This work suggests an alternative approach, the use of an artificial neural network (ANN), for investigation of cohesive sediment flocculation under different flow conditions. The ANN technique provides a powerful utility for input-output mapping if there is sufficient data. A feed-forward network is developed to predict the floc variability from a variety of causative variables. The best network is selected after testing many alternatives. The network is trained and validated by floc size measurements at one vertical level. The developed model allows for estimating the profiles of mean floc size throughout the water column for detailed investigation of flocculation dynamics. The results indicate that the mean size of flocs increases toward the bottom except within the bottom few centimeters, where the floc breakage is observed due to strong turbulence. Simultaneous shear rate and floc size profiles show that low shear promotes flocculation at low concentrations. Increasing turbulence intensity increases the amount of sediment in the water column, but decreases the floc size, indicating the floc breakage.
Estimating transport of suspended cohesive sediments in estuaries and coastal waters is of great importance to coastal morphology, underwater detection, navigation, water quality, and fate of pollutants and biomatter. Cohesive sediment is usually not present as individual grains but rather as flocs or aggregates composed of small particles. Floc hydrodynamic properties differ significantly from those of the primary particles due to complex structure of the floc and its high sensitivity to hydrodynamic forcing. Highly variable properties of mud flocs (e.g., their size, density and strength) affect the settling velocity and hence the vertical and lateral transport. Flocs in suspension experience various processes such as aggregation, fragmentation, repacking, remineralization, deposition, and eventually subsequent resuspension. Thus, it is necessary to predict the size of the flocs and their settling velocities accurately in order for accurate estimations of the transport of cohesive sediment.
Many types of flocculation models have been developed for quantitative prediction of floc size, floc density and settling velocity in the bottom boundary layer. Size class-based models represent the floc distribution in terms of size classes, and are based on the rate of change of particle number concentration due to particle aggregation and breakup (e.g. McAnally and Mehta, 2002; Maggi et al., 2007; Verney et al., 2011). Other approaches use either a characteristic diameter under the assumption of constant floc yield strength (semi-empirical models, Son and Hsu, 2008; Winterwerp, 1998), or an average floc size of a continuous floc size distribution function (distribution based models, Maerz and Wirtz, 2009). These models can resolve many mechanisms causing flocculation and can be potentially included in sediment transport models. However, only a few of applications were achieved in a 1D vertical (Krishnappan and Marsalek, 2002; Winterwerp, 2002) and 2D models (Krishnappan, 1991). This is probably related to the limitations due to computational costs, increasing model complexity, difficulty of parametrization and the lack of knowledge on processes affecting the floc properties like fractal dimension, size and density (Maerz et al., 2011).
