Environmental Engineering Reference
In-Depth Information
To define the reference condition, an extensive dataset was compiled of
macroinvertebrate assemblages, collected using a standardised methodology
(Furse et al.
1981
; Murray-Bligh et al.
1997
) and identified to species level, from
representative sites across the UK which were not subject to pollution or
other environmental stress. These 268 reference sites from England, Scotland
and Wales (later further reference sites were added from Northern Ireland, as
well as additional ones from England, Scotland and Wales) contained 642
species, including species groups for taxa where it was not possible to iden-
tify individuals to species accurately. Using these data, a biological classifica-
tion was derived with TWINSPAN, which successively divided the reference
sites into groups in a hierarchical manner based on the similarity of their
fauna (Moss
1997
). Initially, 16 such end groups were identified varying in
their physical characteristics from naturally acidic, high gradient, upland
streams to large, sluggish rivers draining lowland basins on sedimentary
geologies (Wright et al.
1984
;Wright
1997
). Multiple discriminant analysis
was then used to identify the best set of physical, chemical and geographical
predictor variables to separate the different biological end groups (Moss
1997
). Multiple discriminant analysis does not assume that any individual
characteristic is capable of separating any two or more classes, but that
combinations of characteristics in multidimensional space will produce a
plane of separation that can be used to discriminate between classes
those variables that did not make a statistically significant contribution to
the explanatory power of the model; the original list of 28 physico-chemical
variables was reduced to an optimal subset of 11 variables in the final model
(Furse et al.
1984
;Mosset al.
1987
). It is these pollution insensitive predictor
variables that are then used to match the test site to the most appropriate
reference biological end groups.
In undertaking this process, it was recognised that pigeonholing rivers into a
prescriptive classification was an artificial process as, in reality, rivers repre-
sent a continuous gradient of change from one type to another and do not fall
into distinct biological types (Furse et al.
1984
; Wright et al.
1984
). Thus, in
defining the reference condition, RIVPACS does not assume a perfect match
between the test site and the reference biological end groups, but uses a
probabilistic assignment of a test site to determine end group membership
(based on similarity to test site and frequency of occurrence of that class) to
produce a predicted community based on several end groups, weighted in
the predictions robust with respect to the method of classifying reference sites
(Clarke et al.
2003
). Furthermore, this approach provides a means of automati-
cally identifying test sites that are inadequately represented by the reference
sites, thus preventing unreliable predictions.
