Abstract: Twenty open-pollinated families of Schima superba progenies in Dongjiang Forest Farm and Meinan Forest Farm in Guangdong province were selected for testing growth traits. The height (H), diameter-at-breast-height (D)/collar diameter (C) and individual stock volume (V) were analyzed at ages of 3, 6, 9 and 11 years. Highly significant differences in 9 growth traits were observed among families and between two testing sites, and significant interactions were detected between families and planting sites. To explicitly reveal the linear relationships among growth traits, the V at age of 11 years was taken as dependent variable for the construction of multiple linear regression equations on the basis of taking height and D/C at ages of 3 to 9 years as independent variables. Sixty-three regression equations were established by using backward stepwise elimination method for total population and 20 individual families at single-site and across-site testing. All equations were highly significant on the whole, but each equation displayed obvious differentiation in intercept and independent variable component, amount, regression coefficient and determining coefficient, suggesting that it was appropriate to separately construct a multiple linear regression equation of S. superb for each family at single site. In terms of predicting V11 of S. superb, common precision could be achieved by using single-site or multiple-site regression model from population analysis, whereas higher precision was achievable using single-site regression model of each individual family.

Key words: Schima superba, growth trait, multiple linear regression, backward stepwise elimination

摘要： 选用广东省东江林场和梅南林场20个木荷Schima superba优树自由授粉家系子代，开展生长性状测定试验，分析3、6、9和11年生树高、胸径/地径和单株材积。结果表明，木荷9个生长性状在家系水平上和不同地点间均存在极显著差异，且家系与地点间具有显著的交互作用。以11年生单株材积为因变量，3～9年树高与地径/胸径为自变量，运用反向逐步剔除法，构建群体水平与家系水平的单地点和多地点多元线性回归方程63个，全部方程整体上均达到极显著检验水平，但各方程的截距、自变量数量、组成、系数和决定系数均存在明显差异，说明木荷生长性状多元线性回归方程宜分家系和分地点单独构建。预测木荷家系11 年生单株材积时，群体水平单地点或多地点回归模型可实现普通准确性，单个家系单地点回归模型可实现更高准确性。

关键词: 木荷, 生长性状, 多元线性回归, 反向逐步剔除法