Ecosystem Modeling in Cobscook Bay, Maine: A Boreal, Macrotidal Estuary 2004 Northeastern Naturalist 11(Special Issue 2):225–242 Use of Cost Effective Remote Sensing to Map and Measure Marine Intertidal Habitats in Support of Ecosystem Modeling Efforts: Cobscook Bay, Maine PETER FOSTER LARSEN1,*, SETH BARKER2, JED WRIGHT3, AND CYNTHIA B. ERICKSON1 Abstract - Reliable estimates of the habitat areas of major marine producer groups were needed in support of an ecosystem modeling effort in the macrotidal Cobscook Bay, ME. Results needed to be comprehensive, synoptic, objective, affordable, and on a suitable spatial scale. We chose to address these goals by applying accepted procedures utilizing existing Landsat Thematic Mapper imagery and a computer-generated unsupervised classification. Unsupervised classification of high and low tide Landsat TM images yielded coherent habitat maps that were supported by reference data and independent habitat analyses. The high tide image revealed surface water patterns that supported the existence of a large, Central Bay dipole eddy predicted by a numerical three-dimensional circulation model. Classification of the low tide image resulted in 14 intertidal and water habitat classes being defined. Overall accuracy of the classification was 86%. Good agreement in habitat areas existed between the affordable and easily repeatable satellite survey and four other Cobscook Bay surveys which differed methodologically and temporally. The four studies agreed within 7% of total habitat area and 12% on intertidal habitat area. The area of both brown and green algae in the Bay apparently increased modestly over a 25-year period which saw the introduction of large-scale salmon aquaculture and the advent of intensive dragging for scallops and sea urchins. The increase in both groups is inconsistent with changes induced by nutrient additions observed elsewhere. Landsat imagery appears to be a valuable tool for the management and monitoring of macrotidal environments. Introduction In 1994, an interdisciplinary, multi-institutional team of marine scientists was awarded a grant from The Nature Conservancy and the Andrew W. Mellon Foundation to investigate the ecosystem dynamics of Cobscook Bay, ME. Cobscook Bay is a hydrographically and geologically complex estuary where very high levels of biodiversity and productivity co-exist (Larsen 2004). A primary thrust of this coordinated research program was to construct an energy systems model of the 1Bigelow Laboratory for Ocean Sciences, PO Box 475, West Boothbay Harbor, ME 04575. 2Maine Department of Marine Resources, PO Box 8, West Boothbay Harbor, ME 04575. 3Gulf of Maine Coastal Program, US Fish and Wildlife Service, Falmouth, ME 04105. *Corresponding author - plarsen@bigelow.org. 226 Northeastern Naturalist Vol. 11, Special Issue 2 Bay’s ecosystem to evaluate the annual flows of energy and matter through the system (Campbell 2004). Our approach, given limited resources, was to concentrate on measuring primary production and water column properties to provide a sound basis for future research on higher order ecosystem components. This required contemporaneously acquired, broad-scale area estimates of the dominant habitats (tidal areas, macrophyte beds, microphyte habitat, marshes) so that results of sitespecific productivity studies could be extrapolated to the entire Bay. Data quality estimates needed to be within the model’s general margin of error, i.e., within 10–15% on areas of hundreds or thousands of hectares (D.E. Campbell, US Environmental Protection Agency, pers. comm.). Limited areal data do exist for certain habitats. These data were collected at differing times by differing methods, however, and were deemed to be insufficiently comprehensive and synoptic for the present purposes. Therefore, we set out to produce a large scale, point-in-time, thematic map of Cobscook Bay that would allow our co-investigators to estimate the energy contributions of the principal producer groups. In the coastal zone, with multiple sources of primary production, complex geology, and steep environmental gradients, gaining sufficiently rigorous information on the contributions of ecosystem components can be time-consuming and costly. The advent of airborne and satellite remote sensing technologies offers the potential of synoptic and repeatable data collection on an ecosystem scale that cannot be realized using traditional methods (Muir 1997). The various sensors have a range of capabilities, availabilities, and costs for acquisition and data processing (Mumby et al. 1997). Landsat Thematic Mapper (TM) imagery appealed to us for several reasons: a library of images was available to us at no cost, relieving us of the expense and time delays associated with obtaining new images; the images are synoptic for our entire study area, which saves time and effort in mosaicking and georeferencing a series of smaller images as would occur with the use of aerial photography; the relatively large pixel size (900 m2) is still small compared to the size of the habitat blocks in which we were most interested; and Landsat TM has been shown to be more accurate than other satellite sensors available at the time of this research (Mumby et al. 1997). Once the satellite data have been received, they must be processed to link similar or related elements into spectral classes that can then be related to ecological units on the ground, ultimately producing a thematic map of the study area. One cost-effective method that has been successfully applied where broad land-cover information is needed is called unsupervised classification (USC; Belward et al. 1990). USC involves assigning pixels into a predetermined number of classes based on spectral properties, and proceeding iteratively until maximum statistical separation is achieved. Reference data and site visits are then used 2004 P.F. Larsen, S. Barker, J. Wright, and C.B. Erickson 227 to assign cover types to the defined clusters. This technique has the advantages of needing less input from the user to guide the procedure, i.e., all data used are internal to the image, and the processed classification itself can be used in the field to direct the ground truthing. USC is not as rigorous as other more controlled classification schemes, but it has been used advantageously in intertidal areas where spectral contrasts between habitat types are more pronounced than in most terrestrial situations (Thomson et al. 1998). Site Description Cobscook Bay is located in extreme eastern Maine on the USCanadian border near the mouth of the Bay of Fundy (Fig. 1). Together Figure 1. Map of the convoluted Cobscook Bay, ME, with many place names indicated. Rectangles indicate areas of ground truthing efforts. 228 Northeastern Naturalist Vol. 11, Special Issue 2 with Passamaquoddy Bay, the St. Croix estuary, and the enveloping islands, it constitutes the area known as the Quoddy region. Cobscook Bay is a rock-framed, glaciated, tidally dominated estuary (Kelley and Kelley 2004). The large tidal range, with a mean value of 5.7 m, is a dominating ecological forcing function (Campbell 2004). Mean depth of the Bay is less than 10 m with pockets to about 45 m (Shenton and Horton 1973). The large tidal range and shallowness of the Bay result in about one third of the area being exposed to the atmosphere at low tide (Shenton and Horton 1973). Light reaches the bottom throughout the Bay in spring and summer (Phinney et al. 2004). Freshwater input is small, < 1% of the intertidal volume, whereas the tidal flow is extremely large, being equivalent to the mean outflow of the Mississippi River over the duration of both the ebb and flood tides (Brooks et al. 1999). Peak current speeds are on the order of 2 m/sec. The well-mixed nature of the tidal waters results in moderated seasonal ranges of temperature and salinity. Mean annual temperature variation is less than 10 oC while salinity variation is only about 1 ppt (Shenton and Horton 1973). In addition, the importation of nutrient-rich source water from the Gulf of Maine and the extreme tidal mixing insure that nutrients supporting plant growth are present in excess throughout the Bay (Garside and Garside 2004). Current human impact is largely limited to living resource harvesting. More information on the Cobscook Bay region can be found in the comprehensive bibliography of Larsen and Webb (1997). Methods Image acquisition Two Landsat TM images of Cobscook Bay were obtained from the Maine Geological Survey and the Maine Gap Analysis Program of the University of Maine. The Landsat TM images were sun-synchronous, capturing the area of Cobscook Bay at 0942 local time. Standard radiometric and geometric corrections had been applied. The first image was acquired on June 25, 1991. On that date in Eastport, a neap high tide occurred at 0936 indicating that water levels in the Bay were at or approaching high tide when the image was acquired. The second image was acquired on October 20, 1993. A mean low tide occurred at Eastport at 0812 that day. Because of a tidal phase delay moving up the Bay, the tide in the inner Bay would have been precisely at mean low at the time of image acquisition. Image processing Subsections of both Landsat TM images were synoptic for the entire Bay. Landsat TM image data were organized in pixels 30 m on a side (900 m2) with seven spectral bands in the wavelength range of 440 to 1250 nm. 2004 P.F. Larsen, S. Barker, J. Wright, and C.B. Erickson 229 Images were processed using ERDAS Imagine image analysis software on a UNIX workstation. Land areas were masked to focus the analyses on the marine environment. Masking was accomplished through interactive editing by visual comparisons to NOAA chart No. 13328 (1:40,000) and true color aerial photographs obtained in August 1993 (1:12,000). Interactive editing was necessitated by the very complex geomorphology of Cobscook Bay. The ISODATA algorithm was used to cluster pixels based on minimum spectral distance. This unsupervised processing sorts the image pixel by pixel, groups the pixels together by similar spectral characteristics, and then sorts the pixels into recognizable categories. A spectral set is then calculated using the statistical parameters (e.g., mean and covariance matrix) of the pixels that are in the training sample. This signature set is used to assign each pixel to a class. Pixels that do not fall within a cluster are assigned to the cluster that is closest to its spectral value. The images were classified to 20 unsupervised classes similar to the fashion of Thomson et al. (1998). A color palette that gave the best all-around visual definition was applied to both images resulting in preliminary class maps. These preliminary class maps were converted to thematic maps by assigning environmental characteristics to the unsupervised classes by comparison with field observations and reference data (see below). Through these comparisons, we were able to increase the accuracy of the maps by reassigning selected groups of pixels that were obviously misclassified when considered contextually, e.g., deep water sites classified as mud apparently due to high suspended sediment content. Field and reference data Two subregions of the Cobscook Bay, Dennys Bay and the northeast corner of the Bay including Bar Harbor (known locally as Half Moon Cove), were selected for ground truthing (Fig. 1). Each subregion covered approximately 22 km2 and was chosen because of its habitat diversity, accessibility, and our long-term familiarity with the areas. The imagery of the narrow embayments contained abundant landmarks formed by the complex geology of the shoreline, numerous islands, and ledges making it possible to locate oneself within the imagery with a high degree of positional accuracy. Field data were collected during the spring tides of September 1997, i.e., four years after low tide image acquisition. The preliminary unsupervised classification of the low tide Landsat image was annotated in the field. This process was subsidized by annotation of the 1993 aerial photography in the manner of Berry and Ritter (1995). This photographic survey and the Landsat TM image capture were done within two months of each other and, hence, the photographs may better reflect the distributions of ephemeral plant populations. In brief, photographs were overlaid with acetate transparencies, and environmental features of 230 Northeastern Naturalist Vol. 11, Special Issue 2 sufficient size and character to be distinguished in the imagery, usually homogeneous habitat patches 16 or more pixels in size (1.5 hectares), were outlined and described using an indelible pen. Thirty-five millimeter slides were taken at all sites along with detailed notes on sediment characteristics and structures, dominant plant species present, secondary species, percentage vegetation cover, and area of standing water. Other sources of reference data included National Oceanic and Atmospheric Administration (NOAA) nautical charts, US Geological Survey (USGS) topographic maps, and the Coastal Maine Geological Environments (CMGE) maps, available in paper and digital formats (Maine Geological Survey 1976). The CMGE maps were produced in the early 1970s from interpretation of black and white aerial photography and contain the locations and shapes of 55 defined geological environments found on the coast of Maine. Although widely used for a variety of purposes, specifics on the methodology of their production or accuracy were not documented (B. Timson, Mahoosuc Corp., Augusta, ME, pers. comm.). Details of the categories used were published with paper maps and further described in a companion publication of the Maine State Planning Office (1983). Figure 2. The classified 1991 high tide Landsat Thematic Mapper image of Cobscook Bay. Note the apparent counter-rotating gyres (arrows). 2004 P.F. Larsen, S. Barker, J. Wright, and C.B. Erickson 231 Results and Discussion Class analysis The classification of the 1991 high tide Landsat TM image yielded a coherent pattern of surface water conditions (Fig. 2). These surface water classes, which were not ground-truthed, were undoubtedly derived from the interactions of currents, winds, turbidity, and bottom topography. A noteworthy feature of this analysis is the pattern of classes in the central region of Cobscook Bay which are consistent with the existence of two counter-rotating eddies, i.e., an eddy dipole. The probable development of these eddies by the encounter of the incoming tidal wave with the restriction of the narrows between Denbow and Leighton Points was indicated by a three-dimensional numerical model simulation of Cobscook Bay circulation (Brooks et. al. 1999). This image analysis is suggestive of the actual existence of the eddies and, hence, adds credibility to the output of the computer model simulation. We believe the surface patterns reflect the distribution of surface turbidity, most likely suspended sediments to which Landsat TM is known to be sensitive (Mumby et al. 2004). Additional evidence supporting this initial supposition comes from two sources. First, Kelley and Kelley (2004) note that the areas beneath the predicted gyres are the only subtidal areas in the Bay where mud accumulates. Second, subsequent Figure 3. Tracks of surface drifters released at the beginning of the flood tide. Note the formation of an eddy dipole pattern similar to that seen in Figure 2. (After Brooks 2004). 232 Northeastern Naturalist Vol. 11, Special Issue 2 Table 1. The distribution of ecological categories following the recode of the unsupervised classification of the 1993 low tide Landsat image. These categories, singularly or in combination, were used by co-investigators to input productivity studies into the energy systems model (see Campbell 2004). Unsupervised Number of Area Ecological categories class(es) pixels (hectares) Operational definition Deep water 1, 2, 3, 4, and 5 in part 71337 6420 Water of sufficient depth to mask any signals from the bottom or tidally resuspended sediments. Shallow water/pens 5 in part, 6 5346 481 Shallow water containing perceptible levels of tidally resuspended sediments. Estimated depth range is two meters to several centimeters. Very shallow water 7 2957 266 Lens of water over mud. Knee deep to a shorebird. Shallow water/channels 8 2354 212 Shallow water centered in channels; less turbid than above classes. Dense green algae1 10 4839 436 80–100% coverage of green algae. Moderate green algae 9, 11 6549 589 Moderate coverages of green algae in association with brown algae. Sparse green algae 12 3832 345 50–60% coverage of green algae. Mud 13 6091 548 Pure, clean, glistening mud. Probable presence of benthic diatoms. Mixed sediment 14 3005 270 Mixture of sediments from mud to gravel. Moderate brown algae 15 2749 247 50% coverage of brown algae in mid- and low intertidal. Dense brown algae2 16 6622 596 90–100% coverage by browns in most locations. One area placed in this class only had 60–70% cover. Sparse brown algae 17 3784 341 Less dense coverage of brown algae on upper intertidal ledges. Marsh/upland 18 3885 350 Marginal upper intertidal class; predominantly marsh. Upland/marsh 19 3322 299 Marginal upper intertidal class; predominantly upland. 1This category is the same as the one labeled "Green Algae" in Figures 4 and 5. 2This category is the same as the one labeled "Brown Algae" in Figures 4 and 5. 2004 P.F. Larsen, S. Barker, J. Wright, and C.B. Erickson 233 to our analysis, surface drifter studies clearly show an eddy dipole pattern in the surface circulation (Fig. 3) (Brooks 2004). Examination of the unsupervised 20-class classification of the 1993 low tide Landsat image of Cobscook Bay relative to the field and reference data resulted in the 20 spectral classes being grouped into 14 categories of land/water cover (Table 1, Fig. 4). Detail of the 14-class representation of the TM image in the Bar Harbor (Half Moon Cove) area is presented in Figure 5. Most of the grouping involved spectral classes for deep water, i.e., water clearly not occupying the littoral zone. These classes were similar to those identified in the classification of the high tide image, were contiguous with the open areas of the Bay, and were in locations sufficiently deep as not to show any bottom signal. This combined category was used to define the low tide area of the Bay and the maximum habitat area of subtidal microphytes. There was sufficient variability in other spectral classes related to water to parse them into categories that occur in bands of varying depths based on field observations (Table 1). Coastal environments are characterized by a high degree of spectral variation between surface classes (Thomson et al. 1998), while the relatively low spectral and spatial resolution of Landsat TM will tend to blur within-class variations. Hence, spectral classes, especially those composed using the non-user-controlled USC, will reflect the large spectral discontinuities in the environment. Consistent with our purpose of providing an areal estimate of broad-scale cover types, 11 spectral classes were interpreted as 10 littoral habitat types based on their composition and location in the intertidal zone (Table 1). The broad spectral factors differentiating the classes were non-vegetated vs. vegetated substrates and, within the vegetated substrates, the plant pigments of the green algae (principally Enteromorpha spp.), brown algae (principally Ascophyllum nodosum Le Jolis), and the vascular plants of the marsh (Spartina alterniflora Lois. and S. patens Muhller) and upland areas. Accuracy assessment The ecological classes were amalgamated into five categories comparable to CMGE classes, i.e., water, non-vegetated, brown algae, green Table 2. Error matrix table for grouped ecological categories. Reference data are in columns, classified data in rows. Reference data Brown Green Marsh/upland Unvegetated Water Total Brown 30 30 Green 15 2 17 Marsh/upland 11 11 Unvegetated 2 12 14 Water 6 2 3 11 Total 36 17 11 16 3 83 234 Northeastern Naturalist Vol. 11, Special Issue 2 algae, and marsh/upland. These grouped data were then subjected to a basic accuracy assessment (Congalton and Green 1999). The overall accuracy was a high 86% (Table 2). Examination of producer’s and user’s accuracies indicated that all the individual class accuracies were Figure 4. The 14-class representation of the low tide Landsat Thematic Mapper. Table 3. User’s accuracy and producer’s accuracy for grouped ecological categories. Producer’s accuracy User’s accuracy Brown algae 83% 100% Green algae 88% 88% Marsh/upland 100% 100% Unvegetated 75% 86% Water 100% 27% 2004 P.F. Larsen, S. Barker, J. Wright, and C.B. Erickson 235 high with the exception the user’s accuracy for the water categories (Table 3). Examination of the classified image and the georeferenced aerial photographs used for ground truthing showed that most of the confusion between the water and brown algae categories (Table 2) was caused by differences in tidal heights between the reference data and satellite image. Sites that were observed to be kelp in the field were classified as water at the later tidal stage of the satellite image. It is clear that the Landsat sensors did not penetrate effectively below the surface of shallow water. Hence, the resulting images do not Figure 5. Detailed representation of the low tide Landsat Thematic Mapper image of Bar Harbor (Half Moon Cove) area of Cobscook Bay. Bar Harbor and Vicinity Classified 1993 TM Image 236 Northeastern Naturalist Vol. 11, Special Issue 2 document the distributions of sublittoral and fringing kelps, reds, or submerged aquatic vegetation, all potentially important producer groups (Beal et al. 2004; Vadas et al. 2004a,b). These groups are only exposed at the very lowest tidal levels in the macrotidal environment of Cobscook Bay. Unfortunately, the extreme spring low tides needed for sensor access to these groups do not occur in the Gulf of Maine at the mid-morning satellite overpass times. Future surveys should give special consideration to separating these groups from the currently defined low intertidal classes such as moderate brown algae. Comparisons Four other studies deal, to a greater or lesser extent, with habitat areas in Cobscook Bay. These are the CMGE maps (Maine Geological Survey 1976) described above, the The Research Institute of the Gulf of Maine (TRIGOM) literature review of 1972 (Shenton and Horton 1973), a recent rockweed biomass survey (Crawford 1999), and the hydrographic modeling efforts of Brooks et al. (1999). These studies had different objectives, employed different techniques and scale, and took place at different times. Each used an ecological classification of habitats unique to its purposes. For example, the CMGE maps assigned everything visible on the aerial photographs into 55 defined geological habitats, with the smallest unit mapped equal to 74 m2 and the majority greater than 100 m2 (B. Timson, Mahoosuc Corp., Augusta, ME, pers. comm.). The present study, in contrast, assigned pixels 30 meters on a side (900 m2) to one of 14 classes. Direct comparisons of specific habitats are, therefore, limited. Crawford (1999) planimetered beds of the brown macroalga Ascophyllum using a Maine Department of Marine Resources aerial photographic survey of August 22, 1993. Brooks (2004) determined the Bay’s subtidal area by direct interpolation of NOAA Chart 13328. The CMGE used a feature boundary that was east of the Landsat data, i.e., larger by 232 ha. Brooks et al. (1999) used an outer boundary identical to our own, while the outer boundaries of the TRIGOM and Crawford studies are not known. Four of the studies agree with one another within approximately 7% on total high tide area, and five studies within 12% on subtidal areas of the Bay (Table 4). Some of this difference can be explained by the studies being done at different low tide levels. Remarkably, Brooks subtidal area produced by interpolating the mean low water (MLW) line from a NOAA chart is essentially identical to ours produced from a Landsat image captured at MLW. Only three specific intertidal habitats are defined similarly enough in four of the studies to allow direct comparisons. Three studies show a small amount of marsh area (Table 4). The largest area was registered in the present 2004 P.F. Larsen, S. Barker, J. Wright, and C.B. Erickson 237 study, perhaps because the differentiation between Spartina marsh and upland vegetation was made more difficult by the season. The image was taken in autumn, when the plants did not contain a full complement of pigments which may have blurred the distinction between the marsh plants and deciduous trees. The existence of these surveys allows limited and imperfect longterm comparisons of the two dominant intertidal macroalgal producer communities, i.e., our categories dominated by perennial brown algae and annual green algae. Such comparisons are rare in the literature (Middelboe and Sand-Jensen 2000). Our brown algae categories are predominantly Ascophyllum nodosum while the green algae categories are principally Enteromorpha spp. Each grouping contains various ancillary species (Vadas et al. 2004b). There is excellent agreement between the three studies that measured the area of brown algae, with the areas ranging from 1013 ha in the CMGE survey to 1184 ha in the 1993 Landsat survey (Table 4). Comparison of the CMGEs and the satellite derived habitat maps also showed excellent agreement in terms of the locations, sizes, and shapes of the brown algae beds. This would be expected with long-lived species that principally inhabit stable environments such as ledges. Crawford (1999) focused on commercial beds of Ascophyllum. His brown algae area may have been in even better agreement with the Landsat results had he not excluded areas with a high percentage of Fucus. Comparing the areas of green algal beds over a long period of time must be done cautiously because several variables are involved. These are largely annual and ephemeral species. They respond quickly to nutrient availability and vary in abundance by two or more orders of magnitude over the growing season (Pregnall and Rudy 1985). Obviously, seasonal timing of the surveys to be compared must be considered. The CMGE photographs were taken in April and November whereas our Landsat image was captured in October. The coverage of green algae can be heterogeneous (Vadas et al. 2004b), but peak abundance usually occurs in July and August (Pregnall and Table 4. Comparison of selected Landsat-derived habitat areas with results of other temporally and methodologically different Cobscook Bay studies. Comparable Area (hectares) habitats This study CMGE TRIGOM Crawford Brooks Subtidal area 7379 6828 6475–7770 6936 7380 Green algae 1370 1062 Brown algae 1184 1013 1030 Marsh 350 211 112 Intertidal area 3722 3612 2590 3444 High tide area 11,101 10,440 10,360 10,380 238 Northeastern Naturalist Vol. 11, Special Issue 2 Rudy 1985, Vadas et al. 2004b). Hence, the Landsat image was taken closer to the expected peak of abundance than were the CMGE photographs. Habitat definitions must also be considered before comparisons are made. Discussion with the principal CMGE researcher leads to the conclusion that the CMGE class algal flat is equivalent to our combined three green algae categories (B. Timson, Mahoosuc Corp., Augusta, ME, pers. comm.). The comparison shows that the 1966/69 CMGE photographic surveys identified 1062 ha of green algal beds, whereas 1370 ha of green algal mats were recorded in the 1993 Landsat image (Table 4). The long-term comparison of the areas of macroalgae is of particular interest in Cobscook Bay because of the introduction of large-scale salmon aquaculture in the late 1980s (Sowles and Churchill 2004) and an apparent increase in scallop and urchin dragging. While there are no quantitative data on the dragging, there is speculation that the feed and wastes from salmon farming have increased the nitrogenous load of the Bay causing eutrophication resulting in altered plant growth (Brooks 2004, Sowles and Churchill 2004). In other regions subjected to long-term nutrient increases, the result has been a marked increase in annual green algae abundance with a decrease in long-lived brown algae (Middelboe and Sand- Jensen 2000). The CMGE and Landsat studies, separated by nearly a quarter of a century, bracket the development of salmon farming. During this period, the area of green algae apparently increased by 29% (Table 4). During the same time, the area of brown algae increased by 17%. In terms of the percentage of the total intertidal area occupied by each group, green algae increased by 7.4% while brown algae increased by 3.4%. Hence, apparent changes over time in Cobscook Bay do not parallel eutrophication-induced changes documented elsewhere (Middelboe and Sand-Jensen 2000). The apparent changes observed in Cobscook Bay may be within the natural seasonal and interannual variability of these classes and, hence, may not imply anything about the anthropogenic input of nitrogenous compounds from aquaculture on the Bay’s nutrient budget. After all, we are dealing with only two or three data points over two or more decades. Furthermore, Garside and Garside (2004) show that nitrogen has not been limiting in Cobscook Bay for thousands of years and Phinney et al. (2004) demonstrate that temperature and light are the limiting factors in phytoplankton and microphyte production. The links between nutrients, other environmental factors, and green algae abundance are complex suggesting that more research is needed in Cobscook Bay to document the relationships (Schories et al. 1997). Since this research was completed, other satellite-borne sensors have come on-line. These duplicate the spectral resolution of Landsat TM, 2004 P.F. Larsen, S. Barker, J. Wright, and C.B. Erickson 239 but reduce the pixel size to four meters or less. It should be a priority to validate these new sensors for intertidal and shallow water habitat mapping and monitoring. Summary and Conclusions Technological developments in remote sensing of the environment and storing and manipulating diverse environmental data in a spatial framework have greatly advanced our ability to comprehend heretofore inaccessible and/or incompatible data sets. We used Landsat TM imagery followed by unsupervised classification to produce contemporaneous and synoptic estimates of habitat areas in support of ecosystem modeling efforts. These procedures provided sufficiently detailed, objective, and timely information at an affordable cost. Analysis of a 1991 high tide Landsat image provided area estimates and initial confirmation of the existence of an eddy dipole in central Cobscook Bay. These surface patterns were subsequently confirmed by independent drifter studies and help explain the transport of materials in the Bay. Analysis of a 1993 low tide Landsat image classified to the 20- class level produced a thematic map of 11 ecological categories. When the categories were amalgamated into the principal producer groups, the accuracies were, with one exception, over 75%. Area estimates compared well with other specialized studies done at different times using a variety of methods. Long-term comparisons of the principal intertidal producer groups, bracketing the introduction of large-scale salmon aquaculture and increases in bottom dragging, suggest that both green and brown algae made modest gains in area occupied between the CMGE and Landsat TM surveys. This result is inconsistent with eutrophication-induced changes observed elsewhere and may be within normal interannual variability. These comparisons are based on few points in time made by various methodologies. They highlight the need for more frequent analyses using comparable and consistent methods. These results demonstrate that Landsat TM and USC can be a valuable tool for the management and monitoring of macrotidal environments such as Cobscook Bay. Other satellite sensors now exist, however, with equal or superior spectral resolution and greatly enhanced spatial resolution that is more appropriate for the complex habitat heterogeneity of Gulf of Maine intertidal environments. Satellite remote sensing appears to be an accurate and cost effective method for long-term monitoring of the abundance and distribution of intertidal macroalgae and the implied underlying nutrient conditions. 240 Northeastern Naturalist Vol. 11, Special Issue 2 Acknowledgments This investigation was supported by the Cobscook Bay Marine Research Program underwritten by the Andrew W. Mellon Foundation and The Nature Conservancy. Barbara Vickery of the Maine Chapter of The Nature Conservancy provided administrative leadership. We gratefully acknowledge her skill, enthusiasm, encouragement, and patience. Other aspects of this interdisciplinary, multi-institutional research were lead by (alphabetical order): Brian Beal, University of Maine-Machias; David Brooks, Texas A&M University; Daniel Campbell, US Environmental Protection Agency; Chris Garside, Bigelow Laboratory for Ocean Sciences; David Phinney, Bigelow Laboratory for Ocean Sciences; John Sowles, Maine Department of Marine Resources; Robert Vadas, University of Maine; and Charles Yentsch, Bigelow Laboratory for Ocean Sciences. It is a rare experience to work with such a positive, mutually supportive and good-natured group. Several people contributed to the results of the remote sensing exercise, and the project in general, by their support and encouragement. These include Stewart Fefer and Richard Smith, US Fish and Wildlife Service Gulf of Maine Coastal Program; David Phinney and Emily Chase, Bigelow Laboratory for Ocean Sciences; John Sowles, Maine Department of Marine Resources; Dan Wirtz and Jason Sardano, University of New England; Stephen Dickson, Maine Geological Survey; Ralph Keyes, Wiscasset High School; Susan Caldwell and Jim Dow, the Maine Chapter of The Nature Conservancy; and the many knowledgeable residents of the Cobscook region who provided feedback through their participation in the periodic workshops in Eastport and Lubec. Discussions with Barry Timson were very insightful. The manuscript was improved significantly by Tom Trott’s sharp eyes and pencil. The Maine Department of Marine Resources provided a set of aerial photographs. Landsat images were provided by the Maine Geological Survey and the Maine Gap Analysis Program of the University of Maine. Literature Cited Beal, B.F., R.L. Vadas, Sr., W.A. Wright, and S. Nickl. 2004. Annual aboveground biomass and productivity estimates for eelgrass (Zostera marina L.) in Cobscook Bay, Maine. Northeastern Naturalist 11(Special Issue 2):197–224. Belward, A.S., J.C. Taylor, M.J. Stuttard, E. Bignal, J. Mathews, and D. Curtis. 1990. An unsupervised approach to the classification of semi-natural vegetation from Landsat Thematic Mapper data: A pilot study on Islay. International Journal of Remote Sensing 11:429–445. Berry, H., and R. Ritter. 1995. Puget Sound intertidal habitat inventory 1995: Vegetation and shoreline characteristics classification methods. Unpublished Report, Washington State Department of Natural History, Olympia, WA. 31 pp. Brooks, D.A. 2004. Modeling tidal circulation and exchange in Cobscook Bay, Maine. Northeastern Naturalist 11(Special Issue 2):23–50.. 2004 P.F. Larsen, S. Barker, J. Wright, and C.B. Erickson 241 Brooks, D.A., M.W. Baca, and Y.-T. Lo. 1999. Modeling tidal circulation and exchange in Cobscook Bay, Maine. Estuarine, Coastal, and Shelf Science 49:647–665. Campbell, D.E. 2004. Evaluation and emergy analysis of the Cobscook Bay ecosystem. Northeastern Naturalist 11(Special Issue 2):355–424. Congalton, R.G., and K. Green. 1999. Assessing the Accuracy of Remotely Sensed Data: Principals and Practices. Lewis Publishers, Boca Raton, fl. 137 pp. Crawford, S.E. 1999. Results of a rockweed biomass inventory of Cobscook Bay conducted by Quoddy Spill Prevention Group. Unpublished report. Eastport, ME. 7 pp. Available from author on request. Garside, C., and J.C. Garside. 2004. Nutrient sources and distributions in Cobscook Bay. Northeastern Naturalist 11(Special Issue 2):75–86. Kelley, J.T., and A.R. Kelley. 2004. Controls on surficial materials distribution in a rock-framed, glaciated, tidally dominated estuary: Cobscook Bay, Maine. Northeastern Naturalist 11(Special Issue 2):51–74. Larsen, P.F. 2004. Notes on the environmental setting and biodiversity of Cobscook Bay, Maine. Northeastern Naturalist 11(Special Issue 2):13–22. Larsen, P.F., and R.V. Webb. 1997. Cobscook Bay: An environmental bibliography. Bigelow Laboratory Technical Report #100. Published by the Maine Chapter of The Nature Conservancy, Brunswick, ME. 145 pp. Maine Geological Survey. 1976. Coastal Maine geological environments. Open File, Maine Geological Survey, Augusta, ME. Maine State Planning Office. 1983. The Geology of Maine’s Coastline. Maine State Planning Office, Augusta, ME. 79 pp. Middelboe, A.L., and K. Sand-Jensen. 2000. Long-term changes in macroalgal communities in a Danish estuary. Phycologia. 39:245–257. Muir, J. 1997. Remote sensing as an effective tool for coastal management with special reference to the St. Croix estuary area. M.Sc. Thesis. University of Warwick, Warwick, UK. Mumby, P.J., E.P. Green, A.J. Edwards, and C.D. Clark. 1997. Coral reef habitat-mapping: How much detail can remote sensing provide? Marine Biology 130:193–202. Mumby, P.J., W. Skirving, A.E. Strong, J.T. Hardy, E.F. LeDrew, E.J. Hochberg, R.P. Stumpf, and L.T. David. 2004. Remote sensing of coral reefs and their physical environment. Marine Pollution Bulletin 48:219–228. Phinney, D.A., C.S. Yentsch, and D.I. Phinney. 2004. Primary productivity of phytoplankton and subtidal microphytobenthos in Cobscook Bay, Maine. Northeastern Naturalist 11(Special Issue 2):101–122. Pregnall, A.M., and P.P. Rudy. 1985. Contribution of green macroalgal mats (Enteromorpha spp.) to seasonal production in an estuary. Marine Ecology Progress Series 24:167–176. Schories, D., A. Albrecht, and H. Lotze. 1997. Historical changes and inventory of macroalgae from Konigshafen Bay in the northern Wadden Sea. Helgolander Meeresunters 51:321–341. 242 Northeastern Naturalist Vol. 11, Special Issue 2 Shenton, E.H., and D.B. Horton. 1973. Literature review of the marine environmental data for Eastport, Maine. TRIGOM Publication No. 2A. Volumes I and II. The Research Institute of the Gulf of Maine, Portland, ME. Sowles, J.W., and L. Churchill. 2004. Predicted nutrient enrichment by salmon aquaculture and potential for effects in Cobscook Bay. Northeastern Naturalist 11(Special Issue 2):87–100. Thomson, A.G., R.M. Fuller, and J.A. Eastwood. 1998. Supervised versus unsupervised methods for classification of coasts and river corridors from airborne remote sensing. International Journal of Remote Sensing 19:3423–3431. Vadas, R.L., B.F. Beal, W.A. Wright, S. Nickl, and S. Emerson. 2004a. Growth and productivity of sublittoral fringe kelps (Laminaria longicruris) Bach. Pyl. in Cobscook Bay, Maine. Northeastern Naturalist 11(Special Issue 2):143–162. Vadas, R.L., B.F. Beal, W.A. Wright, S. Emerson, and S. Nickl. 2004b. Biomass and productivity of red and green algae in Cobscook Bay, Maine. Northeastern Naturalist 11(Special Issue 2):163–196.