• LOESS;
  • leave-one-out;
  • flow models;
  • inundated area models;
  • Kolmogorov–Smirnov test


Building of dams and subsequent abstraction upstream of rivers significantly affects major wetland systems. Knowledge of anthropogenic impacts of river regulation, relative to stochastic variation, is essential if managers are able to adequately manage flows to wetlands. We built annual flow and inundated area models for the Macquarie Marshes supplied by the Macquarie River, based on available annual rainfall data (1879–2006). We used LOESS, the flexible local polynomial regression method, ideal for modelling complex processes where no theoretical models exist. To compare effects of river regulation and abstraction, we developed two models, before river regulation (‘natural’ or ‘unregulated’) and after river regulation (‘regulated’). The division was when Burrendong dam, the most significant regulatory structure affecting flows, was built in 1963 and completed in 1967. After this point, water was diverted upstream, predominantly for irrigation. We developed ‘natural’ and ‘regulated’ flow models for three flow gauges Dubbo, Warren and Oxley based on annual rainfall at rainfall stations in the upper catchment and flow, and then inundated area models based on flow and local rainfall to fit annual inundated area data (1979–2006) in the Macquarie Marshes, using LOESS and leave-one-out samples without overfitting. We evaluated the performance of the proposed models and used these models to predict annual flows and inundated areas from 1879 to 2006. Comparison was done in terms of ‘natural’ and ‘regulated’ data using Kolmogorov–Smirnov test, identifying significant reductions in flow and wetland area. Copyright © 2009 John Wiley & Sons, Ltd.