Seagrass mapping using
Landsat TM data Peter J. Lennon Sunmap Remote Sensing centre, Department of Geographic Information, Queensland, Australia Paul Luck Fisheries Branch, Department of Primary Industries, Queensland, Australia. Abstract Seagrass communities are highly productive coastal systems which form important nursery grounds for fishes and crustanceans and provide a primary source of food for dugong and turtles. Using image processing techniques, Landsat TM data was processed to map seagresses In a 95 km long estuarine habitat on the east coast of Australia. Area measurements were made and mapping accuracy assessment was undertaken. Introduction Seagresses are so called because many of them have ribbon-like grassy leaves. Some bear no resemblance to grasses at all and none of them are a true grass. All have prostrate stems buried in sand or mud and produce leaves on erect branches which vary in length from less than a millimetre to half a metre or more. Seagrasses grow in an area between the mid intertidal zone and down to a depth depending largely on clarity of water and form an important part of coastal ecosystems. The leaves of the larger species shelter underlying sediments from erosion and provide a habitat and in some cases food for resident animals such as shrimp and fish, particularly juvenile species. The meadows act as nurseries to a wide range of crustaceans ( especially shrimp). Seagresses are also the primary food resource for dugongs and the green turtle ( Chelonia midas) (Poiner et. al. 1987). Several scientists have demonstrated a clear relationship between the size of nursery area and fishery catches (Taylor and Saloman 1968, Staples 1984). The larger the onshore nursery area the greater the fishery catch. Poiner ( 1989) reported the loss of >100 km1 of dense seagrass meadows in the Gulf of Carpentaria, Queensland, after a tropical cyclone. Examination of fish and prawn catches in subsequent months indicated a significant drop in the size of the catch as a result of the loss. Even lower density seagrass beds are quite important since they appear to be favoured by herebivorous species such as dugoing ( Preen et al. 1989). The production of accurate maps of these valuable ecological areas may better facilitate the development of management strategies and enable the protection of seagrasses for both commercial and environmental reason. The area of study for this seagrass mapping exercise encompasses Great Sandy Strait between the continental mainland and Frasen Island on the east coast of Queensland, Australia. It is an extensive 95 Kilometre long estuarine environment relatively undisturbed by human activity. There are vast areas of intertidal seagrasses, as well as submerged seagrass beds, some of which are not exposed at low tides at all. The environment of Landsat TM data to map seagrass area. Traditional methods of mapping Seagrasses In Australia the mapping of seagrasses by remote sensing techniques has, in the past , involved the use of aerial photography, in particular , colour and colour infrared aerial photography. Sometimes this involves special flights for this purpose. If the coast of these special flights is prohibitive, then the first problem encountered may well be the age of the existing photography of any exists. Other problems with the use of aerial photography include the effects of sun flare and the need to mosaic photographs covering large areas. Extensive field surveys are undertaken in conjuction with the photography. Current field survey methods involve many hours of sampling on the banks, identifying and mapping different species and densities. It is a difficult task to perform especially at low tides. Mud banks are often quite soft and almost impossible to traverse. Tides, currents, and water clarity can of conditions, can be an exhausting task. Quantitative measurements underwater of biomass, densities of cover and species cohabitation, are a less accurate and more time consuming task than measurements in the exposed intertidal areas. Just to look at permanently underwater seagrass meadow can be difficult. If the waters are relatively clear it is sometimes possible to look at the beds by viewing through the bottom of waters are turbid, it is often necessary to scuba dive and in such cases visibility is still limited and so is assessment. Coarse sampling methods can be undertaken by using certain devices that can be dropped overboard and lowered to the bottom to grab a sample of the substrate. It is often necessary to synchronise the survey with tide times as well as good weather. All of these methods are not entirely accurate methods of sampling and assessment. These traditional methods of study are costly and time consuming, involving numerous staff, vehicles and vessels. The relatively small areas that are surveyed are not easily monitored on a regular basis and the relevance of the information dates quickly. Financial resources usually limit the amount of estuarine areas that can be surveyed and monitored using these traditional techniques. The results are not usually quantitative and area not regularly repeated on a monitoring basis. The cost of performing these surveys and of flying to obtain new aerial photography is an important consideration and can be prohibitive. As a consequence, researchers have begun to assess the coasts and utility of other methods and sensors, such as satellite remote sensing using the Landsat satellites. Mapping Seagrasses by Satellite As with most operations, there are advantages and disadvantages with the use of satellite data. Some limitations of acquisition and processing of satellite information include: spatial resolution; cloud cover; satellite passover time and tide; water turbidity and depth: expensive computer trained staff to process and understand data. Some advantages in the use of Landsat Thematic Mapper (TM) satellite data for seagrass mapping are: coast effectiveness: timeliness; quickly sensitivity; monitoring capability; large area mapped quickly; quantitative information obtainable; potential reduction in fields work; and cartographic product easily produced. It is for these reasons that coastal remote sensing has potential in coastal resource management. The most significant of these advantages is perhaps the coast effectiveness of using satellite data. The cost of mapping a large area such as Great Sand Strait using Landsat TM satellite data has been estimated to be US $6700.00 ( Queensland Department of Geographic Information charges fro purchase of data US $1900, image processing charges US $3400, and labour US $1350. Comparing this to the cost of surveying the strait by conventional means US $53200 (estimate made by Department of Primary Industries, 1987 fro use of equipment and lobour), clearly demonstrates the cost effectiveness of using satellite data where possible. However, this methodology has a significant time frame advantage as wel. THe image processing component of mapping Great Sandy Strait seagrasses using Landsat TM data was performed in approximately 15 days. The subsequent field check of the maps that were produced required another 18 days ( three persons for six days) giving a total of 33 days. To produce similar results by conventional field survey methods has been estimated to take somewhere between six and twelve months (Queens) nand Department of primary Industries estimate, 1987). The Landsat TM image used for this study was chosen because it was captured close to low tide. It is most probable that more accurate information will be obtainable about seagrasses exposed exposed above the tide. than seagrasses below water level, where the variables associated with delineation of the beds. The satellite image (path 90, Row 77 centred at G) was captured at 9:18am on 21st September, 1988, by Landsat 5. Low tide did not correspond with the satellite passover at any place in the strait. However, this particular passover was the best and closest to low tide for many months. One of the main problems associated with using satellite data, namely the problem of cloud cover, is evident here. To match satellite passover time in the area of interest with optimum low tide and clear weather conditions is not a common event. The six band TM image (excluding the thermal band, band 6) was then rectified on an image processing system to a 1:50 000 scale hydrographic chart of the Great Sandy Strait area. The two environments where seagrasses occur, (Exposed and submerged), one a wet mud or sandbank covered with a varying depth of water and with seagrass vegetation standing upright, are too different physically and spectrally for treatment together. The differing environments require separate selection of bands and separate treatment of the two environments in image processing. In making the band selections, it was necessary to investigate the range of data in each band. Exposed seagrass areas were examined by first creating smaller images covering only those areas of exposed banks with both known seagrass areas and known bare substrates and then calculating correlation matrices of the spectral bands for these small areas. An examination of the correlation matrices revealed that in all three areas examined, TM bands 4 (near infrared), 5 ( near infrared) and 7 (middle infrared) were the least correlated. The use of these bands for image processing or to create false colour composites should give the greatest spectral separation of features in the exposed intertidal zone. Band selection for use in submerged areas below water areas (the blue, green and red bands [bands 1, 2 and 3]), these bands must be included in the mapping of submerged areas. False colour composite image were created using the optimum bands for both exposed and submerged. Sea grass areas. These false colour images may be useful to resource managers in interpretation of sea grass areas. However, a more quantifiable result is provided from image classification. Classification of Exposed Seagrass Image classification allows for the image to be transferred into some quantifiable form with classes developed to represent the features on the ground. An appropriate technique in image processing before classification is to mask from the image data those areas that are not the primary concern of the task at hand. In this case the area of the image below the tide level at the time ( all water pixele), and the terrestrail vegetation areas leaving only the exposed intertidal areas. The new reduced or masked image allows greater classification accuracies and enables the use of smaller numbers of classes in the classification. An unsupervised Nearest Neighbour classification algroithm was performed of this new masked image using the least correlated TM bands, bands 4, 5 and 7, and the ratio of bands 4.3. Bands 4, 5 and 7 gave the best differentiation in exposed intertidal zones as determined in band selection. The infrared/red ratio was included to assist in the delineation of seagreasses in the exposed intertidal zones as determined in band selection. The infrared/red ratio was included to assist in the delineation of seagrasses in the exposed intertidal zones since the ratio is known to exhibit a good relationship to biomass for vegetation with is known to exhibit a good relationship to biomass for vegetation with simple structures such as grasses (Budd and Milton 1982). The spectral classes obtained were analyzed using the micro BRIAN software analytical techniques such as canonical Variates (CV) analysis, the Minimum Spanning Tree (MST) and clustering methods. Only two classes occupied areas in the exposed intertidal zone and were assumed, by visual interpretation of the false colour imagery and aerial photography of the area, to be possible seagrass area. Classification of Submerged Seagrass Different bands (band 1 blue, 2 green, 3 red and 4 infrared) were used to represent the range of variables of the region for submerged seagrasses. Band 4 was used to mask out land areas of the image using micronBRIAN's software to spectrally digitise in the water areas only and digitise out the land areas. An unsupervised classification was performed on the water areas remaining suing bands, 1,2, and 3. Sixty-eight classes were obtained and the classification using the CV plot and MST produced by micronBRIAN. Several groups of classes were easily identifiable from the plot, namely those that made up deep waters, shallow water and bare sand areas. Five amalgamated seagrass classes became apparent after using the MST and CV analysis techniques in conjuction with visual association of the false colour imagery and aerial photography. User oriented cartographic products in the form of inkjet plots were produced showing the distribution of the seven classes (two exposed and five intertidal) of seagrasses in the 95 kilometre long Great Sandy Strait. The maps were quantified to give area measurements for the total area fo seagrasses in the strait. Plots were produced of four smaller areas of the strait chosen to represent the range of environments with seagrass habitats. Theinkjet plots were taken into the study area and the classification was field checked. The purpose of the field check was to obtain an represented on the ground. The filed observations indicated that the unsupervised classification created the seagrass classes in both the exposed and submerged areas on the basis of density of cover. No species differentiation was evident in the class separations. It appeared that seagrasses had been discriminated from other submerged features to a depth of approximately one and a half metres. Accuracy assessment was made of the classification using both field check and aerial phtography techniques giving a mapping accuracy for the total area of seagrass cover as greater than 83%. Discussion Traditional methods of performing this mapping task over this large area may not have achieved more accurate results, unless a very intensive and costly field sampling methodology was undertaken. The accuracy achieved is quite acceptable in comparison, using the satellite remote sensing techniques. The study area in which the mapping was performed, Great Sandy strait, encompasses a wide range of coastal environmental conditions. Considering this wide range of variables and the acceptable accuracy of the classification, the methodology may also be transferable to other areas (providing that water turbidity or cloud cover is not a serious problem). of considerable advantage to the resource manager arising from this methodology, is the ease of producing the paper maps. Traditional methods require the conversion of fields notes into a cartographic product. This task not only requires certain skills in cartography but also takes considerable time in terms of labour, making it costly. The remote sensing produced image maps were supplied rectified to a map projection, plotted to a desired scale and in the case of the map produced for this study, the subject information was overlain on to a photograph-like image background with relevant nomenclature and legend information. The advantages of this type of quick cartographic product are obvious and could well be one of the main attractions for using remotely sensed data. the map is virtually produced during image processing and analysis. This study has also explained that the purchase of low tide imagery maximizes the amount of seagrass areas that are not subject to the reflectance distortions resulting from being covered by waters. In turbid water cases not all of the seagrass may be mapped but with the purchase of imagery captured near low tide times, the mapping of large areas of exposed intertidal seagrasses is still possible. In this study submerged vegetation was discriminated to a depth of one to two metres depending on turbidity of waters. Very little seagrass was found below that depth. The most important component of this study was the development of a suitable methodology to use Landsat TM data to delineate the areal extent of exposed and submerged seagrasses in a large study area. Some problems were described with using Landsat TM data and problems will be encountered in future studies with the impenetrability of more turbid waters than those encountered in this area. However, the advantages of using Tm data in many areas of the mapping was determined to be the disadvantages. The accuracy of the mapping was determined to be probably better than 83% and this good result indicates that the use of Landsat TM satellite data to map seagrass areas is a time-and coast-effective alternative to traditional methods of mapping these important resources. In conclusion, this study has demonstrated that image processing of Landsat TM data is a most suitable alternative method of accurately mapping the areal extent of seagrasses over a large area. Future for Seagrass Mapping New equipment is coming into use for marine resource managers and others and may vastly improve the capacity for mapping and monitoring. Position fixing in field surveys, may no longer be a serious problem with the advent (onto the market ) of portable Global positioning system obtaining accuracies of 15 metres for the equipments cost of approximately US $5700. The fixing of position of seagrasses and similar resources may no longer involve the expense of undertaking a complete ground survey with the services of a surveyor and accompanying equipment. Field checking of features that are not easily locatable may be much easier. One of the more important aspects of resources monitoring is how the mapping results can be utilized. Traditional paper maps do not easily facilitate the comparison of mapping results over time. Digital remotely sensed products may be transferred and overlaid with other remotely sensed products and with other relevant information in a geographic information system (GIS). The GIS may provide a meshed structure of information and will facilitate enquiry of differences between many surveys and sources of information. To the resource manager this tool may also be of considerable advantage. The computerised systems offered by these geographic information systems may make data storage, retrieval and enquiry much more efficient and may lead to better and more cost efficient resource management strategies. Remote sensing imagery and equipment is also becoming more readily available. New satellites offer the choice of more frequent data acquisition and improving resolutions both spectrally and spatially Personal - computer-based image processing systems are making the purchase managers. These new technologies will transform existing resource management techniques. The future for seagrass mapping and management may well contain the use of all of these new tools. Acknowledgement The authors gratefully wish to acknowledge the support of Mr. Stuart Hyland of the Queensland Department of Primary Industries Fisheries Branch. Bibliography
|