Seasonal Fluctuations and Analysis of Ground-Water Level Using Multivariate Regression Model

Abstract

Ground water is a very important underground resource, the principal source of which is thought to be precipitation and underflow from rivers. In general, ground-water level is subject to seasonal fluctuations, and furthermore, the fluctuation pattern varies locally according to the permeability of the aquifer and the distance from the water recharge area. This paper presents a regression analysis method for prediction of the ground-water level.
The autoregressive model constructed from the time-series data can be used to predict future fluctuations. In this paper, we propose that, by extending the theory of the autoregressive model, the multivariate regression model can be related to two kinds of time-series data. This method was applied to specify the relationship between variations of the ground-water level and precipitation in the Kumamoto plain. The optimum multivariate regression models for the nine observation wells, located in the plateau area, at one end of the plain, and in the lowland area, are determined through the value of AIC. The boring data stored in the geotechnical database for the plain are divided into nine groups by the nearest neighborhood method. Assuming that the water level data at each boring site in a group has thesame stochastic structure as the observation-well data in the group, general and regional fluctuation trends in water level under the plain are clearly shown. Furthermore, by using the trend-surface analysis of the water level, seasonal changes of the water-flow pattern are estimated.