Fanny Ville , Damià Vericat, Ramon J. Batalla, Colin Rennie

Particle entrainment intensity is spatially and temporally variable, making it a complex phenomenon to measure. This paper is the first of a pair, in which we present an automated image processing procedure (PhotoMOB) for monitoring the mobility/stability of gravel river beds. The method is based on local comparison of the shape of the grains identified at the same coordinates between successive photos to identify coincident and new grains. For each grain fraction, the proportion of grains that remained immobile and the proportion of grains newly identified in the study area can be calculated. In this part 1 paper, we present only the GIS-based procedure for identifying and characterising grain shapes in digital images of bed patches to derive a reliable surface Grain Size Distribution (GSD), and for subsequent analyses of bed mobility. The procedure is compatible with different forms of sampling (Area-by-Number i.e., AbN, and Grid-by-Number i.e., GbN) and types of measurements (continuous or real measures of the axes and discrete square holes measurements of the axes). The performance of the GIS procedure is evaluated by comparing estimated percentiles against manually delineated grains in ten 40x40cm image samples, as well as against the real bed grain sizes from the same patches measured with a Pebble-Box (continuous axis value) and two samples measured with a template (discrete axis value). Under optimal condition, the average root mean squared error (RMSE) of the manual procedure compared to the real measurement is 8.2% in AbN and 16.3% in GbN, while PhotoMOB performance is similar with RMSE of 9.5% in AbN and 16.6% in GbN. The paper also analyses how the tool performs when compared to discrete procedures such as measurement with templates. We found that in AbN, the under-estimation of the apparent size due to the imbrication effect is of the same order of magnitude as the under-estimation of the grain size measured by template. In GbN form, results emphasize the need of converting grain axis as a function of the average grain flatness for compatibility with discrete measurement, as coarse grains have more weight in the distribution and are often flatter in shape, hence are more often retained in inferior classes than smaller more spherical particles. A sufficiently large and appropriate sample area could reduce all the above mentioned RMSE by a third for AbN and by half in GbN.