Rapid spatio-temporal pumping volume estimation from electricity consumption big data

Land subsidence due to groundwater over-exploitation is a serious problem worldwide. Acquiring total pumping volumes to assess the stresses imposed that lead to subsidence is often difficult to quantify because groundwater extraction is often an unregulated water source. Consequently, pumping volumes represent a critical step for water resource managers to develop a strategic plan for mitigating land subsidence. In this investigation, we develop a time-dependent spatial regression (TSR) model to estimate monthly pumping volume over a ten-year period based on electricity consumption data. The estimated pumped volume is simplified as the spatial function of the electricity consumption and the electric power used by the water pump. Results show that the TSR approach can reduce the errors by 38% over linear regression models. The TSR model is applied to the Choshui alluvial fan in west-central Taiwan, where hundreds of thousands of unregulated pumping wells exist. The results show that groundwater peak extraction across the region occurs from January to May. Monthly pumping volume, and rainfall information are available to provide a better understanding of seasonal patterns and long-term changes of subsidence. Thus, the temporal regional subsidence patterns are found to respond to variations in pumping volume and rainfall.