Riverine substrate classification using recreational-grade side-scan sonar imagery: trade-offs between accuracy and processing complexity

Recent introduction of recreational multi-beam and side-scan sonar equipment allows rapid, low cost acquisition of bathymetric data and substrate imagery in navigable waters. However, utilization of this data is hindered by a lack of established protocols for processing and classification. We surveyed 3 one-km sites on the Ogeechee River, Georgia, using Humminbird side-scan and multi-beam sonar units. Substrate type was classified and assessed for accuracy using sonar data processed at three levels of effort and complexity; 1) raw images exported as bitmaps, 2) georeferenced images, and 3) images corrected for slant-range distortions. Additionally, substrate type was classified using both heads-up digitizing and automatic feature extraction techniques. Highest accuracy was achieved using raw images, with lower accuracy at greater levels of processing. Heads-up classification yielded higher accuracy than automatic feature extraction for all processing levels. Results indicate a trade-off between processing complexity and classification accuracy. Ecologically relevant habitat variables can be derived at the lowest levels of processing which may be useful for determining relative proportions of substrate types. However due to lack of geometric correction, accurate areal data cannot be calculated. Fully processed images allow for increased measurement accuracy and pleasing maps, at the cost of effort and classification accuracy.