The Horsefly River has several major and numerous minor tributaries within the study area. Tributaries with a significantly different suspended sediment concentration should exhibit significantly different DN values. Moreover, major tributaries should also alter the concentration of the Horsefly itself sufficiently to exhibit significantly different DN values up- and downstream of the confluence point. In addition one expects a heterogeneous picture within mixing zones, i.e. a higher variance. A monitoring system based on the analysis of remotely sensed images of confluence points can then be used to identify tributaries which have an impact on the river. Such a system may reduce the need for field sampling considerably.
Maps 1 to 6 show the confluence of the Horsefly river with 5 major tributaries and one minor one (Prairie Creek). DN values were extracted from scanned 35mm aerial photographs in a random fashion within the extends shown on the map, excluding areas affected by sun glint, shadow, and obvious bottom reflection. The maps also show results of student t-tests (comparison of means) and F-tests (comparison of variances) between the data sets. Since the DN values were assumed to be dependent on the SSC and the latter can only change in one direction, when comparing two sites, all t-tests were performed as one tailed test. In comparisons where an homogeneous distribution of suspended sediments thus and equal variance of the two data sets was assumed a homoscedastic test was performed whereas for all others heteroscedastic tests were used. Reported differences in SSC were based on visual interpretation of the aerial photographs, supported by SSC samples from a limited number of sampling stations.
Moffat Creek: Moffat Creek is a major tributary and at the time of imaging had a higher SSC than the Horsefly. Therefore the Moffat was expected to alter the variance of the Horsefly within the mixing zone, and to significantly alter the spectral response of the Horsefly. The latter could not be tested since no DN values of the Horsefly upstream of the confluence point were available due to sun glint. The statistical results (see map 1) confirm the hypothesis and suggest that in fact the influx of sediment was sufficiently larger to alter the Horsefly's spectral response. While the variance differed in all three bands between all three sample points, t-tests showed no significant difference between the Moffat and the mixing zone in the green band and between the Moffat and the Horsefly further downstream. All other t-tests showed a significant difference.
Little Horsefly: The Little Horsefly is one of the largest tributaries and exhibited extremely low SSC. Therefore one expects this tributary to have a major impact. This confluence area was sampled extensively with multiple samples within each SSC class to serve as null samples. As map 2 shows the results were somewhat surprising. Comparing the Little Horsefly with the Horsefly's upstream, mixing and downstream zones the results were close to what one would expect. Namely that t-tests showed a significant difference between upstream, downstream and the Little Horsefly and that the variance generally differed between the mixing zone and all other areas. The only exceptions to this occurred in the blue band which generally reflects SSC to a lesser extend than red and green do. Similar variances between the upstream area and the Little Horsefly were not unexpected since both show homogenous distribution of sediment, albeit at different concentrations.
A comparison of sub areas within the upstream and downstream zones, however, revealed an unsettling picture. While one expects these sub areas to be very similar, they did exhibit significant differences.
McKinnley Creek: The McKinnley represents a medium sized, shallow tributary and at time of imaging was clear. Here the results pointed in the expected direction. There were significant differences between the Horsefly (upstream) and the McKinnley in terms of means and variances, with the exception of the variance in the blue band. Also, there was a significant difference between the variances of the Horsefly (upstream) and the mixing zone, while, again with the exception of the blue band, the McKinnley is too small to significantly change the mean of the mixing zone from the upstream zone. Similarly one expects means and variances to be different in the McKinnley and the mixing zone, which was the case. Finally, the variances of the upstream and downstream zone did not show significant differences while the ones of the mixing zone differed from both. However, there were also significant differences between the means of the downstream zone on one hand and the mixing and upstream zones on the other.
Prairie Creek: Prairie Creek is a small tributary and though carrying significantly less sediment than the Horsefly was not expected to alter the latter's spectral response much, especially in comparison to larger tributaries. However, there were significant differences in means between the up and downstream zones as well as between the downstream and mixing zone, but not between the upstream and mixing zones. Not surprising were the differences in means between Prairie Creek and all other areas, while the similarities in variance could not be fully explained.
McKusky Creek: The McKusky is a medium to large tributary and at the day of imaging showed SSC similar to the Horsefly's. The lack of difference in variance between the mixing zone and the rest of the Horsefly suggested that the influx from the McKusky had only a minor impact on the sediment regime of the Horsefly. This was supported by the similarities in means between upstream and mixing zones. At the same time the was a significant difference between the downstream area and all other areas. The variance comparison between the McKusky and the Horsefly showed an irregular picture.
McKay River: The McKay must be considered a major tributary. Unfortunately no ground samples were available and no image analysis of the McKay could be performed due to its turbulence and shadows. Therefor the McKay's impact was difficult to predict, however, since it is a major tributary, almost the size of the Horsefly itself a significant impact was to be expected, unless the SSC were equal. The results did not quite confirm this expectation. For example, there was no significant difference in means between upstream and mixing in the green band. Similarly these two areas did not differ in variance in the red band. Most striking was the apparent lack of differences between the mixing and the downstream zone.
When considering only the comparison of different river stretches (i.e. upstream, mixing, downstream and tributary) the results, with some exceptions, seemed to conform with the hypotheses stated above. Namely that tributaries have significantly different DN values and have a significant impact on those of the Horsefly. In a number of cases there was no difference between the areas classified as 'upstream' and 'mixing zone' but between those and the 'downstream' area.
The 'mixing' zone was taken to be the area immediately at or after the confluence point. In reality mixing may only occur further downstream with the tributary either being displaced to the edge or vertical stratification being in place. In this case the 'downstream' zone is the actual mixing zone, while the 'mixing' zone is still part of the 'upstream' area. Such a scenario is more likely with small tributaries, such as the Prairie Creek or the McKinnley than with larger ones like the Little Horsefly or the McKay . In fact the results support this, with some exceptions.
Other potential error sources for this analysis were the presence of bottom reflection, specular reflection form the water surface, floating debris as well as photographic and scanning faults. Considering the relatively low SSC values, ranging from less than 5 mg/l in the McKinnley to up to 100 mg/l in the Moffat, bottom reflection likely occurred despite efforts to exclude extremely shallow areas.
However, the differences between several sub areas within 'upstream', 'mixing', 'downstream', and 'tributary' at the Horsefly - Little Horsefly confluence called the viability of a monitoring system based on above results in question. The same error sources mentioned above might have been responsible for these significant differences in DN values. Given the overall low SSC and the narrow range variations in depth, surface roughness and scanner performance could have an impact as large if not larger than the actual changes in SSC. In other words the analysis performed was too sensitive to subtle variations in order to detect significant sediment influxes from tributaries.
Sediment concentrations in the Horsefly River, it being quite shallow and clear, are difficult to monitor via remote sensing. At the current stage our system is too sensitive to natural variations in the DN values, e.g. caused by bottom reflection, or to errors introduced in the data processing, e.g. due to a low quality scanner. Areas of improvements include:
bathymetry mapping of the entire study area during summer low flow and low SSC periods to facilitate the selection of areas suitable for SSC analysis;
more extensive ground based SSC sampling near confluence points for a limited period of time to better calibrate remote sensing analysis;
use of hyperspectral imagery to determine appropriate wavelength bands.
utilization of a high quality scanner, or high quality digital cameras; and
Bathymetry measurements over several years at selected sites suggest that, although the Horsefly is transporting high sediment loads during early spring the general bathymetry remains fairly constant. Therefore bathymetry maps from remotely sensed images, although not perfectly accurate, will be valid for several years and sufficient to select suitable, i.e. deep areas . Ground based sampling at all confluence points analysed here is planned for the spring 1998 field campaign, as is the use of an imaging spectrometer which is currently being assembled at the University of Dundee, Scotland. Our digital imaging system has recently been upgraded with higher quality progressive scan cameras which not only provide a better spatial resolution but are also less susceptible to electronic noise. The acquisition of a high quality scanner is envisioned for the future but currently not feasible due to budgetary constraints.
Methods
Results
Discussion
Conclusion
Back to SSC main page