Abstract:
Mean sea level (MSL) has been used as a vertical datum for geodetic levelling and mapping in most countries all over the
world. This is because the MSL approximates the geoid and serves as a realist reference surface that could be determined
mostly through tide measurements over a period of time. However, sea levels have been rising over the years due to global
warming and its associated climate change which continuous to melt ice sheets around the Polar Regions. This phenomenon
is likely to affect the reliability of MSL, thus it is important to determine the local MSL at regular time periods. This study
assessed the performance of Artificial Neural Network (ANN) and Multivariate Adaptive Regression Spline (MARS)
models in predicting the MSL. Tide gauge records from the Takoradi Harbour of Ghana were used in the study. Monthly
maximum, minimum and mean tidal values were derived from the secondary data and used for both model formulation and
model testing. A comparative analysis of both models showed that the ANN model performed better than the MARS model.
A Root Mean Square Error (RMSE) of 0.0359 m was obtained for the ANN model, whereas 0.0555 m was obtained for the
MARS model. Mean Absolute Percentage Error (MAPE) of 3.1414% was obtained for the ANN model and whereas the
MARS model yielded 5.6349%. A Mean Absolute Error (MAE) for the ANN model was 0.0284 m as against 0.0446 m for
the MARS model. Correlation coefficient values of 0.9720 and 0.8874 were obtained for the ANN model and the MARS
model respectively. An optimum ANN structure was found to be ANN 2-11-1. Based on the outcome of this study, it is
recommended that ANN model should be adopted for forecasting local mean sea level for the study area.
