B. P. Malone, B. Minasny, A. B. McBratney
Faculty of Agriculture, Food & Natural Resources
The University of Sydney
Telephone: 0011 +61 9568 4113
To benefit from the ecological and economical functions of soil, land holders, corporate stakeholders and, governmental departments need access to quantitative soil information. Such information confers weight to decisions regarding the management of the land and soil resources. To facilitate this need, we must first comprehend the functions and pertinent factors contributing to not only soil variability across a landscape but also soil variability with depth down a profile.
The variation of soil properties down a profile is usually continuous (Ponce- Hernandez et al. 1986). Soil depth functions are often created to represent the depthwise variation of soil properties. However, with traditional sampling of soil profile horizons, it is often assumed that the horizon value of a particular attribute represents the average value for that attribute for the depth interval of that horizon. With this paradigm, in effect what should be a continuous function, the data often appears discontinuous or stepped.
A flexible and accurate method for fitting continuous functions of soil data is the use of smoothing splines (Erh 1972) and equal-area spline functions as proposed by Ponce-Hernandez et al. (1986). Essentially, a spline function is a set of local quadratic functions tied together with ‘knots’ that describe a smooth curve through a set of points. Bishop et al. (1999) demonstrated their superiority over other continuous soil depth functions when they predicted various types of soil properties.
However, in a spatial context, a collection of spline functions for individual site observations will ultimately lead only to point observation data sets. To the parties concerned, such data will be of little use for mapping soil variability. The response to this demand has been answered partly in the way of digital soil mapping, where soil properties are mapped based on their relationship with environmental variables (Minasny et al. 2008). The scorpan factors as proposed by McBratney et al. (2003) provide a valuable predictive framework for determining soil variability in areas with limited soil data.
Given the predictive capabilities of soil depth functions and an explosion in the capabilities of digital soil mapping in areas with limited data (Lagacherie 2008), it seems only logical for there to be an amalgam of both methods to quantitatively predict the vertical and lateral variation of soil properties across a defined area. In this paper we propose a novel method for predicting the vertical and lateral variation of soil properties in areas where limited soil data exists. Using soil carbon as our exemplar soil property we want to firstly determine whether terrain attributes alone are feasible for predicting its lateral and vertical variation or whether it is better described with the inclusion of other environmental factors relating to parent materials and landuse into the predictive models. With the most parsimonious model we want to map carbon storage to a depth of 1m in our defined study area and then demonstrate the functionality of the underlying soil geo-database for data enquiry by mapping the depth at which soil carbon falls below 1%.