (2510.7) (5434.7) (1119.2) B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 ! The maps of macrophyte distribution extrapolated from the acoustic surveys also allowed a coarse estimation of biomass. ! Macrophyte (algae and seagrass) biomass varied with the seasons but the pattern was not uniform among study areas. ! More work is necessary to increase the accuracy of acoustic surveys. Acknowledgements The authors would like to thank the St. Johns River Water Management District who funded this project through grant SF655AA. Additional funding from NOAA/NOS via grant NA03NOS4260046 to NCRI is gratefully acknowledged. G.S. McIntosh captained and made the surveys possible, provided field support and logistics. B.K. Walker and A. Shinohara assisted with data collection. This is NCRI publication # 70. [RH] References Bates, C.R., Whitehead, E.J., 2001. Echoplus measurements in Hopavagen Bay, Norway. Sea Technol., 34 – 43 (June). Brinker, R.C., Minnick, R., 1994. The Surveying Handbook. Kluwer, Dordrecht. 967 pp. Chivers, R.C., Emerson, N., Burns, D.R., 1990. New acoustic processing for underway surveying. Hydrogr. J. 56, 9 – 19. Collins, W.T., Lacroix, P., 1997. Operational philosophy of acoustic waveform data processing for seabed classification. COSU 97. Oceanol. Int., 1 – 7 (Singapore). Davis, J.C., 2002. Statistics and Data Analysis in Geology, 3rd edition. John Wiley and Sons, New York, p. 638. Dierssen, H.M., Zimmerman, R.C., Leathers, R.A., Downes, T.V., Davis, C.O., 2003. Ocean color remote sensing of seagrass and bathymetry in the Bahama Banks by high-resolution airborne imagery. Limnol. Oceanogr. 48 (1), 444 – 455. Ellingsen, K.E., Gray, J.S., Bjørnbom, E., 2002. Acoustic classification of seabed habitats using the QTC view system. ICES J. Mar. Sci. 59, 825 – 835. Greenstreet, S.P.R., Tuck, I.D., Grewar, G.N., Armstrong, E., Reid, D.G., Wright, P.J., 1997. An assessment of the acoustic survey technique, RoxAnn, as a means of mapping seabed habitat. ICES J. Mar. Sci. 54, 939 – 959. Green, E., Mumby, P.F., Edwards, A.J., Clark, C.D., 2000. Remote sensing handbook for tropical coastal management. Coastal Management Sourcebooks, vol. 3. UNESCO, Paris. x+316 pp. Guan, W., Chamberlain, R.H., Sabol, B.M., Doehring, P.H., 1999. Mapping submerged aquatic vegetation with GIS in the Caloo- 103 sahatchee estuary: evaluation of different interpolation methods. Mar. Geod. 22 (2), 69 – 92. Freitas, R., Silva, S., Quintino, V., Rodrigues, A.M., Rhynas, K., Collins, W.T., 2003a. Acoustic seabed classification of marine habitats: studies in the western coastal-shelf area of Portugal. ICES J. Mar. Sci. 60, 599 – 606. Freitas, R., Rodriguez, A.M., Quintino, V., 2003b. Benthic biotopes remote sensing using acoustics. J. Exp. Mar. Biol. Ecol. 285–286, 339 – 353. Hamilton, L.J., 2001. Acoustic Seabed Classification Systems. Department of Defence. Defence Science and Technology Organisation. 150 pp. Hamilton, L.J., Mulhearn, P.J., Poeckert, R., 1999. Comparison of RoxAnn and QTC-view acoustic bottom classification system performance for the Cairns area, Great Barrier Reef, Australia. Cont. Shelf Res. 19, 1577 – 1597. Kenny, A.J., Cato, I., Desprez, M., Fader, G., Schuettelhelm, R.T.E., Side, J., 2003. An overview of seabed-mapping technologies in the context of marine habitat classification. ICES J. Mar. Sci. 60, 411 – 418. Kirlin, R.L., Dizaji, R.M., 2000. Cluster order using clustering performance index rate, CPIR. Proc. NORSIG 2000. Kolmarden, Sweden, pp. 85 – 88. Lawrence, M.J., Bates, C.R., 2001. Acoustic ground discrimination techniques for submerged archaeological site investigations. MTS J. 35 (4), 65 – 73. Legendre, P., Ellingsen, K.E., Bjørnbom, E., Casgrain, P., 2002. Acoustic seabed classification: improved statistical methods. Can. J. Fish. Aquat. Sci. 59, 1085 – 1089. Ma, Z., Redmond, R.L., 1995. Tau coefficients for accuracy assessment of classification of remote sensing data. Photogramm. Eng. Remote Sensing 61, 435 – 439. Matheron, G., 1971. The theory of regionalized variables and its applications. Cah. Centre Morphologie Mathematique, vol. 5. Centre de Geostatistique, Fontainebleau. Middleton, G.V., 2000. Data Analysis in the Earth Sciences using Matlab. Prentice-Hall. 260 pp. Morris, L.J., Hall, L.M., 2001. Estimating drift algae abundance in the Indian River Lagoon, FL. St. Johns River Water Management District Technical Memorandum, 10pp. Morris, L.J., Virnstein, R.W., Miller, J.D., Hall, L.M., 2000. Monitoring changes in Indian River Lagoon, Florida, using fixed transects. In: Bortone, S.A. (Ed.), Seagrass Monitoring, Ecology, Physiology and Management. CRC Press, Boca Raton. 318 pp. Morrison, M.A., Thrush, S.F., Budd, R., 2001. Detection of acoustic class boundaries in soft sediment systems using the seafloor acoustic discrimination system QTC View. J. Sea Res. 46, 233 – 243. Moyer, R.P., Riegl, B.M., Banks, K. Dodge, R.E., in press. Assessing the accuracy of acoustic seabed classification for mapping coral reef environments in South Florida (Broward County, USA). Revta. Biol. Trop. 52 (4). Nielsen, S., Eggers, B., Collins, S., 2000. The influence of seawalls and revetments on the presence of seagrass in the Indian River Lagoon, a preliminary study. Fla. Sci. 63, 48 – 61. 104 B.M. Riegl et al. / J. Exp. Mar. Biol. Ecol. 326 (2005) 89–104 Papritz, A., Stein, A., 2002. Spatial prediction by linear kriging. In: Stein, A., van der Meer, F., Gorte, B. (Eds.), Spatial Statistics for Remote Sensing. Kluwer, Dordrecht, pp. 83 – 113. Preston, J.M., Kirlin, R.L., 2002. Comment on bAcoustic seabed classification: improved statistical methodQ. Can. J. Fish. Aquat. Sci. 60, 1299 – 1300. Preston, J.M., Collins, W.T., Mosher, D.C., Poeckert, R.H., Kuwahara, R.H., 1999. The strength of correlations between geotechnical variables and acoustic classifications. Proc. MTS/IEEE Oceans ’99 Conference, pp. 1123 – 1128. Preston, J.M., Rosenberger, A., Collins, W.T., 2000. Bottom classification in very shallow water. Proc. Oceans 00. Newport 2000. Preston, J.M., Christney, A.C., Bloomer, S.F., Beaudet, I.L., 2001. Seabed classification on multibeam sonar images. Proc. MTS/ IEEE OCEANS 2001: An Ocean Odyssey Honolulu, Hawaii, Nov 2001. Preston, J.M., Christney, A.C., Beran, L.S., Collins, W.T., 2004. Statistical seabed segmentation — from images and echoes to objective clustering. Proc. ECUA. Quester Tangent Corporation,, 2002. QTC IMPACTk acoustic seabed classification, user guide version 3.00. Integrated mapping, processing and classification toolkitQuester Tangent Corporation, Sidney, B.C., Canada. Riegl, B., Purkis, S.J., 2005. Detection of shallow subtidal corals from IKONOS satellite and QTC View (50, 200 kHz) single- beam sonar data (Arabian Gulf; Dubai. UAE). Remote Sens. Environ. 95, 96 – 114. Sabol, B.M., Johnston, S.A., 2001. Innovative techniques for improved hydroacoustic bottom tracking in dense aquatic vegetation. USACE ERDC/EL Report MP-01-2 (20 pp.). Sabol, B.M., Melton, R.E., Chamberlain, R., Doehring, P., Haunert, K., 2002. Evaluation of a digital echo sounder system for detection of submersed aquatic vegetation. Estuaries 25 (1), 133 – 141. Walter, D.J., Lambert, D.N., Young, D.C., 2002. Sediment facies determination using acoustic techniques in a shallow-water carbonate environment, Dry Tortugas, Florida. Mar. Geol. 182, 161 – 177. Virnstein, R.W., Carbonara, P.A., 1985. Seasonal abundance and distribution of drift algae and seagrass in the mid-Indian River Lagoon, FL. Aquat. Bot. 23, 67 – 82. Virnstein, R.W., Howard, R.K., 1987. Motile epifauna of marine macrophytes in the Indian River Lagoon, Florida. II. Comparison between drift macroalgae and three species of seagrasses. Bull. Mar. Sci. 41 (1), 13 – 26. von Szalay, P.G., McConnaughey, R.A., 2002. The effect of slope and vessel speed on the performance of a single beam acoustic seabed classification system. Fish. Res. 54, 181 – 194. Zimmermann, R.C., 2003. A biooptical model of irradiance distribution and photosynthesis in seagrass canopies. Limnol. Oceanogr. 48 (1), 568 – 585.