Application of Artificial Neural Network (ANN) to improve forecasting of sea level

Abstract

To improve on forecasting of tidal water level beyond harmonic analysis requires the incorporation of meteorological variables in the analysis. This suggests the use of Artificial Neural Networks (ANN) as an optimum tool to train and translate the combined influence of meteorological and astronomical forcing to predict sea level variations and reduce the margin of error of close to 50% (from 26% to 12%). To accomplish this, the ANN was trained by using hourly time series of atmospheric pressure, wind, and harmonically derived tides for 1982 as input data and hourly time series of measured tides as output data. The meteorological data were obtained from São Sebastião (SP) and Ponta da Armação (RJ), and the sea level data from Cananéia (SP) and Ilha Fiscal (RJ). Data gaps in the time series were interpolated based on FFT analysis. To forecast water levels, the 1983 meteorological time series was used as input data, and compared the resulting water level outputs to the water level measurements for the same period. The ANN served as a very good forecasting tool for sea level variability. In the case of Cananéia, with several meteorological data gaps, the comparison was less successful as compared to the Ilha Fiscal results, besides this, there is a local influence of the estuary flows, a variable not considered that could answers for the remaining 12% of the correlation. The coefficient of correlation between predicted and measured water level time series at Cananéia was 0.88 and at Ilha Fiscal 0.98. This kind of improvement can be used for port terminals and marinas, for handling incoming and outgoing ships and boats more safely through the navigation channels in the estuaries. It is applicable and useful information for decision makers in management activities in the coastal area.