Suspended sediment dynamics in a tributary of the Saint John River, New Brunswick

Abstract

Historical hydrometeorological and suspended sediment concentration (SSC) data from the Kennebecasis River, a tributary of the Saint John River in New Brunswick, Canada, were investigated to help understand what drives high sediment transport in that system. Analysis of correlation coefficients between SSC and potential drivers at various time steps suggested that multiple regressions would not be optimal for this purpose, and that lagged flow (Q) and precipitation should be taken into account in any model. A frequency analysis involving annual maxima of SSC, Q, and precipitation events revealed there is no systematic unique driver of extreme annual SSC or high annual loads. Finally, artificial neural network (ANN) models were developed to verify whether the variables examined previously would yield better results in a nonlinear context. Network inputs were mean temperature, Q, Q(t–1), Q(t–2), and day-of-year. Using daily loads directly as a target in the network yielded satisfactory results, with 88% of the variance explained by the model and a mean absolute deviation between estimated and real annual loads of 16%. The ANN model systematically outperformed multiple linear regressions.