Publication details for Dr Patrice CarbonneauBlack, M., Carbonneau, P., Church, M. & Warburton, J. (2014). Mapping sub-pixel fluvial grain sizes with hyperspatial imagery. Sedimentology 61(3): 691-711.
- Publication type: Journal Article
- ISSN/ISBN: 0037-0746 (print), 1365-3091 (electronic)
- DOI: 10.1111/sed.12072
- Keywords: Airborne remote sensing, Digital image processing, Fluvial grain size, Grey-level co-occurrence matrix, Hyperspatial imagery, Image texture, Sub-pixel features.
- Further publication details on publisher web site
- Durham Research Online (DRO) - may include full text
Author(s) from Durham
This paper presents an investigation of image texture approaches for mapping sub-pixel fluvial grain-size features from airborne imagery, allowing for the rapid acquisition of surface sand and coarse fraction (>1.41 mm) grain-size information. Imagery at 30 mm resolution was acquired over four gravel bars from the Fraser River (British Columbia, Canada). Combined first-order and second-order image texture approaches (windowed standard deviation filter and the grey level co-occurrence matrix) were used. First-order image texture, through the application of a standard deviation filter and subsequent thresholding was used to detect the presence of surface sand, with optimal accuracy achieved at 91 ± 1.9%. A wide-ranging parameter space investigation was used to derive optimum parameters for the grey level co-occurrence matrix. Subsequently first- and second-order image textures were used in multiple linear regression to achieve good calibrations with several sub-pixel grain-size percentiles; relative error at 1.44%, 3.18%, 6.80% and 10.6% for D5, D16, D35 and D50, respectively. The larger percentiles of D84 and D95 had relative errors of 24.7% and 29.7%, respectively. The breakdown of calibration precision for larger percentiles is attributed to a ‘pixel averaging effect’. It is concluded that multispectral imagery is not required, because sufficient image texture information can be derived from standard colour imagery. Recommendations are suggested for application of this method to other localities and datasets, thus reducing exhaustive parameter searches in future studies.