Abstract: It is very difficult to maintain a stable meteorological observation environment for long time in densely populated areas. Long sequence air temperature data is mixed with a large number of errors caused by the changes of environmental conditions, which affects the application value of observation data. Therefore, it is necessary to find methods to reduce such errors. Based on the actual situation of the observation station, the method of extracting mixed environmental error information from the daily minimum temperature data is developed, and its applicability is evaluated. The results show that the observed environmental changes identified by the method in this paper are fully consistent with the actual scene, most of the affected data segments can be effectively identified on the "seasonal average" scale, and the amount of error caused by environmental changes can also be quantitatively evaluated.