GRAIN-SIZE MEASUREMENTS OF FLUVIAL GRAVEL BARS USING OBJECT-BASED IMAGE ANALYSIS

Traditional techniques for classifying the average grain size in gravel bars require manual measurements of each grain diameter. Aiming productivity, more efficient methods have been developed by applying remote sensing techniques and digital image processing. This research proposes an Object-Based Image Analysis methodology to classify gravel bars in fluvial channels. First, the study evaluates the performance of multiresolution segmentation algorithm (available at the software eCognition Developer) in performing shape recognition. The linear regression model was applied to assess the correlation between the gravels’ reference delineation and the gravels recognized by the segmentation algorithm. Furthermore, the supervised classification was validated by comparing the results with field data using the t-statistic test and the kappa index. Afterwards, the grain size distribution in gravel bars along the upper Bananeiras River, Brazil was mapped. The multiresolution segmentation results did not prove to be consistent with all the samples. Nonetheless, the P01 sample showed an R2 =0.82 for the diameter estimation and R2=0.45 the recognition of the eliptical ft. The t-statistic showed no significant difference in the efficiencies of the grain size classifications by the field survey data and the Object-based supervised classification (t = 2.133) for a significance level of 0.05. However, the kappa index was 0.54. The analysis of the both segmentation and classification results did not prove to be replicable.