Putting numbers to a metaphor: A Bayesian Belief Network with which to infer Soil Quality and Health

Abstract

Soil Quality or Soil Health are terms adopted by the scientific community as metaphors for the effects of differing land management practices on the properties and functions of soil. Because they are metaphors, consistent quantitative definitions are lacking. We present here an approach based on expert elicitation in the field of soil function and management that offers a universal way of putting numbers to the metaphor. Like humans, soils differ and so do the ways in which they are understood to become unhealthy. Long-term experiments such as the Broadbalk Wheat experiment at Rothamsted provide unparalled sources of data with which to investigate the state and changes of soil quality and health that have developed from known management over timescales of one hundred years or more. Similarly, large-scale datasets such as the National Soils Inventory and Countryside Survey provide rich resources to explore the geographical variability of soil quality and health in different places against a background of different observed management practices. We structure experts’ views of the extent to which soil delivers the functions expected of it within Bayesian Belief Networks anchored by measurable properties of soil. With these networks, we infer the likely state of soil (i) on Broadbalk, (ii) at locations throughout England & Wales as well as inferring (iii) the most straightforward ways of improving soil quality and health at the locations in (ii). Our methodology has general applicability and could be deployed elsewhere or in other disciplines.