基于气候重建资料及均生函数-最优子集回归模型的降水预测

Precipitation Short-term Forecasting Based on Climate Reconstruction Data and Mean Generating Function-optimal Subset Regression(MGF-OSR) Prediction Model

利用由科尔沁沙地榆树区域综合年表而重建的181a降水序列,采用均生函数-最优子集回归(MGF-OSR)预测模型对科尔沁沙地降水进行了5a情景的预测,克服了以往序列短的不足。模型建模期相对误差绝对值在3%、5%、10%以内的年数分别占建模年数的36.46%、64.64%、93.37%,建模期模拟精度较高。2007～2010年实测检验期,模拟与实测值保持了方向一致性,各年份相对误差绝对值分别为11.40%、6.73%、22.06%、5.49%,大部分年份预测较理想。经预测,2011～2015年期间,2011年、2012年、2014年和2015年为降水正常年份,2013年为降水偏干年份。与实际情况对比,2011年与预测一致,2012年稍有差别。总体上,2011～2015年降水变化较为平缓。

The future 5-year precipitation forecasting was made for Horchin sandy land by using Mean Generating Function-Optimal Subset Regression （MGF-OSR） based on the reconstructed series of 181years. The years that the absolute values of relative errors less than of 3%, 5%, and 10% respectively account for 36.46%, 64.64%, 93.37% of the modeling years, and the simulation accuracy is higher in modeling pe- riod. The simulated values are consistent with observed data in the experiment period from 2007 to 2010, during which the absolute values of the relative errors are 11.40%, 6.73%, 22.06% and 5.49%, respectively. By forecasting, 2011, 2012, 2014 and 2015 are normal years for pre- cipitation, and 2013 is a dry year for precipitation.