The ability to accurately predict the energy consumption of electric vehicles (EVs) is important for alleviating the range anxiety of drivers and is a critical foundation for the spatial planning, operation and management of charging infrastructures. Based on the GPS observations of 68 EVs in Aichi Prefecture, Japan, an energy consumption model is proposed and calibrated through ordinary least squares regression and multilevel mixed effects linear regression. Specifically, this study focuses on how the ambient temperature affects electricity consumption. Moreover, the interactive effects of ambient temperature and vehicle auxiliary loads are explored. According to the results, the ambient temperature affects the energy efficiency significantly by directly influencing the output energy losses and the interactive effects associated with vehicle auxiliary loads. Ignoring the interactive effects between ambient temperature and vehicle auxiliary loads will exaggerate the energy consumption of the heater during warm conditions and underestimate the energy consumption of the air conditioner during cold conditions. The most economic energy efficiency was achieved in the range of 21.8-25.2 degrees C. The potential energy savings during proper usage of vehicle auxiliary loads is discussed later based on estimated parameters. As a result, a mean of 9.66% electricity will be saved per kilometre by eradicating unreasonable EV auxiliary loads.