The Efficiency of Neural Network Models in Predicting Sea Level During Storm Events at Quinhon, Vietnam

Abstract

Recent researches have shown that neural networks are more effective in modelling and prediction. The coastal communities at large would greatly benefit from such predict sea level as well, especially during tropical storm events. There are few application neural network models for predicting sea level oscillations related to storm conditions in the Vietnam coastline. This paper presents two kinds of neural network model – Back-propagation Artificial Neural Networks (ANNs), and Adaptive Neuro-Fuzzy Inference System (ANFIS), developed to estimate temporary sea level variation caused by meteorologically driven forces related to storm wind and pressure in Quinhon basin in the Central region of Vietnam, a place which is frequently affected by tropical storms. In the study, the two neural network models are developed with 6-hourly series of atmospheric pressure and wind components from the CMA Tropical Cyclone Best Track Dataset of serve storms passing within 300 kilometers of Quinhon city (13.76°N, 109.21°N), as the input parameters; and observed sea level at Quinhon station during these storms are as the target outputs. The root mean square error is used to assess how the models perform. ANN model does well in with two input parameters-wind speed and air pressure. ANFIS models was optimized. It stated that ANFIS model with three input parameters –wind speed, pressure, and direction do the best in predicting surge level during storm. After comparing exiting and confidence interval of ANN and ANFIS, it has been found that the result of ANFIS model is better because it was more accurate and does have lesser assurance.