Seasonal Partition Problem of MOS Forecast for Extreme Temperature in North China

Aiming at seasonal partition problem of MOS (Model Output Statistics) forecast for extreme temperature, experiments are carried out in North China with cluster analysis method. A new seasonal partition way of MOS prediction equations for temperature is proposed on the basis of clustering results. The period from 11 February to 20 March and from 5 November to 4 December is defined as early spring and late autumn class; the period from 1 May to 30 September is defined as summer class; the period from 21 March to 30 April and from 1 October to 4 November is defined as late spring and early autumn class; the period from 5 December to 10 February is defined as winter class. The proposed seasonal partition is significantly different from traditional seasonal partition especially on periods of time from March to May and from September to November. The two kinds of seasonal partition definition are compared and analyzed. MOS prediction equations with new seasonal partition are founded by T213 model data, maximum and minimum temperature data of 154 stations in North China from 2003 to 2008, and verification of extreme temperature forecast in 2009 is conducted.Mean absolution error of maximum temperature forecast from September to November and minimum temperature forecast from March to May and from September to November made by new seasonal partition is less than that by traditional one. Using the new seasonal partition, there are more stations with the absolute error of MOS forecast less than 2℃ for maximum and minimum temperature from March to May and from September to November. Average error of extreme temperature forecast based on two kinds of seasonal partition ways doesn't have great differences, and their absolute error also isn't large at the same time. It shows that the system error is not significant. However, compared with traditional MOS forecast, mean absolute error of maximum temperature forecast made by new seasonal partition from March to May is larger. The cause maybe relates with great changes of temperature in spring of 2009 or cluster analysis program. More study and improvement will be carried out in order to solve the problem. The test result indicates that the overall effect of MOS forecast for maximum and minimum temperature made by new seasonal partition way is better than the traditional one, and shows that the new seasonal partition way is more suitable for MOS extreme temperature forecast.