Geoid determination for national heighting is one of the major research focuses in geodetic sciences. Many studies in the past and recent years have suggested various mathematical techniques for local geometric geoid modelling. This study considered an empirical evaluation of soft computing techniques such as Backpropagation Artificial Neural Network (BPANN), Multivariate Adaptive Regression Spline (MARS), Generalized Regression Neural Network (GRNN), Adaptive Neuro-Fuzzy Inference System (ANFIS), and conventional methods such as Polynomial Regression Model (PRM), and Multiple Linear Regression (MLR). The motive is to apply and assess for the first time in our study area the working efficiency of the aforementioned techniques. Each model technique was assessed based on performance criteria indices such as mean error (ME), mean square error (MSE), minimum and maximum error value (rmin and rmax), correlation coefficient (R), coefficient of determination (R2) and standard deviation (SD). The statistical analysis of the results revealed that ANFIS, GRNN, MARS, BPANN, MLR and PRM, successfully estimate the geoid heights with a good precision for the study area. However, ANFIS outperforms BPANN, MARS, MLR, PRM, and GRNN in estimating a local geoid height. In terms of ME and SD, ANFIS achieved 0.0445 m and 0.0013m as compared to BPANN, MARS, MLR, PRM, and GRNN which achieved 0.1462 m, 0.0059 m, 0.1423 m, 0.0148 m, 0.3117 m, 0.0102 m, 0.1798 m, 0.0208 m, 0.0878 m and 0.0023, respectively. The main conclusion drawn from this study is that, the method of using soft computing is promising and can be adopted to solve some of the major problems related to height issues in Ghana.
Artificial Neural Networks, Geodetic Reference System, Geoid Modelling, Global Navigation Satellite System, Polynomial Regression Model