Developing a cost-effective procedure for investigating within-field variation of soil conditions

Abstract

The aim of this project was to develop a cost-effective procedure for investigating the variability of soils within fields as an aid to farm-level decisions on the adoption of variable rate management of inputs. The project was based on the premise that not all fields will merit variable rate management and so cost-effective investigation, by analysis of cheap and available data, should allow us to identify the fields with the largest potential for variable rate management before substantial resources are invested in soil sampling. It was hypothesized that information to this end could be extracted from past yield maps of the field, and that generalized information on the variability based on the soil parent material would also be indicative of the scope for variable management. It was also hypothesized that division of a field into zones, within which the season-to-season variation of crop yield is more or less uniform should identify the spatial structure of the key variations in soil properties which influence crop performance and so should aid in the identification and mapping of these properties.

Using data obtained by intensive and extensive sampling and examination of the soil at nearly forty fields, 34 fields (10 intensive and 24 extensive) funded in the current project, we were able to validate both hypotheses. We showed that it is possible to emulate expert assessment of the potential for variable rate management of a field (based on an agronomic interpretation of soil information) by applying a classification tree procedure to statistics (spatial and non-spatial) extracted from yield data of the particular field, and generalized assessments of the opportunity for variable management of different inputs over soils developed on different parent materials. The resulting classification tree is a decision support tool which could be used by individual farmers.

We also showed that in all fields at least some measured soil variables were significantly associated with the zonation of the field based on yield data. This allows us to identify the soil factors which may be relevant to variable rate management in the field and to obtain an initial impression of their spatial variation in the field. It is possible, for example, to produce a reasonable map of the principal soil types (e.g. soil series) in a field by using a classification tree, trained with a relatively small number of point observations of the soil, with continuous class memberships (derived in the identification of the within-field zones from yield data) used as predictors. Bulk (composite) soil samples from within each zone may also be tested to look for evidence of limiting soil concentrations of key nutrients. The precision of point predictions of these properties from the zones was generally poor, although not consistently better or worse than point predictions from non-intrusive soil sensors. When detailed information on the variation of a property is needed for management there is probably no substitute for moderately detailed sampling of the soil