Soil attributes as predictors of crop production under standardized conditions

Abstract

To assess soil fertility or quality three controlling components – its physical, chemical and biological nature – have to be considered. In this study a broad spectrum of agricultural soils from Sweden were cropped with ryegrass in pots under standardized conditions in climate chambers. Measurements of physical, chemical and biological attributes of soil were used to predict C and N yields by simple correlation and the multivariate calibration techniques, principal component analysis combined with multiple linear regression, and partial least squares (PLS) regression. The N yields were typically more accurately predicted than the corresponding C yields. The best single predictor of yields was always total soil N, but estimates produced by multivariate models including organic C, total N, C/N ratio, coarse silt, potential denitrification activity, N mineralization, substrate-induced respiration and sample site humidity were, in all cases, substantially more accurate. Coefficients of correlation between predicted and measured C or N yields ranged between 0.61 and 0.80 with total N as predictor, and between 0.69 and 0.97 with the multivariate models. Both quantitative and qualitative aspects of the organic matter were considered to be important with respect to the predictive ability. Both these aspects were accounted for by the multivariate models. The multivariate technique, PLS regression, facilitated the classification of soils into categories of good, normal or poor fertility in relation to their organic matter content.