It is important to understand and quantitatively classify the characteristics of local groundwater flow indicated by the water level fluctuation in wells. In this study a method to evaluate the similarities of water level fluctuation between wells is proposed. Linear regression models are constructed with independent variables such as rainfall and water level of other wells. Well similarity is estimated from model parameters (regression coefficients and model fitness). Regression coefficients are calculated with Genetic Algorithm (GA); with GA identification of parameters is easier even in a complicated model.
The method was applied to twelve wells in the Tono area in central Japan. Although groundwater level fluctuation is primarily affected by rainfall and pumping conditions, different geological conditions should also cause different types of water level. Models using water level in other wells, as well as models using preceding rainfalls and atmospheric pressure, suggest that water level fluctuation data of the wells are classified into groups rAflecting the geological conditions. This is explained by the difference in the property of pressure propagation for rain infiltration among the units. Additionally, comparison of model fitness between the models can be used for estimating the extent of these factors' effect
View full abstract