Maximising the accuracy of image-based surface sediment sampling techniques

Recent years have seen increased interest in automated methods, utilizing photographs collected with a hand-held digital camera, for determining the grain-size distribution of coarse river sediments. Such methods are as precise as traditional field methods, and have considerable time and cost advantages. Nevertheless, several unresolved issues pertaining to their deployment remain to be addressed. Using datasets collected from seven gravel-bed rivers, this paper examines four key issues: (i) the minimum area required to obtain a representative sample; (ii) the effect of lower-end truncation on grain-size percentiles; (iii) the effect of river-bed structure such as imbrication and hiding; and (iv) the potential benefits of using individual particle measurements rather than the number (or mass) of particles per size class to calculate percentiles. It is demonstrated that sampling areas of between 50 and 200- times that of the largest grain are adequate to achieve percentile errors of <10% (in mm). The appropriateness of lower-end truncation depends on the study aims and sediment properties. It has a limited effect on higher percentiles, except where sand is a major constituent. Understanding the influence of bed structure on image-derived size information is complicated by the absence of error-free benchmarks against which accuracy may be evaluated, but it is likely that other errors are more important. The use of individual particle measurements to calculate percentiles in preference to classified data is shown to have a small, but appreciable, effect on precision. These results will assist practitioners in making appropriate operational decisions to maximize data quality using image-based grain-size data capture.