Monitoring electricity consumption (EC) under high spatiotemporal resolution is important for optimizing energy utilization. Although current studies have established various regression models to spatialize EC using Nighttime light (NTL) imagery under the accepted assumption that EC is directly related with NTL, the impact of other factors on the EC-NTL relationship is not fully considered. In this study, we introduced land use types and landscape patterns to optimize EC-NTL model. Specifically, we constructed 12 indicators using NTL and Landsat TM/ETM + imageries to model EC, which considers both human activity and landscape patterns on different land use types. A multi-scale geographically weighted regression (MGWR) model was applied to build the heterogeneous relationship. Experiments were conducted on 292 prefecture-level cities in China. Results showed that the MGWR model gets an adjusted R² as 0.92, which is 13.6% and 8.2% higher than the simple and multiple linear regression. Five indicators obviously correlated with EC were pinpointed, including the total area of urban and industrial-mining land, NTL mean, NTL standard deviation and patch density of rural settlement. It is argued that scale is an effective proxy at urban and industrial-mining land, while NTL distribution is representative at rural settlement for modelling EC.