Multivariate analysis using a discriminant method for evaluating the techniques of weed management in soybean crop

Abstract Background: The analysis of information generated from experiments involving different treatments, can be done by multivariate statistical analysis techniques, such as discriminant analysis, to analyze data obtained from predefined groups. Objective: Verify, through discriminant analysis, the differences among cover crop (Avena strigosa, Chenopodium quinoa, Cichorium intybus, and fallow land) treatments with respect to main crop soybean yield. Methods: For weed control, these cover crops were subjected to different management techniques, namely mowing, the application of glyphosate or the application of paraquat. The experimental design consisted of completely randomized blocks in a 4 × 3 × 2 factorial scheme, with four replications, consisting of the following factors: Factor A: (treatment) cover crops of A. strigosa, C. quinoa, C. intybus, and fallow land; Factor B: (management) plots were subdivided and treated with the application of paraquat or glyphosate, or the mowing of cover plants; Factor C: the plots were sub-subdivided and managed by one or two applications of a post-emergence herbicide. In order to evaluate the percentage of correct classifications of the different management techniques and treatments, a data matrix was elaborated for evaluation of variables relating to the soybean crop and the data were standardized by log - log 10 - log (n; 10). Multivariate analysis was performed using Fisher's linear discriminant method. Results: Discriminant analysis selected four variables with discriminatory power relating to the A. strigosa, C. quinoa, C. intybus and fallow, which contributed to 100% of the explained variance. Conclusions: Treatment with oats used as a cover crop provided higher soybean crop yield, whereas in terms of management, weed control using glyphosate provided the best results with all cover crops.研究背景：针对不同处理实验所产生的信息，可借助多元统计分析（multivariate statistical analysis）技术，如判别分析（discriminant analysis），对预定义分组的实验数据开展分析。研究目的：通过判别分析验证不同覆盖作物（cover crop）处理（包括野燕麦（Avena strigosa）、昆诺阿藜（Chenopodium quinoa）、菊苣（Cichorium intybus）及休耕地）对主栽作物大豆产量的差异影响。试验方法：为实现杂草防控，上述覆盖作物采用了三类不同管理措施：刈割、草甘膦（glyphosate）喷施及百草枯（paraquat）喷施。本试验采用4×3×2析因设计（factorial scheme）的完全随机区组设计，设置4次重复，包含以下3个因子：因子A（处理因子）：覆盖作物，包括野燕麦（Avena strigosa）、昆诺阿藜（Chenopodium quinoa）、菊苣（Cichorium intybus）及休耕地；因子B（管理因子）：将主小区进一步划分为亚区，分别喷施百草枯、草甘膦或对覆盖作物进行刈割处理；因子C（亚区因子）：将亚区再次划分为亚亚区，采用1次或2次苗后除草剂（post-emergence herbicide）喷施的管理方式。为评估不同管理措施与处理的正确分类占比，本研究构建了用于分析大豆作物相关变量的数据矩阵，并采用双对数10转换（log₁₀-log₁₀）对数据进行标准化处理。随后采用费希尔线性判别法（Fisher's linear discriminant method）开展多元统计分析。试验结果：判别分析筛选出4个具备判别效力的变量，分别对应野燕麦、昆诺阿藜、菊苣及休耕地处理，上述变量可解释100%的总方差。研究结论：以燕麦作为覆盖作物的处理可获得更高的大豆产量；而就管理措施而言，对所有覆盖作物采用草甘膦进行杂草防控均可取得最优防控效果。