The Research Of Extracting Stable Components In Extended-Range Weather Forecast Based On Historical Data

The extended-range weather forecast for the coming 10-30 days is a very important topic in meteorological research and business, which relates people’s production and living closely. Although great progress has been achieved in recent days, the forecast skill of extended-range weather forecast for the coming 10-30 days needs further improvement, and it can not provide powerful guarantee for China’s disaster prevention and mitigation work about extreme weather events.How to separate the predictable stable components from actual atmosphere has been an important issue in the research of extended-range weather forecast. In this study, a new method for extracting the stable components of 10-30 days is developed by making use of band-pass filter and Empirical Orthogonal Function based on the historical observation data. Then the stable components are devided into climatic stable components and abnormal stable components by the similarity between weather process and historical atmosphere. The variation characteristics of 10-30 days stable components in intraseasonal time scale and their linkage with extreme weather are investigated. Based on ISVHE multi-model ensemble OLR data, an improved forecasting model for the coming 10-30 days in tropical Pacific is established, the predictable stable components are predicted by dynamical model and the unpredictable random components are predicted by similar statistical method. The major results and conclusions are summarized as follows:(1) The definition and calculation of 10-30 days stable components are proposed. The mathematical definition and physical meaning of 10-30 days stable components are proposed, and the linkage between stable components and intraseasonal oscillation is analysied. The calculation method of stable components including filter and EOF can not only determine their continuity in intraseasonal time scale, but also eliminate small space scale in intraseasonal time scale by principle components analysis. This method simplifies the atmospheric motion.(2) Extracting 10-30 days stable components for case study. The change and distribution features of 10-30 days stable components in extreme weather events are studied, and the relationships between climatic stable components and circulation background, abnormal stable components and circulation anomaly are analysied. The climatological background field represents the large scale circulation pattern background and the abnormal stable components represent the relative anomaly in weather system. (3) Comparing the convergency between stable components and actual atmosphere. The coefficient of variance of climatological background field and actual atmosphere in winter 2010 and 2011 are compared, finding that the predictability decreases along the direction of latitude increasing. The skill improvement of air temperature is better than geopotential height at 500 hPa. This method of extracting stable components and climatological background field can be helpful to increase the forecast skill.(4) Analysising the continuity of stable components in intraseasonal time scale. The explained variance of boreal stable components is tested by statistical extrapolation under different phase of ENSO events based on historical observation data. The stable components of 10-30 days can maintain steadily in intraseasonal time scale and capture the overall trend of atmospheric movement. The explained variance transmission attenuation trends in El Nino and La Nina years are weaker than those in neutral years, and the continuity is better. Which shows that the boreal forecasting skill and the intensity of ENSO are related closely.(5) Tropical intraesasonal oscillation simulation diagnostics to POEM2 model. The overall variance distribution of 850 hPa zonal wind and OLR in POEM2 model is consistent with observation, and the variances simulated by model are higher in some region. Comparing with different seasons, in winter POEM2 simulates the eastward movement and 850 hPa zonal wind better than summer, and in summer the model simulates the northward movement and precipitation better than winter. The speed of eastward MJO signal propagation is faster than observation.(6) Based on historical data, an improved forecasting model for the coming 10-30 days is established by using stable components. The forecasting skill of improved model is better in intraseasonal time scale beyond 30 days and the effect is significant in some case. This improved model’s skill is based on numerical model, the forecast skill will appear when numerical model lose its forecast skill. Comparing with multi-model ensemble results, the PCC of improved model is higher than multi-model ensemble at the lead time of 35-60 days and more stable overall, which is benefitial for extended-range weather forecast.(7) Based on historical data, a weather forecast method for the coming 10-30 days is established by improving unpredictable random components. The overall trend of PCC for this method is the same with numerical model, and the method shows some superiority in last period of forecast time. There is a good correlation between the Nino 3.4 index and forecast skill:when east Pacific SST increases (decreases), the forecast skill of numerical model increases (decreases), and the forecast skill of new method decreases (increases). East Pacific SST is one of the most important external forcing signals to influence the effect of dynamic-statistical forecast method in tropic Pacific.