Predicting sea level anomaly in the Gulf of Guinea

Abstract

Over the last decade Satellite Altimetry has become a viable alternative over Tidal Gauge measurements for the determination and prediction of Sea Level Anomalies (SLA) in that it is able to cover a broader area of the ocean and also due to its higher level of accuracy. It can also be used to obtain data in portions of the sea where Tidal Gauges cannot be used. This study examined various models like the Backpropagation Neural Network (BPNN) and the Generalised Regression Neural Network (GRNN) in predicting SLA in the Gulf of Guinea and the best model was determined. The variables considered were heat, humidity, wind flux and SLA in the study area. The wind flux, heat and humidity were the input variables and SLA the output variable for the various models that were developed. The best model for the training algorithms of the BPNN thus, the Bayesian Regularisation (BR), Scaled Conjugate Gradient (ScG) and the Levenberg Marquardt (LM) was 3-2-1, 3-4-1 and 3-4-1 respectively for each of the algorithms and the Correlation Coefficient (R) values respectively obtained for the models is 0.75156, 0.55852 and 0.55086 and the Mean Square Errors (MSE) are 0.000732, 0.001722 and 0.001226. Also, the optimal model of the GRNN has a spread of 0.40 which gave an R of 0.481043 and a MSE of 0.000975. The higher the correlation coefficient and the lower the mean square error of a model, the better it is for predictions, it was therefore realised from the study that the BR of the BPNN is the optimal model for predicting Sea Level Anomalies (SLA) in the Gulf of Guinea.