Hourly Interpretation And Application Of 24-hour Temperature Forecast Product In Single Station

Fine hourly weather forecasts for single stations are more and more important for human life and activities.Most previous linear-regression-based MOS(Model Output Statistics)models can hardly meet the requirement of meteorological service in forecast accuracy due to insufficiently high temporal resolution.Temperature,as one of the most important and basic weather elements,has great value in studying the method for how to apply and interpret fine weather forecast products of single station.By means of hour-to-hour NWP(Numeral Weather Prediction)products from the GRAPES_MESO model developed by China Meteorological Administration,hourly observations,and hourly analysis data from Forcastio for Yunnan and Tianjin of China from June to December 2016,24 h predictions are made hour by hour using a hybrid model newly developed in this thesis based on a combination of linear regression,autocorrelation(AR),diurnal cycle and advection.After 24 hours forecast,the root mean square error(RMSE)of the NWP product interpretation and application in Tianjin could be reduced from 3.2 for the traditional MOS to 1.3 degree Celsius by the new model,and the correlation coefficient increased from 0.57 to 0.94.The AR item contributes most to the prediction accuracy,the diurnal cycle item ranks second and advection term is the least.The new model also shows great prediction skill for the extreme temperature in next 24 hours with the best root mean square error at 1.2 degree Celsius and the estimation of time at which extreme temperature appear has an error within 2 hours.In order to solve the data missing problem,this study merges the grid analysis data and instrumental observation data which achieves better time series harmony at a correlation coefficient of 0.92.Then,the merges data are applied as real observations to make 24-hour prediction by the new model,with the RMSE reduction from 3.0 degree Celsius from the MOS to 1.4 degree Celsius by the new model,and the correlation coefficient increase from 0.66 to 0.93.All of these indicate that the new model is feasible and suitable for interpretation of application of fine single-station temperature forecast product.