in coastal regions and in small islands
Coastal management sourcebooks 3
Guidelines for Busy Decision Makers
The purpose of this Handbook is to enhance the effectiveness of remote sensing as a tool in coastal resources assessment and management in tropical countries by promoting more informed, appropriate and cost-effective use.
A primary objective of the Handbook is to evaluate the cost-effectiveness (in terms of accuracy of resource estimates or habitat maps) of a range of commonly used remote sensing technologies at achieving objectives identified by the user community. The research activities reported were focused on the Turks and Caicos Islands with comparative work in Belize but outputs are seen as applicable in all countries where water clarity permits optical remote sensing.
Sensor technologies evaluated included: Landsat Multispectral Scanner (MSS), Landsat Thematic Mapper (TM), both multispectral (XS) and panchromatic (Pan) SPOT High Resolution Visible, Compact Airborne Spectrographic Imager (CASI) and aerial photographs. Spatial resolution of the digital imagery ranged from 1–80 m, and spectral resolution from 1–16 wavebands in the visible and infra-red.
For habitat mapping, which is the primary objective of users, the achievable accuracy of outputs depends on: i) the level of habitat discrimination required, ii) the type of sensor used, iii) the amount of ground-survey carried out, and iv) the image processing techniques used. Accuracies of outputs are evaluated for different levels of marine or mangrove habitat discrimination for each sensor with varying ground-survey and image processing inputs (costs).
Satellite-mounted sensors are only able to provide information on reef geomorphology and broadscale ecological information such as the location of coral, sand, algal and seagrass habitats with an accuracy ranging from 55–70%. For this type of information, the most cost-effective satellite sensors for habitat mapping are Landsat TM (for areas greater than one 60 x 60 km SPOT scene) and SPOT XS for areas within a single SPOT scene. Both these sensors can deliver overall accuracies of about 70%. The remaining satellite sensors tested (SPOT Pan and Landsat MSS) could not achieve 60% overall accuracy even for coarse-level (only 4 marine habitat classes being distinguished) habitat discrimination.
For fine-scale habitat mapping (9+ habitat classes) only colour aerial photography and airborne multi-spectral digital sensors offer adequate accuracy (around 60% or better overall accuracy). The most accurate means of making detailed reef or mangrove habitat maps involves use of digital airborne multi-spectral instruments such as CASI (Compact Airborne Spectrographic Imager). Using CASI we were able to map sublittoral marine habitats to a fine level of discrimination with an accuracy of over 80% (compared to less than 37% for satellite sensors and about 57% for colour aerial photography). Similar accuracy was achieved for mangrove habitats using CASI. Comparison of the costs of using CASI and colour aerial photography indicates that CASI is cheaper as well as producing more accurate results.
7. Additional field studies were carried out of the capabilities of remote sensing to map bathymetry and assess mangrove resources (Leaf Area Index (LAI) and percentage canopy closure) and seagrass standing crop (g.m-2). Remote sensing offers very cost-effective rapid assessment of the status of the plant resources but passive optical remote sensing of bathymetry is considered too inaccurate for most practical purposes. The practicalities of using remote sensing technologies for i) coastal resources (fisheries, conch (Strombus gigas), Trochus, and seaweed resources) assessment, and ii) monitoring of coastal water quality (including sediment loadings, oil, thermal discharges, toxic algal blooms, eutrophication and other pollution) are reviewed.
Habitat maps derived using remote sensing technologies are widely and increasingly being used to assess the status of coastal natural resources and as a basis for coastal planning and for the conservation, management, monitoring and valuation of these resources.
Despite the fact that Earth resources data from some satellites has been routinely available for over 25 years, there has until now been almost no rigorous assessment of the capacity of the range of operational remote sensing technologies available to achieve coastal management-related objectives.
Digital sensors commonly used for coastal management applications have spatial resolutions ranging from about 1 m to 80 m on the ground and spectral resolutions ranging from a single panchromatic band (producing an image comparable to a black and white photograph) to around 16 precisely defined wavebands which can be programmed for specific applications. Costs of imagery range from about £250 for a low-resolution (80 m pixel size) satellite image covering 35,000 km2 to around £80,000 for a high-resolution (3 m pixel size) airborne multispectral image covering less than half this area. In addition, high-resolution analogue technologies such as colour aerial photography are still in routine use.
Coastal managers and other end-users charged with coastal planning and management and the conservation and monitoring of coastal resources require guidance as to which among this plethora of remote sensing technologies are appropriate for achieving particular objectives. A study of reports of the use of remote sensing in developing countries and research of other available literature indicated that there is wide use of the various technologies but no clear idea of the extent to which different technologies can achieve objectives. In addition, objectives are often poorly defined.
Data which would allow clear guidance to be given have not been available, with most remote sensing research being devoted to potential applications of an often experimental nature and little attention being paid to the operational realities or the costs of applications. This Handbook, based on three years research funded by the UK Department for International Development (DFID), seeks to fill the gap and provide practical guidance to end-users, particularly those in developing countries.
The Caicos Bank was chosen as the site for the research to discover what coastal management objectives are realistically achievable using remote sensing technologies. It offered almost ideal clear-water conditions and a broad mix of coastal habitat types, a very large area (>10,000 km2 ) of shallow (<20 m deep) coastal waters which would be amenable to the technologies. Results from the site can thus be considered as best-case scenarios. To test the generic applicability of the outputs, additional work was carried out in Belize with the UNDP/GEF Belize Coastal Zone Management Project and Coral Cay Conservation. Of particular concern was the quantification of the accuracies achievable for all outputs, since habitat or resource maps of poor or unknown accuracy are of little use as a basis for planning and management. The key research findings are listed hereafter.
Uses of remote sensing in the coastal zone
Chapter 2 of the Handbook examines the objectives of coastal managers in using remote sensing. Coastal managers and scientists around the world were asked a) to identify what they saw as the primary applications of remote sensing and b) to prioritise the usefulness to them of various levels of information on coastal systems.
The most in-demand applications of remotely sensed data were to provide background information for management planning and to detect coastal habitat change over time (Figure 1). The term ‘background information’ reflects the vagueness with which habitat maps are often commissioned and indicates a need for objectives of remote sensing surveys to be defined more rigorously. Some 70% of respondents who were using remote sensing for change detection were concerned with mangrove assessment and/or shrimp farming. The primary uses made of the habitat/resource maps are shown in Figure 1.
Matching available technologies with managers’ objectives
Analysis of the questionnaire responses revealed that managers sometimes had unrealistic expectations of what remote sensing could achieve, notably in the management of seagrass and coral reef habitats. However, for mangroves the outputs which managers considered most useful (mapping boundaries, clearance and density) are readily achievable using current technologies. The Handbook seeks to close the gap between managers’ expectations and what can be realistically achieved with remote sensing in an operational as opposed to research context.
The need to integrate field survey and remote sensing
Remote sensing is often (erroneously) considered as an alternative to field survey. It should be seen as a complementary technology which makes field survey more cost-effective. A solely field-survey based approach to habitat mapping is shown to be extremely cost-inefficient. For example, it would cost about £0.5 million excluding staff salaries to produce a coarse-level habitat map of the ca 15,000 km2 Caicos Bank to reasonable accuracy without remote sensing inputs and take a team of three more than eight years! On the other hand, a remote sensing approach without extensive field survey is shown to be too inaccurate. For example, <50% overall accuracy is achievable for unsupervised classification (i.e. one without field survey inputs) of a four habitat class (reef, macroalgae, sand, seagrass) Landsat TM image subjected to the highest level of processing. The integrated approach is summarised in the following box:
The end-user community has too often failed to appreciate this; the Handbook (particularly Chapters 9 and 10) seeks to remedy this misapprehension, emphasising the need for an integrated field and image-based approach.
Cloud cover is a key constraint
Most projects requiring remotely sensed imagery have to be completed within a timeframe of one to three years. The question of whether appropriate imagery is likely to be obtainable in such a timeframe is sometimes over-looked at the planning stage. A primary constraint which project planners need to consider when using optical imagery is cloud cover. This constraint does not apply to radar imagery such as that from synthetic aperture radar (SAR) sensors, which can operate through cloud. However, radar does not penetrate water and thus is of limited use for coastal management.
The extent to which cloud cover is a constraint to image acquisition was analysed for a range of developing nations from those with arid desert coastlines to those comprised of high islands in the humid tropics (Chapter 5 of the Handbook). Data sources were both archived imagery and local weather stations. In the humid tropics there was a 4–10% chance (depending on location) of obtaining a satisfactory image (defined as one with <12.5% cloud cover) on a given satellite overpass, equivalent to a 41–77% chance of at least one good image per year for SPOT sensors (if not specially programmed) or a 58–91% chance per year for Landsat sensors, assuming these are recording at the time of every overpass. However, sensors are often inactive over tropical areas (particularly small-island states) because of low commercial demand and this is reflected in the relatively low numbers of archived images available for such areas. This means that there is often a less than even chance of obtaining a reasonable image in a year. Planners and managers need to be aware that this can present a significant constraint to applying satellite remote sensing technologies in developing countries, particularly where projects have short timeframes or require up-to-date imagery.
Maps of unknown accuracy are of little value to managers
The manager or other end-user needs to know the accuracy of the habitat maps or other remote sensing products which are to be used in planning and management (Plate 1). We found that only a quarter of 86 tropical studies reviewed included any assessment of accuracy. Although additional field survey costs will be incurred to enable an accuracy assessment of a habitat map to be carried out, one has to ask what value a map of poor or unknown accuracy has as a basis for management and planning? The Handbook emphasises throughout the need for accuracy assessment and describes the statistical and field survey methodologies required (Chapter 4). Such assessment must be an integral part of any remote sensing survey if the results are intended for more than mere wall decoration.
The importance of rigorous methodologies for defining habitats
In order for remote sensing studies to realise their full usefulness, particularly for the key objective of change detection identified by end-users (Figure 1), the field survey methodologies used to define habitats need to be rigorous and preferably quantitative so that habitat types defined in one study can be clearly identified in subsequent ones. Ad hoc methods, though quick and cheap, produce outputs of very limited use and are a false economy. Chapter 9 reviews which methodologies are suitable for different objectives and advocates a hierarchical approach to allow a) descriptive resolution (i.e. the level of habitat discrimination) to be tailored to management objectives and sensor capabilities, and b) habitat categories to reflect natural groupings appropriate to the are a under study.
The need for atmospheric and water column correction
Most remotely sensed images will be supplied already geometrically corrected, so that they can be used like a map (Chapters 5 and 6). For most coastal applications the images will also require radiometric and atmospheric correction. These corrections take into account sensor peculiarities and the effects of atmospheric haze, the angle of the sun at the time the image was acquired, etc. which would otherwise make it impossible to compare one image with another in any meaningful way. For underwater habitat mapping, water column correction is also recommended to compensate for the effect of water depth (due to attenuation of light in the water column) on the signal received by the sensor. If this is not done, then it is difficult to distinguish marine habitats reliably because depth effects dominate the image.
Remarkably, water column correction was carried out in only 9% of 45 underwater habitat mapping studies reviewed, often leading to unnecessarily low accuracies in habitat classification and thus inefficient use of resources. The main reason for the poor uptake of the technique appears to be that users have found the rather mathematical research papers describing the techniques daunting and difficult to implement. The techniques are thus under-utilised despite being essential for cost-effective achievement of many common management objectives. Chapters 7 and 8 seek to present these image processing techniques in a more accessible and intuitive way so that they will be more routinely adopted.
For change detection (one of the two principal objectives of remote sensing identified by the user-community) radiometric and atmospheric correction are essential if information on habitat spectra generated by the supervised classification of one image are to be made use of in classifying subsequent or neighbouring images. Not using such information is very wasteful of resources.
Water column correction significantly improves the accuracy of habitat maps for multispectral sensors with more than two wavebands that penetrate water, such as Landsat TM and CASI. However, for those multispectral sensors with only two wavebands which penetrate water (SPOT XS and Landsat MSS) water column correction is likely to be ineffective. This is because in these sensors any gains from the depth-invariant processing to carry out water column correction seem to be balanced by reduced discrimination resulting from the loss of one of the two dimensions of spectral information.
There are three main approaches to classifying remotely sensed images (assigning pixels in an image to habitat types). These are:
unsupervised classification of multispectral imagery, and
supervised (using field survey information) multispectral image classification.
These approaches are described and evaluated in the Handbook for a range of applications (Chapters 10–13). All have their uses although supervised multispectral classification based on extensive field survey was found to be most effective and generally produced the most accurate habitat maps. Visual interpretation of remotely sensed imagery (identification by eye of habitat types based on their colour, tone, texture and context within the imagery) can often reveal considerable detail on the nature and distribution of habitats. However, the subjectivity and operator dependence of visual interpretation is a major drawback, particularly if comparisons over time are envisaged.
Unsupervised classification (i.e. a computer-based classification or one based upon a good local knowledge of the habitats, local maps and field experience but with-out use of field survey data from known positions) produced maps with unacceptably low overall accuracies which for Landsat TM were less than 50% at coarse habitat resolution (compared to over 70% using supervised classification).
Habitat maps produced using supervised multispectral image classification can be significantly improved by additional image processing such as water column correction (see above) and contextual editing (the addition of context-dependent decision rules into the image classification process – done automatically by the brain in visual interpretation). Gains in accuracy ranged from 6–17% per day’s processing effort. Use of contextual editing and water column correction together significantly improved the accuracy of habitat maps derived from all the multispectral sensors tested.
Operational versus theoretical spatial resolution of aerial photography
Colour aerial photographs at a photographic scale of 1:10,000 to 1:25,000 can offer potential ground resolution of the order of 0.5–1 m (the minimum detection unit). However, conventional mapping techniques do not make use of this resolution because to define polygons at such fine spatial scales would be prohibitively time-consuming and practically difficult. Thus minimum mapping units at the photographic scales above are in the order of 10–20 m in diameter, equivalent to the spatial resolution of SPOT satellite sensors. At least one order of magnitude of spatial resolution is therefore lost during preparation of habitat maps using standard aerial photographic interpretation (API) techniques. However, the high resolution does provide extra detail on the texture of habitats, which facilitates the visual interpretation process and thus allows more accurate assignment of habitats than would be true if the minimum detection unit were the same as the minimum mapping unit.
Mapping coral reefs and macroalgae
Satellite-mounted sensors are able to provide useful information on reef geomorphology and broadscale ecological information such as the location of coral, sand, algal and seagrass habitats with accuracies of around 70% for the best satellite sensors (Landsat TM and SPOT XS). However, even the best sensors could not provide medium to finescale ecological classifications of habitats at better than 50% to 35% accuracy respectively and, unless such low accuracies are acceptable, we do not recommend using satellite imagery for such purposes.
The most cost-effective satellite sensors for broadscale habitat mapping (e. g. four submerged habitat classes) are Landsat TM (for areas greater than one SPOT scene) and SPOT XS for areas within a single SPOT scene (i.e. <60 km in any direction).
Colour aerial photography can resolve more detailed ecological information on reef habitats but for general purpose mapping (4–6 reef habitat classes) satellite imagery is more effective (i.e. slightly greater accuracy and much faster to use). However, at finer levels of habitat discrimination (9–13 reef habitats) aerial photography performs significantly better than the best satellite sensors.
The most accurate means of making detailed reef habitat maps involve the use of airborne multispectral instruments such as CASI (Compact Airborne Spectrographic Imager). In the Caribbean, CASI can map coral reef habitats at a fine level of discrimination with an accuracy of over 80% (compared to less than 37% for satellite sensors and about 57% for colour aerial photography).
Mapping seagrass beds and assessing seagrass standing crop
The location and extent of seagrass beds can be mapped reasonably accurately with satellite imagery (about 60% user accuracy). Landsat TM appears to be the most appropriate satellite sensor although some mis-classification may occur between seagrass habitats and habitats with similar spectral characteristics (e. g. coral reefs and algal dominated areas). Contextual editing significantly improves the accuracy of seagrass habitat maps. Airborne multispectral digital imagery (e. g. CASI) allows seagrass beds to be mapped with high user accuracies (80–90%) and permits changes of a few metres in seagrass bed boundaries to be monitored. Similarly, the interpretation of colour aerial photography (API) can generate maps of reasonably high accuracies (around 60–70%) but map-making may be time-consuming where seagrass patches are small. Although API is excellent for mapping high-density seagrass (standing crop = 80–280 g. m-2 ) , it is poor at mapping low to medium density seagrass (standing crop = 5–80 g. m- 2 ). A non-destructive visual assessment technique for standing crop was found to be very cost-effective and is recommended for ground-truthing (Chapter 12).
Remote sensing is well suited to mapping the standing crop of seagrass of medium to high density and SPOT XS, Landsat TM and CASI performed similarly overall. To map seagrass standing crop,‘depth-invariant bottom-index’ images are created (using the water column correction techniques discussed above) and those which show the closest correlation between bottom-index and seagrass standing crop for ground-truthed pixels are used to predict standing crop over the whole area of interest.
Accurate predictions of low standing crop are difficult due to confusion with the substrate and increased patchiness but, at higher biomasses, measurement of standing crop by remote sensing compares favourably with direct quadrat harvest and, unlike the latter, has no adverse impact on the environment (see Figure 16.7). Chapter 16 describes how to map seagrass standing crop.
Detecting seagrass degradation is a key application. Allowing for known errors in image rectification and classification we estimate that satellite sensors can detect changes in seagrass bed boundaries in the order of 20–90 m (2–3 pixel widths) with confidence. These are huge changes! Airborne digital sensors are capable of more accurately identifying seagrass habitats and providing spatially-detailed maps of biomass with spatial errors <10 m. This allows monitoring of change in seagrass standing crop in two dimensions at a more useful scale and would, for example, allow the impacts of a point source of pollution (e.g. a thermal discharge) on a seagrass bed to be monitored.
Mapping mangroves and assessing the status of mangrove resources
Mangrove areas are often difficult to reach and equally difficult to penetrate. Thus, field survey of mangroves is logistically demanding particularly where the areas are large. Remote sensing offers a very cost-effective method of extending limited field survey to map large areas of mangroves. Mangrove habitat maps are primarily used for three management applications: resource inventory, change detection and the selection and inventory of aqua-culture sites. Chapters 13 and 17 analyse a range of remote sensing approaches to mangrove mapping and the quantitative assessment of mangrove resources using satellite and airborne imagery, and make recommendations as to the most effective options.
Image processing techniques appropriate for mangrove mapping can be categorised into five main types: 1) visual interpretation, 2) vegetation indices, 3) unsupervised classification, 4) supervised classification, and 5) principal components analysis of band ratios. The image processing method (5) based on taking ratios of different red and infrared bands and using these as inputs to principal components analysis (PCA) generated the most consistently accurate maps of mangroves.
In the Turks and Caicos Islands mangrove vegetation could be distinguished from terrestrial vegetation (thorn scrub) with high accuracies (~90%) using Landsat TM or CASI data. However, it is not always readily distinguishable from adjacent habitats. SPOT XS imagery could not separate mangrove from thorn scrub in the Turks and Caicos Islands, although it has been used to map mangrove successfully elsewhere. Similarly, problems of distinguishing mangroves from rainforest have been reported for Landsat MSS and TM sensors in Australia and Indonesia. At coarse levels of habitat resolution (only two types of mangrove habitat being distinguished), Landsat TM data were found to be more cost-effective than CASI. However, CASI data were greatly superior when it came to mapping mangroves at fine levels of habitat discrimination (nine habitat classes) where Landsat TM had an overall accuracy of only 30%. Using CASI, finescale mangrove habitats of the Turks and Caicos Islands could be mapped to an overall accuracy of 83% (Figure 10.9, Plate 11).
Thus, in the context of qualitative mapping, satellite imagery may achieve little more than mapping mangrove occurrence and distinguishing mangrove from non-mangrove vegetation in areas where mangroves are poorly developed. Where mangroves are well-developed and extensive (e. g. Sunderbans, Niger delta, etc.) satellite imagery may be more effective and allow mapping of major zonation. By contrast, digital airborne imagery can provide detailed and accurate maps of the zonation of mangrove ecosystems even where poorly developed. However, in a quantitative context remote sensing has more to offer.
A number of vegetation indices which can be derived from satellite and airborne digital imagery of mangrove areas show a close correlation with the leaf area index (LAI) and percentage canopy closure of mangroves. LAI can be used to predict growth and yield and to monitor changes in canopy structure due to pollution and climate change. It is thus a good measure of the status and productivity of mangrove ecosystems. Similarly, canopy closure can been used as a measure of tree density. LAI and percentage canopy closure can be predicted with reasonable confidence (Table 1) from Landsat TM, SPOT XS (using the TM image to mask out non-mangrove vegetation) and CASI airborne images using the correlation between Normalised Difference Vegetation Index (NDVI) images and these parameters. Other image-derived indices including the Global Environment Monitoring Index (GEMI) and Angular Vegetation Index (AVI) gave equally good correlations.
|Table 1 The accuracy of LAI and percentage canopy closure prediction using three image types.|
|Landsat TM||SPOT XS||CASI|
|Leaf area index||71%||88%||94%|
Higher spatial resolution SPOT XS imagery (20 m pixel size) was more successful than Landsat TM (30 m pixel size),with the 1 m resolution CASI being best of all. Thus, both satellite and airborne digital imagery can provide useful forestry management information on mangroves which would be very difficult to obtain by field survey.
Monitoring coastal water quality
Chapter 14 of the Handbook reviews how remote sensing may aid coastal managers in assessing and monitoring coastal water quality. Pollution of coastal waters (leading to poor water quality) by, for example:
sediment run-off as a result of deforestation or unsustainable agricultural development,
fertilisers and pesticides from agriculture,
oil spills from land-based or marine installations,
thermal discharges from power-stations or other industry requiring cooling water,
discharges of insufficiently treated sewage,
toxic and other industrial wastes, and
nutrient-rich effluents from coastal aquaculture ponds,
are just some of the issues which coastal managers may be called upon to assess or monitor. Such pollutants (loosely defined here as substances in the water column which are likely to have detrimental effects on the environment at the levels observed – see below for a more formal definition) may have adverse impacts on coastal ecosystems and the economically important resources that these support. Thus, monitoring of ‘water quality’ is in practice about detecting, measuring and monitoring of pollutants.
|Pollution means the introduction by man, directly or indirectly, of substances or energy into the marine environment (including estuaries) resulting in such deleterious effects as harm to living resources, hindrance to marine activities including fishing, impairment of quality for use of seawater and reduction of amenities (GESAMP, 1991).|
Water quality is far more ephemeral and dynamic than habitat status, often changing dramatically over a tidal cycle, or after a monsoon downpour, or following the flushing of an aquaculture pond or the discharge of a tanker’s ballast water. Thus, the usefulness of water quality measurements at a single point in time (as available from a single satellite image) is likely to be far less than, say, knowledge of the habitat type at a particular location.
Since most incidents of pollution are likely to be short-lived (hours to days) and relatively small in size (tens to hundreds of metres across) one needs high temporal resolution and moderately high spatial resolution in order to monitor them. Further, since the pollutants can only be detected and measured if they can be differentiated from other substances in the seawater, one usually also needs a sensor of high spectral resolution.
Such a combination of characteristics is at present only available from airborne remote sensors. Satellite sensors are thus only likely to be useful in coastal water quality assessment where mapping large scale sediment plumes from estuaries or huge oil spills such as that released during the Gulf War of 1990/1991. This means that application of effective remote sensing techniques in pollution monitoring is likely to be expensive (costing one thousand to several thousand dollars a day) because of the costs of hiring sensors and aircraft to fly them and the probable need for repeat surveys to establish the dynamics of the pollution in space and time. These high costs, the relatively small areas which need to be investigated (particularly where point-sources are involved), and the need to collect field survey data for calibration (establishing a quantitative relationship between the amount of pollutant and some characteristic measured by the remote sensor) and legislative purposes anyway, may make the cost-effectiveness of using remote sensing to aid pollution monitoring questionable.
Remote sensing can, at present, only directly detect some of the pollutants listed above, for example, oil, sediment and thermal discharges. The others can in some cases be detected by proxy but only via ephemeral site-specific correlations between remotely sensed features and the pollutants in question. Thus, fertilisers, pesticides, nutrients and colourless toxic and industrial wastes (containing, for example, heavy metals, aromatic hydrocarbons, acid, etc.) cannot be detected but in some instances their concentrations (measured by field survey at the time of image acquisition) may be correlated with levels of substances which can be detected. Such experimentally demonstrated ephemeral relationships are still far from allowing operational detection and monitoring of these pollutants in a cost-effective way.
Whilst remembering these shortcomings, the primary benefits from adding a remote sensing perspective are:
rapid survey for pollutants over large or remote areas which could not feasibly be surveyed from the surface,
a synoptic overview of the pollution and clear demarcation of its boundaries, and,
once reliable calibration has been achieved, the potential for measurement of the pollutant without the need for more field survey.
The Handbook critically reviews the operational use of remote sensing for the study of sediment loadings in coastal waters, oil pollution, sewage discharges, toxic algal blooms and eutrophication, and thermal discharges.
Optical remote sensing has been suggested to offer an alternative to traditional hydrographic surveys for measuring water depth, with the advantage that data are collected synoptically over large areas. However, bathymetry can only be derived from remote sensing to a maximum depth of about 25 m in the clearest water, and considerably less in turbid water. Bathymetric mapping is also often confounded by variation in seabed albedo (e.g. change from white sand to dark seagrass).
Chapter 15 of the Handbook describes and investigates three commonly used methods for mapping bathymetry from satellite imagery, with worked examples from the Turks and Caicos Islands. The accuracy of depth prediction of each method was tested against field data. Predicted depth correlated very poorly with actual depth for all but one method. Even this method does not produce bathymetric maps suitable for navigation since the average difference between depths predicted from imagery and ground-truthed depths ranged from about 0.8 m in shallow water (<2.5 m deep) to about 2.8 m in deeper water (2.5–20 m deep).
By contrast ,airborne LIDAR can produce very accurate (±30 cm) and detailed bathymetric charts to 50–60 m in clear waters and about 2.5 times the depth of penetration of passive optical remote sensing technologies. Boat-based acoustic surveys using single or multibeam depth sounders can produce bathymetric maps of similar accuracies and can operate to depths in excess of 500 m. The drawback of such boat-based techniques is that it may not be feasible to survey shallow water less than 2–3 m deep because of sounder saturation and/or the draught of the survey boat. If bathymetric charts are an objective then airborne LIDAR or boat-based acoustic techniques (depending on depth range required) appear to be the remote sensing technologies to use. However, there may be instances where crude bathymetric maps which indicate major depth contours are useful (see Figure 15.10, Plate 17).
Assessment of marine resources
With rare exceptions (such as some near-surface schooling fish and marine mammals, which can be remotely sensed from aircraft, and seaweeds, which can be mapped from both satellites and aircraft) most marine resources cannot be assessed directly using remote sensing. However, there is potential for assessing some stocks indirectly using a combination of field survey and remote sensing technologies. Chapter 18 of the Handbook briefly considers how remotely sensed characteristics of coastal waters, such as chlorophyll concentration and sea surface temperature, and the ability to discriminate habitats, may help in assessing potential levels of economic resources such as finfish and shellfish.
The progression from mapping of habitats or ocean colour to assessment of living aquatic resources requires the establishment of a significant quantitative relationship between some feature which the remote sensor can reliably detect and levels of the living resource being studied. Establishing such a relationship requires a large amount of field survey data and thus any attempts to assess resources using remote sensing are likely to be costly.
In terms of cost-effectiveness, the question which needs to be asked is whether the costs of the addition of a remote sensing component to the stock assessment study are outweighed by the benefits. Remote sensing does not replace traditional methods of stock assessment, it only augments existing methods and may allow a) a more accurate determination of stocks, and b) some reduction in the need for field survey (once a clear relationship between a feature which can be distinguished on the imagery and levels of the stock being assessed has been established).
The Handbook examines these issues in relation to the assessment of:
phytoplankton and primary production,
coastal fisheries resources (using both direct and indirect remote sensing approaches),
Queen Conch (Strombus gigas) resources,
Trochus resources, and
Cost-effectiveness of remote sensing
The cost-effectiveness of a remote sensing survey is perhaps best assessed in relation to alternative means of achieving the same management objectives. The three primary objectives of remote sensing surveys identified by end-users were: providing a background to management planning, detecting change in habitats over time, and planning monitoring strategies (Figure 1). In all cases the expected outputs are habitat/resource maps of varying detail. Once the political decision has been taken that coastal resource management needs to be strengthened and that as part of this process an inventory of coastal ecosystems is required, the question becomes: ‘Does a remote sensing approach offer the most cost-effective option to achieve this objective?’ Chapter 19 of the Hand-book examines both this question and that of which remote sensing technologies are most cost-effective for which objectives.
The only real alternative to using remote sensing to map the marine and shoreline habitats is to carry out the mapping using boat-based and land-based survey alone. Such surveys would involve inordinate amounts staff time as well as boat and vehicle expenses. We calculate that, even with staff salaries excluded, mapping a large area (e.g. the Caicos Bank) using an integrated remote sensing/field survey approach using Landsat TM for coarse-level habitat discrimination would cost less than 3% of the costs of using field survey alone. Even for a small area of 16 km2 (the median size of marine protected areas), a detailed survey of marine habitats using boat-based surveys would cost nearly double that of hiring CASI in hire charges, not to mention involving an extra 16 person-months of staff time just to collect the data!
The savings achieved from an integrated remote sensing/field survey approach rely on reducing the amount of costly field survey needed and using the imagery to extrapolate reliably to unsurveyed areas. For example, the project collected field survey data from 1100 km2 of the Caicos Bank, about 7% of the area mapped, at a boat-survey cost (excluding staff time) of approximately £2700 over about 6 weeks.
So, although almost 70% of questionnaire respondents considered the cost of remote sensing as a main hindrance to using it for coastal habitat mapping, the issue is not that remote sensing is expensive but that habitat mapping is expensive. Remote sensing is just a tool which allows habitat mapping to be carried out at reasonable cost.
Four types of cost are encountered when undertaking remote sensing: (1) set-up costs, (2) field survey costs, (3) image acquisition costs, and (4) the time spent on analysis of field data and processing imagery. The largest of these are set-up costs, such as the acquisition of hardware and software, which may comprise 40–72% of the total cost of the project depending on specific objectives. However, it should be borne in mind that most of the institutional set-up expenditure (computer hardware and software, maps/charts, almanacs, remote sensing textbooks and training courses, etc.) needed to establish a remote sensing capability is likely to be necessary anyway if an institution is to make effective use of the output habitat maps.
For coarse-level habitat mapping with satellite imagery, the second-most important cost is field survey, which accounts for about 25% of total costs and over 80% of costs if a remote sensing facility already exists (i.e. in the absence of set-up costs). Field survey is a vital component of any habitat mapping programme and may constitute approximately 70% of the time spent on a project (e.g. 70 days for mapping more than 150 km2 of reef with satellite imagery).
The next two sections show how time invested in image processing and in field survey can dramatically improve the accuracy of outputs.
Investment in image processing is time well spent
The relationship between accuracy of outputs and amount of image processing input is examined in Figure 2.
Two key points emerge from Figure 2. Firstly, to achieve acceptable accuracy a considerable investment in staff time is required to carry out image acquisition, correction and subsequent processing; approximately one month for Landsat TM (and other satellite imagery such as SPOT XS), and about 1.5 months for CASI or similar airborne imagery. Secondly, increased image processing effort in general leads to increasing accuracy of outputs. This said, merging Landsat TM data with SPOT Pan to increase spatial resolution, despite producing visually pleasing outputs (see Figure 19.5, Plate 24), produced no benefits in terms of improved accuracy (in fact reducing it). Given the costs of inputs such as field survey data and set-up costs, extra effort devoted to image processing is clearly very worthwhile.
Accurate habitat mapping is dependent on adequate field survey
Field survey represents a major component of total costs. To assess how increasing field survey affects the accuracy of outputs, 25 % , 50 %, 75% and 100% of field survey data (from 157 sites) we re used as inputs to direct supervised classification of a Landsat TM image. The accuracy of the resulting habitat maps was then assessed using independent field data. Figure 3 shows that the amount of field survey used to direct supervised classification profoundly influenced the accuracy of outputs although the increase in accuracy between 75% and 100% of field survey inputs was not significant. As a goal when extrapolating to large areas (several hundred km2 ) we recommend that around 30 sites per class be surveyed in detail for use in directing supervised classification of multispectral imagery. To assess the accuracy of outputs we recommend that around 50 independent accuracy-assessment sites be visited per class. Using 75% of the signature file in Figure 3 would give about 30 sites per class at coarse-level (four habitat classes). For smaller study areas fewer survey and accuracy-assessment sites will be needed.
Since the set-up costs, field survey costs and analyst time do not vary appreciably between satellite sensors, the selection of cost-effective methods boils down to map accuracy and the cost of imagery, the latter depending on the size of the study area and choice of sensor. For coarse-level habitat discrimination, Landsat TM is generally the most cost-effective option, allowing habitat maps of over 70% user accuracy to be created over large areas at relatively modest cost. However, for areas contained within one 60 x 60 km SPOT scene, SPOT XS is likely to be more cost-effective, the small decrease in accuracy being offset by the lower cost of a single SPOT image (see Figure 4). Airborne remote sensing will allow maps of greater accuracy to be produced but it would be a waste of the imagery’s capabilities to use it for coarse habitat mapping alone. SPOT Panchromatic images may be useful for preparing base maps of areas because of its relatively high spatial resolution (10 m) but is not a cost-effective option for habitat mapping. Landsat MSS is very cheap but to achieve an accuracy in excess of 50%,the cost of ground-truthing and processing are likely to be 10–20 times the cost of the imagery, thus buying it may be a false economy (Figure 4).
At fine descriptive resolution, only digital airborne scanners (e.g. CASI) and interpretation of colour aerial photography can deliver overall accuracies in excess of 50% (the former providing accuracies of around 80% and the latter of 57%). Thus, detailed habitat mapping should only be undertaken using these techniques. The cost of commissioning the acquisition of such imagery can be high (£15,000–26,000 even for small areas of 150 km2 ) and may constitute 33–45% of total costs (64–75% if set-up costs are excluded). For a moderate-sized study larger than ca 150 km2, about a month will be required to derive habitat classes from field data and process digital image data – irrespective of the digital data used (though, if staff need to be trained, this will increase time requirements considerably).
The relative cost-effectiveness of digital airborne scanners and aerial photography is more difficult to ascertain because they depend on the specific study. Where possible, it is recommended that professional survey companies be approached for quotes. In our experience, however, the acquisition of digital airborne imagery such as CASI is more expensive than the acquisition of colour aerial photography. However, this must be offset against the huge investment in time required to create maps from aerial photograph interpretation (API). If habitat maps are needed urgently, s ay in response to a specific impact on coastal resources, API might take too long and therefore be inappropriate. For small areas of say 150 km2, a map could be created within 120 days using CASI but might take almost twice this time to create using API. We estimate that API is only cheaper if the staff costs for API are less than £75 per day. If consultants are used, this is unlikely to be the case. Further more, as the scope of the survey increases in size, the cost of API is likely to rise much faster than that arising from digital airborne scanners, making API less cost-effective as area increases. Where the costs of API and digital airborne scanners are similar, the latter instruments (digital scanners) should be favoured because they are likely to yield more accurate results than API. Possible drawbacks are the large data volumes which must be handled and the relative novelty of the technology, which means that institutions are less likely to be equipped to handle it and transfer of the technology may be less easy.
In summary, the costs of airborne imagery are only justified if finescale habitat maps are required and, if they are required, only airborne imagery (CASI or API) can deliver outputs of adequate accuracy. If broadscale habitat maps are required then multispectral satellite imagery (either Landsat TM or SPOT XS depending on the area to be covered) provides the most cost-effective option.