Book of Abstracts
11th IFOAM Scientific Conference
11-15 August 1996, Copenhagen, Denmark
1)Department of Plant Physiology, Swedish University of AgriculturalSciences, P.O. Box 7047, S - 750 07 Uppsala, Sweden. 2) Department of FoodScience, Swedish University of Agricultural Sciences, P.O. Box 7051, S -750 07 Uppsala, Sweden.
In a field experiment carried out for 32 years the influences of 8 manuring systems (biodynamic (1), organic (3) and mineral (4)) on the development of wheat, ley, potato, and fodder beet were compared. Crop and weather datawere available from 10, 7 and 8 years for potato (cv. King Edward and Grata) and for wheat (cv. Drabant) respectively.
Because of lack of repetition of the treatments within each year, it was not possible to estimate any random error, which in turn is necessary for evaluation of statistical significant differences - in any responseparameter - between treatments.
For crops as different as potato and wheat we developed a predictive statistical modelling method, based on multiple linear regressions using principal components from daily weather data of whole growing seasons. The technique copes with different sowing time and different length of growing seasons in different years, and with interactions between weather parameters. Sufficient weather parameters are daily mean temperature and daily precipitation. To avoid models which are by chance descriptive (high R2) but not predictive (high Q2) cross-validation-values were calculated for possible MLR-models. Models with best predictive power were checked for statistically significant factors.
Conceptually the technique is based on these assumptions: every day plants react on weather parameters which are often inter-correlated those weather correlations can be caught in principal components of the weather data the plant expresses a »biological« and the weather principal components a »mathematical integration« of the weather. the relationship between these two expressions of weather conditions can be described by multiple linear regression analyses.
The method improved the regression coefficient (R2) considerably when wevaluated yield of potato and of wheat. The predictive power (Q2) of the technique was found high, and statistically significant differences between treatments could be detected.
Pettersson, Bo D.; Reents, H.J. and Wistinghausen, E.v. (1992): Gödslingoch markegenskaper, (Düngung und Bodeneigenschaften), NordiskForskningsring Meddelande nr. 34.