Analyzing the characteristics of weather and climate memory in temperature has positive significance for improving climate prediction. Based on the daily,monthly,and annual mean temperature data from 1961 to 2020 in Jinan and Qingdao,the characteristics of temperature memory and probability distribution for these time series are analyzed by using autocorrelation function and normalized frequency distribution. And the fractional derivative relationships between monthly / annual mean temperature anomaly and daily mean temperature anomaly are established by using structure function. The results are as follows. (1)The monthly and annual mean temperature anomalies in Jinan and Qingdao show different degrees of memory characteristics,and the annual mean temperature anomaly has better memory than the monthly mean temperature anomaly. (2)There are fractional derivative relationships between monthly / annual mean temperature anomaly and daily mean temperature anomaly both in Jinan and Qingdao. The corresponding derivative orders on monthly and annual scales are 0. 529 and 0. 665 in Jinan,while 0. 553 and 0. 791 in Qingdao,respectively. On the monthly scale,the derivative orders in Jinan and Qingdao are similar with the value in Qingdao slightly larger,while the derivative order on the annual scale in Qingdao is larger than that in Jinan,namely the memory of monthly and annual mean temperature anomalies in Qingdao is better than that in Jinan. (3)Compared with the daily mean temperature anomaly,the monthly and annual mean temperature anomalies in Jinan and Qingdao have different degrees of long-tail characteristics,which reflect the probability of extreme temperature.