Next 2-5 Year Average Summer Rainfall Pattern Hindcast Assessment And Prediction Over East China

Near-term climate prediction is the international issue,especially for next 2-5 year average prediction.In China,it is focused on next 2-5 year summer precipitation pattern prediction over the East in particular.At present,the capacity in precipitation hindcast of initialized near-term climate prediction system is still limited.Correspondingly,it is urgent in reasonable verification of models' hindcasts and improvement of the prediction performance by post-processing methods.Therefore,the physical fundmentals for application of multi-mode ensemble and dynamic-statistic similar error correction in improvement of next2-5 year summer average precipitation pattern prediction over East China are investigated.Considering that there are lots of deficiencies in the point-to-point diagnosis adopted usually to evaluate the ability of climate model's hindcasts,calculate the weights of different models and measure the analogue of factors,such as the low skill score for predictions consistent with the observations in spatial structure due to a little diviation in locations and lack of phycial meaning due to separate the association between grids,which could lead to be inable to meet the physical fundamentals for application of the post-processing methods in comprehensive,the new object-based spatial verification method of MODE is introduced to carry out the research in hindcast performance evalution,multi-mode ensemble prediction and dynamic-statistic similar error correction of precipitation pattern,so as to improve the performance of precipitation pattern hindcasts between 1979 and 1999 optimally.The main conclusions are as follows:(1)After reconstruction of object-based diagnosis scheme for precipitation pattern,the hindcast ability in rainfall pattern of 8 annual real-time near-term climate prediction systems is evaluated reasonably by MODE.According to the spatial characteristic difference of four typical kinds of summer rainfall over East China,the object-based scheme of rainfall pattern analogue is designed initially.The optimal scheme that the accuracy rate of real rainfall pattern classification reaches 95% is applied in analogue diagnosis of rainfall pattern between8 models' hindcasts and observation.The average MODE analogue from 1960 to 1999 is only0.17-0.35 and low in general,which is agreement with the assessment results by ACC point-to-point verification.But the assessment results by MODE show that the performance of MPI model is the best,which is in disagreement with the results by ACC.Analysis on the course of diagnosis show that the assessment results by ACC is coincided with the real because of the deviation in location of large value,whereas the assessment results by MODE which overcome the problem from excessive dependence on location is more reasonable and abstract the model's potential in rainfall pattern hindcast,since it calculates the analogue of rainfall pattern between hindcasts and observation based on the objects of the positive anomaly region according to different spatial characteristics.(2)According to the MODE similarities of rainfall pattern between 8 models and observation,after optimal design of the weight reflecting the uncertainty in spatial structure prediction in rainfall pattern,the better improvement in hindcast of rainfall pattern is achieved by un-even weigthed multi-mode ensemble.The maximum average analogue of rainfall pattern between single model and observation is 0.39.The equal multi-mode ensemble by overlay of the results within positive anomaly region increases the analogue to 0.45,and the improvement is significant in early 1990 s.The uneven multi-mode ensemble weighted by the ranks based on similarities between hindcasts and observations in prophase improves the negative results of the equal ensemble in late 1990 s and its average analogue reaches 0.47.The optimal multi-model ensemble weighted by rank improves the analogue to 0.51,which is more stable and confidant around decadal change from late 1980 s to early 1990 s.Finally,probabilistic ensemble of anomaly percentage larger than 5% is done.The optimal probabilistic ensemble weighted by rank increases the analogue to 0.55,especially better in late 1980 s before decadal change than multi-mode ensemble.After the weight according to the object-based diagnosis instead of the point-to-point diagnosis,the ACC between optimal ensemble results and observation changes from negative to positive,but the further improvement should be done.(3)In the condition of limited model resource,the feasible predcition strategy turns to the dynamic and statistic correction of precipitation prediction error according to the analogue in initial boundary condition of snow content(SST)over the key region,so as to improve decadal change information of rainfall pattern.Since hindcasts from GEOS-5 model and observations also show that the decadal component accounts for high proportion of next 2-5year average rainfall,the error of precipitation hindcast from model's incorrect respond to decadal change influence factors should be diagnosed.Considering that the anomaly of spring snow content over Eurasian continent(summer SST over the North Pacific)which is the influence factor of precipitation decadal change in intial time is related with precipitation anomaly and its hindcast error,the scheme of precipitation similar error correction based on snow content(SST)over the key region is designed.(4)After reconstruction of object-based analogue diagnosis scheme in MODE for snow(SST)pattern,the object region of spring snow content over Eurasian continent(summer SST over the North Pacific)with anomaly percentage more than 10%(1%)and minimum area of 6(6)grids is abstracted around initial time.According to the spatial characteristics of main EOF modes,the analogue of snow content(SST)anomaly pattern over the object region is diagnosed by the area ratio and center distance with the weight of 0.3:0.7(0.5:0.5).This objected-based diagnosed method distinguishes the decadal background of snow content(SST).The historical years similar with the present in anomaly pattern of snow content(SST)are founded to do the similar error correction in precipitation.The hindcast errors in historical similar years are used to enhance the present hindcast confidance of the decadal change in rainfall pattern.Results show the average and maximum ACC between correction results and observations increase from-0.09 to 0.21(0.17)and from 0.47 to 0.70(0.76)respectively after similar error correction.These corrected results become more stable and confidant around decadal change from late 1980 s to early 1990 s,which is more skillful than corrected results based on point-to-point analogue measurement.The optimal improvement is obtained by the average errors from 6(4)similar years.Its average and maximum ACC reaches 0.21(0.18)and 0.77(0.79).