High-resolution grain-size characterisation of gravel bars using imagery analysis and geo-statistics

The scarcity of grain-size data from gravel-bed rivers has traditionally hindered hydraulic, sediment transport and river habitat studies. A new remote sensing methodology to estimate grain-size distribution is presented. It combines textural digital images of the riverbed at 1 : 1000 and 1 : 40 scales with grain-size sampling. It was applied to a 12-km reach of the Isلbena River (Central Pyrenees NE Spain). First, textural patterns for each grain-size range were obtained, selecting the most closely related texture variables, including the use of semivariograms. Second, multiple linear regression equations were derived from the textural variables to estimate each value of the grain-size distribution. The highest correlation values (r2) were obtained from the central part of the distribution (D50 with a RMS error of 12.7%). Finally, new multiple linear regression equations to estimate the D50 and D84 were obtained from 1 : 1000 images and four textural variables. These were used to derive D50 and D84 maps of the riverbed, re-sampled at a resolution of 1.5 m pixels, with RMS estimation errors of 26% and 32%, respectively. Downstream change in grain-size is also well reproduced by the method. The mean D50 of 72 and 32 mm were estimated in the upper and the lower reaches of the river, respectively. The methodology shows great potential for application, the relation between the spatial resolution of the images and the mean grain-size of the riverbed sediment being the main issue for future development.