Closing the energy budget at flux measurement sites is problematic, even when the fetch extends over flat, homogeneous surfaces with low vegetation cover. We used the residual energy balance and ordinary least square (OLS) linear regression methods to quantify spatial variability in soil heat flux contributing to energy balance closure (EBC), by deploying a mobile energy system within the footprints of three Eddy-covariance towers located in the steppe of Inner Mongolia, China. The EBC at the study sites had a daily average residual of 8–19Wm_2 with OLS slopes of 0.83–0.96. The EBC was better achieved at the wet site than at the dry site. The spatial variability in soil heat flux was 48Wm_2 (13% of Rn) during the day and 15Wm_2 (34%) at night, with an average of 29Wm_2 (24%) across the three sites. A 9% OLS slope difference due to this variability was recorded from our eight plot measurements. A large amount of missing energy (110Wm_2 at peak) could occur with decreasing OLS slope of 23% across the three grassland sites when soil heat flux is not taken into account. In particular, heat storage in the top soil layer not only influenced the magnitude of EBC, but also adjusted soil heat flux to match the ‘truth schedule’. Heat storage in the top soil layer comprised half of the soil heat flux when the heat flux plate was deployed at a depth of 30 mm. If this part of heat storage was neglected, the residual of EBC would increase as large as 60Wm_2 with OLS slope decreasing 9%. Comparing them with the multiple-location soil heat flux measurements, the single-location measurements fromnear the Eddy-covariance towers obtained a slightly better EBC with the OLS slope increasing by 4%.