Temporal convolutional networks for subsidence prediction in snowy regions

Satoshi Tajima 

This paper introduces a model based on a temporal convolutional network (TCN) for predicting future land subsidence caused by groundwater pumping for snow melting. Developed using historical snowfall and cumulative subsidence data from Joetsu City, Japan, the model demonstrates satisfactory performance in predicting observed land subsidence. The results suggest that TCNs are effective for real-time predictions of land subsidence associated with snow melting, thanks to their efficient computational capabilities, broad applicability to practical problems, and minimal data requirements. The proposed approach facilitates responsive and effective land subsidence prevention through proactive pumping management in regions with heavy snowfall.