The Index of Wetland Condition (IWC) is a method for assessing the condition of Victoria’s wetlands. It has 13 variables that measure wetland catchment, salinity, nutrients, soil disturbance, water regime, physical form, extent and vegetation condition. Between 2009 and 2011, 830 wetlands were assessed using the method. To inform the management of wetlands not assessed, modelling was trialled to predict their condition. Several ordinal regression and classification techniques were explored. Modelling procedures were tested using IWC condition scores from the 830 wetlands and variables that could (1) affect wetland condition (e.g. land use) and (2) be measured remotely.
A principal components analysis of the individual IWC measures was performed to identify those which best correlated with wetland condition. From these, remote sensed surrogate variables and others that were likely to influence wetland condition were identified. Data for all the variables were obtained from spatial datasets for the 830 wetlands. Models were built by first randomly splitting IWC data into training data (used to calibrate) and testing data (used to assess predictive ability.
The models did not successfully predicted more than 50% of the condition categories of the test data. This demonstrates that the information contained in the variables was not sufficient to accurately classify most wetland condition. The relationships between these variables and wetland condition could be too complex and variable among wetlands to enable good predictions. Predictions may be improved by (1) basing the modelling on a coarser classification (i.e. poor and not poor), (2) ensuring there are additional wetlands in poor condition to increase the sample size of this category and (3) using updated spatial data as it becomes available to attribute wetlands.