Abstract:
It is an important way to improve the accuracy of short-term climate prediction by developing numerical prediction models and enhancing the interpretation and application ability of model products. Based on three climate model products supplied by National Climate Center, studies are conducted in downscaling interpretation and integration techniques in predicting monthly precipitation from March to September, as well as summer precipitation of Yunnan. By utilizing three indexes such as anomaly sign consistency rate (PC), trend anomaly synthesis (PS) and anomaly correlation coefficient (ACC), prediction performances of different models and methods for precipitation in Yunnan are compared and analyzed. The result shows that for the direct outputs of models, the monthly average scores of PC and PS of NCEP model are the highest, followed by EC and NCC models. As for the ACC scores, EC model is the highest followed by NCEP and NCC models. To summer precipitation, the scores of PC and PS of EC model are the highest and so is the score of the ACC of NCC model. The PC scores of the downscaling interpretation results of models are mostly lower than the corresponding ones of direct output of the original models, while the scores of PS and ACC are higher than the ones of original models. This conclusion shows that the prediction ability of downscaling interpretation method for monthly precipitation anomaly signs in Yunnan raining season is not as good as that of the direct output of the original models, but the prediction of spatial distribution and anomalous level of precipitation is closer to the actual situation. The prediction performances of various integration schemes are different. The PC score of model average integration is the highest, which is related to the higher PC score of precipitation from direct outputs of models than that from the downscaling interpretation of models. The PS and ACC scores of the optimized integration are the highest, and monthly scores are stable, which indicates that optimized integration can take advantage of both the direct outputs of models and downscaling interpretation results of models, which will improve the prediction performance.