Gray Models for Real-Time Groundwater-Level Forecasting in Irrigated Paddy-Field Districts

Abstract

This study presents two gray models with which to forecast groundwater levels at two study sites; one is located on an island area, and one is in an inland area. These models used the moving average of daily infiltration depth, historical groundwater level, and water demand from irrigated paddy-field districts as variables, taking into account the characteristics of the sites. For the site on an island area, the sea water surface level was additionally used as one of the variables. The model parameters were estimated by integrating the models with a global search method, the annealing-simplex method. The comparisons between the observed values and the forecasts in calibration and validation show that the forecasts are in close agreement with the observations. The sensitivity analyses of the models revealed that the variable for water demand affects the groundwater levels, depending on irrigation practices, and that the variable for sea water surface level is one of the factors affecting groundwater discharges. In comparing the predictive capabilities of the models with those of the other gray models, it was proved that the models have appropriate variables for forecasting groundwater levels with 1–10 days of lead time. Moreover, the accuracies of the models are higher than those of the base models. Finally, these results demonstrate that the models are slightly superior to other models and can reasonably forecast groundwater levels and the effects of groundwater pumping on groundwater states in real situations.