Multivariate spatial variability in soil variables and grain yields at fine scale

CÓRDOBA, M.; BRUNO, C.; BALZARINI, M.; COSTA, J.L.

Kobe

Conferencia; XXVIth International Biometric Conference; 2012

Biometric Society of Japan y Japanese Region of the International Biometric Society

Massive georeferenced data are common in precision agriculture, where soil properties and grain yield are intensively measured at different sites within a crop field. The correlations of variables have been evaluated through classical principal component analysis (PCA). However, PCA was not specifically developed to handle spatial data and the covariation of soil and yield. Recently developed multivariate analyses allow taking into account spatial autocorrelation among neighbouring sites. In this paper we compare PCA with an spatially constrained multivariate analysis method based on Moran´s Index (MULTISPATI-PCA) as tools to derive synthetic variables to be used in the identification of homogeneous soil zones within field. Their use is illustrated in a data base involving apparent electrical conductivity at 30 cm (ECa30) and 90 cm (ECa90) deep, elevation, hardpan depth as soil variables and also soybean and wheat yields measured at several sites within a field. The synthetic variables obtained by each PCA-method were used as input for fuzzy k-means clustering (FKM) to identify three homogeneous zones in a multivariate sense. Additionally, a canonical correlation analysis (CCA) was used to evaluate the relationship between the first three principal components (PC), as well as with the first three spatial principal components (PCe) of MULTISPATI-PCA of soil variables with the normalized grain yield variables. New synthetic variables were obtained from the first significant canonical correlation, using both the PCA (CCA-PC) and MULTISPATI-PCA (CCA-PCe). Canonical variables were categorized in three classes (by using the 33 and 66 percentiles) and used as another way to identify homogeneous zones within-field. Finally, we compared the average yield among the homogeneous zones obtained by different methods. The first axis score obtained by MULTISPATI-PCA showed a high and positive correlation with elevation (variable with the highest spatial autocorrelation) and with ECa30, whereas the first axis score obtained by PCA was correlated only with ECa30. Hardpan variable was correlated with the second MULTISPATI-PCA axis score and not with the second PCA axis. The ANOVA results showed greater yield differences between the zones obtained with MULTISPATI-PCA of soil variables and with the CCA-PCe. Furthermore, these methods had lower standard error of mean yield differences between zones. The results show that MULTISPATI-PCA and its combination with CCA are useful tools to map spatial soil variability that correlates with yield variability enhancing the identification of homogeneous zones in a multivariate sense, within fields.