List of Tables
6. RESULTS
6.6 SEASONAL AND SPATIAL DISTRIBUTION OF SURFACE WATER
can be separated in the thermal infrared. Based on this evidence it can be concluded that the presence of fire scars has a confounding effect on the accuracy of the spectral mixture analysis and that indications of water within fire scars need to be treated with caution.
Traditional accuracy assessment methods often provide a numerical measure of accuracy, be it producer's accuracy, user's accuracy or even an overall accuracy percentage. In this study, the use of cumulative rainfall as an accuracy test did not lead to such a quantifiable measure but led, instead, to a situation were it could be asserted that the results were 'probably' accurate. This is clearly not an ideal situation, but was the best that could be achieved given the lack of effective methods for assessing the classification accuracy of historical images and/or soft classifications.
The dry-wet rainfall pattern also manifested itself in the distribution of surface water on the Eastern Shores (see Figure 6.9 on page 107), emphasising the close link that exists between the two. This is most clearly shown in Figure 6.17 which illustrates the similarities in the water distribution from the three study dates in the wet period (shown in blue). The curves for these three dates all show elevated amounts of water in the 11 to 50% range, indicating the presence of extensive, vegetation-dominated areas of water.
Figure 6.17: Areas, in hectares, covered by the water classes for each study date.
Curves are grouped according to rainfall regime. Also shown are the affects of the large fire in September 2002. The 'no water' class is not shown.
The picture presented by the results from the drier period in 2002 is confused by the effects of the fire scars. The curves for April and July (shown in brown) are very similar and indicate the presence of very little water, particularly in the 31 to 80% range. Increases are noted in classes 2 and 3 although these are a lot lower than the values for the wet period.
The curves for September (shown in red) and, to a lesser extent, that of October 2002 (green) are unquestionably anomalous and reflect the influence of the large fire scar to the north of Lake Bhangazi. The September curve shows elevated amounts of water in classes 6, 7 and 8 while both the September and October results indicate increases in water in classes 2-5 as compared to April and July 2002. The differences in the September and October curves can be explained by the freshness of the fire scar. In September the fire scar was new and dominated by burnt vegetation. By October the vegetation had started recovering, providing an explanation for why the October curve resembles the September curve for classes 2, 3 and 4 but is more like the dry April and July curves for classes 6 to
10. The April 2002 fire scar (see Figure 6.15a) was very small and did not appear to influence the results shown in Figure 6.17.
A final trend visible in Figure 6.17 is that classes 9 and 10 (81 to 100% water) exhibit similar patterns for all seven study dates. This is not surprising as Lake Bhangazi dominates these two classes, especially class 10 where it makes up over 98% of the area for each of the study dates.
The patterns apparent in Figures 6.16 and 6.17 also manifest themselves in Figure 6.9 on page 107, which illustrates the spatial distribution of water in the study area. The three images in the top row are from the wetter period in 1991 and 2001 and all show similar patterns of water distribution. High concentrations of water are found in the well-defined Mfabeni swamp south of Lake Bhangazi as well as in the wilderness area to the north. The water in the wilderness area is diffuse but widespread during the wetter period, when many regions show concentrations in the 20 to 60% range and some areas reach 90%. Also apparent are regions of water in the Brodies Crossing area, where concentrations of 60%
were measured in May 2001. The four images from the drier 2002 all show low concentrations of surface water throughout the study area. Exceptions to this are Lake Bhangazi and the burnt regions discussed in the previous section. Lake Bhangazi, with its fringe of reeds, shows up clearly in all the images.
The distribution of water in the Mfabeni swamp is shown in Figure 6.18. Water concentrations in the core of this swamp reached 80% during the wet period, decreasing
1991/07/23 2001/03/20 2001/05/07 2002/04/24 2002/07/13 2002/09/15 2002/10/17
Percentage Water
Kilometres L
0-10% 11-20% 21-30% 31-40% 41-50% 51-60% 61-70% 71-80% 81-90% 91-100%
Figure 6.18: Spatial and temporal distribution of water in and around Lake Bhangazi and the Mfabeni swamp.
outwards to the margins of the swamp. During the 2002 drier period the wet core of the Mfabeni was almost absent. However, a secondary linear core at the southern end of the Mfabeni swamp is present throughout, with concentrations in the 21-30% range being recorded even in the dry months. More diffuse patterns of water, discernible to the west of Lake Bhangazi in the wetter periods, were absent or much reduced in 2002.
To conclude this section on the distribution of surface water, it is illuminating to examine those areas that remained the same throughout the study period. Of the 102.8 km2 that were analysed, 19.6 km2 were always classed as class 1 while 2.5 km2 were always mapped as class 10. These areas are shown in Figure 6.19. The pixels in the remaining 80.7 km2 were assigned a different class classification at least once over the seven study dates. A detailed examination of these pixels that changed class proved difficult given that each pixel could
Figure 6.19: Results of change analysis showing areas of change in pink and areas of no change in blue and green.
have assumed any one of ten class values for each of the seven study dates, resulting in a possible 10 million permutations per pixel.
Areas that remained as class 10 throughout the study (i.e. were always water) are confined to Lake Bhangazi while regions that were always vegetation are found mainly to the west and south-west of the Mfabeni as well as in the southern parts of the study area. The Mfabeni swamp, the wilderness area north of Lake Bhangazi and the area near Brodies Crossing are dominated by change, reflecting areas where fluctuations in surface water occurred. These results, in providing insight into the connectedness and spatial distribution of surface water in the study area, allow the structure of wetlands systems and their relationship with biodiversity to be investigated.