A novel classification regression method for gridded electric power consumption estimation in China

Spatially explicit information on electric power consumption (EPC) is crucial for effective electricity allocation and utilization. Many studies have estimated fine-scale spatial EPC based on remotely sensed nighttime light (NTL). However, the spatial non-stationary relationship between EPC and NTL at prefectural level tends to be overlooked in existing literature. In this study, a classification regression method to estimate the gridded EPC in China based on imaging NTL via a Visible Infrared Imaging Radiometer Suite (VIIRS) was described. In addition, owing to some inherent omissions in the VIIRS NTL data, the study has employed the cubic Hermite interpolation to produce a more appropriate NTL dataset for estimation. The proposed method was compared with ordinary least squares (OLS) and geographically weighted regression (GWR) approaches. The results showed that our proposed method outperformed OLS and GWR in relative error (RE) and mean absolute percentage error (MAPE). The desirable results benefited mainly from a reasonable classification scheme that fully considered the spatial non-stationary relationship between EPC and NTL. Thus, the analysis suggested that the proposed classification regression method would enhance the accuracy of the gridded EPC estimation and provide a valuable reference predictive model for electricity consumption.