INTRODUCTION

Background Increased pressures on the Ord River water resource as a result of proposed developments have necessitated the preparation of a Water Allocation Plan by the Department of Environment (previously Water and Rivers Commission). Under the Environmental Water Provisions Policy for Western Australia (WRC, 2000), the Commissions allocation planning process must provide for the protection of water dependent ecosystems while allowing for sustainable use and development to meet the needs of current and future users. When the Department of Environment (DoE) first undertook the development of a water allocation plan for the Ord (WRC, 1999), the approach taken in determining the environmental provision was a rule of thumb 20th percentile of the natural flows. Little ecological data was available to justify a more sophisticated approach. Public comments on the 1999 Draft Water Allocation Plan considered that the environmental values that had arisen in the 30 years since regulation altered the hydrology, had not been adequately protected. Strategic advice from the Environmental Protection Authority (EPA) recommended that DoE review the proposed environmental water provisions and that maintenance of the riverine environmental values established since the construction of the Ord River Dam (ORD) should be the basis of that review. After seeking advice from a panel with expert knowledge of tropical river ecosystems and undertaking further community consultation, DoE undertook to determine ecological water requirements by comparing changes in the dry season wetted perimeter relative to different discharge levels. DoE concluded that the decrease in depth and wetted perimeter associated with the maintenance of a minimum flow rate of 45 m3.sec-1 from the Kununurra Diversion Dam (KDD) to 57.5 km downstream and 40 m3.sec-1 below that point was an acceptable estimate of the ecological water requirement; i.e. the 45/40 environmental allocation. This was viewed as limiting the change to the dry season flows and hence to the risk of triggering the adverse dry season ecological impacts described by the Scientific Panel. This estimate of the ecological water requirement is expected to provide the basis for a revised Interim Water Allocation Plan for the lower Ord River (in preparation), but which may be reduced in times of drought. The Environmental Water Provisions Policy for Western Australia (WRC, 2000) requires that the Commission apply effective monitoring and evaluation to ensure that water provisions are met and that environmental values are protected. This is paramount in the Ord River given the interim nature of the environmental flow provision and the lack of scientific certainty on which it is based. DoE is concerned that reduced flows in the Ord River downstream of the KDD (i.e. the lower Ord River) could result in a decrease in the quantity and quality of habitats identified as important for fish and invertebrates (functional habitat, sensu Storey 2002 & 2003), and this could possibly lead to reduced species diversity and abundance. Therefore, as part of an adaptive management programme, the flow provision will be monitored and evaluated.

To assess the impact of the 45/40 allocation on the river, DoE intends to survey and map the extent of key habitats in the lower Ord River under current dry season flows using remote sensing techniques. This will allow the distribution and extent of important habitats under the 45/40 flow regime to be compared to current dry season flows (~60 m3.sec-1), and also any future changes in extent detected if lower flows are prescribed under a drought scenario. The Functional Habitat Concept The functional habitat approach considers the river channel as being composed of distinct habitat units that can be recognised and classified on the basis of their physical and biological attributes. The Functional Habitat Concept proposes that it is easier to monitor visible changes in habitat distribution and extent than it is to detect change in the distribution, abundance and biomass of fish and invertebrates based on infrequent sampling of populations, especially in systems with diverse communities, for which there is little information on the life history requirements of individual species. Storey (2002 & 2003) therefore recommended the Functional Habitat approach as a method for management and future monitoring of faunal communities in the lower Ord River, on the precept that it is easier to manage critical habitats, rather than the resident species for which there is little biological information. Important functional habitats for aquatic macroinvertebrates and fish of the lower Ord River have been identified by Storey (2002 & 2003) (Table 1). They indicate the most important habitats in terms of species preferences/usage and likely susceptibility to changes in flow discharge. Examples of key habitats are illustrated in Plates 1 12. Mapping of the extent and distribution of functional habitats in the Ord River downstream of the KDD is yet to be undertaken. Prior to undertaking mapping, it is necessary to identify the best remote sensing platform and imaging technique for identifying and mapping aerial extent of functional habitat units. Functional habitat may be a sub-unit of large reach-scale geomorphic unit. For example, whilst pools may form relatively discrete habitat units, all pools will not be equivalent in terms of the range of micro-habitats, and therefore biota, found within them. Therefore, an inventory of larger scale habitat/geomorphic units is unlikely to allow an assessment of the risk of losing species from the Ord River system. Functional habitat therefore will need to be measured at the scale of metres rather than 100s of metres or even 10 of metres. To apply the functional habitat approach to monitor changes in the extent and distribution of key habitats for fish and invertebrate fauna it is first necessary to develop an appropriate technique for monitoring changes in functional habitats. The current project investigated the applicability of various methods of remote sensing to map functional habitat in the lower Ord River. Project Objectives The overall aim of the current project was to identify the most efficient method to remotely survey the aerial extent and distribution of functional habitat in the lower Ord downstream of the KDD.

0.2m Airborne

1.0m Panchromatic (Simulated Ikonos)
4.0m Multispectral (Simulated Ikonos)
Figure 6. Comparison between scanned high-resolution (0.2m) film-based aerial imagery and simulated panchromatic (1.0m resolution) and multispectral (4.0m resolution) Ikonos imagery.
Figure 7. Example of scanned film based aerial imagery 1.0m image resolution.
Acquisition / Processing Evaluation Based on the initial screening of sensors using resolution capabilities, a number of sensors were selected for further assessment of image acquisition and processing. These included the high-resolution satellite sensors (Ikonos and Quickbird) and the aerial photography options (full format aerial, small format aerial and digital aerial). Table 4 identifies the advantages and limitations of each proposed system. The processing component did not include a review of the extraction of the functional habitat components as this would be fundamentally the same for all imagery, independent of acquisition method. The processing assessment considered the processes required to produce a suitable image for analysis. While the option of aerial digital photography offered some advantages in terms of ease and flexibility of image acquisition, the method was not considered suitable to survey extensive areas such as the entire ~70km reach of the lower Ord. This was due to the envisaged complexity of image processing. If the method was to be utilised only over shorter, replicate reaches of several kilometres (the approach taken for the Phase II field trial), the method would have significant advantages. However, for the implementation over the whole of the lower Ord River, there was a high probability that the complexity of image processing would mean that image technical specifications were not met. The high resolution satellite systems (Ikonos and Quickbird) offered significant advantages in terms of acquisition and processing compared to the film-based photographic systems when applied to the whole lower Ord. However, because of the relatively large minimum acquisition area, the same costs still applied for application to multiple shorter reaches, as the whole region needed to be acquired. The film-based systems had the potential for significantly higher image resolutions and hence improved reliability in the extraction of functional habitat components. In addition, they offered flexibility with the ability to vary image output to suit analysis requirements. Images could be designed on a needs basis with respect to radiometry (panchromatic or multispectral) and resolution. Depending upon the method of application (replicate reaches versus whole of lower Ord River) there were not clear benefits of one approach over the other. Therefore, both the high-resolution satellite systems and the film-based photographic systems were further assessed with respect to acquisition and processing cost. Acquisition / Processing Cost Acquisition and processing cost were estimated for the Ikonos, Quickbird, aerial photographic and small format aerial photographic options. The costs did not include the analysis of the imagery to extract the functional habitat components nor did they include DoE costs in managing the project. The tabulated costs should not be used to budget options, as there will be significant variations depending on contractors used, area of coverage and required accuracy. The costs derived have been done consistently to enable comparison of methodologies only. Table 5 details the estimated costs for the acquisition of imagery from the sensors defined in Table 4.

Figure 11. Panasonic NV-MX500 digital video camera (inset) and position of mount on Cessna 206 aircraft.
Initial digital acquisition at 1000 ft altitude was considered unacceptable because of high levels of atmospheric turbulence as a result of flying at midday (intentionally selected so that the sun was overhead to minimise shadow) and the inability to keep the aircraft centrally over the river channel without being able to see the channel immediately below (i.e. difficulties with narrow field of view relative to channel width) made accurate acquisition difficult. Digital imagery was subsequently acquired at an altitude of approximately 2,000 feet1 (609.6 metres) with the camera mounted from the foot-step of a Cessna 206 aircraft.
Table 6. General Digital Video Camera Specifications for Habitat Mapping.
Function Mode Lens Calibration Digital
General Specification Glass Lens; typically Zeiss, Schneider, Leica, Rollei. The lens should be calibrated for radial and tangential distortion. The image chip should be calibrated for sensor distortions. (Only required for monitoring applications where accurate image mosaicing and image georeferencing is required). Manual over-ride required, with aperture of F2.8 to F16 and shutter speed minimum of 1/1000th sec. Both digital tape (video) and digital card (image still) recording. Image resolution should be better the 3.0 x 106 pixels. Make sure that card record time does not exceed 5 to 7 seconds; to achieve this image resolution may need to be sacrificed. Preference is for three individual CCD sensors; one each dedicated to the Red, Green and Blue spectrums. An interlaced sensor with R, G, B recording can be used but does not give as good radiometric recovery. Operation for a minimum of 2 hours in record mode without the need for battery change. This will be more critical for configurations where the battery cannot be changed in-flight. An IR remote and (if possible) a free-style remote. The free-style remote will be critical in cases where the camera mount precludes access to the IR receiver. For an IR remote only, the camera mount and external camera configuration becomes critical. The remote should be able to start, stop and record both digital still and digital video modes. A change in mode between digital card and digital tape recording is an advantage. An option to output raw images (i.e. uncompressed) is an advantage if critical variances in image radiometry are expected. Compression (e.g. JPEG) is acceptable with options to vary image compression quality. The digital memory card should be able to store a minimum of 50 uncompressed images. Verify that image record time does not degrade with increased memory size on the card. Facility to: - turn off sound, - turn off / activate image stabiliser functions, - facility to quickly download images (USB 2.0 or Firewire) and - flexible image capture software (digital stills from either card or tape).

Aperture Settings Image Capture

Battery Life

Remote Control

Output Image Format

Digital Card

General Options

Feet are the standard unit used to measure aircraft altitude.
Any future programmes utilising similar equipment configuration would require a rigid camera mount and an in-cockpit video link to aid with navigation, whereby the pilot may view the actual area seen by the camera and ensure adequate channel coverage. Image positioning was undertaken using a GPS without differential input. A Garmin Map76S was used to track the flight path and the images were referenced to position using time synchronised on both the digital camera and the GPS sensor. Figure 12 shows the track from Kununurra and the image acquisition epochs in the vicinity of the study area. Images were acquired from both the high resolution SD card data set and captured from the magnetic tape. A sample image with an output resolution of 0.5 metres is shown in Figure 13. For the reach between Carlton Crossing and Maccas Barra Camp (adjacent to House Roof Hill), images were captured, georeferenced and mosaiced. The resulting image mosaic is shown in Figure 14. Image mosaicing was undertaken using the ENVI software package, with georeferencing based on the GPS centroid positions measured during flight and with image-to-image constraints. There were significant shifts in image location due to the instability of the selected camera mount on the aircraft and any future data acquisition programmes would require the fabrication of a rigid mount on the aircraft. At an altitude of 2,000 feet (609.6 m) above ground level, the coverage of the river floodplain was limited. Future data acquisition programmes should be undertaken at an altitude approaching 4,000 feet (1,219.2 m) above ground level. While this would push the ground sampling distance out to 1.0 metre it is not envisaged that this would have any significant impact on habitat recognition or mapping. The benefit would be that navigation tolerances would be lower and approximately twice the surrounding river floodplain would be covered. Figure 15 shows the results of the mosaicing process in terms of both radiometric and spatial accuracy. Spatial positioning has high precision, due to the nature of the imageto-image mosaicing process adopted, however absolute accuracy is a function of the accuracy of the GPS positions recorded during flight. It was estimated that absolute positioning accuracy was between five to ten metres with the selective availability function of the GPS in the off position. Improvements in absolute accuracy could readily be achieved through the use of differential GPS or identifiable ground control. Radiometric balance was maintained across the images however, there was some imbalance at the image joins due to the vignetting process. Images acquired at a higher frequency, effectively removing the image edges from the mosaicing process, would reduce the effect of radiometric imbalance.

Plate 14. Eucalyptus spp. with scattered Melaleuca spp., along a steep river bank remote sensing unit #9 in Table 6.
Plate 15. Large woody debris (snags; remote sensing unit #10) with emergent macrophyte (Typha sp.; remote sensing unit #2 ) in the background.
Plate 16. Emergent macrophyte (Typha sp.) at the entrance to a small backwater zone remote sensing unit #2.
Plate 17. Flooded riparian vegetation habitat (snags) remote sensing unit #10 in Table 6.
Plate 18. Inundated backwater (unit #11) with the entrance colonized by emergent macrophyte (Typha sp., unit#2), with submerged Ribbonweed (unit #5) in the foreground. Silver Cadjebut (Melaleuca argentea), Pandanus sp. and Freshwater Mangrove (Barringtonia arcutangula) line bank.
Plate 19. Inundated backwater zone (unit #11).
Plate 20. Shallow backwater (unit #11), with floating Ribbonweed (unit #5) across the mouth.
Plate 21. Exposed sand bank (unit #6) with isolated dead vegetation and snags (unit #10).
Plate 22. Extensive dead vegetation on an exposed sand bar (unit #6).
Plate11. Carlton Crossing showing gravel run habitat (unit #3).
Plate 23. Rapids downstream of Carlton Crossing (unit #4).
Plate 24. Exposed gravel bed (unit #3) with the initial colonization of exposed macrophyte. Riparian forest (Melaleuca spp. and Sesbania sp.; units #7 & 8) occupies the adjacent bank.
Plate 25. Emergent macrophyte (Typha; unit #2) with floating Ribbonweed beds (unit #5) in the foreground.
Plate 26. Small open backwater (unit #11) with exposed snags (unit #10) on the surrounding sandbank.
Plate 27. Steep bank adjacent to the river with sparse Melaleuca spp. and Eucalyptus spp. (units #8 & 9).
Plate 28. Riparian vegetation on a small island overgrown with Phragmites karka and Passiflora foetida (Passion Vine) (unit #1).
Plate 29. Small colony of Typha on an exposed sandbar (unit #2).
In the primary classification, the software image correlation process (using ENVI) defined regions that correlated with the radiometric responses for each habitat group. This process resulted in a complex pattern of habitat types across each zone. Figure 16 shows a typical scatter from the classification process compared with the classification image.

(a) (b)

Figure 16. Example of results of (a) primary classification of functional habitats using ENVI defined regions compared to (b) the source (Quickbird) image.
Based on the primary classification results, a full-classification was undertaken where the classification groups were vectorised into dominant habitat communities. In many cases, the habitats formed a complex mix within any single zone. For example, the riparian forest zone was typically not a discrete entity with well-defined boundaries, but in many cases, the emergent macrophytes and the eucalypt overstorey merged into the riparian vegetation zone. In such instances, the classification was based on the dominant habitat. In addition to the aerial classifications, this secondary phase identified point (snags and backwaters) and lineal features (sandy margins along the waters edge), classified on visual assessment rather than objective classification. A full classification was undertaken on the Quickbird imagery (full scene) and on the Digital Video imagery for the reach between Maccas Barra Camp and Carlton Crossing. Despite limited coverage of the floodplain vegetation communities, the classification of the Digital Video imagery was undertaken in order to compare the definition of functional habitats from both image sources. Results are shown in Figures 17 20. Figure 17 defines the legends for the classification images to follow as well as the classification units. The final vegetation transitions/vectors chosen for inclusion in any given classification unit were subjective, but were consistently applied to all data sets.

Figure 21 illustrates the difference between the Quickbird imagery and the Digital Video imagery with respect to snag definition. Only significant snags, possibly in the more complex arrangements, were identifiable in the Quickbird imagery. Snags that were clearly evident in the digital video image were not detectable in the Quickbird imagery. Assessing changes in snag count from Quickbird imagery, and the consequences for habitat, would have low accuracy and reliability. Quickbird acquisition was achieved with an off-nadir view angle of approximately 20 degrees. This oblique view angle means that at least one bank of the river was obstructed by the vegetation on the bank. In some cases this was severe enough to obstruct snags, riparian vegetation, sandy margins and submerged macrophyte beds. This is a significant disadvantage, given that much fish habitat is close to the bank. Figure 22 shows the extent of obstruction due to the high off-nadir view angle in the Quickbird imagery. Based on these results, the ideal angle would be between degrees.
Figure 21. Snag habitat (arrowed) captured as (a) Quickbird image and (b) Digital Video image. Refer Plate 17 for ground-truthing photo of snag habitat.
Figure 22. Quickbird satellite imagery illustrating impact of high off-nadir view angle.
It is possible to obtain Quickbird images with lower off-nadir view angles. Quickbird offer options to acquire imagery at two off-nadir ranges; 0 - 25 degrees (standard) and 0 - 15 degrees (special request). However, if imagery were to be ordered at degrees there would be no guarantee that the maximum degrees would be achieved. In addition significant time delays are possible with the special request. The frequency of passes by the satellite with this nadir is less and therefore it may take longer (i.e. several months) to obtain images allowing for cloud cover etc. Images with lower degrees off-nadir can be specifically tasked, however tasking Quickbird flights incurs substantial additional cost. Although acquisition costs may be higher and acquisition time significantly extended, any future Quickbird data sets should be acquired within a range of 5 - 10 degrees off-nadir. Possibly the biggest problem with the Quickbird imagery was the radiometric penetration beneath the water surface. Submerged macrophyte extent was difficult to extract and had low reliability, particularly in zones where the radiometric response of deep water was similar to that of submerged ribbonweed. Although a full pansharpened image with a composite spectral range of red, green, blue and near infra-red (NIR) was ordered, the image supplied was a three-band composite of the four available bands, with the NIR-band included. Figure 23 shows a single Digital Video scene inter-cut into the corresponding Quickbird scene to demonstrate the difference in degree of water penetration between the two sensors, with water penetration obvious for the Digital versus the Quickbird imagery. The water appears dark in the Quickbird image due to the absorption of energy in the visible red and NIR bands. It is recommended that future analysis is undertaken on a four-band (R,G,B & NIR) separated GeoTIFF image, with the NIR band excluded for optimum water penetration.

Figure 23. Comparison of water penetration characteristics of a reach coverage of Quickbird satellite sensor (pan-sharpened image without NIR) compared to a single Digital Video scene inter-cut into the corresponding Quickbird scene (overlay).
In order to analyse the Quickbird imagery a degree of image analysis needed to be undertaken. This processing aimed to enhance the water penetration capabilities of the Quickbird imagery by passing enhancement filters over the region of interest. Figure 24 shows a comparison of the Quickbird imagery before and after image enhancement. Although submerged macrophyte colonies could be identified, this was limited to shallower water and small colonies were readily missed.
Figure 24. (a) Quickbird source image and (b) same image with enhancement filters.
Similarly, Figure 25 presents image classification results from un-enhanced and enhanced images for the same river reach as shown in Figure 23, allowing direct comparison of enhanced water penetration.
Figure 25. Classification results from un-enhanced and enhanced Quickbird satellite images.
Digital Video Digital Video imagery was demonstrably superior to the Quickbird satellite imagery for the following technical reasons: 1. With nadir acquisition and a narrow view angle, any problems associated with offnadir obstruction were negated. 2. Image resolution could be varied as a function of the type of habitat that required discrimination. At a similar resolution to the Quickbird imagery, the Digital Video gave superior feature discrimination and definition. 3. Digital Video sensor provided high quality radiometric definition below water. No image enhancements were required and both small and isolated habitat groups were readily identifiable in the imagery. The major deficiencies with the Digital Video pertained to coverage, cost of acquisition and image processing: The coverage of the Digital Video image, at the 2,000 feet altitude employed in this trial, was at the limit of acceptable lateral coverage for the historic floodplain, but gave almost acceptable coverage for in-channel habitats. A combination of bracket movement and the difficulty of keeping the aircraft directly over the channel without an in-cockpit monitor resulted in poor coverage of large sections of the historic floodplain (though only small areas of in-channel habitat were missed). Additional imaging runs would need to be undertaken to ensure complete coverage of the historic floodplain. However, the need for wider coverage of the floodplain is debatable, since the river is now confined within the channel due to regulation and seldom floods the historical bankfull floodplain. Critical in-channel areas that were missed were mostly on the insides of meander bends where the river may flood to several hundred metres wide in the wet season. With an increase in flying height to 4,000 feet, the corresponding ground resolution would be approximately 1.0m and the coverage doubled, providing total coverage of the in-channel habitats. It is not envisaged that the reduced resolution would have any impact on habitat assessment and is recommended as the optimal combination for any future habitat mapping programmes. Image acquisition costs can be high, and need to include aircraft hire, equipment and mount purchases, personnel costs and travel costs. Costs were estimated to be 2 - 3 times that of the basic Quickbird acquisition costs (though tasking Quickbird would increase costs). Image mosaicing and georeferencing can be time consuming and requires skilled operators. It was estimated that half an hour per image would be required to georeference, colour balance and insert into a mosaic. For a reach of 20 images, this would equate to approximately 10 hours work prior to any classification being undertaken. The cost for such analysis, including purchase of mosaicing and image processing software, should be included in any final method recommendation.

Aerial Coverage of Habitats The aim of the Functional Habitat mapping was to provide remote sensor coverage (i.e. aerial coverage) of each habitat in specified reaches in order to monitor temporal changes. Ideally, representative reaches would be selected and routinely mapped on a regular basis to assess temporal changes. To demonstrate the ability of the remote sensing platforms/classifications to provide aerial coverage, the area of each physical unit between Carlton Crossing and Maccas Barra Camp was estimated for each platform (refer Table 10). Differences in area estimates between the two platforms are also discussed in Table 10, however the main points to note were: The Digital Video did not provide total in-channel coverage due to bracket movement and flight path inaccuracy. As such, areas of extreme lateral habitats were under-estimated (viz. area of low-frequency inundation eucalypts on the higher banks and high-frequency inundation Melaleuca/Sesbania vegetation on the low benches); The area of in-channel habitats was likely underestimated by Quickbird due to limitations related to the vertical penetration of the water column and effects of 20 degrees off-nadir.
It should be noted that the area estimates were preliminary and must be viewed with caution. Pre-classification ground-truthing was used to train the classification to recognize specific examples of each physical unit. However, post-classification ground-truthing is needed to confirm the ability of the method to correctly classify physical features outside of the area of original ground-truthing.
Table 10. Aerial coverage of each physical habitat unit between Carlton Crossing and Maccas Barra Camp (House Roof Hill) as calculated from classifications of Quickbird satellite and Digital Video imagery. Area estimates (m2) are given for most classification units except backwaters and snags, which were estimated by abundance (number/count) and sandy margins, which were estimated by length (m). Classification Unit Rapids Backwaters Quickbird 0 mDigital Video 850 mComments Rapids did not appear at all on the Quickbird imagery. The extent and definition on the Digital Video was adequate. Backwaters were readily discernable on both Quickbird and Digital Video imagery. Obstructions due to off-nadir viewing obscured some backwaters in the Quickbird imagery. Although the results are within 4%, the Digital Video offers superior water penetration and Ribbonweed beds are readily discernable. Quickbird imagery had very poor water penetration characteristics and Ribbonweed bed radiometry was similar to that of deep water zones. However, this limitation may be overcome by removing the near infrared band into the pan-sharpened Quickbird images. Approximately 5,370 m2 of floating Ribbonweed was incorrectly classified as Typha sp. in the Quickbird imagery. The radiometry and texture of Typha sp. in the Digital Video imagery offers unambiguous classification for large communities. See classification error above. The radiometry and texture of Typha sp. in the Digital Video imagery offers unambiguous classification for large communities as opposed to the Quickbird platform. The agreement between the classification areas of the two images is not an indication of classification accuracy. Both types of imagery give adequate results with classification accuracy estimated to be within several percent of actual. Preference Digital Video Either

Submerged macrophtyes (submerged Ribbonweed beds)

254,700 m2

244,040 m2

Digital Video

Submerged macrophtyes (floating Ribbonweed beds)

23,520 m2

32,940 m2
Emergent macrophyte (Typha spp.) Emergent macrophyte (Phragmites karka)

11,190 m2

6,570 m2

7,170 m2

Classification Unit Snags/submerged woody debris

Quickbird 335

Digital Video 510
Comments Based on adjusted nominal resolution the resolution of the Quickbird pan-sharpened imagery was 0.7 metres derived from a 2.8 metre resolution multispectral image. It was evident that many of the smaller snags and smaller features were not detected by the sensor. In addition several snags were identified in the vicinity of Maccas Barra Camp that were in reality part of the gravel rapids in the area. The final resolution of the Digital Video was 0.7 metres and at this resolution, the majority of snags were identified. There was some ambiguity in the case of complex snags or where the snags were close to the river bank. Both the Quickbird image and the Digital Video image were suitable for the detection of the flooded riparian vegetation. There was some ambiguity in the demarcation between the riparian vegetation and the eucalypts/river gums, however this was not a function of the imagery but of the complex interface between the two regimes. The Quickbird imagery has significant advantages in terms of lateral coverage extent and hence the ability to readily cover the vegetation to the floodplain boundary. See Above Although the Quickbird imagery readily identified floodplain lagoons it could not discriminate vegetation cover within the lagoon. In the Digital Video imagery, stands of Typha sp. and Phragmites sp. could readily be identified. Sandy margins were readily identifiable from both Quickbird and Digital Video imagery. In isolated cases, the effects of high off-nadir acquisition angles obstructed the edges of the river, however this was a minor issue.

Preference Digital Video

Flooded riparian high frequency of inundation

764,730 m2

771,060 m2
Flooded riparian low frequency of inundation Floodplain lagoons

282,800 m2 26,420 m2

241,070 m2 9,430 m2

Quickbird Digital Video

Sandy Margins

2,900 m

2,930 m

Either

CONCLUSIONS
While there were significant limitations to the technical quality of the Quickbird imagery, particularly for some of the more important in-channel habitats (i.e. snags and macrophyte beds) it is feasible that several of these limitations may be addressed through modifications to the Quickbird acquisition methods. In the first instance, image acquisition should be restricted to within 5 - 10 degrees off-nadir to address the issues of off-nadir obstructions. Also, the NIR (near infra-red band) should be removed in the generation of the RGB (red, blue, green) image during the pan-sharpening process to enhance water penetration. If this substantially increases the ability of Quickbird imagery to classify and quantify habitats such as snags and macrophyte beds, this would make this platform comparable in performance to aerial video, otherwise, aerial video would be preferable. Estimated acquisition and image analysis costs for the reach between Maccas Camp and Carlton Crossing are given in Table 11. This assumes a minimum area for satellite acquisition of 12 x 5 km2, and an equivalent flight distance of ~ 12 km for acquisition of aerial digital imagery. Analysis estimates will vary as a function of image quality, image coverage and the quality of definition of habitat groups within each scene. It is expected that an additional cost of up to $500 may be incurred for geo-referencing and transformation of the recommended four-band imagery. Comparison of costs shows a cost saving for Quickbird over digital video, especially if the one-off equipment costs for establishing digital video capability are included (viz. camera = $3000, mount = $1000 and in-cockpit display = $1000; total = $5000). The logistical and cost limitations associated with the Digital Video platform are greater than the basic Quickbird satellite image acquisition set-up and may limit the frequency and extent to which this technique of habitat mapping can be applied. There is little doubt that the ease and simplicity of ordering satellite imagery has significant operational advantages. However, satellite acquisition costs will increase dramatically if specific tasking is required to achieve 5 - 10 degrees off-nadir. Also, for a design consisting of replicate ~ 5 km reaches dispersed along the lower Ord, satellite acquisition costs for total coverage of the lower Ord versus aircraft time to fly the replicate reaches will alter the cost comparison. Suitability of Remote Sensors for Use in Other River Systems The Ord River seems particularly well suited to the application of remote sensing techniques (satellite or aerial photography) for mapping habitats for several reasons. These include a relatively wide river channel with large scale mesohabitats (metres to 10s of metres) that are distinguishable and distinct at the available levels of resolution, high water clarity (in dry season when habitat mapping would occur), minimal canopy cover over the water to obscure habitats, vegetation zonations equating to habitat types, with vegetation types distinguishable based on texture and colour, and a current active channel and floodplain occupying a narrow zone allowing image capture on single image runs. Systems with narrow channels, low water clarity, extensive and wide floodplains, indistinct zonation across habitat types, small scale habitat units below the level of resolution and extensive canopy cover obscuring in-stream habitats would minimize the effectiveness of these remote sensing techniques on other river systems. These issues would need to be assessed when considered application of these remote sensing approaches to other river systems.

these data will provide a baseline to assess any further change in habitats should a dry season flow scenario be implemented. Data obtained will be a combination of aerial coverage (km2) of each definable habitat and counts of individual habitats (i.e. snags). To assist with future application of remote sensing to monitor functional habitats of the lower Ord River, the digital imagery acquired from satellite and video platforms during this study are included on a CD in the back cover of this report.
Table 11. Comparison of estimated acquisition and image analysis costs for the lower Ord River between Maccas Camp and Carlton Crossing, based on 2003 costing, with salary rate of $800 per day. Component Ground Truthing - field time Boat/vehicle hire + fuel for ground truthing Aircraft Hire Scene Selection, Planning Consumables (Data Storage) Flight Planning Image Acquisition Costs Special Tasking of Satellite Image Processing / Capture Image Mosaicing / Georeferencing Georeferencing / Transformation Primary Classification Full Classification Reporting, Analysis, Transfer to GIS 3 Digital Video Platform Person days 2 $200 per day 3 hrs @ 345 --0.5 2
Quickbird Platform Person days 2 $200 per day -0.5 ------1 2a 3b 2 Cost $1,600 $200 -$400 --$2,000 $2,000 --$800 e $1,600 $2,400 $1,600 $12,600
Cost $1,600 $200 $1,035 -$200 $400 $1,600 -$1,600 $2,400 -$1,600 $2,400 $1,600 $14,635

2 Total

Notes: a Primary classification duration will vary as a function of image quality, image resolution and habitat characteristics. This estimate is for habitat communities that are well defined and where radiometric characteristics allow unambiguous classification. b Full classification duration will vary as a function of the number of classification communities and as a function of the spatial complexity of each community. c Image acquisition duration will vary as a function of climatic conditions and required coverage extent. d Image georeferencing duration estimates indicated are based on an existing image / map defining spatial extent and location. e Additional costs to process a four-band (R,G,B,NIR) separated GeoTIFF are estimated to be $500.

Acknowledgments

We would like to thank Slingair for their assistance whilst undertaking the aerial surveys. Staff at the DoE, Kununurra Office are thanked for their logistical support during ground-truthing, in particular Leith Bowyer, Scott Goodson, Mike Harris and Duncan Palmer. Kerry Trayler and Rob Donohue are thanked for their support during project management on behalf of the Department of Environment. Kerry Trayler is also thanked for constructive criticism of the draft report. Sue Creagh assisted with editing the final report.

References

Storey, A.W. (2002) Lower Ord River: Invertebrate Habitat Survey. Final unpublished report to Water & Rivers Commission by Department of Zoology, The University of Western Australia, Perth. Pp 57. Storey, A.W. (2003) Lower Ord River: Fish Habitat Survey. Draft Final unpublished report to Water & Rivers Commission by Department of Zoology, The University of Western Australia, Perth.