TSP Version 4.0

Abstract

The statistical methods underpinning analysis of streamflow data from paired-catchment studies have not changed much since the 1960s. Whilst such analyses are widespread in hydrologic practice and research, attention is rarely given to the problems of heteroscedacity, seasonality, serial correlation, and non-normally distributed variates. Each of these problems can potentially invalidate the basic assumptions upon which traditional statistical methods are based. We describe methods to contend with some of these problems and apply them to mountain ash (Eucalyptus regnans) forested catchments in the Maroondah Basin, south-eastern Australia. A seasonal regression model with lag-one auto-regressive (AR1) error was developed to predict monthly streamflow at treated catchments based on streamflow data from a control catchment. It is particularly well-suited to situations where little pre-treatment data is available. Differences between observed and predicted streamflow were used to quantify the effect of forest treatment on streamflow. Results from two catchment groups broadly matched the trend predicted by a previous regional model, with 2–3 year increases in streamflow, followed by decreases over the following one or two decades. A third group of catchments also showed initial increases, but expected subsequent decreases in streamflow were offset by the flow-increasing effects of insect infestations.