Abstract: Using the threshold-based methods of meteorological elements on frost-free days, the quality-controlled daily observation data were firstly established, using the observed data from 23 typical meteorological stations in Anhui province from 2003 to 2017. The automatic identification models of frost based on Logistic regression were constructed for each meteorological station in Anhui province, and the performance of the frost identification model was evaluated. The results showed that: (1) the daily frost-free phenomena could be correctly determined for each station in Anhui province by the thresholds of meteorological elements such as temperature, wind speed and precipitation amount. (2)Temperature, humidity and wind speed were selected as the identification elements in the frost identification models for all meteorological stations. The observation time of model elements mostly occurred at the stage from 4：00 to 8：00. (3)The accuracy rate, hit rate, miss rate, empty judgement rate and TS score of Logistic regression model based on independent sample test were 89.0%, 91.6%, 8.4%, 15.8% and 78.2%, respectively. This indicates that the frost recognition model established by Logistic regression has good ability to identify frost in Anhui province. (4) Compared with Bayes discriminant model, it was found that the frost identification model based on Logistic regression had higher accuracy rate and TS score, and lower empty judgement rate. Therefore, the proposed Logistic regression model can be applied to the automatic identification of frost in the future.

Key words: Frost, Logistic regression, Automatic identification