Ground surface temperature (T_s) was manually measured using mercury thermometers in the early stage, and the snow surface temperature was substituted for T_s in areas with thick seasonal snow covers due to measurement difficulty. However, this practice produced T_s data that were inconsistent with more recent automatic measurements that determine T_s in the soil beneath the snow cover using modern instrumentation. In this study, we used observations from northeastern China to explore and correct this T_s inconsistency. Our results showed that the inconsistency varies from a few degrees to tens of degrees depending on the snow conditions and time periods considered. This inconsistency between earlier and modern measurements can lead to the overestimation of soil temperature warming rates in recent decades. We remove the T_s inconsistency using a numerical model that considers the effect of snow insulation on ground surface temperatures. This approach provides a means to improve T_s records that have similarly switched from manual to automatic measurements, which is critical for soil and climate investigations.