Science Wednesday: the shapes of soils

December 15th, 2021

One of our key science challenges over the course of the Digiscape Future Science Platform has been to characterise Australia’s agricultural landscapes with sparsely-measured data: climate, soils, vegetation and management. Soil moisture characteristic data are in particularly short supply.

Historically, spatial inference of soil attributes for modelling analyses has taken one of two paths:

  • continuous spatial interpolation, attribute-by-attribute (e.g. GlobalSoilMap and its Australian implementation, the Soil and Landscape Grid of Australia); or
  • classifying each location to a modestly-sized set of categories, each of which has a “representative” (and often manually-selected) measured soil profile assigned to it.

There is a tradeoff here between the need to reduce the uncertainty in the estimates of the individual attributes describing a soil profile and the need to ensure that the estimates of the different attributes are coherent with one another.

A recent paper in Geoderma by Dan Gladish, Di He and Enli Wang has opened up a path to intermediate – and perhaps more effective – solutions to that tradeoff. By carrying out pattern analysis of the shapes and sizes of a library of measured soil moisture characteristics, they derived a small set of classes or “patterns” for which a rigorously-defined representative profile can be described.

This better way of defining the “representative” soil profiles raises the possibility of profile-centric digital soil maps, in which the landscape is mapped to the soil “patterns” and then continuously-interpolated maps of widely-measured attributes such as clay content and bulk density can be used to adjust the reference profile to improve the accuracy of the attribute estimates at each depth.