On the statistical forecasting of groundwater levels in unconfined aquifer systems

The efficacy of multiple linear regression (MLR) as a modeling tool was investigated for forecasting groundwater levels in unconfined aquifer systems. The monthly groundwater level data of 17 sites and monthly data of rainfall, river stage and temperature and seasonal dummy variables were considered as input variables for MLR modeling. Three different approaches based upon plausible combinations of these input variables were considered to develop MLR models for individual sites. The regression coefficients of each MLR model following all the approaches were determined and the performance of these regression models was assessed using multiple correlation coefficient (R), multiple R ², adjusted R ², F-statistic, p-level, and standard error of estimate goodness-of-fit statistics. The best MLR models obtained for individual sites were then calibrated and validated to forecast groundwater levels over the study area. The analysis of the modeling results indicated that the MLR models developed in this study are able to predict groundwater levels with a reasonable accuracy at almost all the sites under study. It is concluded that the MLR technique can serve as a cost-effective and easy-to-use groundwater modeling tool for hydrogeologists, especially under inadequate field data condition.