水稲の収量・品質決定要因解析手法としての回帰木の適用

日本作物学会紀事

田中 恒大¹  井上 英二²  光岡 宗司²  岡安 崇史²  平井 康丸²  猿田 恵輔³ 

[1] 九州大学大学院生物資源環境科学府;九州大学大学院農学研究院;ヤンマー株式会社

关键词: Determinants;    決定要因;    Grain yield;    精玄米収量;    Multiple regression analysis;    重回帰分析;    Protein content of brown rice;    玄米タンパク質含有率;    Regression tree;    回帰木;    Rice production;    水稲生産;    1000-grain weight;    千粒重;   

DOI  :  10.1626/jcs.83.143

学科分类：农业科学（综合）

来源: Nihon Sakumotsu Gakkai / Crop Science Society of Japan

【 摘 要 】

In recent years, it is becoming important to collect and utilize different data on rice production to achieve a high yield and high quality, and to transfer the skill to the next generation. The objective of this study was to investigate the validity of a regression tree for analysis of yield and quality determinants. Determinants of grain yield (yield), 1000-grain weight, and protein content of brown rice (protein) were analyzed by focusing on explanatory variables for the meteorological environment and conditions of rice growth and nutrition after the heading period. The spikelet number/m2 was identified as a yield determinant by a regression tree and multiple regression analysis. However, the predictive accuracy was higher in multiple regression analysis due to a strong linear correlation between yield and spikelet number/m2. For the analysis of the 1000-grain weight with nonlinear data structure, no reasonable determinant was identified in multiple regression analysis, while a regression tree implied the 1000-grain weight determinant by clarifying its relationship with SPAD readings about 20-d after heading (SPAD20). The SPAD20 was also identified as a protein determinant. Furthermore, it was indicated for data under the condition with high SPAD readings that protein was lowered by the high temperature during 5 to 30 days period after heading. A regression tree can clarify the relationship among parts of data by splitting data hierarchically, but multiple regression analysis can not.

【 授权许可】

Unknown