T.F.A. Bishop

An overview of pedometric techniques for use in soil survey

Quantitative techniques for spatial prediction in soil survey are developing apace. They generally derive from geostatistics and modern statistics. The recent developments in geostatistics are reviewed particularly with respect to non-linear methods and the use of all types of ancillary information. Additionally analysis based on non-stationarity of a variable and the use of ancillary information are demonstrated as encompassing modern regression techniques, including generalised linear models (GLM), generalised additive models (GAM), classification and regression trees (RT) and neural networks (NN). Three resolutions of interest are discussed. Case studies are used to illustrate different pedometric techniques, and a variety of ancillary data. The case studies focus on predicting different soil properties and classifying soil in an area into soil classes defined a priori. Different techniques produced different error of interpolation. Hybrid methods such as CLORPT with geostatistics offer powerful spatial prediction methods, especially up to the catchment and regional extent. It is shown that the use of each pedometric technique depends on the purpose of the survey and the accuracy required of the final product.

A comparison of prediction methods for the creation of field-extent soil property maps

We compare various prediction methods for mapping of soil cation exchange capacity using different combinations of secondary information. The prediction methods used are statistical analysis (generalised additive model, regression tree, multiple linear regression), geostatistical interpolation (ordinary kriging) and the hybrid techniques (regression-kriging and kriging with external drift). The secondary spatial information used are terrain attributes, bare soil colour aerial photograph, bare soil LANDSAT TM imagery, crop yield data and soil apparent electrical conductivity (ECa). A modification of jackknifing was used as the validation method. This involved 100 jackknife partitions to examine the stability of the validation indices with different realisations of the data set. Root-mean-square error (RMSE) was used as the validation index, with the mean RMSE used to judge the prediction quality. The best prediction methods were kriging with external drift, multiple linear regression and generalised additive models. They were best in combination with soil ECa or the bare soil colour aerial photograph.

Modelling soil attribute depth functions with equal-area quadratic smoothing splines

Abstract The objective of this paper is to test the ability of equal-area quadratic splines to predict soil depth functions based on bulk horizon data. In addition, the possibility of improving the prediction quality by the use of additional samples from the top and/or bottom of soil profiles along with horizon data is examined. The predictive performance of the splines is compared with that of exponential decay functions, and 1st and 2nd degree polynomials. In addition, the predictive quality of the conventional horizon data is examined. The measure of predictive performance used is the root mean square error values calculated from differences between the ‘true’ depth function and the fitted depth function. The ‘true’ depth functions were derived from the intensive sampling and laboratory analysis of soil profiles. Three soil profiles were sampled; a Red Podzolic Soil (Red Kurosol), Podzol (Aeric Podosol) and Krasnozem (Red Ferrosol). The soil attributes that were measured included; pH, electrical conductivity (EC), clay %, sand %, organic carbon %, gravimetric water content at −33 kPa and air dry. The results clearly indicated the superiority of equal-area quadratric splines in predicting depth functions. Such splines depend on a parameter, λ that controls goodness-of-fit vs. roughness. Their quality of fit varied with the λ value used and it was found that a λ value of 0.1 was the best overall predictor of the depth functions. The results also showed that using additional samples from the top and/or bottom of the soil profiles improved the prediction quality of the spline functions.

Measuring the quality of digital soil maps using information criteria

One of the purposes of a soil map is to present information regarding the spatial variation of soil to the end user. In the past, statistical methods of uncertainty have been used to indicate map quality. In this paper, the information content of a map is proposed as a more suitable measure of map quality. The theory has its basis in information theory, more particularly Shannons information criterion. A modification of Shannons information criterion is presented. Other than being used as an indicator of map quality, the method is particularly useful as an aid during the map production process, for example, choosing block size or grid spacing. Four examples illustrate the concept and its potential use in soil science.

Uncertainty analysis for soil‐terrain models

The aim of the study was to examine how robust soil‐terrain models are to uncertainty in the source elevation data. The study site was a 74 ha agricultural field in Australia. A global positioning system was used to measure elevation and the uncertainty of the measurement, therefore allowing maps of elevation and its uncertainty to be created. Monte‐Carlo simulation with a modified version of Latin Hypercube Sampling was used to create 100 realizations of a slope map. Clay content was measured at 111 sites, and kriging with external drift was used to map clay content where each slope realization was used as a secondary information source. Maps of the mean and standard deviation of clay content across all realizations were created. The standard deviations of clay content were generally small (<4 dag kg−1) and in most parts of the field less than the analytical accuracy of the hydrometer method which was used to measure soil‐clay content in the laboratory. The values in the map of elevation uncertainty were multiplied by 5 and the entire error propagation process was repeated to create a second set of 100 realizations of the clay content. The ratio of the uncertainty in the original DEM was 5:1 when compared with that in the perturbed DEM, i.e. it was multiplied by 5. The ratio between the standard deviation in the two clay‐content maps was 3.79:1, which indicates a reduction in uncertainty through the modelling process. The results showed that the soil‐terrain model performs well for the study area, and it is not very sensitive to DEM errors. We conclude that input uncertainty tests as shown in this study should accompany soil mapping studies where secondary information is used in the prediction mdoel.

Creating Field Extent Digital Elevation Models for Precision Agriculture

This paper presents a comparison of techniques for the interpolation of elevation data collected using a global positioning system. The techniques examined included global and local regression models and kriging of the residuals, global and local kriging, and the TOPOGRID tool in Arc Info. The results demonstrated the superiority of the TOPOGRID tool.

Two soil profile reconstruction techniques

This paper presents two different soil profile reconstruction techniques; the fitting of equal-area quadratic splines (EQS) through bulked horizon data and the use of Tikhonov regularisation (TR) on layered apparent electrical conductivity readings measured from an electromagnetic induction instrument. A Red Kurosol was used to demonstrate the two techniques. The results revealed some problems with both. For the spline reconstruction problems occur when the morphological horizons do not reflect accurately the depth function of a particular soil attribute. For TR a major problem is choosing the regularisation order without a priori information. Both methods are useful in different situations, the spline reconstruction when horizon data are available and TR when no or little soil information is available as the reconstruction relies only on instrument readings from an electromagnetic induction instrument.

Evaluation of spatial variability of soil arsenic adjacent to a disused cattle-dip site, using model-based geostatistics.

This study investigated the spatial variability of total and phosphate-extractable arsenic (As) concentrations in soil adjacent to a cattle-dip site, employing a linear mixed model-based geostatistical approach. The soil samples in the study area (n = 102 in 8.1 m(2)) were taken at the nodes of a 0.30 × 0.35 m grid. The results showed that total As concentration (0-0.2 m depth) and phosphate-extractable As concentration (at depths of 0-0.2, 0.2-0.4, and 0.4-0.6 m) in soil adjacent to the dip varied greatly. Both total and phosphate-extractable soil As concentrations significantly (p = 0.004-0.048) increased toward the cattle-dip. Using the linear mixed model, we suggest that 5 samples are sufficient to assess a dip site for soil (As) contamination (95% confidence interval of ±475.9 mg kg(-1)), but 15 samples (95% confidence interval of ±212.3 mg kg(-1)) is desirable baseline when the ultimate goal is to evaluate the effects of phytoremediation. Such guidelines on sampling requirements are crucial for the assessment of As contamination levels at other cattle-dip sites, and to determine the effect of phytoremediation on soil As.

Enhancing the value of field experimentation through whole-of-block designs

Precision agriculture (PA) offers opportunities for the development of new approaches to on-farm experimentation to assist farmers with site-specific management decisions. Traditional agricultural experiments are usually implemented in fields with the least possible soil heterogeneity under the assumption that responses to inputs and inherent variation of the soil are additive components of yield variation. However, because the soil in typical fields is not homogeneous, PA has much to offer. Farmers faced with variable conditions need to optimize their management to the variation over space and time on their farm, a problem that is not solved by conventional approaches to experimentation. New designs for on-farm experiments were developed in the 1990s for cereal production in which the whole field was used for the experiment rather than small plots. We explore the extension of this type of experiment to a vineyard in the Clare Valley of South Australia aiming to evaluate options to increase grape yield and vine vigour. Manually sampled indices of vine performance measured on georeferenced ‘target’ grapevines were analysed geostatistically. The major advantage of such an approach is that the spatial variation in response to experimental treatments can be examined. Linear models of coregionalization, pseudo cross-variograms and standardized ordinary cokriging are used to map treatment responses over the experimental area and also the differences between them. The results indicate that both treatment responses and the significance of differences between them are spatially variable. Thus, we conclude that whole-of-block on-farm trials are useful in vineyards.

Pragmatic models for the prediction and digital mapping of soil properties in eastern Australia

To help meet the increasing need for knowledge and data on the spatial distribution of soils, readily applied multiple linear regression models were developed for key soil properties over eastern Australia. Selected covariates were used to represent the key soil-forming factors of climate (annual precipitation and maximum temperature), parent material (a lithological silica index) topography (new topo-slope and aspect indices) and biota (a modified land disturbance index). The models are presented at three depth intervals (0–10, 10–30 and 30–100 cm) and are of variable but generally moderate statistical strength, with concordance correlation coefficients in the order of 0.7 for organic carbon (OC) upper depth, pHca, sum of bases, cation exchange capacity (CEC) and sand, but somewhat lower (0.4–0.6) for OC lower depths, total phosphorous, clay and silt. The pragmatic models facilitate soil property predictions at individual sites using only climate and field-collected data. They were also moderately effective for deriving digital soil maps over the state of New South Wales and a regional catchment. The models and derived maps compared well in predictive ability to those derived from more sophisticated techniques involving Cubist decision trees with remotely sensed covariates. The readily understood and interpreted nature of these products means they may provide a useful introduction to the more advanced digital soil modelling and mapping techniques. The models provide useful information and broader insights into the factors controlling soil distribution in eastern Australia and beyond, including the change in a soil property with a given unit change in a covariate.