Performance of classic multiple factor analysis and model fitting in crop modeling

Multivariate statistical analysis and regression, which are typical methods for crop modeling, have direct influence on the accuracy of model, but the applications of these methods usually depend on experiences. In this research, the performances of some common methods of statistical analysis and regression model were compared and verified, in order to avoid the blindness in crop modeling. The monitoring data of growth environment and photosynthesis of tomato, pumpkin and cucumber were obtained by PTM-48A. For the object variable of CO2 exchange rate, selectivity on the main environmental factors by correlation analysis and path analysis were quantitatively compared. The performances of four kinds of multivariate binomial regression equations were compared using a comprehensive aggregative indicator, and the effectiveness of modeling was verified with the selected optimized multivariate statistical analysis and regression equation. Results showed that path analysis was more comprehensive and effective than correlation to discrimination of the variables, especially the path analysis ruled out some suspected independent variables which were not really independent, and the pure quadratic was more suitable to crop modeling because of its simple structure and high accuracy when the data set was small. The conclusion of this research has a general applicability, and offers a useful reference and guide for the other study and application of crops’ modeling.