A crucial part of any site-specific management is the identification of causes of yield variability and assessment of crop requirements. Therefore, relationships between yield and soil properties must be identified. In this study, relationships between sorghum yield and soil properties on a verbosols within a field located in Moree, in northern NSW, Australia, were examined. Measured soil properties included pH; available phosphorus; percent clay, silt and sand; gravimetric moisture content of air-dry soil and at matric potentials corresponding to −1 500 kPa and −33 kPa; percent organic carbon; CEC and exchangeable calcium, magnesium, sodium and potassium and copper, zinc, manganese and iron contents. The exchangeable sodium percentage (ESP) and the Ca/Mg ratio were calculated. We used a number of empirical methods and found that neural networks, projection pursuit regression, generalized additive models and regression trees are good techniques for modeling yield response. However, further comparison of these techniques is needed. By modeling yield response to individual soil properties and using kriging to map yields predicted from these models, it was possible to identify which soil properties limited production in different areas of the field.
Read full abstract