Prediction of land subsidence in a heavy-snowfall area is one of the m
ost important problems in the determination of optimum use of ground w
ater for melting snow. In this study, land subsidence is predicted by
using a regression-equation model that includes past land subsidence,
ground-water level, and snowfall. The components of the long-period va
riation and the short-period variation in land subsidence are predicte
d by the parameters using the least-square method and Kalman filtering
, respectively. The memory length of the regression-equation model is
justified by comparing the value of Akaike's information criterion. Th
e proposed method has accurately predicted land subsidence one and two
months ahead in snow country in Japan. The input data for this predic
tion model (regression equations) are past land subsidence, ground-wat
er level, and snowfall. This method to predict land subsidence is appl
icable to any snow country in the world.