The variability of the observed sea level and the meteorological tide in Paranaguá Bay-PR was analyzed with the tide gauge station time series used in the Part 1 and reanalysis data set of the "National Centers for Environmental Prediction" (NCEP) and the "National Center Atmospheric Research" (NCAR), on some grid points over the oceanic area, near the Bay to the same period. The Thompson low-pass filter was adapted for 6 hours intervals to remove the high frequency oscillations present in teh reanalysis data set. Remote influence of the meteorological variables, in the rises and lowing of the coastal sea level, are analyzed, statistically, in the time and the frequency domain according to the Part 1. Tide gauge station time series from Cananéia (SP), used to verify the correlation with Paranaguá data set, confirmed the Mesquita (1997) research to the southeastern coastal region. Correlation between the variability of the meteorological tide in both cities were made due to the point 1 is near Cananéia. Artificial Neural Network (ANN) with the same architecture developed in Part 1 was applied to the reanalysis data. The maxima correlations between the input/output vectors were also used, adjusting the learning rate and momentum for improving the algorithm to reach the best performance. As the Part 1, the network performed very well at 6 and 12 time lag simulations. The results to 18 and 24 time lag simulations were lower than these ones presented to the surface station, than these ones, suggesting also, others ANN architectures to improve the predictions for larger periods. The results suggest the using of reanalysis data where the lack of conventional station is significant.
Artificial Neural Network; sea level variability; meteorological tide (surge); time series forecasting; reanalysis dataset
