Prediction Of Rainy Season Precipitation In China Based On Analog Prediction Of The Principal Components Of Modei Errors

This work is proposed mainly to improve the prediction ability of rainy season precipitation in china for operational seasonal prediction model. Based on the basic principle of analogue-dynamical prediction and current model, from the point of inverse problem, historical data is utilized to estimate current unknown model errors using known historical analogical information. Considering the problem of prediction the model error and its local characteristics, a new idea is proposed, which transports the problem of directly estimating the model errors by analogue prediction into the problem of analogue prediction of the principal component of model errors. The principal components of model errors are objectively divided into two parts, predictability and unpredictability. The predictability ones are predicted by analogue prediction of evolving analogues of optimal configuration of multiple predictors, while the rests instead by past climate average. Taking into account the important impact of historical evolution of the external forcing on the climate change, the concept of evolving analogues is brought forth and a new scientific and reasonable criterion similarity coefficient is developed through which the temporal and spatial evolution information of a predictor in a single season could be included. When analogue prediction of the principal component of the model errors in the forecast years, firstly the dominant predictor of this principal component of the year should be determined by sorting the prediction ability of all the potential single predictors through cross-validate prediction this principal component; Subsequently, based on this predictor, predictors of sets of relatively independent and have certain predict skill in predicting this principal component are picked out through the correlation analysis and other methods, based on these factors, then an optimal configuration of multiple predictors is set up through optimal multi-predictor configuration; Finally, predicting the time coefficient of principal component in the forecast year using this optimal configuration of multiple predictors.Based on the National Climate Center (NCC) of China operational seasonal prediction model and dynamical extended range forecast model results for the period1983-2009and the US National Weather Service Climate Prediction Center merged analysis of precipitation in the same period, together with the74circulation indices of NCC Climate System Diagnostic Division and40climate indices of NOAA of US during1951-2009, the key technical issues in this method are discussed firstly. Subsequently, based on geographical partition compartmentalize china region as8small regions such as South China, the Yangtze River, North China, northeast china, eastern of Northwest, western of Northwest, Tibet and south-west respectively, and actual operational forecasting experiments are implemented on seasonal precipitation, temperature, and monthly precipitation as well as temperature for all these regions. To compare the prediction ability of the dynamical model, dynamical-statistical and statistical method in predicting seasonal and monthly precipitation and temperature, of cause, similar technique is also carried on predicting the principle components of seasonal and monthly precipitation and temperature, and then, the spatial and temporal distribution characteristics of predictability of precipitation and temperature are discussed. The main results and conclusions are as follows:1. One of the features of this work is the information of evolving analogues of the predictors is utilized and a new scientific criterion similarity coefficient is developed. Practice confirmed that the rationality of the similarity criteria on evolving analogous judgment. While on the problem of selecting predictors, a method to determine the optimal configuration of predictors called dynamic and optimal multi-predictor configuration is proposed. The method overcomes two common problems in nonlinearity prediction as degrees of freedom of prediction factors and the relationship between the prediction factors respectively through determining the dominant predictor and predictors of evolving analogues, correlation analysis and dynamic and optimal multi-predictor configuration scheme. Results in prediction have proved that the scheme dynamic and optimal multi-factor configuration can improve forecast skill.2. Seem from the forecast skill of actual operational forecast summer rainfall in china in2005-2011, generally the four kinds of analogues prediction are more skillful than model systematic correction of error forecast. The average anomaly correlation coefficients(ACC) for model systematic correction of errors prediction is0.02in2005-2010, while the four kinds of analogues prediction are0.11,0.19,0.15,0.10respectively. The most skillful method is analogue prediction of principle component of CMAP based on predictors of analogue prediction of principle component of model errors; the analogue prediction of principle component of CMAP next, then is analogue prediction of principle component of model errors. After joining the2011forecast, the average ACC for model systematic correction of errors forecasting and four kinds of analogue forecast in the nearly seven-year is0.06,0.14,0.19,0.15and0.13.3. On the predictability of rainy season precipitation, in recent years, model systematic correction of errors forecast is skillful in prediction of rainy season precipitation in Yangtze River and North China. Analogue prediction of principal component of the model errors has certain improvements on model systematic correction of errors forecast except in North China, especially in Yangtze River, Northeast, western of northwest, Tibet and southwest. In these regions analogue-dynamical prediction is skillful in prediction rainy season precipitation except in western of northwest, while for analogue prediction of principle component of CMAP based on its predictors, which is skillful in prediction rainy season precipitation in most regions except South china and North china, especially in Yangtze River, eastern of northwest and Tibet, the average ACC in the nearly6years is above0.4; analogue prediction of principle component of CMAP is skillful in prediction precipitation in South China, Northeast and Tibet, and for analogue prediction of principal component of model errors based on its predictors, which prediction skill is located in the South china, Northeast, Eastern of Northwest, Tibet and southwest.4. Opinion from the dominant factor in recent years to forecast the first principle component of model errors in regional rainy season precipitation prediction, the error of model forecast of rainy season rainfall is mainly influenced by the Northern Hemisphere subtropical high in China's eastern region, for North China which is also impacted by the polar vortex, while the western region is mainly affected by the Atlantic subtropical high and sea surface temperature in Nino region. The error of CGCM prediction of summer temperature in china region is mainly affected by the abnormal circulation and surface features in previous winter, and the Tibetan Plateau (25°N-35°N,80°E-100°E), one of the main factors, and since2011, it seems that the impact is weakening while the impact of polar vortex in the northern hemisphere is enhancing.5. On Seasonal predictability, for the precipitation prediction in china region, the forecast ability, in winter and spring was significantly higher than in summer and autumn, but it is better in summer and winter for the dynamical-statistical forecasting and statistical forecasting. For temperature, model forecast in autumn and winter is significantly higher than in spring and summer, maybe the model itself is already skillful in autumn and winter temperature forecast, so the dynamic-statistical forecasting practically has little improvement in seasonal temperature forecast of the two seasons, only in autumn the forecast RMSE has reduced comparing to model itself. But for the relatively less skill of spring and summer of model forecast, dynamic-statistical forecasting makes certain improvement in ACC for these two seasons.6. On prediction of the precipitation in the first and second raining season in South China region in recent years, comparing to model prediction of DERF, the analogue prediction of principle components of model error almost has no improvement in the region precipitation prediction. Possible reasons are summed up in the following:(1) the complexity of the regional precipitation;(2) model itself has no forecast skill in these two seasons;(3) maybe the defects of the program itself. However, on China's regional, comparing to model itself forecast, the analogue prediction of principal component of the model error shows a slight improvement in these two months of precipitation prediction, and in the nearly six years, the average ACC for model prediction and dynamical-statistical prediction in this two months is0.07,0.08and0.02,0.03, respectively. The most obvious improvement exists in north china, northeast, northwest and so on.7. On the predictability of monthly precipitation and temperature,2005-2010, the average ACC of model prediction for temperature and precipitation is0.38and0.13, and very consistent with previous conclusions. As a whole, analogue prediction of principle component of the model error is relatively limited on the improvement of the DERF model on prediction monthly precipitation and temperature, but the improvements for precipitation is greater than that of temperature, and the improvements is obviously in some regions and months such as summer rainfall in South China and spring and summer precipitation in Northeast. Statistical forecasting are not capabilities than the analogue prediction principle components of model error and DERF model in the monthly precipitation and temperature forecasts because of the important effect the initial value of DERF, especially for the weather forecasts in the predictability limit, while external forcing has a certain impact on monthly forecast, but not so crucial as the initial value does.8. Both analogue prediction of the principal component of model errors based on the dynamic and optimal multi-factor configuration and analogue prediction of the principal component of precipitation and temperature based on its predictors, show certain prediction skill in seasonal and monthly precipitation and temperature prediction, and in some areas the precipitation and temperature prediction levels are improved compared to model systematic correction of errors forecasting, which demonstrated the superiority of combination of dynamic and statistical. From the impact of the numbers of analogues and mode of predictability on analogue-dynamical forecasting, both have great impact on prediction skill. Either the suitable number of analogues or modes of predictability can greatly improve the forecast skill of regional rainy season precipitation and temperature. For analogue-dynamical prediction, in different regions, different meteorological elements have their own optimal numbers of analogues and they should be given due to specific time, space and objects.