Multistep Ahead Groundwater Level Time-Series Forecasting Using Gaussian Process Regression and ANFIS

Abstract

Groundwater level is regarded as an environmental indicator to quantify groundwater resources and their exploitation. In general, groundwater systems are characterized by complex and nonlinear features. Gaussian Process Regression (GPR) approach is employed in the present study to investigate its applicability in probabilistic forecasting of monthly groundwater level fluctuations at two shallow unconfined aquifers located in the Kumaradhara river basin near Sullia Taluk, India. A series of monthly groundwater level observations monitored during the period 2000–2013 is utilized for the simulation. Univariate time-series GPR and Adaptive Neuro Fuzzy Inference System (ANFIS) models are simulated and applied for multistep lead time forecasting of groundwater levels. Individual performance of the GPR and ANFIS models are comparatively evaluated using various statistical indices. In overall, simulation results reveal that GPR model provided reasonably accurate predictions than that of ANFIS during both training and testing phases. Thus, an effective GPR model is found to generate more precise probabilistic forecasts of groundwater levels.