AN EVALUATION OF LEAST SQUARE MODELS FOR TIDAL HARMONIC ANALYSIS AND PREDICTION IN APAPA DOCKYARD, LAGOS, NIGERIA

Abstract

The harmonic method of tidal analysis has been considered as more accurate than the non-harmonic methods. This is because the precise knowledge of the astronomic tidal constituents of a river channel enhances a comprehensive analysis of the tidal behavior in relation to the astronomic constituents which are the predominant factors responsible for tidal patterns and variation. Over the years, the Ordinary Least Squares technique (OLS) has been used for tidal harmonic analysis. However, due to issues of performance speed, computational effectiveness and operational limitations base on confidence interval, the use of the Iteratively Re-weighted Least Squares (IRLS) is becoming increasingly prominent. This study presents a comparative analysis of both methods of tidal harmonic analysis using 12months sea level observational data taking at 10minutes interval at Dockyard Tide Gauge (Lagos state). A computational experiment based on some sort of hierarchical data input scenario has been conducted. The 12months data has been spited into four different data types being; (i) short term data without gaps (5months), (ii) short term data with gaps (4% missing data), (iii) multi-months data without gaps (12months) and (iv) multi-months data with gaps (5% missing data). The harmonic constants for each tidal constituent in all four observational scenarios were computed using both methods and the corresponding confidence interval and SNR determined for each method using the UT-tidal analysis hydrographic tool in MATLAB as designed by Codiga (2012). Tidal values at specific times based on the determined constituent values were thereafter predicted. It was discovered that there is no significant difference in tidal prediction between the results obtained from the use of both the OLS and the IRLS method for analyzing data not exceeding five months at 95% confidence interval with the OLS method performing better than the IRLS for short term data period. However, in the multi- months data period, the IRLS method performed better that the OLS method in both scenarios (complete data and data set with omissions). The study concludes that while long term data might be better for river tidal characteristics determination with values of 1.457m and 1.6495m for the Mean Low Water Springs (MLWS) and MHWS respectively, short term data is best for tidal prediction.