A Bayesian approach toward the use of qualitative information to inform on-farm decision making: The example of soil compaction

Abstract

As the development of predictive tools to aid decision making in agriculture increases, it must be acknowledged that the ability to incorporate management decisions as input data is limited. The data is either not recorded, or highly inadequate in terms of the volume, variety or relevance. Some circumstances are further hampered by the lack of benchmark data for comparison, and soil compaction is an example of this. The premise of this work was to take a probability based approach to decision making, utilising both qualitative and quantitative data to provide a probability distribution of risk against decisions made, in the context of grains production systems as an example of an agricultural enterprise. A Bayesian Belief Network (BBN) was constructed for soil compaction risk, as an exemplar. The BBN conditional probability tables for nodes were populated via a combination of biophysical model output (namely SoilFlex for soil stress distribution, and the APSIM package for soil-water and crop parameters) and expert opinion. Input nodes were parameterised with measured soil data, and the risk of soil compaction, given the soil stress at the wheel of a particular vehicle, was provided as the output. Potential effect on yield was subsequently calculated on the basis of percent change in soil bulk density, which was determined using literature based information (expert opinion). The tool broadly estimated yield impacts due to various agricultural traffic scenarios, providing means to highlight the financial consequences of failing to adopt controlled traffic farming management for a particular agricultural enterprise. Of significance, the BBN approach was determined useful for data limiting environments where empirical models struggle, thus providing a pragmatic and novel approach to on farm decision making incorporating management nuance.