Error Diagnosis And Correction For Monthly Dynamical Extended Range Forecast Of BCC_AGCM

Monthly extended range forecasting has made great progresses with the unremitting efforts of many meteorological researchers, but there still exist prediction errors; especially, the increasing frequency of extreme climate events under the background of present climate change requires a higher accuracy of the monthly extended range forecast. The precision of monthly extended range forecasting is still far from meeting the demands of the country and society for the reason that the time-scales of month-scale climate phenomena exceed the upper limit of the predictability of daily weather forecast as well as the impacts of initial/boundary conditions on the prediction model. Therefore, the further investigation and development of the efficiency correction strategies and methods are necessary and have great potential value of application.We explore how to efficiently apply the existing error correction skill to the operating models, in order to improve the accuracy of weather prediction and also provide some useful information for better interpretation and application of the model forecast products. We examine the characteristics of winter temperature forecast error from this model and the relationship between this error and external forcing, by using the National Climate Center Atmospheric General Circulation Model (BCC_AGCM) from the second generation monthly Dynamic Extended Range Forecast System (DERF2.0) and some other reanalysis datasets in this paper. Moreover, a series of numerical model simulations based on the re-prediction of summer precipitation are developed to explore the efficient strategy and scheme for using the error correction method. The results are listed as following:1. The model can simulate the trend of Eurasian winter temperature as a whole and present the principal spatial modes of the inter-annual variability of wintertime temperature. Overall, the model has considerate forecast ability for wintertime temperature over the East Asian winter monsoon regions. The spatial distribution and temporal evolution of the forecast error indicate that the errors over land and over high latitude region are larger than those over ocean and over low latitude region, respectively, and the error is correlated to altitude. The principal modes of model forecast error are highly correlated to the sea surface temperature (SST) over some key ocean regions, which indicates the model lacks ability to respond to the external anomaly. These results can provide the basis for improving model forecast ability for wintertime temperature over the East Asian winter monsoon regions with the model response to SST and sea ice anomaly over key ocean regions.2. The model error of summer precipitation is mainly caused by the large scale circulation in the Northern Hemisphere and SST anomalies. We build a regression model based on the high correlation between the key factors and model forecast error to correct model errors. The results show more than 60% of total years have noteworthy improvement on anomaly correlation coefficient (ACC) against raw and systematic corrected results. From 2003 to 2011, the raw average model ACC is only 0.03 and systemic corrected ACC is 0.07, whereas error corrected ACC exceeds 0.15. The mean square skill scores (MSSS) score after error correction is positive over most of the regions, indicating the error correction scheme can somehow reduce the error due to the lack of model response to the external forcing. The previous circulation and climate factors, which are highly correlated to the principal components from different model error modes, can reveal the sensitivity of model forecast ability to those factors in a certain extent. This will provide some useful ideas to further improve summertime precipitation forecast and application of model forecast products.3. To obtain the error correction scheme based on empirical orthogonal function (EOF), we compare the correction effects of using different number of modes on observation and forecast fields in the regression models by cross validation method. The results indicate that it’s not best correction scheme when observation and forecast fields have same number of modes. Meanwhile, the correction efficiency changes mainly with the number of modes of forecast fields, relatively less influenced by the number of modes of observation modes. When introducing the high order modes of forecast field, the correction efficiency decreases rapidly. Considering both ACC and variance, the best error correction scheme based on EOF is to use 30 modes as a cluster, which will overcome the forecast instability from certain mode.4. The test on the error correction scheme based on singular value decomposition (SVD) shows that correction efficiency is the best when using the first six modes in the correction. Thereafter, ACC will decreases rapidly with increasing modes, which indicates the high order modes may provide fake signal that can have bad influence on the correction efficiency.5. The error correction schemes based on EOF and SVD can both remarkably improve model forecast efficiency on summer precipitation. EOF scheme has relatively higher ACC score and is more stable than SVD scheme, and has larger positive region and higher value on MSSS score than SVD scheme as well. In addition, these two schemes both have significant effect on reducing original model forecast error. Overall, EOF scheme is relatively better than SVD scheme. EOF scheme is also more stable than SVD scheme by using the mean value of mode cluster.