Module 7
Caribbean Regional Node

To demonstrate the effects of image spatial resolution (ranging from 1 m to 80 m) on the ability of images to discriminate between different coastal habitats.
 
 

A wide variety of remotely-sensed imagery are available for coastal habitat mapping and therefore one of the most important questions to consider when planning a remote sensing project is, "Which imagery will allow me to map the habitats of interest?". The answer to the question depends largely on the size of the habitat of interest. Remote sensing instruments divide the Earth's surface into a grid of sampling units called "pixels" whose spatial dimensions depend on the altitude and design of the sensor. The pixel size roughly equates to the spatial resolution of the sensor. For example, each pixel in a SPOT multispectral image (XS) measures 20 m by 20 m and thus the smallest linear dimension on the Earth's surface which can be resolved is roughly 20 m. In order to map the boundaries of a given habitat, the spatial resolution of the imagery should be smaller than the minimum size of the habitat. This means that SPOT XS imagery is incapable of detecting habitats whose dimensions are less than 20 m by 20 m.

The spatial resolutions of a range of widely-used remote sensing sensors are listed in Table 2.1. Although spatial resolution is a key consideration when selecting a remote sensing method (and we dwell upon it in this lesson), other considerations must also be taken into account. In terms of sensor specifications, the most important of these are the number of useful spectral bands and the area covered by the image. In short, the greater the number of spectral bands, the greater the likelihood that habitats will be identified correctly during image processing. For example, two different habitats might have the same overall colour when viewed in a single band of blue light, but their appearances might differ markedly if viewed in blue, green and red bands simultaneously.

An important aspect of spatial scale is texture. In a remote sensing context, texture describes the structure or pattern of a feature in the imagery. Texture can be fairly simple and uniform or complex and patchy. Many habitats have a characteristic texture (patchiness) which, if considered, can greatly improve the chances of correctly identifying habitats in image data. To measure the texture of habitats using remote sensing, the spatial resolution of the imagery must be similar in size to the smallest patch that makes up a particular pattern or texture. For example, imagine a sparse seagrass bed comprising small patches with an average width of 5 m. Although the entire bed of sparse seagrass may have a width of, say, 100 m, the imagery would need a spatial resolution of 5 m, or less, to detect its texture. Statistical measures of texture are available which measure the local variability in pixel reflectance, but you will make a qualitative assessment of texture in this lesson.

¹ One band for monochrome aerial photography, three bands if colour aerial photography is scanned in the red, green and blue.

This lesson will focus on the importance of matching the spatial scale of coastal habitats to the spatial scale of imagery. To begin with, you will measure the spatial frequency (average patch size) of some key habitats. These measurements will be based on airborne imagery which usually has the highest spatial resolution. With these measures of spatial frequency in mind, you will then predict whether various types of remotely-sensed data are capable of detecting these habitats. You will then see for yourself by comparing images whose spatial resolution varies between 2 m and 30 m.

It is worth noting in passing that the term "descriptive resolution" has been coined to describe the detail to which a sensor will map a given area. For mapping benthic habitats, a coarse descriptive resolution would only separate habitats into broad classes, e.g. coral, algae, sand, and seagrass. A finer descriptive resolution would also distinguish reef zones, variations in seagrass standing crop and so on.

The capability of different sensors for mapping coral reef, seagrass and mangrove habitats are discussed in detail in Chapters 11-13 of the Remote Sensing Handbook for Tropical Coastal Management and papers by Mumby et al. and Green et al. (see References section).

Familiarity with Bilko for Windows 2.0 and Lesson 1 of this module are required to carry out this lesson. Readers who are not familiar with the Turks and Caicos or the appearance of coral reef, seagrass and mangrove habitats should undertake Lesson 1 before continuing further.

All images used in this module were acquired for around the island of South Caicos, Turks and Caicos Islands (see Figure 1.1, Lesson 1 for a map). Images are supplied at a range of spatial resolutions (Table 2.2), the most detailed of which is CASI. Note: When geometric correction and other processing is carried out the pixel size may be changed from the original size (a process called resampling). Spatial resolution and actual pixel sizes are shown in Table 2.2.

You are supplied with three spectral bands of subscenes of multispectral satellite images at three different spatial resolutions (Landsat MSS: Appendix 2.1; Landsat TM: Appendix 1.1; SPOT XS: Appendix 1.2). These can be viewed individually or connected to make false colour composites. You are also supplied with a subscene of a SPOT panchromatic (single-band recorded across the visible spectrum) image (Appendix 1.2) at 10 m resolution and some subscenes from monochrome aerial photographs scanned so that each pixel is 2 x 2 m on the ground. To save time, the airborne multispectral CASI data (Appendix 1.3) have been stored as contrast-stretched composite images (i.e. the individual bands are not provided).

Table 2.2. Images used in this lesson and their principal specifications. Monochrome aerial photographs © Ordenance Survey.
¹ The actual pixel size in the images is included in brackets.

The CASI was mounted on a locally-owned Cessna 172N aircraft using a specially designed door with mounting brackets and streamlined cowling. An incident light sensor (ILS) was fixed to the fuselage so that simultaneous measurements of irradiance could be made. A Differential Global Positioning System (DGPS) was mounted to provide a record of the aircraft's flight path. Data were collected at a spatial resolution of 1 m² in 8 wavebands (Table 2.3) during flights over the Cockburn Harbour area of South Caicos. Further details are given in Clark et al. (1997). The files CASIHARB.GIF and CASIMANG.GIF are false colour composites of bands 1, 3, and 5.

The activities in this lesson are divided into four parts:

For this part of the lesson you will use imagery at three spatial resolutions: aerial photography (2 m), SPOT Pan (10 m), and Landsat MSS (80 m). The aim is to estimate the width of the runway at South Caicos airport using each image type and in so doing, gain a practical feel for various spatial resolutions. Measurements will be made by recording the number of pixels that cross the runway and then multiplying this number by the width (in metres) of each pixel.

You are provided with a false colour CASI image of Cockburn Harbour and the mangroves on the west coast of South Caicos (CASIMANG.GIF). Although a variety of habitats exist here (see Lesson 1), we will focus on two levels of descriptive resolution. The first is the entire fringe of mangrove forest, which encompasses a variety of sub-habitats, and is simply termed "mangrove". The second is a sub-class of mangrove dominated by the red mangrove, Rhizophora mangle. An isolated circular patch of mangrove trees just off the coast of South Caicos is a good example of this habitat.

Activity: Open the pre-stretched composite CASIMANG.GIF. The mangrove appears as red and at the bottom of the image (scroll down as necessary) you can see the settlement and several piers. Use Edit, GoTo to place the cursor on the circular patch of Rhizophora lying just offshore at coordinates (241, 425). Position the mouse pointer on the cursor and double-click 4 times to zoom in. Calculate the diameter of the patch in the horizontal plane starting at coordinates (241, 425) using the transect tool . (Hint: once you have dragged the transect over the patch, use Edit, GoTo to see the number of pixels, DX:). Enter the count in Table 2.5. Move the cursor to coordinates (256, 412) and measure the vertical diameter of the patch (this time enter the value ofDY: into Table 2.5). Convert the numbers of pixels into diameters in metres and calculate the average diameter of the patch, entering your results in Table 2.5. The area of a circle is calculated using the formula below where the radius is half the diameter and p = 3.142.

To make a rough estimate of the area of this mangrove fringe on the northern part of the image (north of the gap in the mangrove fringe) one can measure its length by using the mouse pointer to lay a transect parallel to the coast in the middle of the mangrove. Such a transect might begin around (166, 4) and end around (362, 268).

In this part you will work with a pre-stretched colour composite CASI image (1 m spatial resolution) of Cockburn Harbour (CASIHARB.GIF) showing seagrass and reef habitats. These habitats were described in some detail in Lesson 1 and photographs may be opened in the Bilko for Windows viewer for further familiarisation (files blowout.gif, seag_den.gif, mont_col.gif). The first exercise is to measure the width of three important features which have their own spatial frequencies; seagrass blowouts (blowout.gif), dense seagrass beds (seag_den.gif) and colonies of the hard coral, Montastraea annularis (mont_col.gif).

Activity: Open CASIHARB.GIF and use the Edit, GoTo command to place the cursor on a seagrass blowout (coordinates 272, 119). Double-click twice to zoom in. Using the transect tool  and Edit, GoTo commands described above, measure the horizontal widths of the three seagrass blowouts listed in Table 2.6. The coordinates are the left hand edges of the blowouts. [Hint: Treat each pixel as being 1 m square (ignore fact that they are 1.1 m long) so the number of pixels = distance in metres]. Enter your results in Table 2.6.

The seagrass blowouts lie within a large bed of dense seagrass. Use the procedures set out above to estimate the width of this seagrass bed at three points. The upper coordinates of the bed are given in Table 2.6 and you should measure the width in the vertical plane (i.e. drag the transect towards the bottom of the image). Enter your results in the table.

Use the Edit, GoTo command to place the cursor on a colony of Montastraea (341, 244) which contrasts strongly with the pale green of sand and seagrass. Measure the horizontal width of the dark colonies at the coordinates listed in Table 2.6 and enter your results in the table.

To check your predictions entered in Table 2.7, you are provided with monochrome aerial photography (AP_SCAIC.GIF), SPOT Pan imagery (XPSUBSET.GIF), and Landsat TM imagery (TM1_SC.GIF, TM2_SC.GIF, TM3_SC.GIF).

The lesson as a whole should have indicated that while satellite imagery at resolutions of 20-80 m is able to map habitats to a coarse descriptive resolution, for finer discrimination and mapping of patchy habitats airborne imagery with a spatial resolution of <5m is likely to be needed. The key thing is to make sure that your imagery can achieve your mapping objectives.
 

Clark, C.D., Ripley, H.T., Green, E.P., Edwards, A.J., and Mumby, P.J. (1997). Mapping and measurement of tropical coastal environments with hyperspectral and high spatial resolution data. International Journal of Remote Sensing 18: 237-242.

Green, E.P., Mumby, P.J., Edwards, A.J., Clark, C.D., and Ellis, A. C. (1998). The assessment of mangrove areas using high resolution multispectral airborne imagery. Journal of Coastal Research 14 (2): 433-443.

Green, E.P., Mumby, P.J., Clark, C.D., Edwards, A.J., and Ellis, A.C. (1997). A comparison between satellite and airborne multispectral data for the assessment of mangrove areas in the eastern Caribbean. Proceedings of the Fourth International Conference on Remote Sensing for Marine and Coastal Environments, Florida, 17-19^th March 1997, 1: 168-176.

Mumby, P.J., Green, E.P., Clark, C.D., and Edwards, A.J. (1998). Digital analysis of multispectral airborne imagery of coral reefs. Coral Reefs 17 (1): 59-69.

Mumby, P.J., Green, E.P., Edwards, A.J., and Clark, C.D. (1997). Coral reef habitat mapping: how much detail can remote sensing provide? Marine Biology 130: 193-202.
 
 

1. The quickest way to determine the width of the AP_MANG.GIF aerial photograph is to use <Shift>+<Spacebar> (or Edit, Select All, or <Ctrl>+A) to select the whole image and then use Edit, GoTo dialog box to read off the width of the image (DX:) in pixels. It is 600 pixels wide so the width of the image is 600 x 2 m = 1.2 km.
2. You can either read off the width of XPSUBSET.GIF in pixels by selecting the whole image and inspecting the Edit, GoTo dialog box as above, or you can select View, Coords before inspecting the Edit, GoTo dialog box in which case the width (DX:) will be expressed in metres and UTM coordinates of the top left of the image will be displayed. For the former you will find the image is 720 pixels across, so its width is 720 x 12 m = 8.64 km. For the latter you can read off the DX: of 8640 m directly.
3. Even on the aerial photography it is not absolutely clear where the edge of the runway pavement is. The runway appears to be between 38 and 40 pixels across, giving a mean width of 78 m. On the SPOT Pan image it appears to be 7-8 pixels across, giving a mean width of 90 m. The difference is about 12 m, which happens to be roughly the resolution of the SPOT sensor. "Mixels", that is pixels which are half vegetation and half runway pavement, on either side of the runway make it very difficult to be sure exactly where the runway edges are.
4. You can either read off the width of MSS2.GIF in pixels by selecting the whole image and inspecting the Edit, GoTo dialog box as above, or you can select View, Coords before inspecting the Edit, GoTo dialog box in which case the width (DX:) will be expressed in metres. For the former you will find the image is 501 pixels across, so its width is 501 x 66 m = 33.1 km. For the latter you can read off the DX: of 33066 m directly.
5. The runway is visible in the Landsat MSS scene. But would you have recognised it as such if you did not know it was there?
6. No, one can see that some linear structure is there but all one can say is that it is similar in width to the spatial resolution of the sensor (80 m). Both aerial photography and SPOT Pan appear to be suitable for mapping roads with the aerial images picking out narrower tracks which cannot be seen on the SPOT imagery.

Part 2.

7. The average diameter is about 31.4 m giving a radius of 15.7 m. This gives an area of about 774 m? for the mangrove patch.
8. The average width of the mangrove fringe is 82 m.
9. a) The mangrove transect is about 260 pixels long. b) Its length in metres is about 350 m (see equation to right). c) The average width is 82 m so the area is 350 x 82 = 28,700 m?, which is equivalent to about 2.9 ha.
10. Because it is less than a half SPOT pixel width (about 8 m) offshore and the shallow water area between the mangrove islet and the coast cannot be resolved by the SPOT imagery.

11. The average width of the blowouts is about 7 m, the average width of the dense seagrass bed is about 62 m, and the average width of the Montastraea colony clusters is 3 m.

12. Seagrass beds have a fairly uniform texture whereas coral reefs have a complex patchy texture. The seagrass beds tend to be fairly uniformly dark whilst the reef areas tend to be lighter and of varied texture internally but have dark bands around their edges.
13. The patch widths are approx. 33, 42, 25, 24, 32, 25, 23, 31, 41 pixels giving an average width of 31 pixels or 62 m (each pixel is 2 x 2 m). The MSS imagery (80 m) resolution would be ineffectual at resolving the patches but most should be visible on all the other images.

The Landsat Multispectral Scanner (MSS) was operational from 1972, following the launch of Landsat-1, until 1993. The spatial resolution for the sensor is 79 m. The radiometric resolution is 6 bits (64 grey levels). The swath width for the sensor is 185 km. Multi-Spectral Scanner data ceased to be collected in 1993.

The spectral characteristics of the Landsat MSS bands are as follows:

Images are produced by reflecting the radiance from 79 m wide scan lines on the Earth's surface to detectors on board the satellite using an oscillating mirror. Each scan line is 185 km long (thus the swath width or width of ground covered by the sensor in one overpass is 185 km). The Instantaneous Field of View (IFOV) of the sensor is a 79 x 79 m square on the Earth's surface. This area can be regarded as the spatial resolution of the sensor, that is the smallest area which can be sampled by the sensor. Each picture element or pixel in a MSS digital image is thus a measurement of the brightness of the radiance from a 79 x 79 m square on the Earth's surface.

Because the satellite is moving so fast over the Earth's surface, it has to scan 6 lines at a time (thus covering 474 m along track). Since the MSS sensor measures the radiance in four different wavebands at the same time, it thus has 24 detectors in total. Each detector converts the recorded irradiance into a continuous electrical signal which is then sampled at fixed time intervals (approximately every 10 ´s) and converted to a 6 bit number (64 grey levels). During this interval the ground distance covered by the oscillating mirror is only 56 m. Thus the picture elements or pixels making up an MSS image are 56 x 79 m rectangles.

The 6 bit numbers representing reflectances recorded from each point (79 x 79 m on the Earth's surface) are either recorded on magnetic tape on board the satellite (in earlier Landsats) or transmitted to an Earth receiving station where they are recorded on high-density digital tapes (HDDTs).

A full Landsat MSS scene consists of measurements of the radiance from a ground area of 185 x 185 km and thus consists of 2340 scan lines each consisting of 3240 pixels. There are thus about 7.5 million pixels per scene, each of which has radiances recorded in four wavebands, giving a total of 30 million bytes of information. It takes about 25 seconds to record one scene. The digital information for a scene is supplied on Computer Compatible Tapes (CCTs) or on Exabyte tapes to users who then need to process it on computers according to their needs.

Back to main page

Back to Module 7

USDOC | NOAA | NESDIS | CoastWatch