Bayesian analysis of within-field variability of corn yield using a spatial hierarchical model

Understanding relationships of soil and field topography to crop yield within a field is critical in site-specific management systems. Challenges for efficiently assessing these relationships include spatially correlated yield data and interrelated soil and topographic properties. The objective of this analysis was to apply a spatial Bayesian hierarchical model to examine the effects of soil, topographic and climate variables on corn yield. The model included a mean structure of spatial and temporal co-variates and an explicit random spatial effect. The spatial co-variates included elevation, slope and apparent soil electrical conductivity, temporal co-variates included mean maximum daily temperature, mean daily temperature range and cumulative precipitation in July and August. A conditional auto-regressive (CAR) model was used to model the spatial association in yield. Mapped corn yield data from 1997, 1999, 2001 and 2003 for a 36-ha Missouri claypan soil field were used in the analysis. The model building and computation were performed using a free Bayesian modeling software package, WinBUGS. The relationships of co-variates to corn yield generally agreed with the literature. The CAR model successfully captured the spatial association in yield. Model standard deviation decreased about 50% with spatial effect accounted for. Further, the approach was able to assess the effects of temporal climate co-variates on corn yield with a small number of site-years. The spatial Bayesian model appeared to be a useful tool to gain insights into yield spatial and temporal variability related to soil, topography and growing season weather conditions.