Using a logistic regression model to delineate channel network in southeast Australia

Abstract

The potential to delineate the location along a slope at which channels initiate is important for understanding hydrologic and geomorphic processes governing headwater streams. Most work assumes a uniform input of precipitation across the catchment, and every cell would receive the same volume of water. In reality, sites at higher elevations receive more rainfall, and tend to have smaller contributing area and stream length. In this paper, a channel initiation point (CIP) model is developed. The CIP model estimates channel initialisation based on a logistic regression (LR) technique. An LR relationship is applied because of its flexibility in assumptions where a discrete variable can be considered. By incorporating the accumulated rainfall surface into the LR, resulting drainage areas reflect hydrologic and geomorphic influences on channel initiation. The study area is part of the Lower Cotter experimental catchment, a headwater alpine catchment located in the Brindabella region in south-eastern Australia. The aim is to test the capability of the CIP model in estimating the channel network, capturing channel heads and disconnected channels. The estimated channel network is compared to that obtained using a classical method on the basis of a constant area threshold. The CIP model performs well in identifying channel and non-channel cells while improving channel head localisation and extraction of channel continuity. Overall, the CIP model can be considered as a valid alternative to commonly-applied traditional methods for channel network extraction from Digital Elevation Models (DEMs), in addition to considering hydrologic impacts on channel initiation.