Chapter 7 . Stream Ecosystem Analysis
113
We usually used this tool, but provide only one example here. Similarities
between sampling and reference sites of different streams were visualised with
respect to water quality, habitat features and also by colonisation patterns of
benthic macro-invertebrates and adult aquatic insects (Fig. 7.1.). Reference sites
are a theoretical construct based on expert knowledge of typical chemical and
hydro-morphological quality classes: unpolluted and natural (Ref. 1), moderately
polluted and modified (Ref. 2), heavily polluted and modified (Ref. 3), extremely
polluted and completely degraded (Ref. 4) (LA WA 1998; Hessisches Ministerium
für Umwelt, Landwirtschaft und Forsten 2000).
The U-matrix (Fig. 7.1.) (Ultsch 1993) of the SOM visualises the similarities
between sampIes by location and shading of the space between the neighbouring
codebook vectors, i.e. prototypes. More than one sampling site may be mapped on
the same codebook vector, e.g. Ku2 and Ku3. Small distances between codebook
vectors and light grey shades indicate similar chemical and hydro-morphological
quality. For instance, the sampies We 4,5,6 and the reference Ref. 4 (top left) are
clearlyseparated by their marginal position and dark border. The other sampIes
show no clear separation, but rather transient regions. As expected, the most
contrasting references Ref. 1 and 4 take opposite locations.
Trajectories of subsequent sampling sites along streams can be visualised
within a SOM (Fig 7.2.). Some streams, particularly RW, show only small
differences between sampies, whereas others vary considerably. For example, Ku2
and Ku3 are mapped onto the same codebook vector, their large distances to Kul
and to Ku4 indicate environmental impact. A fishpond between Kul and Ku2 and
a storm water overflow between Ku2 and Ku3 caused damages, mainly to the
hydro-morphological status (see plane hydro-morphological quality class). Two
additional storm water basins between Ku3 and Ku4 affected a further decrease of
the water quality, whereas the hydro-morphological quality improved. The waste
water treatment plant between Ku4 and KuS caused no change of the water
properties. Between Ku5 and Ku6, river bed morphology deteriorated whereas
water quality remained unaffected. Both recuperated downstream Ku6. Another
option to visualise the similarities of the sampies is the Sammon map (Sammon
1969) of the SOM codebook vectors (Fig. 7.3.), approximately preserving their
distances.
The values of the components of the codebook vectors (water or hydromorphological quality) can be visualised by grey levels on the SOM - the darker
the grey shade, the lower the value (Fig. 7.4.). The interrelations of variables
become evident, e.g. the oxygen and the water temperature planes are
complementary .
In conclusion, SOMs are an excellent exploration tool for multidimensional
variables, providing visual aids for inspecting unknown data, outlier detection, and
initial grouping of data. In contrast to many other cluster analysis methods, SOMs
also handle data with smooth transitions, which are often typical for ecological
data (e.g. Vannote et al. 1980). The visualisation capability (grey shades and
distances) of SOMs shows similarities between objects and groups of objects. This
may inspire hypothesis generation and analysis of hybrid networks (see below).
113
We usually used this tool, but provide only one example here. Similarities
between sampling and reference sites of different streams were visualised with
respect to water quality, habitat features and also by colonisation patterns of
benthic macro-invertebrates and adult aquatic insects (Fig. 7.1.). Reference sites
are a theoretical construct based on expert knowledge of typical chemical and
hydro-morphological quality classes: unpolluted and natural (Ref. 1), moderately
polluted and modified (Ref. 2), heavily polluted and modified (Ref. 3), extremely
polluted and completely degraded (Ref. 4) (LA WA 1998; Hessisches Ministerium
für Umwelt, Landwirtschaft und Forsten 2000).
The U-matrix (Fig. 7.1.) (Ultsch 1993) of the SOM visualises the similarities
between sampIes by location and shading of the space between the neighbouring
codebook vectors, i.e. prototypes. More than one sampling site may be mapped on
the same codebook vector, e.g. Ku2 and Ku3. Small distances between codebook
vectors and light grey shades indicate similar chemical and hydro-morphological
quality. For instance, the sampies We 4,5,6 and the reference Ref. 4 (top left) are
clearlyseparated by their marginal position and dark border. The other sampIes
show no clear separation, but rather transient regions. As expected, the most
contrasting references Ref. 1 and 4 take opposite locations.
Trajectories of subsequent sampling sites along streams can be visualised
within a SOM (Fig 7.2.). Some streams, particularly RW, show only small
differences between sampies, whereas others vary considerably. For example, Ku2
and Ku3 are mapped onto the same codebook vector, their large distances to Kul
and to Ku4 indicate environmental impact. A fishpond between Kul and Ku2 and
a storm water overflow between Ku2 and Ku3 caused damages, mainly to the
hydro-morphological status (see plane hydro-morphological quality class). Two
additional storm water basins between Ku3 and Ku4 affected a further decrease of
the water quality, whereas the hydro-morphological quality improved. The waste
water treatment plant between Ku4 and KuS caused no change of the water
properties. Between Ku5 and Ku6, river bed morphology deteriorated whereas
water quality remained unaffected. Both recuperated downstream Ku6. Another
option to visualise the similarities of the sampies is the Sammon map (Sammon
1969) of the SOM codebook vectors (Fig. 7.3.), approximately preserving their
distances.
The values of the components of the codebook vectors (water or hydromorphological quality) can be visualised by grey levels on the SOM - the darker
the grey shade, the lower the value (Fig. 7.4.). The interrelations of variables
become evident, e.g. the oxygen and the water temperature planes are
complementary .
In conclusion, SOMs are an excellent exploration tool for multidimensional
variables, providing visual aids for inspecting unknown data, outlier detection, and
initial grouping of data. In contrast to many other cluster analysis methods, SOMs
also handle data with smooth transitions, which are often typical for ecological
data (e.g. Vannote et al. 1980). The visualisation capability (grey shades and
distances) of SOMs shows similarities between objects and groups of objects. This
may inspire hypothesis generation and analysis of hybrid networks (see below).