Current velocity forecasting in straits with artificial neural networks, a case study: Strait of Istanbul

Owing to their complex character, modeling flow patterns of narrow straits has always been a challenge, even with the numerical techniques of today. This study was aimed at predicting vertical current profiles of a given point in a narrow strait, the Strait of Istanbul. On account of the speed and simplicity it offers, and of its remarkable success in solving complex problems, the feed forward back propagation (FFBP) artificial neural network (ANN) technique was chosen for this study. The model was built on 7039 hours of concurrent measurements of current profiles, meteorological conditions, and surface elevations. The model predicted 12 outputs of East and North velocity components at different depths in a given location. Various alternative models with different inputs and neuron numbers were evaluated attaining the best model by trial and error. Predictions from proposed ANN model were in accordance with the observations with average root mean square error of 0.16 m/s. The same input parameters were then used to build models that predicted current velocities 1-12 h into the future. Results of these predictions show good overall agreement with observations and that FFBP ANN can be used as a reliable tool for forecasting current profiles in straits.