Comparison of artificial neural networks and harmonic analysis for sea level forecasting (Urias coastal lagoon, Mazatlan, Mexico)

Abstract

Urías Estuary, a coastal lagoon in northwestern Mexico, is impacted by multiple anthropogenic stressors. Its hydrodynamics (and consequent contaminant dispersion) is mainly controlled by tidal currents. To better manage the coastal lagoon, accurate tidal-level forecasting is needed. Here we compare the predictions of sea level rise simulated by a conventional harmonic analysis, through Fourier spectral analysis, and by nonlinear autoregressive models based on artificial neural networks, both calibrated and validated using field data. Results showed that nonlinear autoregressive networks are useful to simulate the sea level over a time scale of several days (<10 days), in comparison to harmonic analysis, which can be used for longer time scales (>10 days). We concluded that the joint use of both methods may lead to a more robust strategy to forecast the sea level in the coastal lagoon.