Monitoring River Health Initiative Technical Report Number 25
L.A. Barmuta, L. Emmerson and P. Otahal - University of Tasmania
Environment Australia, 2002
- Australian River Assessment System: AusRivAS Errors Analysis (Phase I Final Report) (PDF - 1,567 KB)
About the report
Of the final models available for testing, there was a wide variation in the types of variables to which AusRivAS is potentially sensitive. The most consistent pattern across most models was that AusRivAS is intolerant of variations in the input of categorical variables. (Categorical variables are those variables whose values consist of a small number of integer values.) Varying the input values of categorical variables by just ± 1 usually changes the values of the output index O/E Families substantially. There were no consistent patterns for the other input environmental variables, even within a State or Territory. This means that each model has a unique suite of variables to which it is most sensitive. As final models are developed for each State or Territory and tested for their sensitivity using our procedures, we can provide advice to each agency about which environmental input variables have the greatest impact on AusRivAS outputs for each of their models. These variables can then be targeted in training and protocol improvements.
It is unclear how much of a threat these sensitivities pose to AusRivAS because there is insufficient information to judge whether the errors we introduced into the models are realistic: some variables may be measured with little or no variation between operators (e.g. those derived from computerised GIS sources), whereas others may be subject to considerable inter-operator variation. We suspect that the variables most vulnerable to such errors are those where visual estimates are made under field conditions (e.g. percentage cover of substrate types or of overhanging vegetation). These estimates are often converted to categorical variables prior to input to AusRivAS, and so minor variations in the original estimate could result in a change on the value of the input categorical variable.