Methodology to predict the spatial distribution of cattle dung using manageable factors and a Bayesian approach

Abstract The aim of this study was to predict the spatial distribution of cattle dung in paddocks based on Bayesian estimation using generalized linear mixed models with an added intrinsic conditional autoregressive term. The predicted herbage green biomass (GBM) and distance from a water trough ( D w ), which can be controlled by farmers, were considered as predictors in the models. This study was conducted in three mixed sown pastures (I and II, 1.02 ha; III, 0.85 ha) in Hokkaido, Japan. After a four-day grazing trial using 20 Japanese Black cows, the paddocks were divided into 10 m × 10 m grid cells (I and II, 102 cells; III, 85 cells) and for each grid cell the number of dung deposits ( N d ) was counted and the mean values of the GBM and D w were computed. The results of Markov Chain Monte Carlo simulations indicated that a higher N d tend to be associated with a higher GBM and locations closer to the water trough. N d had spatial autocorrelation and it is likely that the grid cells that have large residual values could be affected by the difference between cattle activities in the daytime and nighttime. Based on our results, we suggest that the spatial distribution of cow dung can be predicted from two controllable factors in short term grazing trials.