CAJ | 학술논문

利用1892—2003年共112 a 重庆主城区年降水量资料，采用均生函数、经验模态分解和最优子集回归等方法组合建立了两种降水统计预测模型，在两种模型中加入集合预报重新建立两种模型，利用建立的4种统计预测模型对2004—2013年重庆主城区年降水量进行预测及验证。结果表明：集合预报的加入对重庆城区降水的建模和预测效果均有明显改善。加入集合预报建立的降水预测模型优于未加入的模型，采用同一种方法所用序列长度不同的多种预测结果的集合平均降低了预测的随机性，改善了单一均生函数和最优子集回归的预测效果。在 EMD 分解和预测中加入集合预报，使集合平均方法能对经验模态分解的效果进行一定修正；且对未来每个分量的周期预测进行了集合平均，降低了每个分量预测的随机性，在一定程度上对最后各分量叠加后的预测结果进行了修订。从预测尺度来看，预测5 a 尺度的效果较理想。虽然4种模型在建模期对年降水极值有较好的模拟效果，但验证期对极端降水异常年的预测效果较差，未来极端降水年的预测仍具有较大的不确定性。在此基础上利用1892—2013年共122 a 资料，采用加入集合预报的两种统计预测模型预测了未来10 a 即2014—2023年重庆主城区年降水的变化，其中未来前5 a 的预测结果参考性更强。
이용1892—2003년공112 a 중경주성구년강수량자료，채용균생함수、경험모태분해화최우자집회귀등방법조합건립료량충강수통계예측모형，재량충모형중가입집합예보중신건립량충모형，이용건립적4충통계예측모형대2004—2013년중경주성구년강수량진행예측급험증。결과표명：집합예보적가입대중경성구강수적건모화예측효과균유명현개선。가입집합예보건립적강수예측모형우우미가입적모형，채용동일충방법소용서렬장도불동적다충예측결과적집합평균강저료예측적수궤성，개선료단일균생함수화최우자집회귀적예측효과。재 EMD 분해화예측중가입집합예보，사집합평균방법능대경험모태분해적효과진행일정수정；차대미래매개분량적주기예측진행료집합평균，강저료매개분량예측적수궤성，재일정정도상대최후각분량첩가후적예측결과진행료수정。종예측척도래간，예측5 a 척도적효과교이상。수연4충모형재건모기대년강수겁치유교호적모의효과，단험증기대겁단강수이상년적예측효과교차，미래겁단강수년적예측잉구유교대적불학정성。재차기출상이용1892—2013년공122 a 자료，채용가입집합예보적량충통계예측모형예측료미래10 a 즉2014—2023년중경주성구년강수적변화，기중미래전5 a 적예측결과삼고성경강。
Using a long sequence annual precipitation data from 1892 to 2003 in Chongqing,two precipitation fore-cast statistical models were established by methods of a MGF (mean generating function),an EMD (empirical mode decomposition)and an OSR (optimum subsets regression).On the basis of these two models,ensemble pre-diction process was added and then two new models were established.Annual precipitation in main urban area of Chongqing from 2004 to 2013 was estimated using the four models and the results were verified against observa-tions.The results indicate that adding ensemble prediction improves prediction and verification of precipitation,and the model with ensemble prediction is better than other models.Ensemble average of various forecasting results which comes from different time series using the same method can reduce the randomness in forecasting,and im-prove the forecasting results of unitary MGF and OSR.Adding ensemble prediction process to the EMD decompo-sition and prediction makes EMD results adjusted by the ensemble average method.Ensemble average of cycle pre-dictions of each IMF reduces the randomness for prediction of each IMF.Final predict results of the superimposed components are revised to some extent.For forecasting scale,forecast effect is better for a 5-year scale.Although the four models in the modeling process have a better simulation results for annual precipitation extreme,the pre-diction effect during predictive validation period for the years with extreme precipitation anomalies is worse.There is still great uncertainty for future extreme precipitation prediction.On this basis,the two models adding ensemble prediction based data from 1892 to 2013 are used to forecast annual precipitation in future 10 years from 2014 to 2023 in main urban area of Chongqing.Among all results,prediction accuracy from 2014 to 2018 is even better, which could provide references.