摘要
利用新疆33站1961—2011年冬季气温和前期108项月环流特征量指数,用滑动相关—逐步回归—集合分析方法建立预测模型,在建立模型时考虑各站冬季气温及前期月环流特征量的线性变化趋势影响,设计了原始序列和去线性趋势两种建模方案,对比两种方案下新疆冬季气温预测效果。结果表明:两种建模方案均具有一定的预测能力;相比之下,去线性趋势方案的预测效果好于原始序列方案;用独立试报相关系数为指标,得到集合方案预测结果,集合方案预测效果明显优于原始序列方案,与去线性趋势方案相比略有改进。对比10月、11月起报的集合方案预测结果,得到的滚动综合结果能进一步改进新疆冬季气温预测的大部分指标。分离预测量和前期环流特征量因子线性趋势和集合分析等方法,都能有效地改进新疆冬季气温预测效果。
Two schemes of original series and filter linear trend were designed using the data of temperature in winter at 33 climatic stations in Xinjiang as the predictands and 108 previous characteristic indexes of circulation as the predictors and considering their linear trends. The running correlation, stepwise regression and ensemble analysis were used to develop the multiple regression prediction models with two different schemes at the 33 stations. The predicted values of temperature of two schemes were compared. The resuhs showed that certain predictive capability. In contrast, the prediction of filter linear trend was better Taking the correlation coefficients of independent test as the indicators, the ensemble the both two schemes had a than that of original series. analysis prediction was obtained. Its effect was superior to that of the scheme of original series. Contrast to the prediction of ensemble analysis from October and November, the running prediction results of temperature in Xinjiang in winter could be improved. The prediction effect of temperature in Xinjiang in winter was improved.
出处
《干旱区研究》
CSCD
北大核心
2014年第5期882-890,共9页
Arid Zone Research
基金
国家自然科学基金(41065006)
国家重点基础研究发展计划973项目(2010CB951001)联合资助
关键词
增暖趋势
线性趋势
滑动相关
预测效果
冬季气温
新疆
warming trend
linear trend
running correlation
prediction
temperature in winter
Xinjiang
