Application of image texture analysis to grain sorting in a braided river physical model

Abstract:

Gravel braided rivers have complex lateral and vertical patterns of grain size sorting characterizing the river deposit and patchiness within the channel. Sorting has direct interactions with bed material transport processes for which it is important to know both the surface grain size relative to morphology and also the distribution of sizes in the morphological active layer involved in sediment transport over longer time periods. In the field, analysis of vertical sorting from direct sampling is either extremely laborious or requires long-term tracer studies of burial and mixing. We approached the problem from a new direction using a small-scale physical model of a braided channel to map bed elevation changes and grain size over time. The braided channel was developed at a slope of 1.5% and discharge of 2.1 l/s. The grain size distribution in the model is an approximately 1:30 scaled distribution of medium fluvial gravel with median size 1.3mm (D₁₀ 0.3 mm, D₉₀ 4mm). Vertical stereo images (with pixel resolution of approximately 1mm) of the dry bed were taken at one hour intervals over 40 hours of flume running time during which a large area of the river was re-worked. DEMs were derived photogrammetrically with mean elevation error of about 4mm. Surface grain size was mapped using a 7x7 pixel window and based on a calibrated relationship with the entropy values of the grey level co-occurrence matrix for the images. Consequently each pixel in the DEM also has an associated grain size. Over the 40 hour period the range of elevation and grain size variations over the river bed can be used to analyse the vertical particle-size sorting pattern within the morphological active layer (the layer between the maximum and minimum elevations at each point) and the presence of any vertical stratification in particle size. In braided channels with active bed scour and bar migration particle exchange occurs in a morphological active layer with thickness of the order of 10D_90. The layer is well-mixed with no strong correlation between depth in the bed and the grain size. The size distributions of the upper-most and lower-most layers are very similar. At the flume scale, all grain sizes occur with almost equal probability over the bed depth. Reciprocally, at any bed elevation all grain sizes occur with almost equally probability.