Modeling Short-Term Groundwater-Level Fluctuations Using Multivariate Adaptive Regression Spline

The study investigates accuracy of two machine learning methods, neuro-fuzzy system with grid partition (ANFIS-GP) and multivariate adaptive regression spline (MARS) in prediction of 1-day- to 6-day-ahead groundwater levels (GWLs) using data from two wells, USA. The outcomes indicate that the ANFIS-GP provides inferior results compared to regression-based simple MARS method. The MARS method which is much simpler than the ANFIS-GP is recommended for short-term GWL prediction.