Comparison of Multiple Machine Learning Methods for Estimating Digital Elevation Points

Numerous engineering applications need topographic surfaces, and topography should be adequately determined. This study evaluates the estimation accuracy of three different machine learning methods in digital elevation model (DEM). Accurate DEM estimation is vital in water resources engineering, management, and planning. In this study, elevation values (Z) of the Mert River Basin of Samsun were estimated by deep learning (DL), gaussian process regression (GPR), and support vector machine regression (SVMR) from multiple machine learning methods. Estimates were tried with X and Y coordinate information as the input scenario. In addition, root mean square error (RMSE), mean absolute error (MAE), nash–sutcliffe efficiency (NSE), and coefficient of determination (R2) were utilized as comparison criteria. When the results were examined, it was determined that the best estimation method was GPR (R2 = 0.997), then SVMR (R2 = 0.975), and the worst modelling was DL (R2 = 0.972). This result shows that an improved model by machine learning methods can be used in the DEM modelling.