RIVER CHARACTERISTICS

Reach Classification

Stream reaches are classified as a homogeneous section of the stream channel characterized by discharge, gradient, channel morphology, channel confinement and stream bed and bank materials. Reaches also have a specific repetitive pattern of features. This procedure has previously been defined as a predominantly photointerpretive task. With the use of photogrammetric tools, accurate lengths and gradients can be measured and used in place of, or as a supplement to, field observations. Photogrammetry provides reliable measurements given adequate ground control. The time required to survey an area with suitable photogrammetric procedures and an adequate sampling effort is considerably less than would be required for the same survey using only ground survey techniques.

Figure 6 shows an uncontrolled aerial photographic mosaic of a section of Sowaqua Creek. This type of image is useful in interpreting features along the stream as it gives a broader overview without sacrificing spatial detail.

Reach Subdivision

Reach subdivision involves the partitioning of larger reaches into smaller and more homogeneous sections. This procedure is based on specific attributes that are important to a specific study. For example, different riparian structural stages or changes in substrate might be used to further divide a reach. Many of the criteria used to perform this classification task can be derived from aerial photographs. In this case, reconnaissance photography (normal colour, colour infrared, digital, or black and white panchromatic), flown with optimal illumination conditions (sun angle and time of year), can provide excellent data for this type of distinction. Figure 7 is an example of the same area shown in Figure 6 except that this image is a multispectral digital reconnaissance colour infrared composite image rather than a scanned colour aerial photograph.

Because of the specific spectral properties of vegetation, data captured in the infrared region of the spectrum yields valuable information for the discrimination of different vegetation types. This difference is most evident in the contrast between deciduous and coniferous vegetation. For example the old logging road (next to the current logging road) is distinctly seen on the colour infrared image (Figure 7) while it is less noticeable on the normal colour aerial photograph (Figure 6).

Channel Type

Stream channel type describes the channel typology and morphology. Channel typology is determined based on the shape of the channel (step-pool, cascade-pool and riffle-pool) with modifications based on substrate and debris. In areas where the channel is clearly visible, and at times when the suspended sediment concentrations are low, aerial photographic interpretation is an ideal tool for this type of classification. Specific substrate size determinations should be conducted at selected accessible locations in order to provide suitable ground truth for this type of classification. For further information on substrate classification see the section on substrate below.

Channel Type and Disturbance Indicators

These indicators are used to identify specific channel morphologies that are suitable habitat for spawning and evaluate disturbance levels. Of specific importance to this determination is the effect of logging activities on habitat condition. Areas that were formerly prime areas for spawning and have since been degraded through logging activity are of particular interest. Aerial photographic interpretation is an appropriate tool for this type of analysis. In areas with previous historic coverage, the effect of logging on spawning habitat can be quantitatively evaluated through a study of change over time. In areas where canopy or shadow limit the use of aerial photographs, field surveys should be substituted. Figure 8 shows a recent cut-block by Sowaqua Creek. The effects of logging activities in the area can be clearly seen. There are several slope failures around the road causing an increased sediment load into the creek.

Potential Barriers

Features that might prevent or constrict juvenile or adult salmonids are mapped as potential barriers. These features are then field checked to asses the nature of the barrier. The following areas are mapped as potential barriers: 1. culverts and disused bridges (Figure 8); 2. landslides and bank slumping (Figure 8); 3. log jams (Figure 9) ; 4. beaver dams; 5. falls greater than 2m; 6. cascades or chutes (white water), and; 7. gradients greater than 20%. Many of these features can be identified on aerial photographs. Parameters that require measurement (falls and gradient barriers) should be measured using photogrammetric procedures for greater accuracy. If the area in question is in an area where tilt or relief displacement are evident or likely, measurements should be made using analytical photogrammetry; as both tilt and relief displacement (see photogrammetry) can introduce an unacceptable level of error into these measurements.

Barriers

The evaluation of barriers is similar to the evaluation of potential barriers. Barriers are features that can be identified as restricting access to salmonid habitat. In this case, aerial photographic interpretation is a useful tool in the determination of barriers, providing the area of interest is not obstructed from an overhead view. Close attention must be paid to time of year and time of day considerations for flight planning in order to reduce the effect of shadows over the stream (obstructs viewing of features).

Stream Order Classification

Stream order classification is a scale dependant measure. In most circumstances, stream order is determined within a single drainage basin and is therefore derived from aerial photographs at a smaller scale (i.e. 1:50000) than is required for stream habitat evaluation. This is a straight forward photographic interpretation procedure since specific information about each stream is not part of the classification system. Most of these systems (for example the Strahler stream ordering system) are based solely on the numbering of streams. In the Strahler system, all streams that have no up-stream tributaries (smallest detectable channel that would intermittently carry water) are labelled as first order streams. The stream order is not incremented until two streams of the same order meet. For example, the stream that results from the joining of two first order streams is a second order stream, two second order streams makes a third order stream and so on. If a first order stream meets a second order stream, the stream remains a second order stream.

Pool Type

Pool types can be mapped from aerial photographs as the type of pool (scour, dammed or unknown) is a qualitative assessment based on the surrounding attributes of a specific pool. For example, a scour pool is formed by scouring of bed material around or adjacent to an obstruction. If the obstruction is visible on the aerial photograph, the pool can be accurately typed. With the use of large scale aerial photography, many subaqueous features can be interpreted in clear water. Pools appear as darker tones when compared to lighter (shallower) riffles. In many cases, subtle variations in bathymetry, for example salmon spawning redds, are visible in optimal imaging conditions (Figure 10).

Long Profile of Channels

The calculation of profiles along channels is possible using remotely sensed data and digital techniques. Channel profiles require information on the depth and bedform of a specific area for which the profile is desired. Using photogrammetric techniques, the slope and length of the channel can be computed. Using these data in conjunction with shallow water bathymetry can yield comprehensive channel measurements. Shallow water bathymetry techniques, using reconnaissance multispectral imagery and digital image processing, require experienced personnel are in more detailed experimentation (see section on shallow water bathymetry below).

Channel Gradient

Channel gradient is the rise of the channel over a specific distance. The measurable variables (elevation and distance) require field measurement or the use of an analytical stereoplotter and suitable ground control. The use of a stereoplotter, over a parallax bar (see section on photogrammetry) for these measurements is required as the distortions present in tilted aerial photographs can lead to unacceptable levels of accuracy (rivers appearing to run up hill are not uncommon when a gradient is measured from unrectified photography). The primary reason for this difficulty is that stream gradients are often very subtle and a small amount of tilt displacement can give an erroneous sense of slope. Standard photogrammetric techniques require that some positional and elevation control exists for the desired area. In areas where these data are not available, field survey methods must be employed to obtain the necessary ground control. Figure 11 is a diagram showing the geographical location of cross sectional and long stream channel profiles. Figure 12 shows the channel gradient as measured by parallax bar and analytical stereoplotter. Figure 12 shows the discrepancies in height measurement between these two methods. While the curves have similar shapes, the parallax bar heights are less precise. The effect of the correction computed with the stereoplotter is a lowering of height measurements and a more consistent slope.

Cross-sectional Channel Profiles

Cross-sectional channel profiles can be constructed using a combination of photogrammetry and shallow water bathymetry. With the aid of an analytical stereoplotter, all features above water level (including bars) can be accurately surveyed. This information can then be combined with shallow water bathymetry to yield a cross-sectional channel profile. Shallow water bathymetry techniques are experimental in many contexts and are covered more fully below. Figure 13 is a graph of the heights measured across transect 1 (see Figure 11). The width of the river is indicated on the graph, with the addition of shallow water bathymetry data, the bathymetric water profile could be completed.

Mean Wetted Width

Mean wetted width is a measurement of the horizontal distance perpendicular to the channel axis from the water's edge on one side of the stream to the water's edge on the opposite side of the stream. Providing the edge of the stream is in full view on the aerial photography, this is a straight forward measurement for photographic interpretation and photogrammetry. If the stream is relatively small, on the aerial photograph, this distance can be measured directly from the photograph using a suitable precision device as distortions will be negligible. If, however, channel size is relatively large, an analytical stereoplotter should be used as to remove the tilt, radial displacement and scale distortions. Figure 13 shows the difference in the heights measured across transect 1 (Figure 11). The stream elevations were corrected by the stereoplotter; the parallax height measurements show the river as level, but since the measurement travels upstream, the slope for the water indicated by the stereoplotter is correct. The other notable feature on this transect is the more accurate depiction of the bank height on the east side of the stream. This bank has an elevated road section that is more accurately represented by the stereoplotter.

Mean Bankfull Channel Width

Similar to the measurement of mean wetted width, the mean bankfull channel width describes the distance between the banks of the river at the tops of the river banks. This measurement is more difficult to obtain using photogrammetric methods as the tops of the banks are often obscured by vegetation. If vegetation is not a factor, this measurement can be obtained in the same manner described for mean wetted width. Figure 13 shows the mean bankfull width of transect 1, see sections on cross-sectional channel profiles and mean wetted width for further discussion.

Elevation Determinations

Elevation determination is a primary task of photogrammetry. All topographic maps compiled in Canada and the United States rely on photogrammetry for their accurate depictions of elevation and area. Elevations can be measured in forest environments as the trees tend to mirror the underlying topography and an experienced photogrammetrist can measure the average height of the trees and subtract this value when measuring elevation at the tops of the trees to get a measurement of the topography.

Floodplain Topography

Floodplain topography can be characterized using aerial photographic interpretation. Geomorphologists specifically trained in landform evaluation traditionally have used aerial photographic interpretation as a primary tool. When combined with ground truth, the landforms present in the flood plain can be accurately mapped. Figure 14 is a stereogram of the confluence of Sowaqua Creek and the Coquahalla River. When viewed with the aid of a pocket stereoscope the area of the stereogram can be seen stereoscopically. A stereoscopic view allows superior identification and more accurate delineation of landforms and produces more accurate maps.

DISTURBANCE

Large Woody Debris

The identification and classification of large woody debris (LWD) is a straight forward photo interpretative task. The detection of LWD is scale dependant. If the scale of the aerial photographs is too small, accurate identification of LWD is not possible. On large scale aerial photography (1:5000 - 1:20000), LWD present within the bankfull channel width can be easily seen on aerial photography provided the logs have not become so sun bleached that they fade into the background. The number of trees and logs can be estimated and the size of the pile measured using simple photogrammetric procedures. Figure 15 shows an area of LWD along Sowaqua Creek. This 35mm reconnaissance aerial photography colour image was flown in June 1997 and scanned at 1200 dpi for digital display and analysis. On the original colour diapositive, the individual logs are much more clearly visible. Scanning of aerial photography for digital analysis alters the contrast characteristics (usually increases contrast) and reduces the spatial resolution of the original photography.

Erosion Parameters

An experienced geomorphologist/photogrammetrist can assess erosion parameters using photogrammetric measurements and photo interpretative techniques. One of the most important criteria for erosion susceptibility is slope angle. Slopes can be easily measured from aerial photographs. In areas where relief displacement or tilt are factors, slopes should be calculated with the aid of an analytical stereoplotter (see section on photogrammetry). In many cases geomorphologists can identify and classify the slope material by looking at the vegetation patterns and other site associations (drainage, landforms, erosion features). Most photographic interpretation of this type should be ground checked to validate the observation when ever it is deemed necessary (see also sediment sources).

Upslope Impact Potential

Upslope impact potential can also be determined using aerial photographic interpretation. In all instances involving important or critical interpretations, they should be done by an experienced photo interpreter (see also sediment sources).

BATHYMETRY

Bathymetry generally involves the measurement of water depth in marine or riverine environments. Water depth is an important factor for many parameters in riparian evaluations and fish habitat assessment.

Shallow Water Bathymetry

Aerial photographic interpretation can be used to intuitively assess water depths. Generally, the deeper the water is, the darker it looks on an aerial photograph. If the water is shallow, the bottom of the river can be seen on the photography. In deeper water, the texture of the bottom will not be visible; deeper water will look smoother and darker. This relationship between image brightness and texture and water depth can be used to find the locations of pools, and/or identify deeper and shallower reaches along a river.

The use of colour infrared film, and related digital imagery, has potential for qualitative water depth assessment. Water is a strong absorber of near infrared (NIR) electromagnetic energy. Starting at a certain depths, depending on water and substrate conditions (30-90 cm depth), the incident near infrared energy will be mostly absorbed and any water deeper than the absorption depth will appear black on the imagery. This could be used to easily distinguish between water shallower and deeper than the NIR absorption depth. Ground truth measurements should be used to accurately establish the absorption depth.

Figure 16 shows a comparison between the reflectance characteristics of water for normal colour film and colour infrared film. The shallower areas in the river delta can be distinguished from the deeper areas as they appear brighter. The underwater topography of the delta can be clearly seen.

The relationship between water depth and image brightness can also be evaluated quantitatively when digitally scanned colour aerial photography or multispectral digital electro-optical data are used. Certain brightness values (called digital numbers, or DNs) on the digital images can be directly associated with specified water depths. Ground truth observations (water depth measurements along selected calibration sites) are essential to accurately establish the quantitative calibration relationship between DN values and water depth. Simple regression techniques can be used to establish the calibration curve that becomes the basis for further water depth estimations on the same imagery.

Figure 17 shows an example of such a curve from a study site along the Harrison River. Image brightness values were selected from the imagery for the locations of the water depth measurements and a regression curve was established. It is evident that there is a strong relationship between the two variables (e.g. the DNs are lower the deeper the water becomes). This is used to produce a calibration curve that can then be used for water depth prediction in other areas on imagery from the same riverine environment, flight conditions and exposure values. For example, a DN of 25 would predict a water depth of 6 cm. A map of the water depth area for an image can be established by applying the regression curve to the digital numbers of the image. A map of the topography of the river bed can be developed from such a procedure to show estimated bottom contour lines.

These types of procedures are still highly experimental and require detailed and precise ground observation data and suitably trained experienced personnel.

Procedures have been developed for acquiring the necessary ground truth observations for coordination with pixel DN values from the multispectral digital images. It is crucial to accurately identify the DNs that represent the locations of the ground truth water depth measurements to be used in the calibration. If the correct location is not accurately determined, the correlation between the two variables (DNs and depth) is degraded. In most circumstances, it is very difficult to identify an exact location to within a few cm. As a consequence, the DN values used are averages of several pixels in the target area. This procedure has performed quite well and is the approach currently being used. Further research is necessary to improve calibration as well as identification of target locations and to assess the accuracy of the procedure for predicting water depth.

Apart from the usual limitations (weather and atmospheric conditions), this quantitative procedure is challenged by several factors. Surface roughness will cause sunlight to be reflected directly into the camera while imaging and cause specular reflection (see sediment below) that will increase the average DN values as well as alter the calibration curve and subsequent depth estimations. As well, large numbers of spawning fish in the water, at sub pixel resolution, can be a problem because the water will appear shallower than it actually is (see fish below). Changes in substrate (see substrate classification) and turbidity (see sediment) will change overall brightness levels and influence the depth estimations. Most of these problems can be solved by using appropriate calibration procedures. Selecting the right time of the day for image acquisition, avoiding times of heavy spawning activity in the river and not imaging during periods of high sediment concentrations will reduce water depth estimate errors.

Mean Water Depth

In fish habitat assessment, the water depth within a habitat unit is usually determined by averaging three depths measurements along a transect to portray average conditions within that habitat unit. This is outlined as the most efficient procedure, although using the average of only three measurements reduces data accuracy (Johnston and Slaney, 1996).

The above procedure for shallow water bathymetry will improve the measuring of mean water depth significantly since a map of water depth, covering the entire habitat unit, can be generated once the calibration has been established.

Mean Bankfull Channel Depth

This parameter is usually determined by first measuring the vertical distance from a horizontal line, at the height of the bankfull width, to the water surface and then measuring the mean water depth. The two measurements are then added to yield the mean bankfull channel depth. The first parameter can be measured from aerial photography using photogrammetry. Shallow water bathymetry procedures can be used to measure mean water depth.

Maximum Pool Depth

A pool is measured or estimated at its deepest point. The procedure for obtaining shallow water bathymetry estimates can be used to measure this parameter.

Figure 10 shows a stretch along the Horsefly River. The sinusoidal patterns in the water are the redds from the spawning sockeye salmon. In a similar fashion, pools will be visible on multispectral digital imagery and their depths can be estimated (Figure 18).

Pool Outlet Depth

These procedures are the same as in shallow water bathymetry and maximum pool depth above.

Residual Pool Depth

This parameter consists of the maximum pool depth minus the outlet pool depth (see above for these procedures).

SEDIMENT

High sediment loads adversely affect fish spawning and rearing habitats (Groot & Margolis, 1991). Sediment introduced into streams through human activity, such as agriculture or logging, is therefore of considerable concern for fisheries management (DFO, 1986). The sediment load of a river is commonly measured as its suspended sediment concentration (SSC). Another readily available measure is turbidity which is influenced not only by SSC but also by dissolved substances. At high SSC, turbidity has been shown to be predominantly determined by sediment (Goodin et al., 1993).

There is considerable interest in remote sensing of water quality in the literature ( e.g. Gitelson et al., 1994; Van Stokkom, 1993). Generally researchers focus on marine or lacustrine environments (Gitelson et al., 1993) with some examples from large, deep streams (Ferrari et al., 1996). Our current research focuses on the detection and monitoring of SSC in the highly dynamic riparian environments of salmon spawning and rearing habitats.

The simplest way to employ airborne remote sensing for SSC monitoring is through a visual interpretation of suitable aerial images. Figure 19) shows the confluence of the Little Horsefly River with the Horsefly River. Despite the relatively high flying height and resulting low spatial resolution, one can clearly see that the Horsefly River carries significantly more sediment than the Little Horsefly. Figure 20 shows the confluence of the Coquitlam River with the Fraser River. At this lower flying height, subtle changes in SSC are visible. At the very least, one can distinguish between low, medium, and high concentrations.

Visual interpretation only allows relative estimates: For example, the Horsefly carries more sediment than the Little Horsefly. After its confluence with the Coquitlam, the Fraser carries more sediment close to its right bank. No intuitive estimates of SSC quantity are practical, and visual interpretation does not allow for the monitoring or mapping of actual SSC concentrations in terms of mg/l. It is, however, of considerable practical value because it can significantly reduce the amount of field work when looking for sediment sources and variations in SSC. For a watershed like the Horsefly, one can identify the tributaries that introduce sediment into the river, as well as dynamic changes in SSC. A subsequent detailed field study to determine actual SSC values can then be limited to a few locations of interest. Similarly, photo interpretation helps for monitoring streams that have not been identified as problem watersheds. Then, after inspection, and only if the visual inspection reveals changes in the sediment regime, one can more closely examine a particular stream and location(s). The British Columbia Forest Service, Caribou Forest Region, currently use visual inspection from the air to monitor remote streams before logging activities start (P. Teti, pers. com., 1997).

In addition, qualitative monitoring can be applied in retrospect utilizing historic aerial photography and imagery to establish trends in sediment loads for specific watersheds. Since for most streams there are no historic sediment data available, a visual comparison between different images over time is the only way to examine past SSC variations. From a resource management point of view, streams historically carrying high sediment loads can then be separated from those with recent SSC increases.

In addition to relative estimates of SSC using visual inspection, airborne remote sensing can provide quantitative estimates of SSC and turbidity. Our previous research reveals that quantitative estimates are difficult at low concentrations (< 25 mg/l), especially in shallow streams where depth and bottom substrate play an important role (e.g. Roberts et al., 1995). Under these conditions, the estimated values are generally more variable than the measured SSC and require adjustment for bottom reflection and depth variations. If the SSC is sufficiently high to obscure bottom reflection, the estimated SSC values become more stable. These parameters are dealt with in separate sections (see bathymetry and substrate).

Multispectral digital and photographic imagery of the Horsefly River watershed during the 1997 freshet were flown for SSC analyses. The digitally converted 35 mm colour reconnaissance aerial photography has been analysed and used to illustrate these values. The ground truth consisted of sediment samples integrated over the whole water column (integrated SSC), sediment samples from near the water surface (surface SSC) and turbidity measurements. Due to limited access and the strong river current, all ground truth samples and measurements were taken from bridges. These ground truth data were compared with digital numbers (DNs) extracted from the scanned aerial photographs. Figure 21 shows a typical example of such an image with several pixel read out locations.

We initially compared raw DNs with the ground truth data in an attempt to correlate the two. The DNs show little reaction at low concentrations but respond well to higher ones until the image saturates. One can see quite a range of DN values at each turbidity level, stemming from the extraction of several values for each sampling location (Figure 22 a).

In a second step (Figure 22 b), all DN values for each sampling location were averaged. This resulted in an improved correlation. Figure 23 shows the results for the SSC estimates with lower r² values, especially from the integrated samples. The expected function does not fit well here and the best fit function was a reciprocal logarithm for surface SSC and a reciprocal model for the integrated SSC. The sample size for integrated SCC was much smaller than for turbidity, and sampling occurred only at times of low concentration. This would have a considerable impact and partially explains the curve differences. Similarly, the values for surface SSC were concentrated at low SSC levels and this would alter the shape of the curve.

Some of the DN values included in this analysis probably contained an error component due to imaging conditions and should be excluded. These errors can be due to a number of factors, such as bottom reflection, shadow, sun glint and atmospheric conditions. Some of these commonly encountered problems are illustrated on Figure 24 which shows an image of the study area affected by cloud shadow, sun glint, uneven illumination and vignetting (light fall-off across the image due to camera lens parameters).

In this study, obvious problem areas have been excluded from much of the analyses but, because of changing weather conditions and low SSC throughout the study area, significant errors were introduced and resulted in reduced estimate accuracies. Figure 25 shows a slight improvement in the correlation between turbidity and DN values after applying a simple radiometric calibration technique, which accounts for unequal illumination of different images (Pellika, 1996).

In conclusion, based on raw DN values, turbidity can be predicted in most situations with reasonable accuracy and performed best. Surface SSC has adequate potential for estimation calibration, while integrated SSC does not perform to the same levels of accuracy. Turbidity is an optical measure, whereas SSC is a physical one. Since remote sensing also is an optical technique, the better correlation with turbidity is not surprising. Unless the water is quite clear, remote sensing can only detect features at or near the surface. This suggests that surface SSC should produce a more accurate calibration than integrated SSC values. However, most salmon streams like the Horsefly are quite clear for most of the year. In fact, the integrated SSC samples for this study were taken under relatively clear water conditions. Bottom reflection would have affected the values and would have contributed towards the reduction in integrated SSC performance.

Quantitative remote sensing of suspended sediments and turbidity in shallow streams has the potential of becoming a readily available procedure using reconnaissance multispectral imagery. For most studies, it would be initially an experimental procedure and would require experienced personnel and a detailed research programme. Important considerations will include:

- the application and testing of more further radiometric calibration techniques, such as white region normalization, to reduce ambient light flux problems;

- establishing and testing criteria for the selection of appropriate pixel read out locations for improved correspondence with ground truth;

- analysis of multispectral electro optical digital imagery and comparisons with the performance of digitally converted reconnaissance aerial photography;

- since CCD electro-optical cameras are more responsive to subtle changes in electromagnetic energy than photographic film, an improvement in correspondence with surface observations would be expected; however, these higher contrast characteristics make accurate image exposures more difficult and such CCD based systems would also be more sensitive to light flux problems;

- development and testing of correction procedures for varying water depth and bottom substrate conditions (see shallow water bathymetry and bottom substrate).

Sources

For the identification and mapping of sediment sources, two approaches exist. Conventionally, aerial photographic interpretation as well as field surveys are used to identify and map all potential sources, such as roads, clear cuts or land slides within the watershed (see Figure 26). A large proportion of these potential sources, however, may not introduce any sediment into the river under most circumstances. Whether a potential source is in fact a real sediment source needs to be verified in the field. Digital image analysis of multispectral remote sensing data may be used to monitor suspended sediment concentration (SSC). When a change in SSC within the river or a sediment introducing tributary is detected, one can search for the sediment source within a comparatively small area. This approach is liable to miss small sediment sources but allows for faster and more efficient identification of significant sources.

Examples for potential or real sediment sources are given in Figures 27 to 29. Figure 27 shows a natural cut bank along Moffat Creek, a major sediment contributor to the Horsefly River. A recent slope failure is visible on the image and it has introduced sediment and large woody debris into the creek. A major land slide is visible on Figure 28, and an encroaching clear cut can be seen on Figure 29. While Figure 27 quite clearly shows a significant sediment source, the low spatial resolution of such scanned high altitude images can make a decision as to whether these are potential or real sources difficult. In such cases, one needs to examine the original photographic diapositive as it will have superior resolution and a broader contrast range. In the case of the landslide on Figure 28, the original diapositive image shows that the slide has reached the river and introduced a significant amount of sediment. In the case of the clear cut on Figure 29, the original diapositive does not allow a clear decision and a field inspection is warranted.

In summary, at this time, qualitative monitoring of suspended sediment concentrations and sources using airborne multispectral reconnaissance remote sensing is operational. In addition, such remote sensing data have the potential to provide detailed and comprehensive quantitative estimates of SSC. However, for any particular study, an experimental approach involving calibration and further research is essential for reliable performance and to fully explore and develop this potential. We expect such reconnaissance remote sensing to play a significant role in future quantitative sediment monitoring projects in coastal riparian environments. Although it has a larger margin for error, the synoptic capability and access to remote areas, where ground inspection is difficult, provides an important contribution.

SUBSTRATE

Classification

As discussed in the sections on bathymetry and suspended sediment, the type of bottom substrate affects the accuracy and precision of quantitative estimates of suspended sediment concentrations and water depth. Several attempts at classifying substrate types have been made by other researchers (e.g. Lyzenga, 1978; Peddle et al., 1996), some with considerable success. A comparison of the images in Figure 18 and 30 shows that the distinction between spawning gravel (Figure 18), sandy bottom (Figure 30 a) and coarse gravel or boulders (Figure 30 b) is possible without field observations. Under the right conditions, a more precise classification of bottom substrate is possible; however, at this time, no research has been undertaken towards making it operational.

In addition to size and type of the bottom material, algae cover and the presence of aquatic vegetation will have a considerable impact. Changing water depth, overhanging vegetation, sun glint, and other light flux factors have an impact on the classification of bottom substrate in the same way that they can hamper other procedures.

Area Estimates

Since remote sensing provides a synoptic view, successful substrate classifications can easily be translated into area estimates. The photogrammetry section provides a discussion of techniques for area estimates or measurements from remotely sensed images. Sun glint, overhanging vegetation and shadow can prevent a successful substrate classification and it is not possible to expect to accurately map substrate along the entire length of a river using only remote sensing data. However, remote sensing allows convenient substrate mapping of substantial portions of most rivers. Considering limited access, especially in remote locations, remote sensing has considerable advantages over field surveys.

VEGETATION

Cover Type and Percent

Cover type and percent are components in all vegetation classifications. Vegetation classification has traditionally be conducted with the aid of aerial photographic interpretation and field verification. The spectral response of varying vegetation shows the greatest separation in the near infrared region (Figure 5). It is therefore advantageous to obtain remote sensing imagery in the near infrared (photographic or digital). Stressed plants often appear as a different colour than healthy plants on colour infrared images. This property has been routinely exploited to accurately map vegetation.

Until recently, colour infrared film for small format reconnaissance camera systems was difficult to obtain and expensive to process. Eastman Kodak has just released a new colour infrared film. This CIR film can be processed using the E-6 process (rather than E-5A), which considerably reduces the expense and processing turn around time, as it permits local processing in most photo finishing laboratories. Figure 31 shows a normal colour and a false colour infrared (CIR) image acquired over the Coquitlam River. This example shows the ability of the CIR film to distinguish between water and plants.

Riparian Vegetation

The mapping of riparian vegetation types and structural stages is a standard application of aerial photographic interpretation. It is usually done in several stages, preliminary mapping and interpretation, field reconnaissance and subsequent finalization of the interpretations. An example can be found in Morantz and Haefele (1996) where, in addition to type and stage, levels of health are reported as well. Figure 32 represents a portion of their classification system which includes vegetation, land use, bank erosion, encroachments and other parameters.

Aquatic Vegetation

Images taken within the visible spectrum can generally detect objects in clear water up to several metres in depth. So it is possible to identify locations with aquatic vegetation. The colour image in Figure 33 shows a portion in the Horsefly River with aquatic vegetation clearly visible in the channel.

With adequate ground truth, it is possible to identify the species of the vegetation by using digital image classification procedures. Furthermore, colour infrared film has the ability to detect more subtle vegetation differences than normal colour film. It can be used to detect aquatic and submerged vegetation that is close to the surface. The colour infrared image in Figure 34 is an example from the Horsefly River flown in the Spring of 1997. The channel has flooded over the banks and covered the adjacent floodplain. Vegetation that is not submerged appears in shades of red. Vegetation that is submerged in very shallow water (on top of the levee) appears in bluish tones. The deeper water looks dark. Another example (Figure 35) shows a series of flooded meander scars with vegetation that is just at the water surface and appears in pink tones.

FISH

Over Winter Dewatering

The extent of over winter dewatering is a critical measure for the availability of fish habitat. It can be determined through the mean water depth at the lowest water level compared to the mean water depth at other times during the year. The lowest water level at which mean water depth is conveniently measurable usually occurs shortly before freeze up. Obtaining several mean water depth measurements throughout the year, however, is redundant unless the bed form changes significantly. Bedform changes are most likely in the spring freshet or during major flooding events. Conditions for measuring mean water depth are best in late summer when water levels are relatively low, SSC is low and weather conditions are fair for reconnaissance flights and collecting field measurements.

Remote sensing based methods to determine the bathymetry of a river not only yield mean water depth but allow mapping of the river's bedform, provided it is not too deep (see bathymetry). Once established, the bedform can be assumed constant at least until the next major flooding event, which would not likely occur until the next spring. Therefore, monitoring of the water level will suffice for determining mean water depth on subsequent dates by taking into account previous bedform data. In remote areas, multiple imaging flights to determine the location of the waterline may replace the monitoring of actual water levels to provide an estimate.

Mapping the waterline can be achieved through aerial photographic interpretation as well as through digital image processing. The generally high contrast between dark water and brighter land lends itself to detection using digital classification procedures. This land-water contrast is highest in the NIR portion of the spectrum and colour infrared film or NIR channels of multispectral imaging systems are well suited for this application. Figure 36 presents a portion of the Horsefly River on three different dates, spring, early fall and early winter. Generally, this kind of image analysis can be carried out without any field work. Note that overhanging vegetation as well as shadow from vegetation can introduce significant error. Since there is no need to know the water level and the water line at all locations, the analysis may be restricted to areas with favourable conditions. Also note that the extreme contrast between dark water and fresh snow (Figure 36 c) makes proper image exposure difficult and may prohibit shallow water bathymetry estimates.

Off Channel Habitat

Off channel habitats are back waters, lakes and ponds that may or may not be connected with the river's main channel. These habitats play an important role as rearing or over wintering habitat. In addition flooded areas during the spring freshet or after major rain storms in summer provide important feeding grounds for smolts. Aerial photographic interpretation and digital image processing provide convenient means of mapping such habitats.

Water bodies away from the river are easily detected on all kinds of imagery, with NIR imagery providing the best contrast. Depending on the image scale, one may not always be able to identify connections between off channel habitats and the main channel. In such cases, field visits are necessary to determine the level of utility of the off channel habitat. Figure 21 presents an example of such flooded areas. For mapping these flooded areas, the date of image acquisition is crucial.

Fish Population Estimates

Currently the Department of Fisheries and Oceans estimates the number of spawning salmon using manual counts on the ground, supported by occasional helicopter reconnaissance flights (D. Lawrence, DFO, pers. com., 1996). Ground based manual count is probably the most accurate and precise method available. It is labour intensive and time consuming, therefore costly. Similar to the monitoring of sediments, airborne multispectral remote sensing promises to reduce the amount of field work necessary to conduct fish population estimates.

Aerial photography can already now provide fish counts if the following conditions are met:

very low flying height (high resolution);

shallow, clear water;

no obstructions ( e.g. overhanging trees);

high sun angle (to avoid sun glint and facilitate water penetration).

These conditions are met in the image presented on Figures 18 and 37 which show the same image of a spawning habitat, scanned at different resolutions. On Figure 18, the fish are barely visible, whereas on Figure 37 they can be counted with ease. This comparison illustrates the need for high quality scanners and large data storage facilities for digital processing of such images. For manual counts, the original colour film diapositives are preferred.

These conditions severely limit the use of photo interpretation for population estimates. Furthermore, the use of aerial photography may increase the turnaround time compared to field methods due to processing and interpretation constraints. An automated estimate of the number of spawners from multispectral imagery or from scanned photographs has, to our knowledge, never been attempted. We are, however, currently implementing three different experimental methods to estimate the density of spawners in riparian habitats. They range from the analysis of local variance within an image, to the use of spatial statistics, to a simple model to simulate an image, under fish free conditions, which is then compared to an actual image with fish present. These methods will be tested for the first time on the 1997 Horsefly River sockeye run.

REFERENCES

Avery, T.E., and G.L. Berlin (1992): Fundamentals of Remote Sensing and Air Photo Interpretation, 5th Edition, Maxwell MacMillan Canada, Toronto.

Department of Fisheries and Oceans (1986): Policy for the Management of Fish Habitat., Ottawa.

Dickinson, C.G. (1979): Maps and Air Photographs, Edward Arnold, London.

Ferrari, G.M., N. Hoepffner and M. Mingazzini (1996): Optical Properties of the Water in a Deltaic Environment: Prospective Tool to Analyse Satellite Data in Turbid Waters. - Remote Sensing of Environment, vol. 58, pp. 96-80.

Gitelson, A., G. Garbuzov, F. Szilagyi, K.-H. Mittenzwey, A. Karnielli and A. Kaiser (1993): Quantitative Remote Sensing Methods for Real-Time Monitoring of Inland Waters Quality. - International Journal of Remote Sensing, vol. 14, pp. 1269-1295.

Gitelson, A., F. Szilagyi and K.-H. Mittenzwey (1994): Improving Quantitative Remote Sensing for Monitoring of Inland Water Quality. - Water Research, vol. 27, pp. 1185-1194.

Groot, C., and L. Margolis (ed.) (1991): Pacific Salmon Life Histories, UBC Press, Vancouver.

Goodin, D.G., L. Han, , R.N. Fraser, D.C. Rundquist, and W.A. Stebbins (1993): Analysis of Suspended Solids in Water using Remotely Sensed High Resolution Derivative Spectra. - Photogrammetric Engineering and Remote Sensing, vol. 59, pp. 505-510.

Lawrence, D., Department of Fisheries and Oceans, Williams Lake, pers. Comm. October 1, 1996.

Lillesand, T.M. and R.W. Kiefer (1994): Remote Sensing and Image Interpretation, 3rd Edition, John Wiley & Sons, Toronto.

Morantz, D. and M. Haefele (1996): Nicola River Drainage Riparian Condition Survey and Mapping Report. - Department of Fisheries and Oceans, Fraser River Action Plan, unpublished.

Peddle, D.R., E.F. LeDrew and H.M. Holden (1996): Remote Estimation of Coral Reef Abundance and Depth from SPOT Image Spectral Mixture Analysis and Ocean Profile Spectra, Fiji. - Proc 26th Int. Symp Remote Sensing of Environment.

Pellika, P. (1996): Multitemporal Relative Spectral Calibration of Airborne Video Data. - Proc. 2nd Airborne Remote Sensing Conference and Exhibition, vol. 3, pp. 17-26.

Roberts, A., C. Kirman and L.Lesack (1995): Suspended Sediment Concentration Estimation from Multispectral Video Imagery. - International Journal of Remote Sensing, vol. 16, pp. 2439-2455.

Teti, P. British Columbia Forest Service, Cariboo Forest Region, pers. comm. May 30, 1997.

Van Stokkom, H.T.C., G.N.M. Stokman and J.W. Hovenier (1993): Quantitative Use of Passive Optical Remote Sensing over Coastal and Inland Water Bodies. - International Journal of Remote Sensing, vol. 14, pp. 541-563.