Comparison of Automated Image-Based Grain Sizing to Standard Pebble-Count Methods

This study explores the use of an automated image-based method for characterizing grain-size distributions (GSDs) of exposed open-framework gravels by comparing the GSDs measured with the image-based method to distributions obtained with two pebble-count methods. Selection of grains for the two pebble-count methods was carried out using a gridded sampling frame and the heel-to-toe Wolman walk method at six field sites. At each site, 500-particle pebble-count samples were collected with each of the two pebble-count methods and digital images were systematically collected over the same sampling area. For the methods used, pebble counts collected with the gridded sampling frame were assumed to be the most accurate representations of the true grain-size population. Therefore, results from the image-based method were compared to the grid-derived GSDs for accuracy estimates; comparisons between the grid and Wolman walk methods were conducted to give an indication of possible variation between commonly used methods for the particular field sites used in the study. The grain-size comparisons were made at two spatial scales. At the larger scale, results from the image-based method were integrated over the sampling area required to collect the 500-particle pebble-count samples. At the smaller sampling scale, the image derived GSDs were compared to those from 100-particle, pebble-count samples obtained with the gridded sampling frame. The comparison shows that the image-based method performed reasonably well on five of the six study sites. For those five sites, the image-based method slightly underestimated all grain-size percentiles relative to the pebble counts collected with the gridded sampling frame, but the method performed well in estimating the median grain size (the average bias for ${\psi }_5$ , ${\psi }_{50}$ , and ${\psi }_{95}$ was $0.07\psi$ , $0.04\psi$ , and $0.19\psi$ , respectively). The Wolman pebble-counts yielded coarser results than the pebble counts obtained with the gridded sampling frame, especially for the smaller percentiles (the average bias for ${\psi }_5$ , ${\psi }_{50}$ , and ${\psi }_{95}$ was $0.20\psi$ , $0.16\psi$ , and $0.04\psi$ , respectively). Oversegmentation of large pitted grains in the image-analysis procedures was identified as a leading cause for failure of the image-based method at one of the sites. It is likely that lower degrees of oversegmentation and physical particle orientation contributed to the slight underestimation of all grain-size percentiles in the image-based method.