大尺度小样本情况下数据预测方法对比研究.docVIP

大尺度小样本情况下数据预测方法对比研究 　　摘要：采用反距离加权法（IDW）、协同克里格法（Cokriging)、回归模型法，利用云南省134个县（市）气象站观测数据对全省≥10°积温区域进行预测.结果表明，由于气象站点稀少，用不同的方法结合不同的辅助信息得到的预测结果不同.反距离加权法强调了空间距离尺度的影响，但未能较好地体现云南省局部积温的分布规律；协同克里格法则优于反距离加权法，但在样本数偏少的情况下预测结果会出现明显的 　　凹凸现象；回归模型法预测结果能很好地体现云南省北低南高、西高东低的气温总体变化规律，又能体现峡谷地带中局部干热河谷的特点，是3种方法中效果最好的预测方法. 　　关键词：积温；地统计学；空间插值;反距离加权法（IDN）；协同克里格法（Cokriging）;回归模型法 　　中图分类号：S758.4 　　文献标识码：A文章编号：1671-3168(2010)05-0007-05 　　 　　Comparison of Data Forecasting Methods in large-scale Small Samples 　　LI Jun1，YANG Rui2，ZHOU Ru-Liang1 　　(1.Key Laboratory of Forest Fire Disaster Warning and Control, Southwest Forestry University, Kunming 650224, China; 2. Department of Environmental Science , Southwest Forestry University, Kunming 650224, China) 　　Abstract: Using inverse distance weighted (IDW), Cokriging (Cokriging), regression modeling, weather observation data of 134 counties (cities) in Yunnan Province are used to predict weather of regions accumulated temperature ≥ 10 °. The results show that the scarcity of meteorological stations, using different methods to combine different auxiliary information, the prediction results are different. inverse distance weighted emphasized the impact of distance, but failed to better reflect thelocal accumulated temperature distribution of Yunnan Province, Cokriging is better than the inverse distance weighted method, but in the case of too few samples the prediction results appear significant bump phenomenon; The results of Regression Model Forecast can not only well reflect the high temperature in southern Yunnan Province and low in the north, as well as high in west to low in east, the overall change in the temperature, but also reflects the characteristics of local dry-hot valley area in the valleys.It is the best prediction method of the 3 methods . 　　Listen Read phonetically 　　Key words: Accumulated temperature; Geostatistics; Spatial interpolation; IDN；Cokriging; 　　regression model method 　　 　　在大尺度的研究项目中，往往存在样本点偏少或是不足的情况，由于项目的需要，对数据