Climate Change: Defining Reference Conditions and Restoring Freshwater Ecosystems - Climate Change Impacts on Freshwater Ecosystems - page 208

Table 9.1 Standard deviations of the observed/expected ratios of biotic indices for
RIVPACS-type model reference sites based on null models, WFD system-A models and
the RIVPACS-type models used in Great Britain, Sweden and the Czech Republic. (Results
are taken from Davy-Bowker et al . 2006.)

Index Country Season Null model* System-A RIVPACS-type model

Number
of taxa

Great Britain Autumn 0.263 0.251 (−5) 0.218 (−17)

Sweden Autumn 0.355 0.345 (−3) 0.312 (−12)

Czech
Republic

Autumn 0.383 0.317 (−17) 0.276 (−28)

ASPT Great Britain Autumn 0.132 0.113 (−14) 0.086 (−35)

Sweden Autumn 0.139 0.122 (−12) 0.112 (−19)

Czech
Republic

Autumn 0.091 0.089 (−2) 0.085 (−7)

The per cent reduction in SD(O/E) compared to null models is shown in parentheses. ASPT (average
score per taxon of BMWP families) (National Water Council, 1981).
*The null model, the predicted reference condition index value for a site, was calculated as the
average observed value of the index for all reference sites (Van Sickle et al . 2005).

was relatively high (c. 57.3% prediction error). Taking this one step further,
Davy-Bowker et al . (2006) compared the efficacy of typology- (system-A) and
model-based (RIVPACS-type models) approaches for establishing reference
conditions. Comparisons were made using predictive models calibrated for
Great Britain (RIVPACS, e.g. Clarke et al . 2003), Sweden (SWEPAC SRI , R.K.
Johnson, unpublished) and the Czech Republic (PERLA, Kokes et al . 2006).
The effectiveness of the system-A typology was compared directly with the
RIVPACS-type model predictions of expected index values. The null model,
the predicted reference condition index value for a site, was calculated as the
average observed value of the index for all reference sites (Van Sickle et al .
2005), and prediction accuracy was measured as the standard deviation (SD) of
the ratios of the observed (O) to expected (E) values of each biotic index for
the reference sites. Davy-Bowker et al . (2006) found that typology-based
approaches were inferior to model predictions. For all four indices studied
(only the results for two indices are shown here) and for all seasons and season
combinations (only results from autumn shown), SD(O/E) ratios were
consistently highest for the null models, indicating relatively high uncertainty,
and lowest for the RIVPACS-type models (Table 9.1).
On average, RIVPACS-type models resulted in an 11% decrease in uncertainty
compared with system-A. Although the environmental variables used by the three
RIVPACS-type predictive models differed, several variables, such as latitude,
longitude, altitude and substratum type, were used in either two or all three of
the models. In all three country models, the variables explained a larger proportion
of the variation in invertebrate communities than the WFD system-A variables.
The authors argued that predictive models were more effective due to the use of
continuous rather than categorical predictor variables and because the models