变量筛选方法对郁闭度遥感估测模型的影响比较.pdfVIP

第43卷 第 12期 林 业 科 学 Vo1．43．No．12 2 0 0 7年 12月 SCIENTIA SILVAE SINICAE Dec．，2 0 0 7 变量筛选方法对郁闭度遥感估测模型的影响比较* 琚存勇 邸雪颖 蔡体久 (东北林业大学林学院 哈尔滨 150040) 摘 要： 比较基于偏最小二乘回归的Bootstrap方法与传统的平均残差平方和(RMS。)准则所选变量建立模型的精 度差别。结果表明：Bootstrap方法是一种更优秀的变量筛选方法，比RMS。方法精度提高约5％；而且它不受变量多 带来的运算困难的限制，更便于实际应用。 关键词： 郁闭度估测模型；遥感；RMS。准则；Bootstrap方法；偏最小二乘回归 中图分类号：$718．45 文献标识码：A 文章编号：lOOt一7488{2007)12—0033—06 Comparing Impact of Two Selecting Variables Methods on Canopy Closure Estimation Ju Cunyong Di Xueying Cai Tijiu (College of Forestry．^bnk∞f Fomstry University Harbin 150040) Abstract： Change patterns of each ecological factor，such as spatial and periodic distribution of wind，sun light and temperature，redistribution of precipitation，are closely relate to canopy closure in stand forest．To properly estimate the distribution of canopy closure is a~undation of recognizing and utilizing ecological service function of forest．Due to the complexity of objective world and uncertainty of remote sensing data，we don’t always find out the variables that significantly impact the estimation of canopy closure but in term of common sense select sufficient variables to analyze．In this paper，Bootstrap approach based on partial least squares regression and RMSg principle based on least squares estimate were used to find out optimal variables to construct the estimation model of canopy closure．The results showed using the Bootstrap approach attributed to improve the estimation precision of regression models．Additionally，despite of more variables，the Bootstrap approach worked on well while the RMSg carried out slowly． Key words： canopy closure estimation models；remote sensing；RMS。principle；Bootstrap appr