Digital Mapping Of Landscape Units And Salinity Across The Bourke Irrigation

District

Ehsan Zare

A thesis submitted in

fulfilment of the requirements

for the degree of

Master of Philosophy

School of Biological, Earth and Environmental Sciences

March 2017

Abstract

In Bourke Irrigation district (BID), one of the most productive agricultural soil types is the Vertosol which has been extensively developed for irrigated agricultural production.

However, there have been increasing instances of decline in productivity as a result of secondary soil salinization due to injudicious location of water storages and conveyance channels in irrigated farmlands. Therefore, it is important to map the soil in this area to understand the long-term sustainability issues and spatial variability in soil types. A great deal of soil data is required to characterise large areas such as BID. One approach is to use (DSM) methods because they rely on the use of low-cost ancillary data to value add to limited soil information via the use of spatial and non-spatial numerical methods. In this thesis, two different DSM approaches are used in order to identify soil landscape units and a soil property (i.e. ECe dS/m).

Firstly ancillary data, including remotely sensed air-borne gamma-ray (-ray) spectrometer (i.e. potassium-K, uranium-U, thorium-Th and total counts-TC) and proximal sensed EM38 in the horizontal (EM38h) and vertical (EM38v) mode of operation are used with a non-spatial numerical clustering algorithm (fuzzy k-means: FKM). The FKM analysis

(using Mahalanobis metric) of the kriged ancillary (i.e. common 100 m grid) data revealed a fuzziness exponent () of 1.4 was suitable for further analysis and that k = 4 classes was smallest for the fuzziness performance index (FPI) and normalised classification entropy

(NCE). Using laboratory measured physical (i.e. ) and chemical (i.e. CEC, ECe and pH)

2 properties revealed k = 4 was minimized in terms of mean squared prediction error (i.e.  p,C)

when considering (0-0.3 m) and (0.9-1.2 m) Clay, CEC, ECe and pH (i.e. only for topsoil).

Secondly, in order to map a soil property limiting agricultural productivity, ancillary data, including remote and proximal, were used with a spatial numerical model (linear mixed modelling-LMM and restricted likely - REML) to map the spatial distribution of the saturated soil paste –extract (ECe dS/m). DSM of ECe using the same sources of ancillary data and empirical best linear unbiased prediction (E-BLUP) showed elevation, radioelement of thorium (Th) and logEM38v were the most statistically useful ancillary data. It was also found that the development of an error budget procedure, enabled the quantification of the relative contributions that model, input, combined and covariate error made to the prediction error of the map of ECe. The combined error is approximately 4.44 dS/m, which is relatively large compared to the standard deviation of measured ECe (3.61 dS/m). Of this error, most of it is attributable to the input error (1.38 dS/m) which is larger than the model error (0.02 dS/m). In terms of the input error, it is determined that the larger standard deviation is attributable to the lack of ancillary data, namely the ECa in areas adjacent to the Darling

River and also on the aeolian dune where data collection was difficult owing to dense native vegetation.

ACKNOWLEDGEMENTS

I would like to acknowledge my supervisor Dr. John Triantafilis for his fabulous guidance, persistence, patience and contribution to the ideas and the writing of this thesis and during the past two years and whilst I have been enrolled as an MPhil student within the School of Biological, Earth and Environmental Sciences (BEES) in the Faculty of Science, UNSW

Australia.

I would also like to greatly thank Mr. Jingyi Huang (Ph.D. student, UNSW) for his invaluable contributions in the statistical analysis of the results of the FKM analysis, LMM, and

REML. I also thank him for introducing me to the R software package and for providing me with the necessary templates and ideas to analyse and interpret the results in terms of calculating the expected value of the mean squared prediction error from classes (i.e. 2 p,C) and regression

(2 p,R). I am also indebted to Mr. Triven Koganti (MPhil student, UNSW) for his contribution to the ideas of the thesis structure and format.

Finally, I would like to thank two examiners, Professor Cristine L. S. Morgan (Texas

A&M Univesity, USA) and Professor Eric C. Brevik (Dickinson State University, USA) for their kind suggestions on the revisions of the thesis.

Table of Contents

1 Introduction ...... 7 1.1 References ...... 10

2 Literature Review ...... 11 2.1 Introduction ...... 11 2.1.1 Ancillary data used in DSM ...... 12 2.1.2 systems ...... 14 2.1.3 Proximal sensing systems ...... 15 2.2 Review of -ray spectrometry ...... 17 2.2.1 The -ray theory ...... 17 2.2.1.1 Isotopes ...... 18 2.2.1.2 Basic radioactivity ...... 19 2.2.1.3 The -Radiometric ...... 20 2.2.1.4 Airborne -ray spectrometry survey ...... 25 2.2.2 The -ray applications ...... 26 2.2.2.1 ...... 27

2.2.2.2 Salinity (ECe) ...... 29 2.2.2.3 Parent materials and soil types ...... 29 2.2.2.4 zones and classes ...... 32 2.2.2.5 Other soil properties ...... 33 2.3 Review of electromagnetic methods ...... 33 2.3.1 Theory of electromagnetic induction ...... 33 2.3.2 EM instruments ...... 35 2.3.2.1 Geonics Series ...... 35 2.3.2.2 EM 38 ...... 37 2.3.2.3 DUALEM ...... 38 2.3.3 EM Applications ...... 40

2.3.3.1 Salinity (ECe) ...... 40 2.3.3.2 Clay content and soil texture ...... 44 2.3.3.3 ...... 47 2.3.3.4 Soil Acidity (pH)...... 49 2.3.3.5 Soil cation exchange capacity (CEC)...... 50 2.3.3.6 Soil management classes ...... 52 2.3.3.7 Other properties ...... 55 2.4 Conclusions ...... 56 2.5 References ...... 57

3 Study area ...... 80 3.1 Location ...... 80

1

3.2 ...... 81 3.3 ...... 82 3.4 ...... 82 3.5 Soil types ...... 83 3.5.1 Cracking Clay (II1)( (Isbell 1996)) ...... 84 3.5.2 Deep Grey Soils (CC19)(Vertisols (Isbell 1996)) ...... 85 3.5.3 Neutral Reaction (My1) (Kandosol (Isbell 1996)) ...... 85 3.5.4 Crusty Loamy Soils with Red Clayey (Nb4) (Kandosol (Isbell 1996)) ...... 85 3.6 Physiography ...... 86 3.6.1 Floodplain Unit (Qrs) ...... 87 3.6.2 Aeolian Dunes (Qd and Qrd) ...... 89 3.7 Refrences ...... 91

4 Identifying soil landscape units at the district scale by numerically clustering remote and proximal sensed data ...... 92 4.1 Introduction ...... 92 4.2 Materials and methods ...... 94 4.2.1 Ancillary instruments, data collection and interpolation ...... 94 4.2.2 Fuzzy k-means (FKM) analysis ...... 96 4.2.3 Soil sampling and laboratory analysis ...... 98 4.2.4 Linear mixed model (LMM) ...... 99 4.2.5 Computation of the prediction error variance for class means ...... 100 4.3 Results & discussion ...... 100 4.3.1 Preliminary data analysis of ancillary and laboratory measured data ...... 100 4.3.2 Spatial distribution of proximally sensed data ...... 101 4.3.3 FKM analysis...... 105 4.3.4 Spatial distribution of the FKM classes ...... 107 4.3.5 Mean squared prediction error of digital soil maps ...... 109 4.3.6 REML analysis of FKM class map = 4 ...... 110 4.3.7 Fuzzy canonical analysis of FKM class map = 4 ...... 114 4.4 Conclusions ...... 116 4.5 Refrencess ...... 118

5 An error budget for mapping using different ...... 121 5.1 Introduction ...... 121 5.2 Materials and methods ...... 123 5.2.1 Ancillary data ...... 123 5.2.2 Soil sampling data ...... 127 5.2.3 Digital mapping ...... 127

2

5.2.4 Predicting ECe ...... 128 5.2.5 Error budget calculation ...... 131 5.3 Results and discussion ...... 132 5.3.1 Exploratory data analysis ...... 132 5.3.2 Variograms of ancillary data ...... 134 5.3.3 Spatial distribution of kriged ancillary data ...... 136 5.3.4 Model selection ...... 138

5.3.5 Spatial distribution of predicted ECe using E-BLUP ...... 140 5.3.6 Error budget evaluation: model, input and combined ...... 146 5.3.7 Error budget evaluation: covariate ...... 150 5.4 Conclusions ...... 153 5.5 References ...... 154

6 Conclusion ...... 158

List of Figures Literature Review Fig. 2.1.1.1 The electromagnetic spectrum used for soil and proximal sensing (after McBratney et 14 al. 2003). Fig. 2.1.3.1 Classification of on-the-go proximal soil sensors (Adamchuk and Viscarra Rossel, 15 2010). Fig. 2.2.1.1.1 The three most stable isotopes of hydrogen. 18 Fig. 2.2.1.3.1 Components of a -ray spectrometer and including the detector and photomultiplier 21 chamber and counting device (i.e. spectrometer). Fig. 2.2.1.3.2 Recommended -ray spectrum windows (Grasty et al., 1991). 22 Fig. 2.2.1.3.3 Factors affecting -ray response in regolith (Wilford et al., 1997). 24 Fig. 2.2.1.3.4 ground based -ray spectrometer being towed on a PVC sled. 25 Fig. 2.2.2.1.1 Bagging-PLSR maps of a) clay and b) coarse content and for the 15–50cm soil 28 layer at Nowley (a) and Stanleyville (b). Fig. 2.2.2.3.1 Bivariate comparisons between ground-measured γ-ray counts from 40K and selected 31 properties of the topsoil (0–10 cm) layer (Wong and Harper, 1999). Fig. 2.2.2.4.1 a) Soil mapping units of Edgeroi district and Narrabri sub-catchment with respect to 33 the lower Namoi valley of New South Wales, Australia (after Northcote et al., 1965); b) Soil classes associated with the eroded landscape in Narrabri sub-catchment (after Triantafilis et al., 2013). Fig. 2.3.1.1 Schematic principal of the functioning of an EMI sensor, specifically the DUALEM-1 34 in which one receiving coil is coplanar (Rx1) and the other is perpendicular (Rx2) to the transmitting coil (Tx) (Visconti and de Paz , 2016). Fig. 2.3.2.1.1 EM 34-3 37 Fig. 2.3.2.2.1 Schematic diagram of the EM-38, which is 1 m in length. Tx is the transmitting coil 38 and Rx is the receiving coil. (after Robinson et al., 2004). Fig. 2.3.2.3.1 DUALEM-21 being towed with the help of a vehicle on a PVC sled. 40 Fig. 2.3.3.2.1 Validation of predictions. a) modeling efficiency (ME) and b) mean absolute error (MAE) calculated for the validation dataset of eight different predictor sets. PLS-R=partial least 47 squares regression; kNN = k nearest neighbor prediction. ECa = apparent electrical conductivity; photo = aerial photo DN, gamma = gamma ray spectrometry (40K, 232Th and TC). The hatched lines represent the values for ordinary of calibration soil samples (after Piikki et al., 2013).

3

Fig. 2.3.3.3.1 Distribution of predicted change in soil  (cm3/cm3) from day 1 to day 12. 49 Fig. 2.3.3.4.1 Relationship between EM31 and EM38 readings and soil pH (CaCl2) at 0–10 cm (x) 50 and 0–30 cm (◦) soil depths (averaged over 0–10, 10–20 and 20–30 cm depth increments) for three rice fields in the Murrumbidgee Irrigation Area (fields 1–3) and one rice field in the Wakool Irrigation District (field 4) (after Dunn et al.). Fig. 2.3.3.5.1. Spatial distribution of predicted average (0–2 m) cation exchange capacity (CEC, 51 cmol(+)/kg) with (a) stepwise, and (b) standard least-squares hierarchical spatial regression (HSR) model with ancillary data available on 24-m transect spacing. Fig. 2.3.3.6.1 Spatial distribution ancilary data including; a) total count (TC – counts per second), 54 b) potassium (K - %), and c) thorium (Th – ppm) e) 1mPcon (ECa – mS/m) and, f) 1mHcon (ECa – mS/m), and c) altitude (m) (after Huang et al. 2014a). Fig. 2.3.3.6.2 Spatial distribution of FKM classes for K=7 (after Huang et al. 2014a). 54

Study Area Fig. 3.1.1 Regional setting. Bourke is located in the mid north-western section of the Murray- 80 Darling Basin. Fig. 3.2.1 Rainfall and temperature averages for the township of Bourke (after BoM 2006). 81 Fig. 3.4.1 The geological provinces in the Bourke District (after Kingham 1998). 83 Fig. 3.5.1 Spatial distribution of the four soil types identified in the Bourke Irrigation District (after 84 Northcote 1966). Fig. 3.6.1 Approximate coverage of the three major types of physiographic units in the Bourke 86 irrigation District (after Brunker, 1971). Fig. 3.6.1.1 Waterlogging and salinity from the bank of the circular storage. Image was taken at 87 E385000, N66725000 looking in a westerly direction. (Note: Water storage is located at the left side of the image). Fig. 3.6.1.2 Waterlogging and salinity from the bank of the circular storage. Image was taken at 88 E385000, N66725000 looking in northerly direction Fig. 3.6.1.3 Waterlogging and salinity from the bank of the circular storage. Image was taken at 88 E385000, N66725000 looking in a north-westerly direction. Fig. 3.6.2.1 An example of the aeolian dune physiographic unit (Image location E385000, 89 N6675500). Note the elevated nature of the dune which looks down onto the floodplain unit in the distance. Fig. 3.6.2.2 A further example of the aeolian dune physiographic unit which occurs as an isolated 90 outcrop between the dual-cell reservoir and the circular reservoir (E382500, N6670000).

Identifying soil landscape units at the district scale by numerically clustering remote and proximal sensed data Fig. 4.2.1.1 (A) Traditional soil landscape unit map (Northcote, 1966), (B) -ray survey transects 95 and EM38 survey locations and (C) Google map image of the study area (obtained on June 20, 2015), physiographic units (Brunker, 1971) and soil sampling points. Fig. 4.3.2.1 Spatial distribution of -ray spectrometry data including; (A) potassium (K– counts per 103 second), (B) uranium (U– counts per second) and (C) thorium (Th– counts per second). Fig. 4.3.2.2 Spatial distribution of -ray spectrometry data including; (A) total count (TC – counts 104 per second) and EM38 electrical conductivity including; (B) EM38h (ECa – mS/m) and (C) EM38v (ECa – mS/m). Fig. 4.3.3.1 Plot of; (A)fuzziness performance index (FPI), (B)normalized classification entropy 106 (NCE) versus classes (k = 2 to 6) and (C) fuzziness exponent () versus –J(M,C)/.

Fig. 4.3.4.1 Spatial distribution of fuzzy k-means (FKM) derived classes for digital soil maps of k = (A) 3, (B) 4, and (C) 5. 108

4

Fig. 4.3.6.1 Plot of mean and standard deviation of : (A) topsoil clay content (%), (B) subsoil clay 112 content (%), (C) topsoil CEC (cmol(+)/kg), (D) subsoil CEC (cmol(+)/kg), (E) topsoil ECe (dS/m), (F) subsoil ECe (dS/m), (G) topsoil pH and (H) subsoil pH. Fig. 4.3.7.1 Plot of first two fuzzy canonical axes for k = 4 classes and fuzziness exponent () = 1.4. 115

An error budget for soil salinity mapping using different ancillary data Fig.5.2.1.1 A) Air-photo and B) infrastructure of the Bourke study area and C) locations of the 126 EM38/34, -ray spectrometry and DEM surveys and soil sampling sites. Fig. 5.2.4.1 Flowchart of E-BLUP and error budget calculation. 130 Fig. 5.3.2.1 Variograms of the raw ancillary data and including; a) elevation; b) thorium (Th) and 135 c) logEM38v. Fig. 5.3.3.1 Spatial distribution of kriged A) elevation (m), B) thorium (Th) (ppm) and C) 137 logEM38v (mS/m) across the Bourke study area. Fig. 5.3.5.1 A) Spatial distribution of predicted ECe (dS/m) using E-BLUP; B) predicted ECe 144 (dS/m) vs. measured ECe (dS/m) using leave-one-out cross-validation. Fig.5.3.6.1 Spatial distributions of the standard deviation of predicted ECe (dS/m) due to A) model 149 error, B) input error and C) combined error across Bourke study area. Fig.5.3.7.1 Spatial distributions of the standard deviation of predicted ECe (dS/m) due to A) 152 elevation error, B) -ray error and C) EM error across Bourke study area.

List of Tables

Literature Review Table 2.2.1.3.1 Average value radioelement content of Australian rocks and soils (modified from 23 Dickson and Scott, 1997). Table 2.2.2.1 Some of the studies using -ray spectrometry in digital soil mapping. 27 Table 2.3.2.1.1 Parameters of the common EM instrument 36 Table 2.3.2.3.1 Parameters of DUALEM sensors (DUALEM-421S Manual, 2008). 39 Table 2.3.3.1.1 Some of the studies on salinity mapping using EM instrument. 41 Table 2.3.3.1.2 Regression relationships between measured soil paste ECe and recorded meter 43 readings of apparent conductivity (ECaW and ECa) ( after Cameron et al. 1981). Table 2.3.3.2.1 Some of the studies on texture mapping using EM instrument. 44 Table 2.3.3.3.1 Some of the studies on mapping soil moisture using EM instrument. 48 Table 2.3.3.7.1 Studies on other soil properties using EM instruments. 55

Identifying soil landscape units at the district scale by numerically clustering remote and proximal sensed data Table 4.3.1.1 Pearson correlation coefficient between remote and proximally sensed ancillary 101 data along the survey transects and sampling points. Note: values are shown for potassium (K- cps), uranium (U-cps), thorium (Th-cps), total count (TC-cps) and soil apparent electrical conductivity (ECa – mS/m) for EM38h and EM38v. Table 4.3.1.2 Summary statistics of soil physical and chemical properties collected at sampling 101 points. Note: values shown for Clay (%), CEC (cmol(+)/kg), ECe (dS/m), pH, respectively. All values rounded to nearest integer.

5

Table 4.3.4.1 Mahalanobis centroid values of remotely and proximally sensed ancillary data 107 clustered using FKM and for classes k = 3, 4 and 5. Note: centroids shown for potassium (K-cps), uranium (U-cps), thorium (Th-cps), total count (TC-cps) and soil apparent electrical conductivity (ECa – mS/m) for EM38h (h) and EM38v (v). 2 Table 4.3.5.1 Mean Squared Prediction Errors (MSPE) ( p,C ) for topsoil (0-0.3 m) and subsoil 110 (0.9-1.2 m) for the measured soil properties used in the REML analysis. Different rows represent the different numbers of classes as categorized by the fuzzy k-means analyses from k = 2-6. Bold highlighted values indicate the lowest values of each row (i.e. lowest MSPE for each measured soil property).

An error budget for soil salinity mapping using different ancillary data Table 5.3.1.1 Summary statistics of soil samples and reduced ancillary data. 133 Table 5.3.4.1 Summary statistics of kriged ancillary data and Pearson’s r between soil ECe and 139 kriged ancillary data. Table 5.3.4.2 Summary statistics of selected LMM comprising elevation, -ray spectrometry and 140 ECa data. Table 5.3.5.1 Summary statistics of selected LMM comprising elevation and -ray spectrometry 145 data. Table 5.3.6.1 Calculated error budget for different sources of error. 146

6

1 Introduction In the Murray-Darling Basin, one of the most productive agricultural soil types is the

Vertosol (Weaver et al. 2005). It has been used to produce cereal crops by dryland farming methods and it has also been extensively developed for irrigated agricultural production.

However, there have been increasing instances of decline in productivity as a result of soil salinization (Council, M.D.B.M., 1999). In one area, that is the Bourke Irrigation District, this is because of the large store of natural salts deep in the regolith and associated with a saline

Cretaceous Marine Mudstone (Williams, 1993). Owing to the development of the area to extensive irrigated cotton production, large water storage reservoirs dot the landscape and are connected to the Darling River by an extensive network of supply and conveyance channels.

Unfortunately, these have been constructed without regard to the location of prior stream or buried paleochannels. Because these near-surface physiographic (prior streams) and sub-surface stratigraphic (paleochannels) units can be spatially and laterally continuous, they may act as significant hydrological features, owing to their high porosity and permeability. Therefore, each offers a pathway for deep drainage of the water to interact with primary salts.

It is important, therefore, that areas characterised by these Vertosols are mapped intensively to understand the long-term sustainability issues and spatial variability in soil types.

A major problem is that to characterise large areas, such as the Bourke Irrigation district in northwest New South Wales which has been affected by shallow water tables and soil salinity, a great deal of soil data is required and including on physical (e.g. clay) and chemical properties

(e.g. salinity). A major limitation for soil and/or landscape mapping is a lack of data at sufficient scale and resolution that is required to make meaningful and useful management decisions

(McBratney 2003). This limitation comes from the fact that costs associated with soil sampling and laboratory analysis are high (i.e. ~ $100 per sample). Webster and Oliver (1992) suggest that approximately 100 sample points are required to estimate a spatial statistical model. In

7 recent , applying pedometics and/or digital soil mapping techniques have significantly improved sampling efficiencies and simultaneously reduced costs of soil mapping.

By definition, digital soil mapping or is the development of a numerical or statistical model of the relationship between environmental variables and soil properties, which is then applied to a geographic data base to create a predictive map (Scull

2003). Substantial progress has been made in this field in latest years which allowed soil scientists to predict soil properties at a scale and resolution which was not previously possible.

This has been obtained by the growth in accessibility of remote and proximal ancillary data such as gamma-ray (-ray) spectrometry and electromagnetic (EM) induction as well as the growth in computer-assisted statistical science. Nevertheless, advanced study is required to develop innovative approaches to fit these emerging fields to soil and landscape clustering and accordingly sustainable land-use management.

The objectives of this work are to address some research and knowledge gaps in digital soil mapping and show the application of -ray spectrometry and EM data for identifying the soil types and mapping the physical and chemical soil properties using fuzzy k-means (FKM) and Linear Mixed Model (LMM) with Residual Maximum Likelihood (REML). This information can then be used to explore how these two sources of ancillary data aid the study of prediction efficiency of soil management classes and soil properties variation at the district scale. In the following, the uncertainty caused by using different ancillary data is studied for mapping salinity.

The thesis is structured as follows; Chapter 2 presents an overview of the literature and previous studies in digital soil mapping procedures. It is described with the various components explored in detail and including the different sources of ancillary data, the different methods to mathematically and statistically analyse and relate the ancillary data with each other and relative

8 to the soil properties and types that might be mapped. Chapter 3 presents an overview of the biophysical setting of the study area, Bourke Irrigation District.

Chapter 4 aims to use remote (airborne -ray spectrometry) and proximal sensed EM38 data acquired across the geologically and geomorphological diverse aeolian and alluvial clay plain of the Bourke Irrigation district, to identify soil landscape units (Northcote, 1966).

Specifically, it will discern the merit of using FKM analysis to identify soil landscape units using both -ray (i.e. K, Th, U radioelements and TC) and ECa (i.e. EM38 in horizontal

[EM38h] and vertical [EM38v] modes of operation) data and how the units are tested.

In Chapter 5, an error budget procedure is developed to quantify the relative contributions made by model, input (for all the ancillary data) and particularly the individual covariate (for each of the ancillary data) error when combining remotely sensed (DEM) or -ray spectrometry data and/or using ECa data collected using EM38 or EM34 data to map soil ECe. The spatial distribution of different sources of errors will also be studied by using REML analysis across the predominantly irrigated cotton growing area of the Bourke district of Darling River Valley in New South Wales, Australia.

Chapter 6 contains the conclusions of the thesis and includes potential future research directions of digital soil mapping using ancillary data such as -ray spectrometry and electromagnetic induction data. Some suggestions are made to improve the accuracy of the results.

9

1.1 References Council, M. D. B. M., 1999. The salinity audit of the Murray-Darling Basin: a 100-year perspective.

McBratney, A. B., Santos, M. M., & Minasny, B. 2003. On digital soil mapping. Geoderma, 117(1), 3-52.

Scull, P., Franklin, J., Chadwick, O. A., & McArthur, D. 2003. Predictive soil mapping: a review. Progress in Physical , 27(2), 171-197.

Weaver, T. B., Hulugalle, N. R., & Ghadiri, H. (2005). Comparing deep drainage estimated with transient and steady state assumptions in irrigated vertisols. Irrigation Science, 23(4), 183-191.

Webster, R., & Oliver, M. A. 1992. Sample adequately to estimate variograms of soil properties. Journal of , 43(1), 177-192.

Williams, R.M., 1993. Saline Inflows to the Darling River at 'Glen Villa"near Bourke, NSW. Department of Water Resources.

10

2 Literature Review

2.1 Introduction Knowledge about the spatial distribution of soil properties at various depths is required for effective soil use and management. This is particularly the case in highly productive agricultural areas and including the Murray-Darling Basin of Australia which is considered

Australia’s bread basket. This is especially the case in irrigated areas such as the Bourke

Irrigation Australia, where soil salinization is jeopardising long-term sustainability. The problem is that to acquire soil property information at a level useful to soil use and management, extensive soil sampling and laboratory analysis is required which is -consuming and labour intensive. This is, unfortunately, expensive when considering the fact that information about soil properties such as the electrical conductivity of a saturated soil paste extract (ECe – dS/m) and cation exchange capacity (CEC – cm(+)/kg) are required. This is because the former provides information about the current status of soil salinity and the latter an indication of the fertility and structural resilience. Added to this is information about soil pH and clay content, the scope of the expense and the problem become apparent.

To value-add to limited soil property data that can be collected, the use of ancillary data is increasingly utilised. This includes both proximal and remotely sensed data which are collected using either passive or active sensors. One of the most commonly used sensors to measure and map ECe is the suite of electromagnetic (EM) induction sensors. This is because the instruments measure soil apparent electrical conductivity (ECa – mS/m) which has been shown again to be highly correlated with ECe (Huang et al. 2015).

Another issue is the fact that in some instance soil properties are not directly correlated with ancillary data. In this case, a different approach might be to look to see if soil types can be mapped, and use ancillary data as surrogates for the soil property data. One way to do this is to

11 use clustering algorithms to generate classes derived from the ancillary data and then test whether those classes are statistically different in terms of measured soil properties. For example, this was the approach used by Altdorff and Dietrich (2012), Van Meirvenne et al.

(2013) and Huang et al. (2014a) to identify management zones on the field scale.

In this literature review the DSM procedure is described with the various components explored in detail and including the different sources of ancillary data, the different methods to mathematically and statistically analyse and relate the ancillary data with each other and relative to the soil properties and types that might be mapped. At the end various research gaps are identified and relative to the application of DSM methods on the field scale to measure and map individual soil properties and to map and identify soil landscape units at the district scale and in the Bourke Irrigation District.

2.1.1 Ancillary data used in DSM One of the main components of DSM is ancillary data; Ancillary data is usually addressed to non-conventional data. Ancillary data is used to assist DSM joint with conventional data along with computational methods for developing a digital . In general, ancillary data is divided into proximal sensed and remotely sensed data.

Proximal sensed data is referred to those types of data which are obtained from a range of instruments used to measure the soil properties when the sensor is located only a short distance (i.e. within 2 m) of the soil surface. These instruments can rely on the sensors which actively measure the apparent electromagnetic (EM) fields from a certain wavelength of the soil components or passively measure the electromagnetic radiations.

Contrariwise, in remote sensing the aerial sensors passively measure the emitted electromagnetic radiations of the soil and/or landscape components from an aircraft (e.g.

AVIRIS, GeoSAR) or satellites (e.g. Landsat and RADARSAT2). The Handbook of

12

Agricultural Geophysics (Allred et al. 2008) has reviewed several geophysical methods for gathering the ancillary data.

The EMI instruments that operate in less than 104 Hz of EM waves are considered as low-frequency instruments. Fig. 2.1.1.1 shows that EMI instruments work at a low frequency of

EM waves. Additionally, some other instruments operate at slightly higher frequencies between

105Hz to 1010Hz. These instruments are described as radar-based proximal sensors, which include ground penetration radar (GPR) and time domain reflectometry (TDR) instruments.

Furthermore, other instruments are being developed to operate in the mid-wavelength infrared

(MIR) (4 × 1013–1 × 1014 Hz) and near-infrared (NIR) (2 × 1014 – 4 × 1014 Hz).

Various remote sensing platforms are being employed with the aim of collecting information about soil surface and crop growth. Landsat TM, MODIS, SPOT, AVHRR, Ikonos and airborne -radiometrics are the main remote sensing platforms. The American earth observation satellite Landsat 8 is one of the popular platforms which consist of 11 bands. These bands are functioning at a range of frequencies from 2 × 1013 Hz to 7 × 1014 Hz. The pixel sizes are 15 m, 30 m and 100 m for Multispectral Band (Band 1-7, Band 9), Panchromatic Band

(Band 8), and Thermal Band (Band 10-11), correspondingly. Lobell et al. (2010) used satellite imagery (e.g. MODIS) for DSM of salinity in district scale for the first time. This study improved the use of remote imagery for salinity assessment significantly. In field scale use of ground based -ray spectrometry systems produce a better resolution. The instrument can be mounted on vehicles or carried by the surveyor. Airborne-EM (AEM) systems are another type of remote sensing system. Airborne-EM systems operate at very low frequency (10-102 Hz) and have a broad application in geological survey mapping groundwater, regolith features and mineral deposits at a depth of 100 m to 300 m.

13

Fig. 2.1.1.1 The electromagnetic spectrum used for soil and proximal sensing (after McBratney et al. 2003).

2.1.2 Remote sensing systems DSM has been benefited from remote sensing methods in the recent decades. This is the result of the vast improvement in the space-borne and airborne -ray spectrometry and aero- electromagnetic systems. However, it is still powerless to map the spatial and temporal variation of the soil due to its indirect determination (ie. absorption bands and other auxiliary information) approach as well as limited penetration depth of the EM waves. For example, in regards to the mapping of salinity several factors limit the application of remote sensing. These factors are described as (a) lack of specific absorption bands of some salt type (e.g. halite); (b) low spectral resolution; (c) vertical, spatial, and temporal variability of salinity in the surface and substratum; and, (d) the presence of other soil chromophores (e.g. structural crusts) (Mougenot et al. 1993,

14

Metternicht and Zinck 2003). Despite all these limitations remote sensing methods have been recently used in different areas of DSM such as salinity (Lobell et al., 2010), clay and sand fractions (Singhroy et al., 2003), moisture (Wagner et al., 2007), and organic carbon

(Bartholomeus et al., 2008).

2.1.3 Proximal sensing systems Proximal sensing systems appear to be more capable of collecting spatial data from different soil depths in comparison with the remote sensing systems. Moreover, they are cost- effective, easy-to-operate and less laborious to collect than soil samples (Adamchuk and

Viscarra Rossel, 2010). Proximal sensing systems are based on measurement of a different component of the soil such as electrical, mechanical, electrochemical, acoustic, pneumatic, optical, electromagnetic and radiometric components (Fig. 2.1.3.1).

Fig. 2.1.3.1 Classification of on-the-go proximal soil sensors (Adamchuk and Viscarra Rossel, 2010).

Electrical methods are used to measure soil apparent electrical resistivity () to reflect soil properties such as salinity and clay content. In these methods, electrical electrodes should be inserted into the soil. Analogously, mechanical methods require flaps to be physically used in contact with soil to measure soil mechanical resistance (Sharifi et al., 2007). Electromechanical methods are based on detection of the activity of specific ions including hydrogen, nitrate or

15 potassium (Adamchuk and Viscarra Rossel, 2010). In these methods, use of an ion-selective electrode (ISE) or an ion-selective field effect transistor (ISFET) is required for the measurements (especially of pH). All the three above-mentioned methods require direct contact of sensors. Therefore, they are time-consuming, destructive and unable to detect subsoil conditions.

Acoustic sensors are used for measuring the soil texture and bulk density. This has been measured by characterizing the change in the level of noise caused by a tool’s interaction with soil particles. For the same purpose, soil-air permeability has been measured by pneumatic sensors. In this method a given volume of air forces into a certain depth of soil profile. This amount of air will be compared to several soil properties, such as and compaction.

Future research is required to establish the relationship between sensor output and the physical state of the soil as the information at the present time is deficient (Adamchuk and Viscarra

Rossel, 2010).

Similar to electrical and mechanical methods, cost associated with optical (i.e. reflectance spectroscopy) methods (e.g. NIRSystems model 6500) is high due to the extensive calibrations. In order to quality control at samples with different characteristics (e.g. soil types, crop rotations) it is required to collect more samples. Consequently, this increases the cost

(Cozzolino and Morón, 2006).

Amongst all the proximal sensors with all the advantages and disadvantages, EM induction methods, and ground-based -ray spectrometry seem to be more capable of gaining spatial data from soil profile as they are fairly easier and cheaper to operate. Particularly, the advantage of -ray spectrometry originates from the nature of -rays in which most of the radiation is derived from the top 0.3-0.45 m of the soil profile. Radioelements of potassium (K), uranium (U) and thorium (Th) and total counts (TC) has been used to map the various soil properties such as soil texture (Pracilio et al., 2006; Van Der Klooster et al., 2011; Petersen et

16 al., 2012; Spadoni and Voltaggio, 2013). However, the deeper depth of survey of EMI systems

(e.g. 0-6 m deep for DUALEM-421) allow these systems to be used to map the subsoil properties such as soil salinity (e.g. Lesch et al., 2005), depth to (e.g. Doolittle et al.,

1994; Jung et al., 2006), soil moisture (e.g. Kachanoski et al., 1988; Sheet and Hendrickx, 1995;

Robinson et al., 2012), CEC (e.g. Triantafilis et al., 2009b), deep drainage (e.g. Woodforth et al.,

2012) and soil phosphorus maximum sorption capacity (e.g. Quenum et al., 2012).

Ground-based-ray spectrometry and EM instruments have been frequently used to study soil types and properties at field scale due to their non-contact nature. -ray spectrometry provides information from topsoil and EM instruments provide information from topsoil and subsoil. In this literature review, the theory of operation of these methods and the description of instrumentation that is currently available is described. Furthermore, the practical approaches which researchers have used to generate DSM of either , soil classes or management zones and individual soil properties are discussed in terms of various computational and statistical methods of analysis.

2.2 Review of -ray spectrometry

2.2.1 The -ray theory Radiometrics is the measure of natural radiation in earth's surface in order to recognize the distribution of certain soil and rock types. Radiometrics is also known as -ray spectrometry.

Geologists/geophysicists routinely use -ray spectrometry surveys as mapping tools to identify geological variations. In recent years, the use of ray spectrometry in the areas of soil science is increasing because it has a worthy application in measuring the distribution and abundance of various radio-elements of different energies, which are emitted during the decay of some naturally occurring elements (Horsfall, 1997). This includes, potassium (K)-40, bismuth-214 from the uranium (U)-238 decay series; and thorium (Th) -232 from the Th-232 decay series

17

(Wilford, 2009) are the elements with emissions of γ-ray of sufficient energy and intensity to be detected at altitudes flown by survey aircraft. Therefore, they have major contribution in geological and soil mapping surveys (Pickup and Marks, 2000).

To understand -ray spectrometry, the following section provides a brief introduction to what is an isotope, basic radioactivity, -radiation, -ray spectrometer and how γ-ray spectrometry works. Applications are also provided as examples.

2.2.1.1 Isotopes Isotopes are atoms of an element which have the same atomic number but different mass numbers. That is, they have a same number of protons but different number of neutrons present in their nuclei. For example, Hydrogen has three isotopes which are hydrogen, deuterium and tritium respectively (Fig. 2.2.1.1.1).

Hydrogen – 1 proton and 0 neutrons.

Deuterium – 1 proton and 1 neutron

Tritium – 1 proton and 2 neutrons.

Fig. 2.2.1.1.1 The three most stable isotopes of hydrogen.

18

The lighter elements generally have neutrons and protons in equal numbers whereas the heavier elements tend to have a higher number of neutrons in order to stabilize the nuclear forces between nucleons in the densely packed nuclei. However, isotopes of an element show similar chemical properties but have different weights.

Also, certain isotopes of an element tend to be unstable in nature and show radioactivity.

These radioactive isotopes disintegrate through the emission of energetic ionising radiations to form more stable daughter nuclei. The half – lives of different radioactive isotopes of an element are different and they decay through distinct decay paths. For example, the most common natural isotopes of uranium which are potasium – 238 and uranium – 235 have half lives 4.47 billion years and 704 million years respectively and they both decay to produce lead isotopes as the end products (Pb - 206 and Pb – 207).

2.2.1.2 Basic radioactivity Radioactivity or radioactive decay is the process by which the nucleus of an unstable atom loses energy by the emission of alpha particles, beta particles, and/or g – radiations in order to attain stability. There are many isotopes of different elements which are unstable in nature and show radioactivity.

The decay rate of a radioactive element is measured by its half-life and different isotopes of different elements have different decay rates which act as an index to determine the ageing of the rocks and soils. This is known as radioactive dating. Also, the different paths by which a radioactive isotope decay leads to the formation another chemical element which may be stable or radioactive in nature.

Alpha decay occurs when an alpha particle is emitted by the radioactive element. An alpha particle constitutes of 2 protons and 2 neutrons and is similar to a Helium nucleus. It generally occurs in heavier isotopes which are radioactive. As an alpha particle is released, 2 protons and 2 neutrons are lost by the original radioactive atom and an atom of a new element is

19 formed which has an atomic number of 2 and mass number of 4 less than the original atom. The best example for an alpha decay is U 238 to Th 234 where the radioactive U atom emits alpha particle to form Th atom. Sometimes the daughter nuclei produced may be radioactive in nature and decays process may still be continued by following one or other paths.

Beta decay can occur as 2 processes which are beta+ and beta- . Beta particles are electrons or positrons (anti-electrons). Beta decay occurs when an atom has too many protons or neutrons in its nucleus. Beta- emission occurs when a neutron gets transformed into a proton, an

− electron and an antineutrino (n→ p + e + νe). This causes the atomic number to rise by one and a new daughter element is produced.

It generally occurs in the by-products of fission reactions in nuclear reactors as they are likely to have an excess number of neutrons. Beta+ decay occurs when a proton in a radioactive

− isotope is converted into a neutron, positron and a neutrino (p→ n + e + νe). This causes the atomic number of the daughter element formed to decrease by 1 and the mass number remains constant.

2.2.1.3 The -rays The -radiation of -ray occurs when a radioactive nuclei undergo alpha or beta decay and are left in an over excited state with excess energy. This is similar to an electron in an excited state releasing a photon to lower its energy and attain stability. The -rays are the most penetrating of all three radiations and can pass through a few centimetres of lead comfortably.

Beta particles are the second most penetrating and are absorbed by a few centimetres of aluminium, whereas alpha particles do not pass through the topsoil or few centimetres of air above it. Gamma radiation is a very useful tool in the radiometric analysis of soil because of its higher degree of penetration, and because it can be comfortably measured at certain heights from the soil surface.

20

Fig. 2.2.1.3.1 Components of a -ray spectrometer and including the detector and photomultiplier chamber and counting device (i.e. spectrometer).

An -ray spectrometer has the capability to distinguish the -rays of different energies.

The three components of an -ray spectrometer are a detector,a photomultiplier and a spectrometer. The detector contains a transparent crystal of sodium iodide (NaI) (Fig. 2.2.1.3.1).

The -ray is emitted from the above-mentioned elements, absorbs in the NaI crystal, and gives rise to a little pulse of light. This light is received by a photomultiplier tube. The photomultiplier tube measures emitted -rays by converting the light flash to a voltage of a proportional intensity. This voltage is separated into a number of magnitude-dependent classes in the counting device. These classes represent the energy spectrum of -rays based on selectivity record windows corresponding to energy levels.

The energy windows recommended by the International Atomic Energy Agency are

1.37–1.57 MeV for K, 1.66–1.86 MeV for U, 2.41–2.81 MeV for Th and 0.41–2.81 MeV for the total counts (TC) (Fig. 2.2.1.3.2). Background cosmic radiation (i.e. >3 MeV), Compton

21 scattering, atmospheric radiation and survey height can affect the results of the collected data.

Thus, the data correction is usually required. The measuring unit for different forms of -ray varies. The common units are counts per second for the total count (TC) and parts per million

(ppm) for K, U, and Th. However, K is measured into values of a percent (%) as well.

Fig. 2.2.1.3.2 Recommended -ray spectrum windows (Grasty et al., 1991).

It is worth noting, the penetration depth of -ray is up to 0.4 m (Jaques et al., 1997) from dry soil with a density of 1.5 g/cm3 (Grasty, 1975). Based on the relative concentration of radioelements the type of soils and rocks can be identified. However, the radiometric values vary a lot in different types of soils and rocks within small regions. The average radioelements of the soils and rocks of Australia are shown in table 2.2.1.3.1.

22

Table 2.2.1.3.1 Average value radioelement content of Australian rocks and parent matterials of soils (modified from Dickson and Scott, 1997). Rock type Rock Soil K (%) U (ppm) Th (ppm) K U Th (%) (ppm) (ppm) Intrusives granitoids 2.4 3.3 16 2.1 2.7 13 gneissic rock 2.4 2.5 15 1.3 1.3 12 pegmatite 3.7 0.7 2 quartz-feldspar 2.9 1.7 13 porphyry Intermediate 2.7 0.8 2.4 1.6 1.9 5.6 intrusives mafic intrusives 0.4 0.3 1.2 Extrusives felsic volcanics 3.7 2.4 17 2.4 2.1 13 intermediate volcanics 2.7 2.3 9 1.9 2.1 10 low-K andesites 0.8 1.6 5 1.1 1.3 5 mafic volcanics 0.9 0.7 3.0 0.7 1.6 7.9 Ultramafic volcanics 0.4 0.6 1.2 0.6 2.0 6 Sedimentary rocks Archaean shales 0.9 0.9 2.7 0.8 1.2 3 other shales 2.6 2.6 19 1.5 2.3 13 carbonates 0.2 1.6 1.4

It can be interpreted from the table that K (%), U (ppm) and Th (ppm) are high in felsic rocks and soil derived from them. Generally, the -ray concentrations of soils are intensely correlated to their parent rocks. For instance, radioelement contents are relatively high both in granitic rocks (i.e. K – 2.4 %; U – 3.3 ppm and Th – 16 ppm) and granite soil (i.e. K – 2.1 %; U

– 2.7 ppm and Th – 13 ppm). Likewise, low radio element contents are documented both for mafic extrusive rocks (i.e. K – 0.4 %; U – 4 ppm and Th – 1.2 ppm) and soils of mafic parent material (i.e. K – 0.6 %; U – 2.0 ppm and Th – 6 ppm).

23

Fig. 2.2.1.3.3 Factors affecting -ray response in regolith (Wilford et al., 1997).

As shown in Fig. 2.2.1.3.3 the detected -rays by aircraft is a factor of landscape and topography. For instance, due to the weathering of K, U and Th from the parent rocks in upper landscape positions, high radiometric values could be recorded in upper landscape positions owing to the shallower soil and therefore exposure to higher radioelement concentrations of the parent material beneath. This would be the case in an area underlain by granite rocks. As a result of weathering and erosion of K, U, and Th from these upper areas, the lower lying areas may accumulate these radioelements and similarly lead to higher radioelement counts at the bottom of the slope. For example, a primary mineral like orthoclase (i.e. K-feldspar (6KAlSi3O8)) which is found in granite is subject to chemical weathering by hydrolysis in high rainfall areas. The products of this chemical weathering are illite (KAl2(AlSi3)O10(OH)2), orthosilicates (SiO4) and potash (KOH), which are mobile and can be translocated down the slope. Since illite is very resistant to any further chemical weathering, it may concentrate in the lower lying area and the radioelement K may be larger. As a result, for the areas in lower landscape positions the radioelements K may be higher due to the accumulation of higher amounts of K. Fig. 2.2.1.3.4 shows a ground-based -ray spectrometer being towed on a PVC sled.

24

Fig. 2.2.1.3.4 Proximal -ray spectrometer being towed on a PVC sled.

2.2.1.4 Airborne -ray spectrometry survey As discussed above -rays can comfortably penetrate large distances which make it advantageous for detection from certain heights. Airborne -ray spectrometry is widely used in different fields of study as they provide background maps of radiation. It is used by geologists to map the concentrations of K, U and Th in rocks and soils which can improve geological mapping and help to locate mineral deposits. For an environmental physicists background, radiation maps provide a means to measure man-made contamination and risk to the health posed by the contamination.

Airborne -ray data are collected by helicopter or aeroplane flying at an altitude of 60 –

100 m above the surface. The -ray spectrometer device is mounted on an aeroplane and the

Sodium Iodide (NaI) scintillation crystal measures the -rays emanating from the subsurface.

Photomultiplier tubes, present in the -ray spectrometer, record and amplify the ray induced

25 signal (Minty,1997). The spatial resolution of -ray data depends on the spacing between transects at which the airborne survey is carried out. In Australia, regional airborne geophysical surveys or reconnaissance surveys are carried out at 200 - 400 m line spacing, whereas detailed surveys are carried out at 50 – 100 m line spacing. The spacing between the transects at which the airborne surveys are carried out is generally a compromise between data resolution and acquisition costs.

2.2.2 The -ray applications Passive remote sensing is relatively cheap and easy to obtain rapidly across large areas.

Gamma ray surveys across earth's surface tell us about the distribution of certain soil and rock types. Geologists/geophysicists routinely use radiometric surveys as mapping tools to identify where certain rock types change. Use of radiometrics, which is also known as  -ray spectrometry, is increasing for applications in study of geological survey (Galbraith and

Saunders, 1983), mineral exploration (Foote and Humphrey, 1976), radioelement contaminant evaluation (Roca et al., 1989) and space exploration (Metzger and Parker, 1979).

Use of  -ray surveys in soil science has been carried out in many countries rich in minerals. Australia, China, UK, Denmark and the USA are some of the countries which use this technology in soil science and have a lot of publications in this area. The topics are mostly about soil texture-related properties, identifying differences in parent materials and soil types and most recently in identifying soil management zones (Table 2.2.2.1).

26

Table 2.2.2.1 Some of the studies using proximal or aerial data in digital soil mapping. Property Authors Texture class Schwarzer and Adams (1973); Paine et al. (1998); Becegato and Ferreira (2005); Pracilio et al. (2006); Viscarra Rossel et al. (2007); Beckett (2008); Wilford (2009); Vukašinović et al. (2010); Van Der Klooster et al. (2011); Buchanan et al. (2012); Petersen et al. (2012); Piikki et al. (2013); Snellen et al. (2013); Spadoni and Voltaggio (2013) Parent materials Schwarzer and Adams (1973); Cook et al. (1996); Wilford et al. and soil types (1997); Wilford and Minty (2006); McCafferty and Van Gosen (2009); Lacoste et al. (2011); Schuler et al. (2011); Lemercier et al. (2012); Guimaraes et al. (2013) Management Altdorff and Dietrich (2012); Castrignanò et al. (2012); Triantafilis et classes al. (2013); Popp et al. (2013); Van Meirvenne et al. (2013) Potassium Wong and Harper (1999); Haskard et al. (2010); Moisture Grasty (1997); Cassiani et al. (2012) Organic carbon Dierke and Werban (2013 Salinity Wilford et al. (2001) Weathering Wilford (1995) Soil thickness Barišić (1996) Bulk density Costa et al. (2013) Acidity Wong et al. (2008) Compaction Naderi-Boldaji et al. (2013) Land suitability Hyvönen et al. (2003) Accuracy Dickson (2004) Data fusion Ma et al. (2010)

2.2.2.1 Soil Texture The -ray spectrometry has been used for mapping soil texture class. Schwarzer and

Adams used -ray spectrometry for soil mapping for the first time in 1973. Their study was based on this assumption that most of the thorium in the soil was contained in clay/shale and because this Th is more resistant to discharge and redistribution than K and U, they mapped Th as a surrogate of clay. Later, Paine et al. (1998) used a similar concept to map clay content in the arid Trans-Pecos region of West Texas, USA. However, they used the correlation between -ray total counts and clay content. In their study, they found that different calibration relationships should be used for the estimation to be made with any confidence because the correlation between -ray total counts and clay content was not the same at different sites. Viscarra Rossel et al. (2007) used a similar but more complex methodology to map clay and sand content across two study fields located in two geographically and physiographically different sites at Nowley and Stanleyville, NSW, Australia (Fig. 2.2.2.1.1).

27

Fig. 2.2.2.1.1 Bagging-PLSR maps of a) clay and b) coarse sand content and for the 15–50 cm soil layer at Nowley (a) and Stanleyville (b) after Viscarra Rossel et al. (2007).

Another study conducted by Pracilio et al. (2006) at the farm scale suggested a strong correlation was found between clay content and U, and ratios of Th:U, K and K:Th (R2 = 0.70).

Spadoni and Voltaggio (2013) used the same approach to map spatial distribution of sand a larger scale. They collected 289 (3.2 samples km-2) soil samples from the topsoil across a floodplain in Central Italy and 1906 of γ-ray emission measurements (TC) at a density of 21.1 measurements/km2. The results showed a significant inverse correlation existed. Therefore, they used this relationship to map the spatial distribution of sand content by ordinary kriging.

28

2.2.2.2 Salinity (ECe) Soil salinization is mostly a human-made process despite the fact that soil salinity primarily is a natural property of the soil. A lot of different factors contribute to the determination of susceptibility of a land to human-induced salinization. The origin of salts is the main factor that suggests the geographic distribution of the natural source areas and the further salt dissemination. Salts discharge into soil from different resource; Include release of salts from weathering of primary minerals, release of salts from saline sedimentary rocks, release of salts from volcanic activity, deposition of salts in coastal marches and estuaries, deposition of salts in inland depressions and aeolian influx of salts are the main source of discharge of the salts in the soils. This type of information can show where and how the contribution of -ray spectrometry is required. It can also enlighten the nature, variety of crystallization patterns, mineralogy, modes of occurrence in the soil surface, characteristic figures that they usually form at the soil surface and the physical structure.

Wilford et al. (2001) modelled -ray spectrometry data combined with DEM-derived indices in an area of 180 km2 located in the east of the township of Bethungra, NSW, Australia.

They showed it is an effective technique for predicting salt stores and salt outbreaks in erosional landscapes. Thematic maps quickly generated by this approach can be used directly in developing farm management plans and in prioritising areas for remedial action.

2.2.2.3 Parent materials and soil types Using -ray spectrometry is a cost-effective technique, which can be adequately used to map different soil types and parent materials from which the soil is derived. The ability of -ray spectrometer data to discern the spatial variation of soil is shown by Cook et al. (1996) where ground and airborne measurements of -ray emissions were simultaneously compared over a catchment in southwestern Australia. Their studies revealed a significant correlation between

29 variations in the -ray counts and the distribution of soil forming materials over the landscape.

Gamma ray spectrometer data were used to differentiate between newly formed material and highly weathered residuum in granitic outcrops. Also, the -ray spectrometer data clearly demarcates the origin of the parent material from which the soil is formed i.e Granitic, doleritic or lateritic rock types.

Similar investigations have been done by Wong and Harper (1999) where they employed

-ray spectrometry to predict the amount of potassium (K) present. They found that strong relationships exists between total K, plant-available K as well as the clay content which accounts for the total on-ground counts of the -rays derived from 40K radionuclide and the establishment of these relationships can be helpful to devise a cost-effective and easy to operate mapping technique relative to sampling and analysing the whole area (Fig. 2.2.2.3.1).

30

Fig. 2.2.2.3.1 Bivariate comparisons between ground-measured γ-ray counts from 40K and selected properties of the topsoil (0–10 cm) layer (Wong and Harper, 1999).

31

Also, Schuler et al. (2011) showed that ground based -ray spectrometry surveys can be effectively used to differentiate between and in a recent study conducted in Bor

Krai karst area in northern Thailand. Alisols and Acrisols have different clay mineral composition (K containing illite vs K-free kaolinite) and hence showed significant differences in the -ray spectrum emitted by them especially for K and Th signatures. The usage of -ray spectrometer surveys is widely increasing in the field of soil sciences as it is an economical and feasible technique.

2.2.2.4 Soil management zones and classes Using -ray spectrometry is a profitable method for identifying soil use management zones.

The -ray spectrometry can be used individually (Triantafilis et al., 2013) or in combination with

EMI data (Castrignanò et al., 2012) and/or other ancillary data (Popp et al., 2013) for this purpose.

As discussed in the above section, -ray spectrometry surveys can be independently used to classify different soil types and effectively the soil management classes. Recent studies performed by Triantafilis et al. (2013) showed that consistent results can be produced for soil mapping using -ray spectrometer surveys where the research was conducted at a district and a sub-catchment level. In their study, clustering of K, U, Th and TC produced 11 classes which closely reflected the geological and geomorphological interpretations of an eroded landscape, alluvial lands and dust-mantled alluvial lands at the district level. The results produced also showed a high level of relationship with broad soil mapping units at the sub-catchment level and also disclosed the minute amount of differences as compared to the traditional mapping techniques (Fig 2.2.2.4.1).

32

Fig. 2.2.2.4.1 a) Soil mapping units of Edgeroi district and Narrabri sub-catchment with respect to the lower Namoi valley of New South Wales, Australia (after Northcote et al., 1965); b) Soil classes associated with the eroded landscape in Narrabri sub-catchment (after Triantafilis et al., 2013).

2.2.2.5 Other soil properties The -ray spectrometry surveys have acquired a great fame over a period of years because of the ease of operation and since then the usage has widely increased across the globe in different fields of study especially soil sciences. Apart from using the technique to map the soil potassium content (Wong and Harper, 1999; Haskard et al., 2010), it was also used to discern various properties such as moisture differences (Grasty, 1997; Cassiani et al., 2012), organic carbon (Dierke and Werban, 2013), salinity (Wilford et al., 2001), soil thickness (Barišić, 1996), bulk density (Costa et al., 2013), acidity (Wong et al., 2008), (Naderi-Boldaji et al., 2013), land suitability (Hyvönen et al., 2003) and other properties.

2.3 Review of electromagnetic methods

2.3.1 Theory of electromagnetic induction Electromagnetic induction works on the basis of measurement of the change in mutual impedance between a pair of coils and are governed by Maxwell’s laws. It is one of the most recognized and widely used methods to evaluate the bulk electrical conductivity (ECa) of the

33 soil and is measured in (dS m-1). The elements of a ground-based EM instrument are shown in

Fig. 2.3.1.1.

Fig. 2.3.1.1 Schematic principal of the functioning of an EMI sensor, specifically the DUALEM-1 in which one receiving coil is coplanar (Rx1) and the other is perpendicular (Rx2) to the transmitting coil (Tx) (Visconti and de Paz , 2016).

The instrument basically consists of a transmitter and receiver coil placed same distance apart. An alternating current is passed through the transmitter coil which generates a time- varying alternating magnetic field in the perpendicular direction, which is known as the primary magnetic field. This primary magnetic field enters into/and out of the ground, thereby creating circular eddy currents depending upon the nature of the subsurface. These eddy currents generate the secondary magnetic field, in proportion to the distribution of conductive material in the subsurface.

Both the induced secondary magnetic field, along with the primary magnetic field, are detected at the receiver coil. The primary magnetic field and secondary magnetic field created are generally out of phase with each other i.e they do not increase or decrease in unison. It is possible to resolve the secondary magnetic field into two components mathematically. The first component which is in phase with the primary magnetic field is called the real component and the second which is 90 degrees out of phase with the primary magnetic field is called the

34 imaginary component or the quadrature. The secondary magnetic field is created as a function of spacing between the coils, their orientation, the frequency and the distance from the ground at which the electromagnetic (EM) instrument is employed (Hendrichx and Kachanoski, 2002).

The amplitude and phase vary depending upon various soil properties including clay content, type of clay, salinity and water content. Hence, as discussed by (Corwin and Lesch, 2005c) the

EM instrument is helpful to determine different soil properties such as salinity, water content, clay content and type, organic matter, bulk density, temperature and CEC.

2.3.2 EM instruments Electromagnetic (EM) methods of prospecting have been employed extensively in geophysical investigations from the 1960s. Several manufacturers have produced different customized EM instruments which have significant applications in , natural resource management, soil study, mineral exploration, geological mapping, and environment engineering. The two geophysical companies, namely Geonics and DUALEM are the most popular manufacturers of EM instruments at the present time.

2.3.2.1 Geonics Series Geonics Limited is a Canadian-owned company located in Mississauga, Canada, and are pioneers in manufacturing different kinds of EM instruments. The company is established in

1962 with an early commitment to specialize in electromagnetic methods. Since then, Geonics has produced a wide variety of unique instruments to serve for the increasingly diverse range of applications. Geonics introduced EM31 Ground Conductivity Meter in 1976 which assisted geotechnical engineers and environmental professionals to characterize the near-surface geophysical properties. They subsequently developed the EM34-3 and EM38 series of ground conductivity meters, which are helpful for exploration to relatively greater and lesser depths respectively. All these instruments are widely employed in the fields of agriculture, , and groundwater.

35

In the EM31 instrument, the transmitter and the receiver coils are spaced 3.66 m apart and the instrument operates at a fixed frequency of 9.8 kHz. The instrument has two modes of operation which are Horizontal Magnetic Dipole (HMD) and Vertical Magnetic Dipole (VMD), respectively. The effective depth of investigation in (HMD) mode is 0 to 3 m whereas in (VMD) mode it measures up to a depth of 6 m when the instrument is held 1 m off the ground.

EM34-3 (Fig. 2.3.2.1.1) employs the same inductive method as an EM31 but it has two independent coils with 3 inter-coils spacings which are 10 m, 20 m and 40 m respectively. The frequency at which the instrument is used depends on the inter-coil spacing and is 6.4 kHz for

10 m spacing, 1.6 kHz for 20 m spacing and 0.4 kHz for 40 m spacing. This instrument also has two modes of operation (HMD and VMD). In VMD, measurement depth is greater compared to

HMD mode of operation, and the EM34-3 can exploreto a depth of 60 m. The depth of investigation varies with the inter-coil spacing and the mode of operation. Given the depths of investigation, the EM31 and EM34-3 instruments are mostly used in shallow ground water investigations and hydrogeological interactions, respectively. Table 2.3.2.1.1 shows the parameters of different EM instruments and their depths of investigations.

Table 2.3.2.1.1 Parameters of the common EM instrument Instrument Spacing Frequency Effective measurement depth Horizontal Vertical EM31 3.7m 9.8 kHz 0 – 3m 0 – 6m EM 34-3 10m 6.4kHz 0 – 15m 0 – 7.5m EM 34-3 20m 1.6kHz 0 – 30m 0 – 15m EM 34-3 40m 0.4kHz 0 – 60m 0 – 30m EM38-MK2 1m 14.5 kHz 0.75–1m 1.5–2m

36

Fig. 2.3.2.1.1 EM 34-3

2.3.2.2 EM38 The EM38 has a much smaller inter-coil spacing of either 0.75 m or 1 m. The operating frequency of EM38 is 14.5 kHz and the depth of investigation ranges from about 0.75 m to 2 m depending upon the inter-coil spacing and mode of operation. EM38 also has a similar mode of operation as EM31 and EM34-3 which is HMD and VMD modes. The VMD reaches greater depth as compared to HMD. The shallow depth of investigation makes EM38 suitable for applications in agriculture, archaeology and soil sciences. EM38 is especially useful for determining the apparent clay content, salinity and moisture in the soils. It is designed

37 specifically for the measurement of ECa in the root zone (McNeill, 1990). On the increasing demands from scientists and consultants across the globe, new developments have been introduced into the platforms which can collect ECa data in both the modes of operation. New models of EM38-MK2 have been developed which include EM38-DD which collects ECa data in both the horizontal and vertical modes of operation at the same time (Fig. 2.3.2.2.1).

Fig. 2.3.2.2.1 Schematic diagram of the EM 38, which is 1 m in length. Tx is the transmitting coil and Rx is the receiving coil (after Robinson et al., 2004).

2.3.2.3 DUALEM The DUALEM company is located at Milton, Ontario, Canada. They manufacture different versions of the DUALEM instruments, which are DUALEM-1, DUALEM-21 and

DUALEM-421. The advantage of using the DUALEM series is they have a pair of receiver coils, which are mutually perpendicular at different distances. The paired received coils are helpful to determine the ECa values in both horizontal mode and vertical mode in a single survey. Obtaining VDM and HDM simultaneously reduces the difficulty of rotating the EM instrument to determine ECa values in different modes of operation. The receiver coil pairs are placed at a distance of 1 m, 2 m and 4m respectively in the DUALEM series which allows us a

38 choice of penetration depth. Also, as the instrument uses a constant frequency of 9 kHz receiver coil pairs, at different distances, can be combined.

Fig. 2.3.1.1 shows the working and schematic representation of the DUALEM series (i.e

DUALEM-1). In this instrument, there is a pair of receiver coils positioned at 1 m from the transmitter coils. The first coil is placed in the same orientation as the transmitter coil, which is in vertical mode and is known as Horizontal co-planar (HCP). The second coil is placed in the perpendicular orientation with respect to the transmitter coil, which allows the instrument to operate in horizontal mode and is termed as perpendicular co-planar (PERP). The measurements of ECa in these different modes of operation are referred to as Hcon and Pcon, respectively. This reduces the hassle of rotating the instrument at every position and the instrument can be towed on a non-metal sled with the attached to a vehicle. Fig. 2.3.2.3.1 shows a DUALEM-21 being towed with the help of a vehicle on a PVC sled.

Table 2.3.2.3.1 shows the expected depth of investigation for the DUALEM sensors.

The instrument can go up to 6 m deep with the help of DUALEM-4 HCP sensor whereas the

ECa values as shallow as 0.5 m can be measured by DUALEM-1 PRP sensor.

Table 2.3.2.3.1 Parameters of DUALEM sensors (DUALEM-421S Manual, 2008). Instrument Spacing Effective measuring range Pcon Hcon Pcon Hcon DUALEM-1 1.1 m 1 m 0-0.5 m 0-1.5 m DUALEM-2 2.1 m 2 m 0-1.0 m 0-3.0 m DUALEM-4 4.1 m 4 m 0-2.0 m 0-6.0 m

39

Fig. 2.3.2.3.1 DUALEM-21 being towed with the help of a vehicle on a PVC sled.

2.3.3 EM Applications

2.3.3.1 Salinity (ECe) Soil containing salts is relatively more conductive in nature than soil without salt, because of conduction due to ion mobilisation and hence electromagnetic instruments prove to be an effective method to determine the soil salinity. Several research groups across the globe have displayed enormous success in mapping the soil salinity using a wide variety of EM instruments and the usage of EM instruments to map the subsurface soil is continuously

40 increasing. Table 2.3.3.1.1 displays the amount of research going on and the increase in the number of publications to-date, which proves the effectiveness of EM instruments in mapping subsurface soil salinity and as well as other soil properties. (e.g. Corwin and Lesch 2005c;

2013).

Table 2.3.3.1.1 Some of the studies on salinity and soil properties mapping using EM instrument. Property Instrument Authors ECe EM38 Rhoades and Corwin (1981); Corwin and Rhoades (1982); Van Der Lelij (1983); Corwin and Rhoades (1984); Wollenhaupt et al. (1986); Boivin et al. (1989); McKenzie et al. (1989); Rhoades et al. (1989); Rhoades and Corwin (1990); Diaz and Herrero (1992); Hendrickx et al. (1992); Lesch et al. (1992); Cannon et al. (1994); Nettleton et al. (1994); Bennett and George (1995); Lesch et al. (1995a,b); Vaughan et al. (1995); López-Bruna and Herrero (1996); Bourgault et al. (1997) ; Johnston et al. (1997); Mankin et al. (1997); McKenzie et al. (1997); Eigenberg and Nienaber (1998); Lesch et al. (1998); Odeh et al. (1998); Ceuppens and Wopereis (1999); Eigenberg and Nienaber (1999); Triantafilis et al. (2000); Barbiéro et al. (2001); Doolittle et al. (2001); Triantafilis et al. (2001a); Mankin and Karthikeyan (2002); Triantafilis et al. (2002); Barnes et al. (2003); Corwin et al. (2003); Corwin and Lesch (2003); Herrero et al. (2003); Corwin and Lesch (2005a,b,c); Friedman (2005); Horney et al. (2005); Kaffka et al. (2005); Lesch (2005); Lesch et al. (2005); Amezketa (2006); Corwin et al. (2006); Nogués et al. (2006); Yao et al. (2007); Amezketa and de Valle de Lersundi (2008); Corwin et al. (2008); Urdanoz et al. (2008); Thomas et al. (2009); Bakker et al. (2010); Corwin et al. (2010); McLeod et al. (2010); Herrero et al. (2011); Kaman et al. (2011); Cetin et al. (2012); Li et al. (2012); Ganjegunte et al. (2013); Herrero and Hudnall (2013); Li et al. (2013a, b); Scudiero et al., (2013); Taghizadeh- Mehrjardi et al. (2014): Li et al. (2015): Huang et al. (2016) EM34 Williams and Baker (1982); Williams and Fidler (1983); Dixon (1989); Paine (2003); Goes et al. (2009); Triantafilis and Buchanan (2010) EM31 De Jong et al. (1979); Cameron et al. (1981); Kinal et al. (2006) DUALEM Moffett et al. (2010); Urdanoz and Aragüé (2011); Atwell et al. (2013): Zare et al. (2015) EC1:5 EM38 Slavich (1990); Slavich and Petterson (1990); Aragüés et al. (2010); Yao and Yang (2010) ECw EM34 Triantafilis and Buchanan (2009) Salinity EM38 Amezketa (2007); Barbiéro et al. (2008); Lesch and Corwin (2008); Arriola- Morales et al. (2009); Dixit and Chen (2010); Akramkhanov et al. (2011); Aragüés et al. (2011); Feikema and Baker (2011); Urdanoz and Aragüés (2012); Yao et al. (2012); Akramkhanov et al. (2014) EM34 Salama et al. (1994)

Most of the research papers listed are carried out in semi-arid and arid environments, but recent trends show that similar kind of research is being carried out by developing countries over the past few years for salinity assessment and soil management to increase the amount of cultivable land to suffice the needs of increasing population. For example, Chinese researchers

41 displayed a growing interest in employing EM instruments for salinity assessment and soil management in the recent years because of the rapidly increasing population and lack of new arable lands in mountainous topography (Zare et al., 2015).

The application of EM instruments in soil science dates back to 1970’s where de Jong et al. (1979) demonstrated how an EM31 can be helpful in salinity assessment and soil mapping.

An EM31 instrument was used in two modes of operation along a single transect, which was

400 m in length. The EM31 instrument proved to be an easy and rapid method for mapping soil salinity and act as an effective replacement to the labour-intensive and time-consuming, Wenner four probe method. (Rhoades and Ingvalson, 1971; Halvorsen and Rhoades, 1974). However,

Wenner four probe method provided a better resolution of change in soil salinity with depth compared to the non-invasive EM31 instrument.

In years to come, several researchers have employed different EM instruments in field studies to map soil salinity and develope various calibration techniques to correlate data obtained from EM instruments to laboratory data obtained by ground truthing. Cameron et al.

(1981) established a relationship between ECa collected by the EM instruments (EM31 and

EM38) and ECe to the average depths of 0-1.22 m and 0-3.66 m, respectively, to map the soil salinity in a 16-ha field (Table 2.3.3.1.2) . They showed that the EM38 because of shallower investigation depths produced better results in mapping soil salinity compared to the EM31. Due to greater depth of penetration EM31 investigates depths below the traditional soil profile and below typical soil sampling depths therefore, it showed less accuracy. Similar work has been done by Williams and Baker (1982) using an EM34 instrument on a 5-km grid to map salinity to the depth and they concluded that EM instruments are useful for reconnaissance salinity hazard mapping across large areas. Lesch et al. (1995a) developed multiple-linear regression calibration models to simplify the process of calibration. Triantafilis et al. (2000) developed a logistic model for salinity mapping across different profiles. Different calibration techniques were

42 developed by researchers across the globe depending on study site conditions. In more informative words, calibrations of ECa to soil properties is site-specific.

Table 2.3.3.1.2 Regression relationships between measured soil paste ECe and recorded meter readings of apparent conductivity (ECaW and ECa) ( after Cameron et al. 1981).

The process of EM data collection and salinity mapping has become easy in the 1990’s with the advent of global positioning system (GPS). Cannon et al (1994) is one of the first to mobilise EM38 and a GPS unit on an all-terrain vehicle which eased the way of collecting georeferenced EM data. Similar work has been performed by Bourgault et al. (1997) on a broader scale (i.e. 2400 ha.) for salinity mapping where they developed a geostatistical analysis of an EM data set through GIS (Global Information System). With time, more advanced mobile

EM sensing system platforms are developed by the research groups across the globe which helped in quick data acquisition and management. For example, the US Salinity Laboratory staff

(Carter et al., 1993), and the University of Sydney (Triantafilis et al., 2002)

With the advent of these MESS, the large amount of data collected requires soil sampling locations to be computed statistically and considering the range in ECa and also ensure adequate representation as evenly as possible. In this regard, the ESAP software was developed by Corwin and Lesch (2003). The software uses Response Surface Methodology to partition the study area into different sub-groups based on the distribution of a priori ancillary data (mostly

EM data). The ability of the ESAP software to handle large data sets helps to determine the soil sampling locations at field (Amezketa and de Valle de Lersundi, 2008), farm (Herrero et al.,

2003), district (Herrero et al., 2011) and even sub-catchment level (Triantafilis and Buchanan,

2010). Triantafilis and Buchanan (2009) have shown that ESAP software can also be used for

43 mapping soil ECe in the deeper regolith where they mapped the spatial distribution of saline subsurface material using EM34 data and a hierarchical spatial regression model.

The application of EM instruments in salinity mapping has dramatically increased over the past decade of time as it proved to be efficient, fast and cost-effective technique compared to the traditional resistivity methods. Also, the usage and employability of EM instruments have diversified in different fields of study. For example, McLeod et al. (2010) have shown how salinity could be monitored using an EM38 and as a function of time after the December 2004

Indian Ocean tsunami.

2.3.3.2 Clay content and soil texture EM instruments also enable direct/indirect mapping of several other soil properties of the along with the salinity. EM data can map the amount of clay present and hence soil texture.

There have been several publications discussing the salinity either alone or in combination with one of the aspects such as soil texture or particle size (Table 2.3.3.2.1). Some of the examples include identification of soil texture features (Domsch and Giebel, 2004), texture contrasts

(Doolittle et al., 2002) and soil profile heterogeneity (Saey et al., 2009a). Also determining the soil properties such as soil texture and particle size using hydrometer or pipette methods or particle size analysers helps to add more value to salinity mapping by EM instruments.

Table 2.3.3.2.1 Some of the studies on texture mapping using EM instrument. Property Instrument Authors Clay content EM38 Triantafilis et al. (2001b); Doolittle et al. (2002); Sudduth et al. (2003); Domsch and Giebel (2004); Sudduth et al. (2005); Triantafilis and Lesch (2005); Jung et al. (2006); Weller et al. (2007); Cockx et al. (2009); Saey et al. (2009a); Triantafilis et al. (2009a); De Benedetto et al. (2010); Lück et al. (2011); Nelson et al. (2011); Mahmood et al. (2012); Grellier et al. (2013); Piikki et al. (2013); DUALEM Robinson et al. (2008); Robinson et al. (2010); Soil type EM38 James et al. (2003); boundaries

44

Sand content EM38 Islam et al. (2012); Sand deposition EM38 Kitchen et al. (1996); Clay lenses EM38 Cockx et al. (2007); Depth to EM38 Brus et al. (1992); Doolittle et al. (1994); DUALEM Saey et al. (2009b); Saey et al. (2011); Depth slice DUALEM Saey et al. (2012); Depth to EM38 Doolittle and Collins (1998); Zhu et al. bedrock (2010b); Depth of EM38 Vitharana et al. (2008c); substratum Texture EM38 Cook and Walker (1992); Knotters et al. (1995); Brevik and Fenton (2002); Carroll and Oliver (2005); Kerry and Oliver (2007); Hbirkou et al. (2011); Lamandé et al. (2011); De Benedetto et al. (2012); Heil and Schmidhalter (2012); EM31 Inman et al. (2001); Inman et al. (2002); Stroh et al. (2001); DUALEM Allred et al. (2005); N.A. Stroh et al. (1993); Zhu et al. (2013)

Determining the soil texture and soil particle sizes helps in addressing several issues such as for building planning, road constructions, and agriculture. For example, national

Resource Conservation Service (NRCS) in the US was given the responsibility to map the soil throughout the US. Soil texture is one of the main properties in delineating the soil map units and NRCS achieved it through using EM instruments as they provide fast and accurate means of direct soil sampling (Kitchen et al., 1996). Williams and Hoey (1987) were one of the first researchers to study the ability of EM data to map clay content. They were successful to show a

2 strong relationship between EM34 ECa and EC1:5 (R = 0.78) and between ECa and clay content (0.73).

Later on, Brus et al. (1992) tried to study the depth of overburden sand overlying boulder clay in Netherlands using EM38 and/or EM31 but they were unsuccessful. Similar work has been performed by Kitchen et al. (1996) using EM38 instrument where they studied the amount of sand deposited after the midwest floods of 1993 in the USA.

45

Triantafilis et al. (2001b) developed a linear relationship between EM38-ECa and average clay content to a depth of 1.5m in order to map the spatial distribution of average clay content to a depth of 1.5m in a irrigated field (244 ha). They have used various geostatistical methods (e.g. regression, ordinary, co-kringing) and transect spacing (e.g. 24, 48, 72 m) and showed that spatial distribution of clay containing alluvial plains can be delineated from the sandier prior stream channel sediments and regression kriging with a transect spacing of 24 m produces better accuracy in mapping the same. Triantafilis and Lesch (2005) also used a hierarchical spatial regression model to map average clay content to a depth of 7 m across a larger district scale study (~50,000 ha) located in the lower Macquarie valley (Australia) using a combination of EM38 and EM34 ECa data.

Piikki et al. (2013), from Sweden, showed better results for topsoil clay content mapping can be produced by data fusion of proximal sensor data. The covariate datasets used were EM38 and -ray spectrometry data along with terrain attributes such as elevation, slope, cosine and sine, along with the digital numbers (DN) of an aerial photo. They showed that even though - ray data can be used individually for the mapping of the topsoil clay content, better predictions were made by integrating the EM38 data with the digital numbers of an aerial photograph (Fig.

2.3.3.2.1). Similar work was done by Heil and Schmidhalter (2012), where they compared multivariable data using various geostatistical techniques (i.e. regression and co-kriging) to identify fields with differing histories of management (i.e fertilizer treatments) to map weaknesses pertaining to boundaries with depth (e.g. clay lenses covered by sand).

46

Fig. 2.3.3.2.1 Validation of predictions. a) modeling efficiency (ME) and b) mean absolute error (MAE) calculated for the validation dataset of eight different predictor sets. PLS-R=partial least squares regression; kNN = k nearest neighbor prediction. ECa = apparent electrical conductivity; photo = aerial photo DN, gamma = gamma ray spectrometry (40K, 232Th and TC). The hatched lines represent the values for ordinary kriging of calibration soil samples (after Piikki et al., 2013).

2.3.3.3 Soil moisture

Soil Moisture is a temporally and spatially variable soil property which can be estimated using EM instruments. Interestingly, most of the papers have been published in recent times as scientists became interested in the use of water by crops in irrigated areas owing to concerns

47 caused by climate change. Some papers demonstrastiong the ability of ECa measuremets to estimate soil moisture, including content (volumetric or gravimetric) and/or available water content (AWC), are listed below in the table 2.3.3.3.1.

Table 2.3.3.3.1 Some of the studies on mapping soil moisture using EM instrument. Property Instrument Authors Water EM38 Kachanoski et al. (1988); Kachanoski et al. (1990); Hanson and Kaita content (1997); Khakural et al. (1998); Reedy and Scanlon (2003); Akbar et al. (2005); Brevik et al. (2006); Hezarjaribi and Sourell (2007); Hossain et al. (2010); Zhu et al. (2010a); Rodríguez-Pérez et al. (2011); Sun et al. (2011); Guo et al. (2013b); Hedley et al. (2013); Jonard et al. (2013); Sun et al. (2013) EM31 Sheet and Hendrickx (1995) DUALEM Abdu et al. (2008); Robinson et al. (2009a); Robinson et al. (2012) GEM-300 Slavich (1990); Slavich and Petterson (1990); Aragüés et al. (2010); Yao and Yang (2010) Available EM34,EM Triantafilis and Buchanan (2009) water 38, -ray content DUALEM Jiang et al. (2007), Huang et al. (2017a), Huang et al. (2017b) Water EM38 Padhi and Misra (2011) distribution

Breifly, Kachanoski et al. (1988) were one of the first groups to recognize the ability of

EM instruments to map the moisture content. They used EM38 and EM31 data to establish correlation between ECa and average moisture content by a TDR and to a depth of 0.5m and found that EM38 ECa was more strongly correlated and better suited to measure the root zone moisture content (R2 = 0.96) using a second order regression equation similar to the one described in Rhoades et al. (1976).

Most recently, Huang et al. (2017a) developed a relationship between  and estimates of true electrical conductivity () and used this relationship to develop time-lapse images of soil  beneath a centre-pivot irrigated alfalfa (Medicago sativa L.) crop in San Jacinto, California,

USA. They measured the bulk apparent electrical conductivity (ECa–mS/m) using a DUALEM-

421 over a period of 12 days after an irrigation event (i.e. days 1, 2, 3, 4, 6, 8 and 12). Then, they used EM4Soil software package to generate EM conductivity images (EMCIs) (Fig.

2.3.3.3.1). They also used a physical model to estimate  from , accounting for soil tortuosity

48 and pore water salinity, with a cross-validation RMSE of 0.04cm3/cm3. Testing the scenario where no soil information was available, they used a three-parameter exponential model to relate to  and then to map  along the transect on different days. These results allowed monitoring the spatiotemporal variations of  across the surveyed area, over the 12-day period.

In this regard, they were able to map the soil close to field capacity (0.27 cm3/cm3) and approaching permanent wilting point (0.03 cm3/cm3). The time-lapse  monitoring approach, developed using EMCI, has implications for soil and water use and management and will potentially allow farmers and consultants to identify inefficiencies in water application rates and use. It can also be used as a research tool to potentially assist precision irrigation practices and to test the efficacy of different methods of irrigation in terms of water delivery and efficiency in water use in near real time.

Fig. 2.3.3.3.1 Distribution of predicted change in soil  (cm3/cm3) from day 1 to day 12 after Huang et al (2017a).

The amount of research going on this field of interest has enormously increased with time and scientists have tried to estimate the moisture content and other related properties using either EM data alone or in conjunction with the other ancillary data.

2.3.3.4 Soil Acidity (pH) Soil Acidity is one of the major problems as irrigated agriculture might drive the acidification of top and subsoils. Lime is typically applied at a uniform rate to raise the pH of the soil, as there aren’t many indicators which can be helpful to measure the acidity of the soil.

EM instruments have once again proven to be a feasible and rapid method to map the variation

49 of acidity across the field. This helps in variable rates of application of lime to the field depending upon the acidity value, making it a cost-effective approach as shown by Dunn et al.

(2007) (Fig. 2.3.3.4.1). Similar work had been done by Wong et al. (2008) where they have used

-ray spectrometry data in conjunction with EM data in order to map the soil acidity. The -ray spectrometry helped to overcome the inability of ECa-based methods to sense soil depth in highly weathered sandy soil over cemented gravel.

Fig. 2.3.3.4.1 Relationship between EM31 and EM38 readings and soil pH (CaCl2) at 0–10 cm (x) and 0–30 cm (◦) soil depths (averaged over 0–10, 10–20 and 20–30 cm depth increments) for three rice fields in the Murrumbidgee Irrigation Area (fields 1–3) and one rice field in the Wakool Irrigation District (field 4) (after Dunn et al. (2007)).

2.3.3.5 Soil cation exchange capacity (CEC) Cation exchange capacity (CEC) is the total capacity of a soil to hold exchangeable cations.

It is a very important soil property influencing soil structure stability, nutrient availability, soil pH and the soil’s reaction to fertilisers and other ameliorates (Hazleton and Murphy 2007). CEC is one of the most important properties because it acts as an index for the shrink-swell potential and hence is a measure of the soil structural resilience to tillage. According to Rhoades et al.

(1976), Rhoades et al. (1989) and Corwin and Lesch (2005c), there are three pathways of

50 current flow contributing to the ECa of a soil: (i) a liquid phase pathway via dissolved solids contained in the soil water occupying the large pores, (ii) a solid – liquid phase pathway primarily via exchangeable cations associated with clay minerals, and (iii) a solid pathway via soil particles that are in direct and continuous contact with one another. Therefore, can be helpful to map the variations in CEC across the field alone and with the help of other ancillary data. Bishop and McBratney (2001) were one of the first groups who used (ECa – mS/m), terrain attributes, aerial photography and LANDSAT TM to map soil CEC. Officer et al. (2004) had used a similar approach for the determination of CEC using Veris 3100 and digital elevation model data. More recently, Triantafilis et al. (2009b) showed how EM data and other ancillary data sets can be helpful to map CEC (Fig. 2.3.3.5.1) at the field level such as using multiple linear regression models, while Taghizadeh et al. (2015) mapped CEC at different depths using digital

Lit review:elevation model (DEM) and Landsat data.

Fig. 2.3.3.5.1. Spatial distribution of predicted average (0–2 m) cation exchange capacity (CEC, cmol(+)/kg) with (a) stepwise, and (b) standard least-squares hierarchical spatial regression (HSR) model with ancillary data available on 24-m transect spacing.

51

2.3.3.6 Soil management classes EM instruments are also helpful to recognize the soil management classes apart from their potential to map several soil properties as discussed above. Several researchers across the globe attempted to discern the soil management class by using either EM data alone or in combination with other ancillary data. The idea of application of EM instruments in order to understand soil management class gained interest in the early 21st century as the need for widespread cost-effective precision farming has become important because of the growing population.

Cambouris et al. (2006) employed a simple approach by using EM38 data alone, acquired on a regular grid over a 13.8 ha commercial field under potato production in St.

Amable Quebec, Canada. Considering various physical (e.g. available water content and texture) and chemical (e.g. pH) attributes, they used k-means clustering to partition the field after kriging the data and found that two management zones were sufficient to account for within field variance given the disagreement in potato yield (5.9 t per ha) and soil moisture availability.

Vitharana et al. (2006) used EM38 data to map the variations in top and subsoil clay content, and Fuzzy k-means to discern the soil management zones in an 11.5 ha field situated in a polder area of northwest Flanders in Belgium. Some researchers used data acquired using multiple EM instruments in order to discern the soil management zones. Zhu et al. (2010a) compared soil maps generated using different range of instruments (i.e EM38, DUALEM-2 and

EM-31) with the EM data being acquired at different periods of time and in different modes of operation (i.e Horizontal and Vertical) and found that EM31 in horizontal mode of operation can better be used to differentiate zones basing on the soil content whereas it can be used in its vertical mode to best resolve depth to bedrock.

Several researchers also used other types of ancillary data together with the EM data to understand and identify the management zones. Anderson- Cook et al. (2002) were one of the

52 first groups who combined EM38 data with the crop yield in a 24-ha field in the coastal plain of

Virginia, USA. They found that variation in the EM38 data and crop yield data using recursive binary classification trees produced coherent results to understand different soil types (i.e Order

1 using National Cooperative survey manual) in contrast to using either EM38 data or yield data alone.

Elevation data was used in combination with EM38 and Veris3100 data across two fields covering 41 ha near Centralia in north-central Missouri in the USA by Kitchen et al 2005, where they showed productivity zones obtained from FKM clustering are in agreement with the results achieved by the independent clustering of the yield maps. Also by using Kappa statistic and by varying the data used they showed that the best data combination included ECa and elevation variables which gave 60–70% agreement between yield maps and productivity zones.

Data collected from the remote sensing platforms can potentially be used in Digital Soil

Mapping (DSM) to delineate the soil management zones in areas with little or no relief. Huang et al. (2014a) numerically clustered DUALEM-1 data and -ray radiation of potassium (K%), uranium (Uppm), thorium (Thppm) and total counts (TCppm), using FKM and LMM analysis. Fig 2.3.3.6.1 shows the DUALEM-1 and -ray data acquired across the study field located in the valley of the River Trent, east of Nottingham, UK. Results indicated that k=7

(Fig. 2.3.3.6.2) and 8 were statistically different and accounted for most of the soil variation.

They also compared the prediction of soil properties by FKM analysis and regression models.

The results of this analysis indicated that for a sample size of 80, the regression models were able to predict clay content, better than FKM clustering. However, when predicting pH, FKM clustering performs better than the linear regression model when k=6–9. They concluded that both the FKM and LMM methods have merit.

53

Fig. 2.3.3.6.1 Spatial distribution ancillary data including; a) total count (TC – counts per second), b) potassium (K - %), and c) thorium (Th – ppm) e) 1mPcon (ECa – mS/m) and, f) 1mHcon (ECa – mS/m), and c) altitude (m) (after Huang et al. 2014a).

Fig. 2.3.3.6.2 Spatial distribution of FKM classes for K=7 (after Huang et al. 2014a).

The -ray spectrometry data which measures the concentration of radioisotopes of K, U, and Th are used by some researchers in conjunction with the EM data to understand their

54 applicability in mapping the management zones. Altdorff and Dietrich (2012) tested EM38 data and -ray data using k-means clustering collected across a 5 ha field in Germany to study their usefulness to discern management zones. Some of the other works where similar kind of research had been done include Miervenne et al. (2013), Islam et al (2011b) and Castrignano et al (2012).

It is expected that better results could be achieved by a large-scale clustering of proximally sensed EM data with either proximally or remotely sensed ancillary data such as -ray spectrometry or other sensor platforms which have the potential to successfully map individual soil properties.

2.3.3.7 Other properties In addition to the application of EM instruments to understand and map the above- mentioned soil properties, they also proved worthy to study several other soil properties. Some of the publications where EM instruments are used to study various other properties of soil are listed in the Table 2.3.3.7.1.

Table 2.3.3.7.1 Studies on other soil properties using EM instruments. Property Instrument Authors Accuracy EM38 Sudduth et al. (2001); Brevik et al. (2004); Robinson et al. (2004); Abdu et al. (2007); Morris (2009); Beamish (2011); Ma et al. (2011) Survey Design EM38 Johnson et al. (2005a); Brenning et al. (2008); Wetterlind et al. (2010) Yield EM38 Jiang et al. (2009); Serrano et al. (2013) DUALEM Eigenberg et al. (2002); Johnson et al. (2005b); Cockx et al. (2005) Nitrate EM38 Eigenberg et al. (2002); Johnson et al. (2005b); Cockx et al. (2005) Acidity EM38 Dunn and Beecher (2007) DUALEM Serrano et al. (2010) Soil depth EM38 Dunn and Beecher (2007) DUALEM Serrano et al. (2010) Volatile fatty acid DUALEM Woodbury et al. (2011) Sodicity EM38 López-Lozano et al. (2010); Huang et al. (2014b) Nutrient EM38 Cordeiro et al. (2011a,b) DUALEM Woodbury et al. (2010) Organic carbon EM38 Martinez et al. (2009); Werban et al. (2009); Sarkhot et al. (2011)

55

Phosphorus EM38 Quenum et al. (2012) maximum sorption capacity EM38 Herbst et al. (2010) Root distribution EM38 Myers et al. (2007) Seeding depth EM38 Knappenberger and Koller (2012) Soil pattern EM38 Meerschman et al. (2013b) DUALEM Saey et al. (2013a, b) Soil variability EM31 O'Neill et al. (2008) EM38 Eigenberg et al. (2006); Harvey and Morgan (2009); Bramley et al. (2011b); Islam et al. (2011b) DUALEM Papanicolaou et al. (2008); Franz et al. (2011) Ice-wedge EM38 Reid et al. (2003); Cockx et al. (2006); Meerschman et al. (2013a) DUALEM Singleton et al. (2010) Earthworm EM38 Valckx et al. (2009) populations CO2 storage EM31 Arts et al. (2009)

2.4 Conclusions Digital Soil Mapping requires the use of ancillary data (either proximal or remote) which is easy and cheap to collect and in association with spatial and non-spatial inference methods to map either soil types or soil properties. In terms of ancillary data, and with the ever increasing availability of EM instruments which have been used mostly at the field scale, there is scope to explore the use of such data across larger areas. The advantage of these instruments is they can provide depth specific information depending on the instrument (e.g. EM38, DUALEM-421). In addition, they have been used to map soil types and properties of agricultural significance (e.g. salinity, CEC, moisture). There is also scope to use -ray spectrometry data because it can be used as a surrogate to map soil properties such as clay content, with this data also shown to be useful in identifyiung soil landscapoe units. In both cases, larger areas have been able to be mapped as a result of the approach being able to be mobilised onto a airborne platform, which allows large areas to be convered quickly and as a result databases of large spacial extents (e.g. continental Australia) are now available (Minty et al., 2009).

56

In this thesis, the aim is to explore how ancillary data in the form of EM data acquired from a root zone measuring EM38 and in vertical and horizontal modes of operation and collected on a reconnaissance grid of 0.5 – 1.0 km can be used in conjunction with remotely sensed -ray spectrometry data to map soil types and soil properties of relevance and significance to sustainable agriculture and in the highly production Bourke irrigation district. In

Chapter 4, soil landscape units are identified using the proximal EM38 data and the remote sensed gamma ray data and using a non-spatial inference model (FKM clustering). The soil landscape units are tested using REML analysis. In Chapter 5, the same data is tested for its ability to develop a LMM and to map soil salinity (i.e. ECe) across the same area. In order, to assess the errors in this modelling an error budgeting procedure was applied and to identify which produced the largest errors (i.e. model, input, etc).

2.5 References Abdu, H., Robinson, D. A., & Jones, S. B. 2007. Comparing bulk soil electrical conductivity determination using the DUALEM-1S and EM38-DD electromagnetic induction instruments. Soil Science Society of America Journal, 71, 189-196.

Abdu, H., Robinson, D. A., Seyfried, M., & Jones, S. B. 2008. Geophysical imaging of watershed subsurface patterns and prediction of soil texture and water holding capacity. Water Resources Research, 44, W00D18.

Adamchuk, V. I., & Viscarra Rossel, R. A. 2010. Development of On-the-Go Proximal Soil Sensor Systems. Proximal soil sensing, Springer.

Akbar, M. A., Kenimer, A. L., Searcy, S. W., & Torbert, H. A. 2005. Soil water estimation using electromagnetic induction. Transactions of the American Society of Agricultural Engineers, 48, 129- 135.

Akramkhanov, A., Brus, D. J., & Walvoort, D. J. J. 2014. Geostatistical monitoring of soil salinity in Uzbekistan by repeated EMI surveys. Geoderma, 213 , 600-607.

Akramkhanov, A., Martius, C., Park, S. J., & Hendrickx, J. M. H. 2011. Environmental factors of the spatial distribution of soil salinity on flat irrigated terrain. Geoderma, 163, 55-62.

Al-Ain, F., Attar, J., Hussein, F., & Heng, L. K. 2009. Comparison of nuclear and capacitance-based soil water measuring techniques in salt-affected soils. Soil Use and Management, 25, 362-367.

Allred, B., Daniels, J., & Ehsani, R. (eds.) 2008. Handbook of Agricultural Geophysics. CRC Press, Boca Raton, FL. p. 242-248.

57

Allred, B. J., Redman, J. D., McCoy, E. L., & Taylor, R. S. 2005. Golf course applications of near- surface geophysical methods: A case study. Journal of Environmental and Engineering Geophysics, 10, 1-19.

Altdorff, Daniel, & Dietrich, Peter. 2012. Combination of electromagnetic induction and gamma spectrometry using K-means clustering: A study for evaluation of site partitioning. Journal of Plant Nutrition and Soil Science, 175, 345-354.

Amezketa, E. 2006. An integrated methodology for assessing soil salinization, a pre-condition for land desertification. Journal of Arid Environments, 67, 594-606.

Amezketa, E. 2007. Soil salinity assessment using directed soil sampling from a geophysical survey with electromagnetic technology: A case study. Spanish Journal of Agricultural Research, 5, 91-101.

Amezketa, E., & del Valle de Lersundi, J. 2008. and salinity mapping for determining restoration potential of cropped riparian areas. Land Degradation and Development, 19, 153-164.

Anderson-Cook, C. M., Alley, M. M., Roygard, J. K. F., Khosla, R., Noble, R. B., & Doolittle, J. A. 2002. Differentiating soil types using electromagnetic conductivity and crop yield maps. Soil Science Society of America Journal, 66, 1562-1570.

Aragüés, R., Guillén, M., & Royo, A. 2010. Five-year growth and yield response of two young olive cultivars (Olea europaea L., cvs. Arbequina and Empeltre) to soil salinity. Plant and Soil, 334, 423- 432.

Aragüés, R., Urdanoz, V., Çetin, M., Kirda, C., Daghari, H., Ltifi, W., . . . Douaik, A. 2011. Soil salinity related to physical soil characteristics and irrigation management in four Mediterranean irrigation districts. Agricultural Water Management, 98, 959-966.

Arriola-Morales, J., Batlle-Sales, J., Valera, M. A., Linares, G., & Acevedo, O. 2009. Spatial variability analysis of soil salinity and alkalinity in an endorreic volcanic watershed. International Journal of Ecology and Development, 14(F09), 1-17.

Arts, R. J., Baradello, L., Girard, J. F., Kirby, G., Lombardi, S., Williamson, P., & Zaja, A. 2009. Results of geophysical monitoring over a "leaking" natural analogue site in Italy, Washington DC.

Atwell, Melissa, Wuddivira, Mark, Gobin, Judith, & Robinson, David. 2013. Edaphic Controls on Sedge Invasion in a Tropical Wetland Assessed with Electromagnetic Induction. Soil Sci. Soc. Am. J., 77, 1865-1874.

Bakker, D. M., Hamilton, G. J., Hetherington, R., & Spann, C. 2010. Productivity of waterlogged and salt-affected land in a Mediterranean climate using bed-furrow systems. Field Crops Research, 117, 24-37.

Barbiéro, L., Cunnac, S., Mané, L., Laperrousaz, C., Hammecker, C., & Maeght, J. L. 2001. Salt distribution in the Senegal middle valley analysis of a saline structure on planned irrigation schemes from N'Galenka creek. Agricultural Water Management, 46, 201-213.

Barbiéro, L., Filho, A. Rezende, Furquim, S. A. C., Furian, S., Sakamoto, A. Y., Valles, V., . . . Neto, J. P. Queiroz. 2008. Soil morphological control on saline and freshwater lake hydrogeochemistry in the Pantanal of Nhecolândia, Brazil. Geoderma, 148, 91-106.

Barišić, D. 1996. Dose rate conversion factors, soil thickness and their influence on natural background dose rate in air above carbonate . Journal of Environmental Radioactivity, 31(1), 51-64.

58

Barnes, E. M., Sudduth, K. A., Hummel, J. W., Lesch, S. M., Corwin, D. L., Yang, C., . . . Bausch, W. C. 2003. Remote- and ground-based sensor techniques to map soil properties. Photogrammetric Engineering and Remote Sensing, 69, 619-630.

Bartholomeus, H. M., Schaepman, M. E., Kooistra, L., Stevens, A., Hoogmoed, W. B., & Spaargaren, O. S. P., 2008. Spectral reflectance based indices for soil organic carbon quantification. Geoderma, 145, 28-36.

Beamish, David. 2011. Low induction number, ground conductivity meters: A correction procedure in the absence of magnetic effects. Journal of Applied Geophysics, 75, 244-253.

Becegato, V. A., & Ferreira, F. J. F. 2005. Gamma-ray spectrometry, electrical resistivity and magnetic susceptibility of the agricultural soils in the Northwest of the Parana state. Gamaespectrometria, resistividade elétrica e susceptibilidade magnética de solos agrícolas no noroeste do estado do paraná, 23, 371-405.

Beckett, K. 2008. Multispectral processing of high-resolution radiometric data for soil mapping. Near Surface Geophysics, 6, 281-287.

Bennett, D. L., & George, R. J. 1995. Using the EM38 to measure the effect of soil salinity on Eucalyptus globulus in south-western Australia. Agricultural Water Management, 27, 69-85.

Boettinger, J. L., Doolittle, J. A., West, N. E., Bork, E. W., & Schupp, E. W. 1997. Nondestructive assessment of rangeland soil depth to petrocalcic horizon using electromagnetic induction. Arid Soil Research and Rehabilitation, 11, 375-390.

Boivin, P., Hachicha, M., Job, J.O., Loyer, J.Y. 1989. Electromagnetic conductivity and kriging: a tool for of soils salinity. Science du Sol 27, 69-72.

Bork, E. W., West, N. E., Doolittle, J. A., & Boettinger, J. L. 1998. Soil depth assessment of sagebrush grazing treatments using electromagnetic induction. Journal of Range Management, 51, 469-474.

Bramley, R. G. V., Ouzman, J., & Boss, P. K. 2011a. Variation in vine vigour, grape yield and vineyard soils and topography as indicators of variation in the chemical composition of grapes, wine and wine sensory attributes. Australian Journal of Grape and Wine Research, 17, 217-229.

Bramley, R. G. V., Trought, M. C. T., & Praat, J. P. 2011b. Vineyard variability in Marlborough, New Zealand: Characterising variation in vineyard performance and options for the implementation of Precision Viticulture. Australian Journal of Grape and Wine Research, 17, 83-89.

Brenning, Alexander, Koszinski, Sylvia, & Sommer, Michael. 2008. Geostatistical homogenization of soil conductivity across field boundaries. Geoderma, 143, 254-260.

Brevik, E.C., & Fenton, T.E. 2002. The relative influence of soil water, clay, temperature, and carbonate minerals on soil electrical conductivity readings taken with an EM-38 along a catena in central Iowa. Horizons , 9-13.

Brevik, E. C., Fenton, T. E., & Horton, R. 2004. Effect of daily soil temperature fluctuations on soil electrical conductivity as measured with the Geonics EM-38. Precision Agriculture, 5(2), 145-152.

Brevik, E. C., Fenton, T. E., & Lazari, A. 2006. Soil electrical conductivity as a function of soil water content and implications for soil mapping. Precision Agriculture, 7, 393-404.

59

Brus, D. J., Knotters, M., van Dooremolen, W. A., van Kernebeek, P., & van Seeters, R. J. M. 1992. The use of electromagnetic measurements of apparent soil electrical conductivity to predict the boulder clay depth. Geoderma, 55, 79-93.

Buchanan, S., Triantafilis, J., Odeh, I. O. A., & Subansinghe, R. 2012. Digital soil mapping of compositional particle-size fractions using proximal and remotely sensed ancillary data. Geophysics, 77, WB201-WB211.

Cambouris, A. N., Nolin, M. C., Zebarth, B. J., & Laverdière, M. R. 2006. Soil management zones delineated by electrical conductivity to characterize spatial and temporal variations in potato yield and in soil properties. American Journal of Potato Research, 83, 381-395.

Cameron, D. R., De Jong, E., Read, D. W. L., & Oosterveld, M. 1981. Mapping salinity using resistivity and electromagnetic inductive techniques. Canadian Journal of Soil Science, 61, 67-78.

Cannon, M. E., McKenzie, R. C., & Lachapelle, G. 1994. Soil salinity mapping with electromagnetic induction and satellite-based navigation methods. Canadian Journal of Soil Science, 74, 335-343.

Carroll, Z. L., & Oliver, M. A. 2005. Exploring the spatial relations between soil physical properties and apparent electrical conductivity. Geoderma, 128, 354-374.

Carter, L.M., Rhoades, J.D., Chesson, J.H. 1993. Mechanization of soil salinity assessment for mapping. ASAE Paper No. 931557, 1993 ASAE Winter Meetings, 12–17 December 1993, Chicago, IL. ASAE, St. Joseph, MI, USA.

Cassiani, G., Ursino, N., Deiana, R., Vignoli, G., Boaga, J., Rossi, M., . . . Werban, U. 2012. Noninvasive monitoring of soil static characteristics and dynamic states: A case study highlighting vegetation effects on agricultural land. Vadose Zone Journal, 11.

Castrignanò, A., Wong, M. T. F., Stelluti, M., De Benedetto, D., & Sollitto, D. 2012. Use of EMI, gamma-ray emission and GPS height as multi-sensor data for soil characterisation. Geoderma, 175- 176, 78-89.

Cetin, M., Ibrikci, H., Kirda, C., Kaman, H., Karnez, E., Ryan, J., . . . Sesveren, S. 2012. Using an electromagnetic sensor combined with geographic information systems to monitor soil salinity in an area of southern turkey irrigated with drainage water. Fresenius Environmental Bulletin, 21, 1133- 1145.

Ceuppens, J., & Wopereis, M. C. S. 1999. Impact of non-drained irrigated rice cropping on soil salinization in the Senegal River Delta. Geoderma, 92(1–2), 125-140.

Chanzy, A., Chadoeuf, J., Gaudu, J. C., Mohrath, D., Richard, G., & Bruckler, L. 1998. Soil moisture monitoring at the field scale using automatic capacitance probes. European Journal of Soil Science, 49, 637-648.

Cockx, L., Ghysels, G., Van Meirvenne, M., & Heyse, I. 2006. Prospecting frost-wedge pseudomorphs and their polygonal network using the electromagnetic induction sensor EM38DD. Permafrost and Periglacial Processes, 17, 163-168.

Cockx, L., Van Meirvenne, M., & De Vos, B. 2007. Using the EM38DD soil sensor to delineate clay lenses in a sandy forest soil. Soil Science Society of America Journal, 71, 1314-1322.

Cockx, L., Van Meirvenne, M., & Hofman, G. 2005. Characterization of nitrogen dynamics in a pasture soil by electromagnetic induction. Biology and Fertility of Soils, 42, 24-30.

60

Cockx, L., Van Meirvenne, M., Vitharana, U. W. A., Verbeke, L. P. C., Simpson, D., Saey, T., & Van Coillie, F. M. B. 2009. Extracting Topsoil Information from EM38DD Sensor Data using a Neural Network Approach All rights reserved. Soil Science Society of America Journal, 73, 2051-2058.

Cook, P. G., & Walker, G. R. 1992. Depth profiles of electrical conductivity from linear combinations of electromagnetic induction measurements. Soil Science Society of America Journal, 56, 1015-1022.

Cook, S. E., Corner, R. J., Groves, P. R., & Grealish, G. J. 1996. Use of airborne gamma radiometric data for soil mapping. Australian Journal of Soil Research, 34, 183-194.

Cordeiro, M. R. C., Sri Ranjan, R., & Ferguson, I. J. 2011a. Calibration models for electromagnetic induction methods to assess nutrient accumulation beneath confined livestock areas. Environmental Technology, 32, 103-117.

Cordeiro, M. R., Ranjan, R. S., & Cicek, N. 2011b. Assessment of potential nutrient build-up around beef cattle production areas using electromagnetic induction. Environmental technology, 33, 1825- 1833.

Corwin, D. L., & Lesch, S. M. 2003. Application of soil electrical conductivity to precision agriculture: Theory, principles, and guidelines. Agronomy Journal, 95, 455-471.

Corwin, D. L., Lesch, S. M., Oster, J. D., & Kaffka, S. R. 2006. Monitoring management-induced spatio–temporal changes in through soil sampling directed by apparent electrical conductivity. Geoderma, 131, 369-387.

Corwin, D.L., Lesch, S.M., Shouse, P.J., Soppe, R., Ayars, J.E., 2003. Identifying soil properties that influence cotton yield using soil sampling directed by apparent soil electrical conductivity. Agronomy Journal, 95, 352–364.

Corwin, D.L., Lesch, S.M., Shouse, P.J., Soppe, R., & Ayarsm, J.E. 2008. Delineating site-specific management units using geospatial ECa measurements. In: Allred, B., Daniels, J., and Ehsani, R. (eds.) Handbook of Argicultural Geophysics. CRC Press, Boca Raton, FL. p. 242-248.

Corwin, D. L., & Lesch, S. M. 2005a. Characterizing soil spatial variability with apparent soil electrical conductivity: I. Survey protocols. Computers and Electronics in Agriculture, 46(1-3 SPEC. ISS.), 103-133.

Corwin, D. L., & Lesch, S. M. 2005b. Characterizing soil spatial variability with apparent soil electrical conductivity: Part II. Case study. Computers and Electronics in Agriculture, 46(1-3 SPEC. ISS.), 135-152.

Corwin, D. L., & Lesch, S. M. 2005c. Apparent soil electrical conductivity measurements in agriculture. Computers and Electronics in Agriculture, 46(1-3 SPEC. ISS.), 11-43.

Corwin, D.L., & S.M. Lesch. 2010. Chapter 6- Delineating site-specific management units with proximal sensors. In: Oliver, M. (ed.) Geostatistical Applications in Precision Agriculture. Springer, New York, NY. P.139-166.

Corwin, D.L., Lesch, S.M., Segal, E. Shouse, P., Skaggs, T., & Bradford, S. 2010. Comparison of model- and design- based sampling strategies for characterizing spatial variability with ECa-directed soil sampling, Journal of Environmental and Engineering Geophysics, 15: 147-162.

Corwin, D.L., & Lesch, S.M. 2013. Protocols and guidelines for the field-scale measurement of soil salinity distribution with ECa-directed soil sampling, Journal of Environmental Engineering Geophysics, 18: 1-25.

61

Corwin, D. L., & Rhoades, J. D. 1982. An improved technique for determining soil electrical conductivity-depth relations from above-ground electromagnetic measurements. Soil Science Society of America Journal, 46, 517-520.

Corwin, D. L., & Rhoades, J. D. 1984. Measurement of inverted electrical conductivity properties using electromagnetic induction. Soil Science Society of America Journal, 48, 288-291.

Costa, J. C., Borges, J. A. R., & Pires, L. F. 2013. Soil bulk density evaluated by gamma-ray attenuation: Analysis of system geometry. Soil and Tillage Research, 129, 23-31.

Cozzolino, D., & Morón, A. 2006. Potential of near-infrared reflectance spectroscopy and chemometrics to predict soil organic carbon fractions. Soil and Tillage Research, 85, 78-85.

Cresswell, H. P., Green, T. W., & McKenzie, N. J. 2008. The Adequacy of Pressure Plate Apparatus for Determining Soil Water Retention. Soil Science Society of America Journal, 72, 41-49.

Dang, Y. P., Dalal, R. C., Pringle, M. J., Biggs, A. J. W., Darr, S., Sauer, B., . . . Orange, D. 2011. Electromagnetic induction sensing of soil identifies constraints to the crop yields of north-eastern Australia. Soil Research, 49, 559-571.

De Benedetto, D., Castrignanò, A., Rinaldi, M., Ruggieri, S., Santoro, F., Figorito, B., . . . Tamborrino, R. 2013. An approach for delineating homogeneous zones by using multi-sensor data. Geoderma, 199, 117-127.

De Benedetto, D., Castrignanò, A., Sollitto, D., & Modugno, F. 2010. Spatial relationship between clay content and geophysical data. Clay Minerals, 45, 197-207.

De Benedetto, Daniela, Castrignano, Annamaria, Sollitto, Donato, Modugno, Francesca, Buttafuoco, Gabriele, & Papa, Giuseppe lo. 2012. Integrating geophysical and geostatistical techniques to map the spatial variation of clay. Geoderma, 171–172, 53-63

De Jong, E., Ballantyne, A.K., Caneron, D.R., Read, D.W. 1979. Measurement of apparent electrical conductivity of soils by an electromagnetic induction probe to aid salinity surveys. Soil Science Society of America Journal, 43, 810-812.

De Smedt, P., Van Meirvenne, M., Meerschman, E., Saey, T., Bats, M., Court-Picon, M., . . . Crombé, P. 2011. Reconstructing palaeochannel morphology with a mobile multicoil electromagnetic induction sensor. , 130, 136-141.

Diaz, L., & Herrero, J. 1992. Salinity estimates in irrigated soils using electromagnetic induction. Soil Science, 154, 151-157.

Dickson, B. L. 2004. Recent advances in aerial gamma-ray surveying. Journal of Environmental Radioactivity, 76, 225-236.

Dickson, B.L., Scott, K.M. 1997. Interpretation of aerial gamma-ray surveys –adding the geochemical factors. AGSO Journal of Australian Geology and Geophysics, 17, 187-200.

Dierke, C., & Werban, U. 2013. Relationships between gamma-ray data and soil properties at an agricultural test site. Geoderma, 199, 90-98.

Dixit, P. N., & Chen, D. 2010. Modification of a spatially referenced crop model to simulate the effect of spatial pattern of subsoil salinity. Computers and Electronics in Agriculture, 74, 313-320.

62

Dixon, P. 1989. Dryland salinity in a subcatchment at Glenthompson, Victoria. Australian Geographer, 20(2), 144-152.

Domsch, H., & Giebel, A. 2004. Estimation of soil textural features from soil electrical conductivity recorded using the EM38. Precision Agriculture, 5, 389-409.

Doolittle, J. A., & Collins, M. E. 1998. A comparison of EM induction and GPR methods in areas of karst. Geoderma, 85, 83-102.

Doolittle, J. A., Indorante, S. J., Potter, D. K., Hefner, S. G., & McCauley, W. M. 2002. Comparing three geophysical tools for locating sand blows in Alluvial soils of southeast Missouri. Journal of Soil and Water Conservation, 57, 175-182.

Doolittle, J. A., Sudduth, K. A., Kitchen, N. R., & Indorante, S. J. 1994. Estimating depths to claypans using electromagnetic induction methods. Journal of Soil & Water Conservation, 49, 572-575.

Doolittle, J., Petersen, M., & Wheeler, T. 2001. Comparison of two electromagnetic induction tools in salinity appraisals. Journal of Soil and Water Conservation, 56, 257-262.

DUALEM-421S. 2008. User’s Manual. Dualem Inc., Canada.

Dunn, B. W., & Beecher, H. G. 2007. Using electro-magnetic induction technology to identify sampling sites for soil acidity assessment and to determine spatial variability of soil acidity in rice fields. Australian Journal of Experimental Agriculture, 47, 208-214.

Eigenberg, R. A., Doran, J. W., Nienaber, J. A., Ferguson, R. B., & Woodbury, B. L. 2002. Electrical conductivity monitoring of soil condition and available N with animal manure and a cover crop. Agriculture, Ecosystems and Environment, 88, 183-193.

Eigenberg, R. A., & Nienaber, J. A. 1998. Electromagnetic survey of cornfield with repeated manure applications. Journal of Environmental Quality, 27, 1511-1515.

Eigenberg, R. A., & Nienaber, J. A. . 1999. Soil conductivity map differences for monitoring temporal changes in an agronomic field. American Society of Agricultural Engineers Paper, 993173.

Eigenberg, Roger A., Nienaber, John A., Woodbury, Bryan L., & Ferguson, Richard B. 2006. Soil Conductivity as a Measure of Soil and Crop Status. Soil Science Society of America Journal, 70, 1600-1611.

European Commission. 2006. Soil protection – the long story behind the strategy. Office for Official Publications of the European Communities, Luxembourg.

Feikema, P. M., & Baker, T. G. 2011. Effect of soil salinity on growth of irrigated plantation Eucalyptus in south-eastern Australia. Agricultural Water Management, 98, 1180-1188.

Foote, R. S., & Humphrey, N. B. 1976. Airborne radiometric techniques and applications to uranium exploration. Exploration For uranium Ore Deposits, Vienna.

Fraisse, C. W., Sudduth, K. A., & Kitchen, N. R. 2001. Delineation of site-specific management zones by unsupervised classification of topographic attributes and soil electrical conductivity. Transactions of the American Society of Agricultural Engineers, 44, 155-166.

Franz, T. E., King, E. G., Caylor, K. K., & Robinson, D. A. 2011. Coupling vegetation organization patterns to soil resource heterogeneity in a central Kenyan dryland using geophysical imagery. Water Resources Research, 47.

63

Friedman, S. P. 2005. Soil properties influencing apparent electrical conductivity: A review. Computers and Electronics in Agriculture, 46(1-3 SPEC. ISS.), 45-70.

Fulton, A., Schwankl, L., Lynn, K., Lampinen, B., Edstrom, J., & Prichard, T. 2011. Using EM and VERIS technology to assess land suitability for orchard and vineyard development. Irrigation Science, 29, 497-512.

Galbraith, J. H., & Saunders, D. F. 1983. Rock classification by characteristics of aerial gamma-ray measurements. Journal of Geochemical Exploration, 18, 49-73.

Ganjegunte, G., Leinauer, B., Schiavon, M., & Serena, M. 2013. Using electro-magnetic induction to determine soil salinity and sodicity in turf root zones. Agronomy Journal, 105, 836-844.

Goes, B. J. M., Oude Essink, G. H. P., Vernes, R. W., & Sergi, F. 2009. Estimating the depth of fresh and brackish groundwater in a predominantly saline region using geophysical and hydrological methods, zeeland, the netherlands. Near Surface Geophysics, 7, 401-412.

Gooley, L., Huang, J., Page, D., & Triantafilis J., 2013. Digital soil mapping available water content using proximal and remotely sensed data. Soil Use and Management, in press.

Grasty, R.L., 1975. Uranium measurement by airborne gamma-ray spectrometry. Geophysics, 40, 503- 519.

Grasty, R.L., Mellander, H., and Parker, M., 1991, Airborne gamma ray Spectrometer Surveying. International Atomic Energy Agency, Vienna, 97

Grasty, R. L. 1997. Radon emanation and soil moisture effects on airborne gamma-ray measurements. Geophysics, 62, 1379-1385.

Gregory, A.F. & Horwood, J.L., 1961. A laboratory study of gamma-ray spectra at the surface of rocks. Department of Energy, Mines & Resources, Ottawa, Mines Branch Research Report R85.

Grellier, S., Florsch, N., Camerlynck, C., Janeau, J. L., Podwojewski, P., & Lorentz, S. 2013. The use of Slingram EM38 data for topsoil and subsoil geoelectrical characterization with a Bayesian inversion. Geoderma, 200-201, 140-155.

Guimaraes, S. N. P., Hamza, V. M., & da Silva, J. J. 2013. Airborne geophysical surveys in the north- central region of Goias (Brazil): Implications for radiometric characterization of tropical soils. Journal of Environmental Radioactivity, 116, 10-18.

Guo, Y., Shi, Z., Li, H. Y., & Triantafilis, J. 2013a. Application of digital soil mapping methods for identifying salinity management classes based on a study on coastal central China. Soil Use and Management, 29, 445-456.

Guo, Y., Shi, Z., Zhou, L. Q., Jin, X., Tian, Y. F., & Teng, H. F. 2013b. Integrating Remote Sensing and Proximal Sensors for the Detection of Soil Moisture and Salinity Variability in Coastal Areas. Journal of Integrative Agriculture, 12, 723-731.

Halvorson, A. D., & Rhoades, J. D. 1974. Assessing Soil Salinity and Identifying Potential Saline-Seep Areas with Field Soil Resistance Measurements1. Soil Science Society of America Journal, 38, 576- 581.

Hanson, B. R., & Kaita, K. 1997. Response of electromagnetic conductivity meter to soil salinity and soil-water content. Journal of Irrigation and Drainage Engineering, 123, 141-143.

64

Harvey, O. R., & Morgan, C. L. S. 2009. Predicting regional-scale soil variability using a single calibrated apparent soil electrical conductivity model. Soil Science Society of America Journal, 73, 164-169.

Hartemink, A. E., Krasilnikov, P., Bockheima, J.G. 2013. Soil maps of the world. Geoderma, 207–208, 256-267.

Haskard, K. A., Rawlins, B. G., & Lark, R. M. 2010. A linear mixed model, with non-stationary mean and covariance, for soil potassium based on gamma radiometry. Biogeosciences, 7, 2081-2089.

Hbirkou, C., Welp, G., Rehbein, K., Hillnhütter, C., Daub, M., Oliver, M. A., & Pätzold, S. 2011. The effect of soil heterogeneity on the spatial distribution of Heterodera schachtii within sugar beet fields. Applied , 51, 25-34.

Hedley, C. B., Roudier, P., Yule, I. J., Ekanayake, J., & Bradbury, S. 2013. Soil water status and water table depth modelling using electromagnetic surveys for precision irrigation scheduling. Geoderma, 199, 22-29.

Hedley, C. B., Yule, I. J., Eastwood, C. R., Shepherd, T. G., & Arnold, G. 2004. Rapid identification of soil textural and management zones using electromagnetic induction sensing of soils. Australian Journal of Soil Research, 42, 389-400.

Heil, K., & Schmidhalter, U. 2012. Characterisation of soil texture variability using the apparent soil electrical conductivity at a highly variable site. Computers and Geosciences, 39, 98-110.

Hendrickx, J. M. H., Baerends, B., Raza, Z. I., Sadig, M., & Chaudhry, M. A. 1992. Soil salinity assessment by electromagnetic induction of irrigated land. Soil Science Society of America Journal, 56, 1933-1941.

Hendrickx, J.M.H., Kachanoski, R.G., 2002. Solute content and concentration – indirect measurement of solute concentration – nonintrusive electromagnetic induction. In: Dane, J.H., Topp, G.C. (Eds.), Methods of Soil Analysis, Part 4 –Physical Methods. Soil Sci. Soc. Am. Book Ser. 5. Soil Science Society of America, Madison, WI, USA, 1297–1306.

Herbst, M., Prolingheuer, N., Graf, A., Huisman, J. A., Weihermüller, L., Vanderborght, J., & Vereecken, H. 2010. Multivariate conditional stochastic simulation of soil heterotrophic respiration at plot scale. Geoderma, 160, 74-82.

Herrero, J., Ba, A. A., & Aragüés, R. 2003. Soil salinity and its distribution determined by soil sampling and electromagnetic techniques. Soil Use and Management, 19(2), 119-126.

Herrero, J., & Hudnall, W. H. 2013. Measurement of soil salinity using electromagnetic induction in a paddy with a densic pan and shallow water table. Paddy and Water Environment, 1-12.

Herrero, J., Netthisinghe, A., Hudnall, W. H., & Pérez-Coveta, O. 2011. Electromagnetic induction as a basis for soil salinity monitoring within a Mediterranean irrigation district. Journal of Hydrology, 405, 427-438.

Hezarjaribi, A., & Sourell, H. 2007. Feasibility study of monitoring the total available water content using non-invasive electromagnetic induction-based and electrode-based soil electrical conductivity measurements. Irrigation and Drainage, 56, 53-65.

Horney, R. D., Taylor, B., Munk, D. S., Roberts, B. A., Lesch, S. M., & Plant, R. E. (2005). Development of practical site-specific management methods for reclaiming salt-affected soil. Computers and Electronics in Agriculture, 46(1-3 SPEC. ISS.), 379-397.

65

Horsfall, K. R. 1997. Airborne magnetic and gamma-ray data acquisition. AGSO Journal of Australian Geology and Geophysics, 17, 23-30.

Hossain, M. B., Lamb, D. W., Lockwood, P. V., & Frazier, P. 2010. EM38 for volumetric soil water content estimation in the root-zone of deep vertosol soils. Computers and Electronics in Agriculture, 74, 100-109.

Huang, J., Taghizadeh-Mehrjardi, R., Minasny, B., & Triantafilis, J. 2015. Modeling soil salinity along a hillslope in Iran by inversion of EM38 data. Soil Science Society of America Journal, 79(4), 1142- 1153.

Huang, J., Lark, R. M., Robinson, D. A., Lebron, I., Keith, A. M., Rawlins, B., Tye, A., Kuras, O., Raines, M. & Triantafilis, J. 2014a. Scope to predict soil properties at within-field scale from small samples using proximally sensed γ-ray spectrometer and EM induction data. Geoderma, 232, 69-80.

Huang, J., Davies, G. B., Bowd, D., Monteiro Santos, F. A., & Triantafilis, J. 2014b. Spatial prediction of the exchangeable sodium percentage at multiple depths using electromagnetic inversion modelling. Soil use and management, 30(2), 241-250.

Huang, J., Scudiero, E., Clary, W., Corwin, D. L., & Triantafilis, J. 2017a. Time-lapse monitoring of soil water content using electromagnetic conductivity imaging. Soil Use and Management, in press. DOI:10.1111/sum.12261.

Huang, J., McBratney, A. B., Minasny, B., & Triantafilis, J. 2017b. Monitoring and modelling soil water dynamics using electromagnetic conductivity imaging and the ensemble Kalman filter. Geoderma, 285, 76-93.

Hyvönen, E., Pänttäjä, M., Sutinen, M. L., & Sutinen, R. 2003. Assessing site suitability for Scots pine using airborne and terrestrial gamma-ray measurements in Finnish Lapland. Canadian Journal of Forest Research, 33, 796-806.

Inman, Daniel J., Freeland, Robert S., Ammons, John T., & Yoder, Ronald E. 2002. Soil Investigations using Electromagnetic Induction and Ground-Penetrating Radar in Southwest Tennessee. Soil Sci. Soc. Am. J., 66, 206-211.

Inman, D. J., Freeland, R. S., Yoder, R. E., Ammons, J. T., & Leonard, L. L. 2001. Evaluating GPR and EMI for morphological studies of loessial soils. Soil Science, 166, 622-630.

Islam, M. M., Cockx, L., Meerschman, E., de Smedt, P., Meeuws, F., & Van Meirvenne, M. 2011b. A floating sensing system to evaluate soil and crop variability within flooded paddy rice fields. Precision Agriculture, 12, 850-859.

Islam, M. M., Meerschman, E., Saey, T., de Smedt, P., van de Vijver, E., & Van Meirvenne, M. 2012. Comparing apparent electrical conductivity measurements on a paddy field under flooded and drained conditions. Precision Agriculture, 13, 384-392.

Islam, M. M., Saey, T., Meerschman, E., De Smedt, P., Meeuws, F., Van De Vijver, E., & Van Meirvenne, M. 2011a. Delineating water management zones in a paddy rice field using a Floating Soil Sensing System. Agricultural Water Management, 102, 8-12.

James, I. T., Waine, T. W., Bradley, R. I., Taylor, J. C., & Godwin, R. J. 2003. Determination of Soil Type Boundaries using Electromagnetic Induction Scanning Techniques. Biosystems Engineering, 86, 421-430.

66

Jaques, A.L., Wellman, P., Whitaker, A., Wyborn, D., 1997, High resolution geophysics in modern geological mapping. AGSO Journal of Australian Geology & Geophysics, 17, 159-174.

Jaynes, D. B., Colvin, T. S., & Kaspar, T. C. 2005. Identifying potential soybean management zones from multi-year yield data. Computers and Electronics in Agriculture, 46, 309-327.

Jiang, P., Anderson, S. H., Kitchen, N. R., Sadler, E. J., & Sudduth, K. A. 2007. Landscape and conservation management effects on hydraulic properties of a claypan-soil toposequence. Soil Science Society of America Journal, 71, 803-811.

Jiang, P., He, Z., Kitchen, N. R., & Sudduth, K. A. 2009. Bayesian analysis of within-field variability of corn yield using a spatial hierarchical model. Precision Agriculture, 10, 111-127.

Johnson, C. K., Eigenberg, R. A., Doran, J. W., Wienhold, B. J., Eghball, B., & Woodbury, B. L. 2005b. Status of soil electrical conductivity studies by central state researchers. Transactions of the American Society of Agricultural Engineers, 48, 979-989.

Johnson, C. K., Eskridge, K. M., & Corwin, D. L. 2005a. Apparent soil electrical conductivity: Applications for designing and evaluating field-scale experiments. Computers and Electronics in Agriculture, 46(1-3 SPEC. ISS.), 181-202.

Johnston, M. A., Savage, M. J., Moolman, J. H., & Du Plessis, H. M. 1997. Evaluation of calibration methods for interpreting soil salinity from electromagnetic induction measurements. Soil Science Society of America Journal, 61, 1627-1633.

Jonard, François, Mahmoudzadeh, Mohammad, Roisin, Christian, Weihermüller, Lutz, André, Frédéric, Minet, Julien, . . . Lambot, Sébastien. 2013. Characterization of tillage effects on the spatial variation of soil properties using ground-penetrating radar and electromagnetic induction. Geoderma, 207–208, 310-322.

Jung, W. K., Kitchen, N. R., Sudduth, K. A., & Anderson, S. H. 2006. Spatial characteristics of claypan soil properties in an agricultural field. Soil Science Society of America Journal, 70, 1387-1397.

Jung, W. K., Kitchen, N. R., Sudduth, K. A., Kremer, R. J., & Motavalli, P. P. 2005. Relationship of apparent soil electrical conductivity to claypan soil properties. Soil Science Society of America Journal, 69, 883-892.

Kachanoski, R. G., Gregorich, E. G., & Van Wesenbeeck, I. J. 1988. Estimating spatial variations of soil water content using noncontacting electromagnetic inductive methods. Canadian Journal of Soil Science, 68, 715-722.

Kachanoski, R. G., Wesenbeeck, I. J. Van, & Jong, E. De. 1990. Field scale patterns of soil water storage from non-contacting measurements of bulk electrical conductivity. Canadian Journal of Soil Science, 70, 537-542.

Kaffka, S. R., Lesch, S. M., Bali, K. M., & Corwin, D. L. 2005. Site-specific management in salt- affected sugar beet fields using electromagnetic induction. Computers and Electronics in Agriculture, 46(1-3 SPEC. ISS.), 329-350.

Kaman, H., Çetin, M., & Kirda, C. 2011. Monitoring and assessing of changes in soil and groundwater salinity of yemisli irrigation district of Turkey using low quality irrigation water. Scientific Research and Essays, 6, 1388-1396.

Kerry, R., & Oliver, M. A. 2007. Comparing sampling needs for variograms of soil properties computed by the method of moments and residual maximum likelihood. Geoderma, 140, 383-396.

67

Khakural, B. R., Robert, P. C., & Hugins, D. R. 1998. Use of non-contacting electromagnetic inductive method for estimating soil moisture across a landscape. Communications in Soil Science and Plant Analysis, 29, 2055-2065.

Kinal, J., Stoneman, G. L., & Williams, M. R. 2006. Calibrating and using an EM31 electromagnetic induction meter to estimate and map soil salinity in the jarrah and karri forests of south-western Australia. Forest Ecology and Management, 233, 78-84.

Kitchen, N. R., Sudduth, K. A., & Drummond, S. T. 1996. Mapping of sand deposition from 1993 midwest floods with electromagnetic induction measurements. Journal of Soil and Water Conservation, 51, 336-340.

Kitchen, N. R., Sudduth, K. A., Myers, D. B., Drummond, S. T., & Hong, S. Y. 2005. Delineating productivity zones on claypan soil fields using apparent soil electrical conductivity. Computers and Electronics in Agriculture, 46(1-3 SPEC. ISS.), 285-308.

Knappenberger, T., & Köller, K. 2012. Spatial assessment of the correlation of seeding depth with emergence and yield of corn. Precision Agriculture, 13, 163-180.

Knotters, M., Brus, D. J., & Oude Voshaar, J. H. 1995. A comparison of kriging, co-kriging and kriging combined with regression for spatial interpolation of horizon depth with censored observations. Geoderma, 67, 227-246.

Lacoste, M., Lemercier, B., & Walter, C. 2011. Regional mapping of soil parent material by machine learning based on point data. Geomorphology, 133, 90-99.

Lamandé, Mathieu, Labouriau, Rodrigo, Holmstrup, Martin, Torp, Søren B., Greve, Mogens H., Heckrath, Goswin, . . . Jacobsen, Ole H. (2011). Density of macropores as related to soil and earthworm community parameters in cultivated grasslands. Geoderma, 162(3–4), 319-326.

Lemercier, B., Lacoste, M., Loum, M., & Walter, C. (2012). Extrapolation at regional scale of local soil knowledge using boosted classification trees: A two-step approach. Geoderma, 171-172, 75-84.

Leong, E. C., He, L., and Rahardjo, H. 2002. Factors affecting the filter paper method for total and matric suction measurements. Geotechnical Testing Journal, 25, 321-332.

Lesch, S. M. 2005. Sensor-directed response surface sampling designs for characterizing spatial variation in soil properties. Computers and Electronics in Agriculture, 46, 153-179.

Lesch, S. M., & Corwin, D. L. 2008. Prediction of spatial soil property information from ancillary sensor data using ordinary linear regression: Model derivations, residual assumptions and model validation tests. Geoderma, 148, 130-140.

Lesch, S. M., Corwin, D. L., & Robinson, D. A. 2005. Apparent soil electrical conductivity mapping as an agricultural management tool in arid zone soils. Computers and Electronics in Agriculture, 46(1-3 SPEC. ISS.), 351-378.

Lesch, S. M., Herrero, J., & Rhoades, J. D. 1998. Monitoring for temporal changes in soil salinity using electromagnetic induction techniques. Soil Science Society of America Journal, 62, 232-242.

Lesch, S. M., Rhoades, J.D., Corwin, D.L., 2000. ESAP-95 version 2.10R: User manual and tutorial guide, Research Rpt. 146. USDA-ARS, U.S. Salinity Laboratory, Riverside, CA, USA.

Lesch, S. M., Rhoades, J. D., Lund, L. J., & Corwin, D. L. 1992. Mapping soil salinity using calibrated electromagnetic measurements. Soil Science Society of America Journal, 56, 540-548.

68

Lesch, S. M., Strauss, D. J., & Rhoades, J. D. 1995a. Spatial prediction of soil salinity using electromagnetic induction techniques. 1. Statistical prediction models: a comparison of multiple linear regression and cokriging. Water Resources Research, 31, 373-386.

Lesch, S. M., Strauss, D. J., & Rhoades, J. D. 1995b. Spatial prediction of soil salinity using electromagnetic induction techniques. 2. An efficient spatial sampling algorithm suitable for multiple linear regression model identification and estimation. Water Resources Research, 31, 387-398.

Li, H. Y., Shi, Z., Webster, R., & Triantafilis, J. 2013a. Mapping the three-dimensional variation of soil salinity in a rice-paddy soil. Geoderma, 195-196, 31-41.

Li, H.Y., Wu, C.F., Li, F.H., & Shi, Z. 2013b. Three-dimensional variation of electrical conductivity in a paddy rice soil based on the disjunctive kriging method. Research Journal of Applied Sciences, Engineering and Technology, 6, 1486-1489.

Li, X. M., Yang, J. S., Liu, M. X., Liu, G. M., & Yu, M. 2012. Spatio-Temporal Changes of Soil Salinity in Arid Areas of South Xinjiang Using Electromagnetic Induction. Journal of Integrative Agriculture, 11, 1365-1376.

Lobell, D. B., Lesch, S. M., Corwin, D. L., Ulmer, M. G., Anderson, K. A., Potts, D. J., . . . Baltes, M. J. 2010. Regional-scale assessment of soil salinity in the Red River Valley using multi-year MODIS EVI and NDVI. Journal of Environmental Quality, 39, 35-41.

López-Bruna, D., & Herrero, J. 1996. The behaviour of the electromagnetic sensor and its calibration for soil salinity. El comportamiento del sensor electromagnético y su calibración frente a la salinidad edáfica, 16, 95-105.

López-Lozano, R., Casterad, M. A., & Herrero, J. 2010. Site-specific management units in a commercial maize plot delineated using very high resolution remote sensing and soil properties mapping. Computers and Electronics in Agriculture, 73, 219-229.

Lück, E., Ruehlmann, J., & Kirchmann, H. 2011. Properties of soils from the swedish long-term fertility experiments: VI. mapping soil electrical conductivity with different geophysical methods. Acta Agriculturae Scandinavica Section B: Soil and Plant Science, 61, 438-447.

Ma, R., McBratney, A., Whelan, B., Minasny, B., & Short, M. 2011. Comparing temperature correction models for soil electrical conductivity measurement. Precision Agriculture, 12, 55-66.

Ma, R., Short, M., Lobsey, C., McBratney, A., Whelan, B., Minasny, B., & Sun, G. 2010. Integrated multi-sensor hardware system for soil information measurement. Nongye Gongcheng Xuebao/Transactions of the Chinese Society of Agricultural Engineering, 26, 156-161.

Mahmood, H. S., Hoogmoed, W. B., & van Henten, E. J. 2012. Sensor data fusion to predict multiple soil properties. Precision Agriculture, 13, 628-645.

Mankin, K. R., Ewing, K. L., Schrock, M. D., & Kluitenberg, G. J. 1997. Field measurement and mapping of soil salinity in saline seeps, Minneapolis, MN, USA.

Mankin, K. R., & Karthikeyan, R. 2002. Field assessment of saline seep remediation using electromagnetic induction. Transactions of the American Society of Agricultural Engineers, 45, 99- 107.

Mann, K. K., Schumann, A. W., & Obreza, T. A. 2011. Delineating productivity zones in a citrus grove using citrus production, tree growth and temporally stable soil data. Precision Agriculture, 12, 457- 472.

69

Martinez, G., Vanderlinden, K., Ordóñez, R., & Muriel, J. L. 2009. Can apparent electrical conductivity improve the spatial characterization of soil organic carbon? Vadose Zone Journal, 8, 586-593.

McBratney, Alex B., Odeh, Inakwu O. A., Bishop, Thomas F. A., Dunbar, Marian S., & Shatar, Tamara M. 2000. An overview of pedometric techniques for use in soil survey. Geoderma, 97, 293-327.

McBratney, A. B., Santos, M. L. M., & Minasny, B. 2003. On digital soil mapping. Geoderma, 117, 3-52.

McCafferty, A. E., & Van Gosen, B. S. 2009. Airborne gamma-ray and magnetic anomaly signatures of serpentinite in relation to soil , northern California. Applied Geochemistry, 24, 1524- 1537.

McKenzie, N.J. 1992.Soils of the lower Macquarie Valley, New South Wales, CSIRO division of soils. Divisional Report No 117, Canberra, ACT.

McKenzie, R. C., Chomistek, W., & Clark, N. F. 1989. Conversion of electromagnetic inductance readings to saturated paste extract values in soils for different temperature, texture, and moisture conditions. Canadian Journal of Soil Science, 69, 25-32.

McKenzie, R. C., George, R. J., Woods, S. A., Cannon, M. E., & Bennett, D. L. 1997. Use of the electromagnetic-induction meter (EM38) as a tool in managing salinisation. Journal, 5, 37-50.

McLeod, M. K., Slavich, P. G., Irhas, Y., Moore, N., Rachman, A., Ali, N., . . . Caniago, C. 2010. Soil salinity in Aceh after the December 2004 Indian Ocean tsunami. Agricultural Water Management, 97, 605-613.

McNeill, J. D., 1990. Geonics EM38 ground conductivity meter: EM38 op-erating manual: Geonics Limited.

Meerschman, E., Van Meirvenne, M., Mariethoz, G., Islam, M. M., De Smedt, P., Van De Vijver, E., & Saey, T. 2013a. Using bivariate multiple-point statistics and proximal soil sensor data to map fossil ice-wedge polygons. Geoderma, 213, 571-577.

Meerschman, E., Van Meirvenne, M., Van De Vijver, E., De Smedt, P., Islam, M. M., & Saey, T. 2013b. Mapping complex soil patterns with multiple-point . European Journal of Soil Science, 64, 183-191.

Metternicht, G. I., & Zinck, J. A. 2003. Remote sensing of soil salinity: potentials and constraints. Remote Sensing of Environment, 85, 1-20.

Metzger, A. E., & Parker, R. E. 1979. The distribution of titanium on the lunar surface. Earth and Planetary Science Letters, 45, 155-171.

Minty, B. R. S. 1997. Fundamentals of airbone gamma-ray spectrometry. AGSO Journal of Astralian Geology and Geophysics, 17, 39-50.

Minty, B., Franklin, R., Milligan, P., Richardson, M., & Wilford, J. 2009. The Radiometric Map of Australia*. Exploration Geophysics, 40(4), 325-333.

Moffett, K. B., Robinson, D. A., & Gorelick, S. M. 2010. Relationship of Salt Marsh Vegetation Zonation to Spatial Patterns in Soil Moisture, Salinity, and Topography. Ecosystems, 13, 1287-1302.

70

Morari, F., Castrignanò, A., & Pagliarin, C. 2009. Application of multivariate geostatistics in delineating management zones within a gravelly vineyard using geo-electrical sensors. Computers and Electronics in Agriculture, 68, 97-107.

Morris, Evan R. 2009. Height-above-ground effects on penetration depth and response of electromagnetic induction soil conductivity meters. Computers and Electronics in Agriculture, 68, 150-156.

Mougenot, B., Pouget, M., & Epema, G. F. 1993. Remote sensing of salt affected soils. Remote Sensing Reviews, 7, 241-259.

Mulder, V. L., De Bruin, S., Schaepman, M. E., & Mayr, T. R. 2011. The use of remote sensing in soil and terrain mapping - A review. Geoderma, 162, 1-19.

Myers, D. B., Kitchen, N. R., Sudduth, K. A., Sharp, R. E., & Miles, R. J. 2007. Soybean root distribution related to claypan soil properties and apparent soil electrical conductivity. Crop Science, 47, 1498-1509.

Naderi-Boldaji, M., Sharifi, A., Alimardani, R., Hemmat, A., Keyhani, A., Loonstra, E. H., . . . Keller, T. 2013. Use of a triple-sensor fusion system for on-the-go measurement of soil compaction. Soil and Tillage Research, 128, 44-53.

Nam, Soonkie, Gutierrez, Marte, Diplas, Panayiotis, Petrie, John, Wayllace, Alexandria, Lu, Ning, & Muñoz, Juan Jorge. 2010. Comparison of testing techniques and models for establishing the SWCC of riverbank soils. Engineering Geology, 110, 1-10.

Nelson, M. A., Bishop, T. F. A., Triantafilis, J., & Odeh, I. O. A. 2011. An error budget for different sources of error in digital soil mapping. European Journal of Soil Science, 62, 417-430.

Nettleton, W. D., Bushue, L., Doolittle, J. A., Endres, T. J., & Indorante, S. J. 1994. Sodium-affected soil identification in south-central Illinois by electromagnetic induction. Soil Science Society of America Journal, 58, 1190-1193.

Nogués, J., Robinson, D. A., & Herrero, J. 2006. Incorporating electromagnetic induction methods into regional soil salinity survey of irrigation districts. Soil Science Society of America Journal, 70, 2075- 2085.

Northcote, K.H., Isbell, R.F., McArthur, W.M., Van Dijk, D.C., 1965. Atlas of Australian Soils, Sheet 3: Sydney –Canberra– Bourke–Armidale Area. CSIRO Australia and Melbourne University Press, Melbourne.

Odeh, I. O. A., Todd, A. J., Triantafilis, J., & McBratney, A. B. 1998. Status and trends of soil salinity at different scales: The case for the irrigated cotton growing region of eastern Australia. Nutrient Cycling in Agroecosystems, 50, 99-107.

O'Neill, C. J., Humphreys, E., Louis, J., & Katupitiya, A. 2008. Maize productivity in southern New South Wales under furrow and pressurised irrigation. Australian Journal of Experimental Agriculture, 48, 285-295.

Padhi, J., & Misra, R. K. 2011. Sensitivity of EM38 in determining soil water distribution in an irrigated wheat field. Soil and Tillage Research, 117, 93-102.

Paine, J. G. 2003. Determining salinization extent, identifying salinity sources, and estimating chloride mass using surface, borehole, and airborne electromagnetic induction methods. Water Resources Research, 39, HWC31-HWC310.

71

Paine, J. G., Goldsmith, R. S., & Scanlon, B. R. 1998. Electrical conductivity and gamma-ray response to clay, water, and chloride content in fissured sediments, Trans-Pecos Texas. Environmental and Engineering Geoscience, 4, 225-239.

Papanicolaou, A., Elhakeem, M., Wilson, C., Burras, C. and Oneal, B. 2008. Observations of soils at the hillslope scale in the Clear Creek watershed in Iowa, USA. . Soil Survey Horizons, 49, 83-86.

Petersen, H., Wunderlich, T., Attia Al Hagrey, S., & Rabbel, W. 2012. Characterization of some Middle European soil textures by gamma-spectrometry. Journal of Plant Nutrition and Soil Science, 175, 651-660.

Pickup, G. & Marks, A. 2000. Identifying large-scale erosion and deposition processes from airborne gamma radiometrics and digital elevation models in a weathered landscape. Earth Surface Processes and Landforms, 25, 535-557.

Piikki, K., Söderström, M., & Stenberg, B. 2013. Sensor data fusion for topsoil clay mapping. Geoderma, 199, 106-116.

Popp, S., Altdorff, D., & Dietrich, P. 2013. Assessment of shallow subsurface characterisation with non- invasive geophysical methods at the intermediate hill-slope scale. Hydrology and Earth System Sciences, 17, 1297-1307.

Pracilio, G., Adams, M. L., Smettem, K. R. J., & Harper, R. J. 2006. Determination of spatial distribution patterns of clay and plant available potassium contents in surface soils at the farm scale using high resolution gamma ray spectrometry. Plant and Soil, 282, 67-82.

Priori, S., Martini, E., Andrenelli, M. C., Magini, S., Agnelli, A. E., Bucelli, P., . . . Costantini, E. A. C. 2013. Improving wine quality through harvest zoning and combined use of remote and soil proximal sensing. Soil Science Society of America Journal, 77, 1338-1348.

Quenum, M., Nolin, M. C., & Bernier, M. 2012. Digital mapping of soil phosphorus maximum sorption capacity. Canadian Journal of Soil Science, 92, 733-750.

Rawlins, B. G., Lark, R. M., Webster, R., & O'Donnell, K. E. 2006. The use of soil survey data to determine the magnitude and extent of historic metal deposition related to atmospheric smelter emissions across Humberside, UK. Environmental Pollution, 143, 416-426.

Reedy, R. C., & Scanlon, B. R. 2003. Soil water content monitoring using electromagnetic induction. Journal of Geotechnical and Geoenvironmental Engineering, 129, 1028-1039.

Reid, J. E., Worby, A. P., Vrbancich, J., & Munro, A. I. S. 2003. Shipborne electromagnetic measurements of Antarctic sea-ice thickness. Geophysics, 68, 1537-1546.

Rhoades, J. D., Raats, P. A. C., & Prather, R. J. 1976. Effects of liquid-phase electrical conductivity, water content, and surface conductivity on bulk soil electrical conductivity. Soil Science Society of America Journal, 40(5), 651-655.

Rhoades, J. D., & Corwin, D. L. 1981. Determining Soil Electrical Conductivity-Depth Relations Using an Inductive Electromagnetic Soil Conductivity Meter. Soil Science Society of America Journal, 45, 255-260.

Rhoades, J. D., & Corwin, D. L. 1990. Soil electrical conductivity: effects of soil properties and application to soil salinity appraisal. Communications in Soil Science and Plant Analysis, 21, 837- 860.

72

Rhoades, J. D., & Ingvalson, R. D. 1971. Determining Salinity in Field Soils with Soil Resistance Measurements1. Soil Science Society of America Journal, 35, 54-60.

Rhoades, J. D., Manteghi, N. A., Shouse, P. J., & Alves, W. J. 1989. Soil electrical conductivity and soil salinity: New formulations and calibrations. Soil Science Society of America Journal, 53(2), 433-439.

Robinson, D. A., Lebron, I., Lesch, S. M., & Shouse, P. 2004. Minimizing drift in electrical conductivity measurements in high temperature environments using the EM-38. Soil Science Society of America Journal, 68(2), 339-345.

Robinson, D. A., Abdu, H., Jones, S. B., Seyfried, M., Lebron, I., & Knight, R. 2008. Eco-geophysical imaging of watershed-scale soil patterns links with plant community spatial patterns. Vadose Zone Journal, 7, 1132-1138.

Robinson, D. A., Abdu, H., Lebron, I., & Jones, S. B. 2012. Imaging of hill-slope soil moisture wetting patterns in a semi-arid oak savanna catchment using time-lapse electromagnetic induction. Journal of Hydrology, 416, 39-49.

Robinson, D. A., Jones, S. B., Wraith, J. M., Or, D., & Friedman, S. P. 2003. A review of advances in dielectric and electrical conductivity measurement in soils using time domain reflectometry. Vadose Zone Journal, 2, 444-475.

Robinson, D. A., Lebron, I., Kocar, B., Phan, K., Sampson, M., Crook, N., & Fendorf, S. 2009a. Time- lapse geophysical imaging of soil moisture dynamics in tropical deltaic soils: An aid to interpreting hydrological and geochemical processes. Water Resources Research, 45, W00D32.

Robinson, D. A., Lebron, I., Lesch, S. M., & Shouse, P. 2004. Minimizing drift in electrical conductivity measurements in high temperature environments using the EM-38. Soil Science Society of America Journal, 68, 339-345.

Robinson, D. A., Lebron, I., & Querejeta, J. I. 2010. Determining soil-tree-grass relationships in a California oak savanna using eco-geophysics. Vadose Zone Journal, 9, 528-536.

Robinson, N. J., Rampant, P. C., Callinan, A. P. L., Rab, M. A., & Fisher, P. D. 2009b. Advances in precision agriculture in south-eastern Australia. II. Spatio-temporal prediction of crop yield using terrain derivatives and proximally sensed data. Crop and Pasture Science, 60, 859-869.

Roca, V., Napolitano, M., Speranza, P. R., & Gialanella, G. 1989. Analysis of radioactivity levels in soils and crops from the Campania region (South Italy) after Chernobyl accident. Journal of Environmental Radioactivity, 9, 117-129.

Rodríguez-Pérez, J. R., Plant, R. E., Lambert, J. J., & Smart, D. R. 2011. Using apparent soil electrical conductivity (ECa) to characterize vineyard soils of high clay content. Precision Agriculture, 12, 775- 794.

Saey, T., De Smedt, P., De Clercq, W., Meerschman, E., Islam, M. M., & Van Meirvenne, M. 2013a. Identifying soil patterns at different spatial scales with a multi-receiver EMI sensor . Soil Science Society of America Journal, 77, 382-390.

Saey, T., De Smedt, P., Islam, M. M., Meerschman, E., Van De Vijver, E., Lehouck, A., & Van Meirvenne, M. 2012. Depth slicing of multi-receiver EMI measurements to enhance the delineation of contrasting subsoil features. Geoderma, 189-190, 514-521.

73

Saey, T., Simpson, D., Vermeersch, H., Cockx, L., & Van Meirvenne, M. 2009b. Comparing the EM38DD and DUALEM-21S sensors for depth-to-clay mapping. Soil Science Society of America Journal, 73, 7-12.

Saey, T., Van Meirvenne, M., De Smedt, P., Cockx, L., Meerschman, E., Islam, M. M., & Meeuws, F. 2011. Mapping depth-to-clay using fitted multiple depth response curves of a proximal EMI sensor. Geoderma, 162, 151-158.

Saey, T., Van Meirvenne, M., De Smedt, P., Neubauer, W., Trinks, I., Verhoeven, G., & Seren, S. 2013b. Integrating multi-receiver electromagnetic induction measurements into the interpretation of the soil landscape around the school of gladiators at Carnuntum. European Journal of Soil Science, 64, 716- 727.

Saey, T., Van Meirvenne, M., Vermeersch, H., Ameloot, N., & Cockx, L. 2009a. A pedotransfer function to evaluate the soil profile textural heterogeneity using proximally sensed apparent electrical conductivity. Geoderma, 150, 389-395.

Salama, R. B., Bartle, G., Farrington, P., & Wilson, V. 1994. Basin geomorphological controls on the mechanism of recharge and discharge and its effect on salt storage and mobilization - comparative study using geophysical surveys. Journal of Hydrology, 155, 1-26.

Sarkhot, D. V., Grunwald, S., Ge, Y., & Morgan, C. L. S. 2011. Comparison and detection of total and available fractions using visible/near infrared diffuse reflectance spectroscopy. Geoderma, 164, 22-32.

Schuler, U., Erbe, P., Zarei, M., Rangubpit, W., Surinkum, A., Stahr, K., & Herrmann, L. 2011. A gamma-ray spectrometry approach to field separation of illuviation-type WRB reference soil groups in northern Thailand. Journal of Plant Nutrition and Soil Science, 174, 536-544.

Schwarzer, T. F., & Adams, J. A. S. 1973. Rock and soil discrimination by low altitude airborne gamma- ray spectrometry in Payne County, Oklahoma. Economic Geology, 68, 1297-1312.

Scudiero, E., P. Teatini, D.L. Corwin, R. Delana, A. Berti, & F. Morari. 2013. Delineation of site- specific management units in a saline region at the Venice Lagoon margin, Italy using soil reflectance and apparent soil electrical conductivity. Computer and Electronics in Agriculture. 99: 54-64.

Serrano, J. M., Peça, J. O., Marques da Silva, J. R., & Shaidian, S. 2010. Mapping soil and pasture variability with an electromagnetic induction sensor. Computers and Electronics in Agriculture, 73, 7-16.

Serrano, J. M., Shahidian, S., & Marques da Silva, J. R. 2013. Small scale soil variation and its effect on pasture yield in southern Portugal. Geoderma, 195-196, 173-183.

Sharifi A., Godwin R.J., O’Dogherty M.J., and Dresser M.L., 2007. Evaluating the performance of a soil compaction sensor. Soil Use Management., 171-177.

Sheets, Keith R., & Hendrickx, Jan M. H. 1995. Noninvasive Soil Water Content Measurement Using Electromagnetic Induction. Water Resources Research, 31, 2401-2409.

Sherlock, M. D., & McDonnell, J. J. 2003. A new tool for hillslope hydrologists: Spatially distributed groundwater level and soilwater content measured using electromagnetic induction. Hydrological Processes, 17, 1965-1977.

74

Singhroy, V., Barnett, P.J., Assouad, P., Molch, K., 2003. Terrain interpretation from SAR techniques. International Geoscience and Remote Sensing Symposium (IGARSS)'03. Proceedings. 2003 IEEE International, 106-108.

Singleton, A. C., Osinski, G. R., Samson, C., Williamson, M. C., & Holladay, S. 2010. Electromagnetic characterization of polar ice-wedge polygons: Implications for periglacial studies on Mars and Earth. Planetary and Space Science, 58, 472-481.

Slavich, P. G. 1990. Determining ECa-depth profiles from electromagnetic induction measurements. Australian Journal of Soil Research, 28, 443-452.

Slavich, P.G., & Petterson, G.H. 1990 Estimating average rootzone salinity from electromagnetic induction (EM-38) measurements. Soil Research, 28, 453-463.

Snellen, M., Eleftherakis, D., Amiri-Simkooei, A., Koomans, R. L., & Simons, D. G. 2013. An inter- comparison of sediment classification methods based on multi-beam echo-sounder backscatter and sediment natural radioactivity data. Journal of the Acoustical Society of America, 134, 959-970.

Sommer, M., Wehrhan, M., Zipprich, M., Weller, U., Zu Castell, W., Ehrich, S., . . . Selige, T. 2003. Hierarchical data fusion for mapping soil units at field scale. Geoderma, 112, 179-196.

Spadoni, M., & Voltaggio, M. 2013. Contribution of gamma ground spectrometry to the textural characterization and mapping of floodplain sediments. Journal of Geochemical Exploration, 125, 20- 33.

Stroh, J. C., Archer, S., Doolittle, J. A., & Wilding, L. 2001. Detection of edaphic discontinuities with ground-penetrating radar and electromagnetic induction. Landscape Ecology, 16, 377-390.

Stroh, J. C., Archer, S. R., Wilding, L. P., & Doolittle, J. A. 1993. Assessing the influence of subsoil heterogeneity on vegetation in the Rio Grande Plains of south Texas using electromagnetic induction and geographical information system, College Station, TX, USA. The Station, 39-42.

Sudduth, K. A., Drummond, S. T., & Kitchen, N. R. 2001. Accuracy issues in electromagnetic induction sensing of soil electrical conductivity for precision agriculture. Computers and Electronics in Agriculture, 31, 239-264.

Sudduth, K. A., Kitchen, N. R., Bollero, G. A., Bullock, D. G., & Wiebold, W. J. 2003. Comparison of electromagnetic induction and direct sensing of soil electrical conductivity. Agronomy Journal, 95, 472-482.

Sudduth, K. A., Kitchen, N. R., Myers, D. B., & Drummond, S. T. 2010. Mapping depth to argillic soil horizons using apparent electrical conductivity. Journal of Environmental and Engineering Geophysics, 15, 135-146.

Sudduth, K. A., Kitchen, N. R., Wiebold, W. J., Batchelor, W. D., Bollero, G. A., Bullock, D. G., . . . Thelen, K. D. 2005. Relating apparent electrical conductivity to soil properties across the north- central USA. Computers and Electronics in Agriculture, 46(1-3 SPEC. ISS.), 263-283.

Sudduth, K. A., Myers, D. B., Kitchen, N. R., & Drummond, S. T. 2013. Modeling soil electrical conductivity-depth relationships with data from proximal and penetrating ECa sensors. Geoderma, 199, 12-21.

Sun, Y., Cheng, Q., Lin, J., Schellberg, J., & Schulze Lammers, P. 2013. Investigating soil physical properties and yield response in a grassland field using a dual-sensor penetrometer and EM38. Journal of Plant Nutrition and Soil Science, 176(2), 209-216.

75

Sun, Y., Druecker, H., Hartung, E., Hueging, H., Cheng, Q., Zeng, Q., . . . Schulze Lammers, P. 2011. Map-based investigation of soil physical conditions and crop yield using diverse sensor techniques. Soil and Tillage Research, 112(2), 149-158.

Taghizadeh-Mehrjardi, R., Minasny, B., Sarmadian, F., & Malone, B. P. 2014. Digital mapping of soil salinity in ardakan region, central iran. Geoderma, 213, 15-28.

Taylor, J. A., Coulouma, G., Lagacherie, P., & Tisseyre, B. 2009. Mapping soil units within a vineyard using statistics associated with high-resolution apparent soil electrical conductivity data and factorial discriminant analysis. Geoderma, 153, 278-284.

Thomas, M., Fitzpatrick, R. W., & Heinson, G. S. 2009. Distribution and causes of intricate salinesodic soil patterns in an upland South Australian hillslope. Australian Journal of Soil Research, 47, 328- 339.

Triantafilis, J., Ahmed, M. F., & Odeh, I. O. A. 2002. Application of a mobile electromagnetic sensing system (MESS) to assess cause and management of soil salinization in an irrigated cotton-growing field. Soil Use and Management, 18, 330-339.

Triantafilis, J., & Buchanan, S. M. 2009. Identifying common near-surface and subsurface stratigraphic units using EM34 signal data and fuzzy k-means analysis in the Darling River valley. Australian Journal of Earth Sciences, 56, 535-556.

Triantafilis, J., & Buchanan, S. M. 2010. Mapping the spatial distribution of subsurface saline material in the Darling River valley. Journal of Applied Geophysics, 70, 144-160.

Triantafilis, J., Gibbs, I., & Earl, N. 2013. Digital soil pattern recognition in the lower Namoi valley using numerical clustering of gamma-ray spectrometry data. Geoderma, 192, 407-421.

Triantafilis, J., Huckel, A. I., & Odeh, I. O. A. 2001b. Comparison of statistical prediction methods for estimating field-scale clay content using different combinations of ancillary variables. Soil Science, 166, 415-427.

Triantafilis, J., Kerridge, B., & Buchanan, S. M. 2009a. Digital soil-class mapping from proximal and remotely sensed data at the field level. Agronomy Journal, 101, 841-853.

Triantafilis, J., Laslett, G. M., & McBratney, A. B. 2000. Calibrating an electromagnetic induction instrument to measure salinity in soil under irrigated cotton. Soil Science Society of America Journal, 64, 1009-1017.

Triantafilis, J., & Lesch, S. M. 2005. Mapping clay content variation using electromagnetic induction techniques. Computers and Electronics in Agriculture, 46(1-3 SPEC. ISS.), 203-237.

Triantafilis, J., Lesch, S. M., La Lau, K., & Buchanan, S. M. 2009b. Field level digital soil mapping of cation exchange capacity using electromagnetic induction and a hierarchical spatial regression model. Australian Journal of Soil Research, 47, 651-663.

Triantafilis, J., Odeh, I. O. A., & McBratney, A. B. 2001a. Five geostatistical models to predict soil salinity from electromagnetic induction data across irrigated cotton. Soil Science Society of America Journal, 65, 869-878.

Tromp-van Meerveld, H. J., & McDonnell, J. J. 2009. Assessment of multi-frequency electromagnetic induction for determining soil moisture patterns at the hillslope scale. Journal of Hydrology, 368, 56- 67.

76

UNDP. 2007. Human development report 2007 ⁄ 2008. United Nations Development Programme, New York.

UNEP. 2007. Global environmental outlook GEO4 – environment for development. United Nations Environment Programme, Nairobi.

Urdanoz, V., Amezketa Lizarraga, E., Clavería, I., Ochoa, V., & Aragües Lafarga, R. 2008. Mobile and georeferenced electromagnetic sensors and applications for salinity assessment (Vol. 6).

Urdanoz, V., & Aragüé, R. 2011. Pre- and post-irrigation mapping of soil salinity with electromagnetic induction techniques and relationships with drainage water salinity. Soil Science Society of America Journal, 75, 207-215.

Urdanoz, V., & Aragüés, R. 2012. Comparison of Geonics EM38 and Dualem 1S electromagnetic induction sensors for the measurement of salinity and other soil properties. Soil Use and Management, 28, 108-112.

Valckx, J., Cockx, L., Wauters, J., Van Meirvenne, M., Govers, G., Hermy, M., & Muys, B. 2009. Within-field spatial distribution of earthworm populations related to species interactions and soil apparent electrical conductivity. Applied Soil Ecology, 41, 315-328.

Van Der Klooster, E., Van Egmond, F. M., & Sonneveld, M. P. W. 2011. Mapping soil clay contents in Dutch marine districts using gamma-ray spectrometry. European Journal of Soil Science, 62, 743- 753.

Van Der Lelij, A. 1983. Use of an electromagnetic induction instrument (type EM-38) for mapping of soil salinity. NSW Water Resources Commission, NSW, Australia.

Van Meirvenne, M., Islam, M. M., De Smedt, P., Meerschman, E., Van De Vijver, E., & Saey, T. 2013. Key variables for the identification of soil management classes in the aeolian landscapes of north- west Europe. Geoderma, 199, 99-105.

Vaughan, P. J., Lesch, S. M., Corwin, D. L., & Cone, D. G. 1995. Water content effect on soil salinity prediction: a geostatistical study using cokriging. Soil Science Society of America Journal, 59, 1146- 1156.

Viscarra Rossel, R. A., Adamchuk, V. I., Sudduth, K. A., McKenzie, N. J., & Lobsey, C. 2011. Proximal Soil Sensing. An Effective Approach for Soil Measurements in Space and Time. 113, 237-282.

Viscarra Rossel R.A., McBratney A.B. 1998. Soil chemical analytical accuracy and costs: implications from precision agriculture. Australian Journal of Experimental Agriculture, 38, 765-775

Viscarra Rossel, R. A., Taylor, H. J., & McBratney, A. B. 2007. Multivariate calibration of hyperspectral γ-ray energy spectra for proximal soil sensing. European Journal of Soil Science, 58, 343-353.

Visconti, F.,de Paz, J.M. 2016, Trends and Developments in Metrology, Chapter: 5, Publisher: InTech, Editors: Luigi Cocco, pp.99-126

Vitharana, U. W. A., Saey, T., Cockx, L., Simpson, D., Vermeersch, H., & Van Meirvenne, M. 2008c. Upgrading a 1/20,000 soil map with an apparent electrical conductivity survey. Geoderma, 148, 107- 112.

Vitharana, U. W. A., Van Meirvenne, M., Cockx, L., & Bourgeois, J. 2006. Identifying potential management zones in a layered soil using several sources of ancillary information. Soil Use and Management, 22, 405-413.

77

Vitharana, U. W. A., Van Meirvenne, M., Simpson, D., Cockx, L., & De Baerdemaeker, J. 2008b. Key soil and topographic properties to delineate potential management classes for precision agriculture in the European area. Geoderma, 143, 206-215.

Vitharana, U. W. A., Van Meirvenne, M., Simpson, D., Cockx, L., & Hofman, G. 2008a. Agronomic consequences of potential management zones delineated on the basis of EM38DD measurements. Near Surface Geophysics, 6, 289-296.

Vukašinović, I., Dordević, A., Rajković, M. B., Todorović, D., & Pavlović, V. B. 2010. Distribution of natural radionuclides in -type soil. Turkish Journal of Agriculture and Forestry, 34, 539-546.

Wagner, W., Pathe, C., Sabel, D., Bartsch, A., Kunzer, C., Scipal, K., 2007. Experimental 1 km soil moisture products from ENVISTAT ASAR for Southern Africal, SHARE project and the MISAR project. European Space Agency.

Weller, U., Zipprich, M., Sommer, M., Zu Castell, W., & Wehrhan, M. 2007. Mapping clay content across boundaries at the landscape scale with electromagnetic induction. Soil Science Society of America Journal, 71, 1740-1747. doi: 10.2136/sssaj2006.0177

Werban, U., Behrens, T., Cassiani, G., Dietrich, P. 2010 Proximal Soil Sensing. Progress In Soil Science, 1, 103-110. In: R.A. Viscarra Rossel, A. McBratney, and B. Minasny, editors, Springer, Berlin.

Werban, U., Kuka, K., & Merbach, I. 2009. Correlation of electrical resistivity, electrical conductivity and soil parameters at a long-term fertilization experiment. Near Surface Geophysics, 7, 5-14.

Wetterlind, J., Stenberg, B., & Söderström, M.. 2010. Increased sample point density in farm soil mapping by local calibration of visible and near infrared prediction models. Geoderma, 156, 152-160.

Wilford, J. 1995. Airborne gamma-ray spectrometry as a tool for assessing relative landscape activity and weathering development of regolith, including soils. AGSO Research Newsletter, 22, 12-14.

Wilford, J. 2009. Using airborne geophysics to define the 3D distribution and landscape evolution of Quaternary valley-fill deposits around the Jamestown area, South Australia. Australian Journal of Earth Sciences, 56(SUPPL. 1), S67-S88.

Wilford, J., & Minty, B. 2006. Chapter 16 The Use of Airborne Gamma-ray Imagery for Mapping Soils and Understanding Landscape Processes. Vol. 31 (pp. 207-218,609-610).

Wilford, J. R., Bierwirth, P. N., & Craig, M. A. 1997. Application of airborne gamma-ray spectrometry in soil/regolith mapping and applied geomorphology. AGSO Journal of Australian Geology and Geophysics, 17, 201-216.

Wilford, J. R., Dent, D. L., Dowling, T., & Braaten, R. 2001. Rapid mapping of soils and salt stores: Using airbone radiometrics and digital elevation models. AGSO Research Newsletter(34), 33-40.

Williams, B. G., & Baker, G. C. 1982. An electromagnetic induction technique for reconnaissance surveys of soil salinity hazards. Australian Journal of Soil Research, 20, 107-118.

Williams, B. G., & Fidler, F. T. 1983. The use of electromagnetic induction for locating subsurface saline material. Relation of Groundwater Quantity and Quality, IAHS Publication No. 146.

Williams, B. G., & Hoey, D. 1987. The use of electromagnetic induction to detect the spatial variability of the salt and clay contents of soils. Australian Journal of Soil Research, 25, 21-27.

78

Wollenhaupt, N. C., Richardson, J. L., Foss, J. E., & Doll, E. C. 1986. A rapid method for estimating weighted soil salinity from apparent soil electrical conductivity measured with an aboveground electromagnetic induction meter. Canadian Journal of Soil Science, 66, 315-321.

Wong, M. T. F., Asseng, S., Robertson, M. J., & Oliver, Y. 2008. Mapping subsoil acidity and shallow soil across a field with information from yield maps, geophysical sensing and the grower. Precision Agriculture, 9, 3-15.

Wong, M. T. F., & Harper, R. J. 1999. Use of on-ground gamma-ray spectrometry to measure plant- available potassium and other topsoil attributes. Australian Journal of Soil Research, 37, 267-277.

Woodbury, B. L., Eigenberg, R. A., Nienaber, J. A., & Spiehs, M. J. 2010. Soil-crop dynamic depth response determined from TDR of a corn silage field compared to EMI measurements. Journal of Environmental and Engineering Geophysics, 15, 185-196.

Woodbury, B. L., Eigenberg, R. A., Varel, V., Lesch, S., & Spiehs, M. J. 2011. Using electromagnetic induction technology to predict volatile fatty acid, source area differences. Journal of Environmental Quality, 40, 1416-1422.

Woodforth, A., Triantafilis, J., Cupitt J., Malik, R.S. & Geering, H. 2012. Mapping estimated deep drainage in the lower Namoi Valley using a chloride mass balance model and EM34 data. Geophysics, 77, WB245-256.

Yao, R. J., Yang, J. S., & Liu, G. M. 2007. Calibration of Soil Electromagnetic Conductivity in Inverted Salinity Profiles with an Integration Method. , 17, 246-256.

Yao, R., & Yang, J. 2010. Quantitative evaluation of soil salinity and its spatial distribution using electromagnetic induction method. Agricultural Water Management, 97, 1961-1970.

Yao, R., Yang, J., Zhao, X., Chen, X., Han, J., Li, X., . . . Shao, H. 2012. A New Soil Sampling Design in Coastal Saline Region Using EM38 and VQT Method. Clean - Soil, Air, Water, 40, 972-979.

Zare, E., Huang, J., Santos, F. A., & Triantafilis, J. 2015. Mapping Salinity in Three Dimensions using a DUALEM-421 and Electromagnetic Inversion Software. Soil Science Society of America Journal, 79(6), 1729-1740.

Zhu, Q., Lin, H., & Doolittle, J. 2010a. Repeated electromagnetic induction surveys for determining subsurface hydrologic dynamics in an agricultural landscape. Soil Science Society of America Journal, 74, 1750-1762.

Zhu, Q., Lin, H., & Doolittle, J. 2010b. Repeated electromagnetic induction surveys for improved soil mapping in an agricultural landscape. Soil Science Society of America Journal, 74, 1763-1774.

Zhu, Q., Lin, H. S., & Doolittle, J. A. 2013. Functional soil mapping for site-specific soil moisture and crop yield management. Geoderma, 200-201, 45-54.

79

3 Study area

3.1 Location The Murry Darling Basin is one of Australia’s most important national ecological and economic assets as it covers 14% of Australia’s land area (1.06 million km2 ) and supports

30,000 wetlands. Within this basin, 48% of all bird species and 28% of all mammal species of

Australia can be found (DEH 2004).

The study area is located mainly to the west of the township of Bourke (Fig. 3.1.1) covering 270 km2 which includes “Darling Farms”, “Janbeth”, “Ferguson’s Farm” and

“Allambi” located to the north of the Darling River.

Fig. 3.1.1 Regional setting. Bourke is located in the mid north-western section of the Murray-Darling Basin.

80

3.2 Climate Climate at the township of Bourke is semi-arid and has a mean annual rainfall of

355mm. During the month of February, rainfall has been recorded as the highest average.

However, January has the highest recorded rainfall with 185.9 mm. Also, January has the highest potential evaporation of 8.3mm a day with 2 mm a day being the lowest in July.

The mean maximum temperature of 36.3 degrees Celsius falls in January compared to a low 17.9 °C in July. Temperature and rainfall patterns of Bourke increase in a synchronised manner. This allows a balance to happen as with the increase in temperatures increase in rainfall

(Fig. 3.2.1) is observed. Given the rainfall of Bourke is slightly summer dominated, this compensates the effect of high temperatures on crops during the hot season and the higher availability of water offsets crop water stress.

Fig. 3.2.1 Rainfall and temperature averages for the township of Bourke (after BoM 2006).

81

3.3 Land use In 1894 irrigated agriculture started in the Bourke region with the installation of two government-owned bores. Irrigation systems with up to 2500 m3 a day had been installed in 2.6 km2 of land. It is reported that the new colony of Bourke settled by 1902.The dominated crops at the time were cotton, fruits and vegetables including apricots, bananas, pumpkins, lucerne, hay, maize sorghum and buckwheat. By 1918, the first signs of salinization were seen. The alkali salt was depositing on the surface of the land from artesian water. As a result, the land productivity reduced significantly and the Bourke irrigation district was left abandoned.

In nowdays, 14,000 ha of irrigated cotton farms is supplied with good quality water from the Darling River. This good quality supply of water has also revitalized considerable areas of citrus, jojoba, grapes and horticultural products. Due to the application of better management techniques and good quality supply water, the extent of salinization has been controlled.

However, the salinity issues still exist and signs of growing salinity conditions are present. This has been observed around the water reservoirs and supply canals in irrigated cotton farms.

3.4 Geology The BID is located on the edge of the western margin of the Surat Basin and Eromanga

Basin (Kingham 1998). Due to the special characterization of Eromanga Basin the region is predisposed to secondary salinization. The terrestrial sequences of Eromanga Basin that has formed in Jurassic period fallowed by shallow marine and lacustrine influences in the

Cretaceous period have resulted in the accumulation of large salt deposits in this region.

The Bourke District has become a major interest from an agricultural perspective, due to its geological properties. It is located downstream a junction of main inland rivers including, the

Bogan, Barwon and Castlereagh rivers, which all cojoin in a short distance as a result of the

Cobar block formation. Due to its slower erosion and higher elevation, the Cobar block impacts the inland rivers from the normal south-westerly movement into a north-westerly direction. As

82 per Fig. 3.4.1, once driven away from the force of the block, the rivers merge to compose the

Darling River which takes a south-westerly turn and moves through Bourke.

Fig. 3.4.1 The geological provinces in the Bourke District (after Kingham 1998).

3.5 Soil types Northcote (1966) state-wide survey is the best detailed (1:2,000,000) soil study up to present time. The maps resulted from this study is based on some degree of field inspection and mostly based on a combination of soil, ecological and geological surveys. As a result, this study is not useful for management purposes and can generally be referenced for an overall guide to the actual condition. There are four soil units recognised in the study are including: Cracking

83

Clay Soils (II1), Deep Grey Clay Soils (CC19), Earths- Neutral Reaction (My1), Crusty Loamy

Soils with Red Clayey Subsoils (Nb4) (Fig. 3.5.1). These units are discussed in continue.

Fig. 3.5.1 Spatial distribution of the four soil types identified in the Bourke Irrigation District (after Northcote 1966).

3.5.1 Cracking Clay Soils (II1)(Vertisols (Isbell 1996)) Different landscape features characterise this unit including floodplains, residuum, lake bottoms, alluvium of cracking yellow-grey clays (Ug5.28). This unit also falls into three sub- units comprising of Islands of dunes and sheets of brown (Uc5.13), residual islands of brown calcareous earths (Gc1) and crusty loamy soils (Dr1.33).

The yellow-grey cracking clays (Ug5.28) are identified by fine uniform textures with depth and seasonal cracking. They display smooth surfaced peds with grey clay which extend to a depth below 60 cm. The brown sands (Uc5.13) can be described by weak horizonation

84 affected by the variable CaCO3 content, colour variations and weak textural changes. Crusty loamy soils (Dr1.33) are distinguished by a thin crust, formed when dry. Also, it’s identified as alkaline reaction with CaCO3 and/or CaSO4 and fluctuating salinity.

3.5.2 Deep Grey Clay Soils (CC19)(Vertisols (Isbell 1996)) Deep Grey Clay soils are identified by lightly gilgaied cracking clays related with minor and major drainage routes including, chiefly grey clays (Ug5.24), and yellow-grey clays

(Ug5.25), brown clays (Ug5.38) and swampy basins of (Ug5.24). The grey cracking clays are deep >150 cm (Ug5.24) can be distinguished by their fine texture and uniform profile with seasonal cracking.

3.5.3 Neutral Reaction (My1) (Kandosol (Isbell 1996)) This unit can be identified as moderate to more strong undulating tablelands in addition to tableland remainings and plains which have been fragmented by steep hilly areas. Soils types can be determined by neutral red earths (Gn2.12) mantled by ironstone gravels and/or siliceous and further related are the ridge peaks and low hills of shallow (Um1.43, Um5.41). Also,

Soils of CC19 may appear in the drainage paths and low areas of gravel -free red earth (Gn2.12 and Gn2.13) may be found in the low-lying sites.

3.5.4 Crusty Loamy Soils with Red Clayey Subsoils (Nb4) (Kandosol (Isbell 1996)) This unit can be indicated by the plains with many wind-deflated sections of crusty loamy soils (Dr1.33 and Dr1.43). Also related are the most elevated areas of loamy and sandy red earth (Gn2.12 and Gn2.13) and other yellow earth including (Gn2.83), low sandy dunes

(Uc1.23) and lower fields of cracking brown clays (Ug5.38 and Ug5.34) and cracking grey clays

(Ug5.28).

This unit can be greatly distinguished from air photographs due to the differentiating colour with cracking clays. By correlating the expansion of this unit on the map of Northcote

85 with the photograph of Landsat (Fig. 4.2.1.1 C), it depicts the inaccurate spatial amount of the soil units.

3.6 Physiography The physiography of the study area consists of two main physiographic units including alluvial floodplains and aeolian sand dunes (Fig. 3.6.1). The dominant physiographic unit is floodplains and plains (Qrs). This unit consists of low energy clay-dominated floodplains of

Darling River characterized by predominantly unconsolidated Tertiary alluvial material. The other major physiographic unit of the study area is aeolian sand dunes. This unit consists of two sub-units including: Red sand forming undulating plains (Qd) and red sand and silt forming ridges (Qrd). The study area is generally flat with maximum local relief of only 20m. These units are discussed below.

Fig. 3.6.1 Approximate coverage of the three major types of physiographic units in the Bourke irrigation District (after Brunker, 1971).

86

3.6.1 Floodplain Unit (Qrs) The floodplain unit is the main physiographic unit of the study area, containing all the irrigated farms around Darling River. Darling River cuts into this unit with an average depth of

8 meters level below the river-side lands. The depth of the water flow of the river is significant to affect the irrigation system of the irrigated farms. Given that the soils of the unit contain the highest clay content within the district and therefore most suitable for irrigation agriculture it was essential for the farmers to use a mechanism to drive the water upstream in order to irrigate the farms. Thus, they established some dams and pumped the water into these water reservoirs.

As a result, the area shows point source indications of secondary salinization particularly around almost all of the water reservoirs and some supply canals. In some cases the degree of salinization caused the farm to be uncultivable. The example is illustrated in Fig. 3.6.1.1 and

Fig. 3.6.1.2.

Fig. 3.6.1.1 Waterlogging and salinity from the bank of the circular storage. Image was taken at E385000, N66725000 looking in a westerly direction. (Note: Water storage is located at the left side of the image)

87

Fig. 3.6.1.2 Waterlogging and salinity from the bank of the circular storage. Image was taken at E385000, N66725000 looking in a northerly direction.

Fig. 3.6.1.3 Waterlogging and salinity from the bank of the circular storage. Image was taken at E385000, N66725000 looking in a north-westerly direction.

88

3.6.2 Aeolian Dunes (Qd and Qrd) The aeolian Dunes unit is characterized by the aeolian red sandy dunes situated in the northern side of the Darling River. The red colour of this unit is the most distinction feature of the unit compared to the heavy grey colour floodplain unit. This unit contains two sub-units including red sand forming undulating plains (Qd) and red sand and silt forming ridges (Qrd).

The red sand and silt forming ridges (Qrd) unit are identifiable by its poorly to moderately consolidated deposits of light reddish-brown, pale-brown to yellowish-brown gravel, sand, and minor sandy clay. The unit is not developed for irrigation agriculture due to its light texture and low clay content. However, it has been established partially with citrus, jojoba (Simmondsia chinensis) and grapes gardens. The unit is originally poorly covered with the large Wattles

(Acacia mearnsii, Acacia victoriae, Acacia aneura) and Ironbark (Eucalyptus melanophloia) communities. Fig. 3.6.2.1 and Fig. 3.6.2.2 illustrate the vegetation and soils of this unit.

Fig. 3.6.2.1 An example of the aeolian dune physiographic unit (Image location E385000, N6675500). Note the elevated nature of the dune which looks down onto the floodplain unit in the distance.

89

Fig. 3.6.2.2 A further example of the aeolian dune physiographic unit which occurs as an isolated outcrop between the dual-cell reservoir and the circular reservoir (E382500, N6670000).

90

3.7 Refrences BOM, 2006. The Australian Government Bureau of Meteorology- Climate Data. (Ed.

http://www.bom.gov.au/)Canberra, ACT, Australia).

Brunker R.L., 1971, Bourke 1:250 000 Geological Sheet SH/55-10, 1st edition, Geological Survey of New South Wales, Sydney

DEH, 2004. Integrated Water Resource Management in Australia, Department of environment and heritage, Canberra, ACT, Australia

Isbell, 1996. The Australian Soil Classification. CSIRO, Canberra.

Kurian GT, 1989. Geo-Data: The world geographic encyclopedia. Gale Research Co. Detroit.

Kingham R. 1998. Geology of the Murray-Darling Basin- Simplified Lithostratigraphic Groupings.

Australian Geological Survey Organisation. Department of Primary Industries, Canberra,

Australia.

Northcote, K.H., 1966. Atlas of Australian Soils, Sheet 3, Sydney Canberra-Bourke-Armidale Area, with ExplanatoryData

91

4 Identifying soil landscape units at the district scale by numerically clustering remote and proximal sensed data

4.1 Introduction Identifying soil landscape units at a district scale is important as it allows for the sustainable land-use management (Hensen et al., 2009). This is because inappropriate use of land threatens rich and fertile farm land. This has been the case across large tracts of the state of

New South Wales, Australia. For example, the removal of crops for agricultural export has removed nutrients and led to (SoE 2012). In addition, the over irrigation of fields which have small tracts of lighter textured soil types (i.e. prior stream channels) than the surrounding alluvial clay plains (Woodforth et al., 2012) has led to excessive deep drainage

(Triantafilis et al., 2004) and in some cases the creation of shallow saline water tables

(Buchanan and Triantafilis, 2009) and soil salinization (Huang et al., 2015a). What is therefore required is a way to either map these soil properties individually or identify soil types which differ in terms of various soil properties in a given area. Soil properties which are important in terms of land use management include clay, cation exchange capacity (CEC), the electrical conductivity of a saturated soil paste extract (ECe – dS/m) and pH. This is because these properties indicate the ability of soil to: shrink and swell (clay and CEC); affect plant growth or nutrient availability, respectively.

Given a large number of soil properties contribute to land-use management considerations, each one needs to be mapped using a cost-effective method. Increasingly, the approach of digital soil mapping is being used because cheaper to acquire ancillary data is coupled to soil property data using statistical and mathematical approaches. For example, clay content has been mapped using electromagnetic (EM) induction instruments that measure the apparent soil electrical conductivity (ECa – mS/m). This includes places like Australia

(Triantafilis et al. 2001), Germany (Weller et al. 2007) and Belgium (Saey et al. 2009).

92

Equivalent research has also been carried out to map soil properties which also influence the

ECa including; salinity (Huang et al., 2015a), cation exchange capacity (Triantafilis et al.,

2009a), and soil moisture content (Huang et al., 2016). Upscaling of this approach to district scale (i.e. > 5,000 ha) is a little more complicated although recent research has demonstrated various individual soil properties have been mapped, including individual soil particle size fractions (PSF) to map soil texture Buchanan et al. (2012) and to map available water content

(Gooley et al., 2014). In both cases remote-sensed -ray spectrometry and proximal EM38 data were used in conjunction with mathematical models and soil property data.

However, in some instances, individual soil properties of importance in land-use management, such as soil pH, are often not correlated with this type of ancillary data. In this regard, it still may be possible to map these soil properties using ancillary data, but by considering such data as surrogates for air-photo interpretation; whereby the ancillary data is numerically clustered to identify soil landscape units. At the field level ECa from an EM38 instrument has been used in concert with ancillary data derived from a digital elevation model

(Fraisse et al. 2001), crop yield (Anderson-Cook et al., 2002), EM31 with digital numbers of color (i.e. red, green, and blue), from an aerial photograph (Triantafilis et al., 2009b), red reflectance (Bramley et al., 2011), and Quickbird imagery (Guo et al., 2013). More recently,

ECa data has been clustered along with -ray spectrometry data to map management zones in

Germany (Altdorff and Dietrich, 2012), Belgium (Van Meirvenne et al., 2013) and the UK

(Huang et al., 2014a).

The aim of this study is to use remote (airborne -ray spectrometry) and proximal sensed

EM38 data acquired across the geologically and geomorphological diverse aeolian and alluvial clay plain of the Bourke Irrigation district, to identify soil landscape units (Northcote, 1966).

Specifically, the merit of using fuzzy k-means (FKM) analysis to identify soil landscape units

93 using both -ray (i.e. K, Th, U radioelements and TC) and ECa (i.e. EM38 in horizontal

[EM38h] and vertical [EM38v] modes of operation) data will be discerned. The mean squared prediction error of various topsoil (0–0.3 m) and subsoil (0.9–1.2 m) physical (clay) and chemical (CEC, ECe and pH) soil property is used to identify the most appropriate number of classes. These results are compared with the traditional soil landscape unit map.

4.2 Materials and methods

4.2.1 Ancillary instruments, data collection, and interpolation Two sources of ancillary data were acquired. To characterize the topsoil (0–0.3 m), airborne -ray spectrometry data, collected in July 1995 by the NSW Department of Mineral

Resources, was used. The data was collected at a flying height of 60 m was and 250 m spacing between east-west runs (Fig. 4.2.1.1B) and using an Exploranium GR820 multi-channel -ray spectrometer coupled to two NaI crystal detectors with a volume of 33.6 L. The measured environmental levels of radiation were given in counts per second (cps) for the total counts

(TC), potassium (K), uranium (U) and thorium (Th).

Soil apparent electrical conductivity (ECa) was collected using a Geonics EM38. The data was collected an approximate 0.5 km spacing in irrigated areas and a 1 km spacing elsewhere (Fig. 4.2.1.1B). In total, 1,236 sites were surveyed (Buchanan et al., 2012). The

EM38 was selected because it can be used to collect ECa in horizontal (EM38h) and vertical

(EM38v) dipole readings to depths of about 0.75 and 1.5, respectively. Both modes of ECa data were collected at the same time. The -ray and ECa were interpolated onto a common 100 m grid by ordinary kriging (OK) using a neighbourhood of 20-30 and a local variogram using VESPER

(Minasny et al., 1999).

94

Fig. 4.2.1.1 (A) Traditional soil landscape unit map (Northcote, 1966), (B) -ray survey transects and EM38 survey locations and (C) Google map image of the study area (obtained on June 20, 2015), physiographic units (Brunker, 1971) and soil sampling points.

95

4.2.2 Fuzzy k-means (FKM) analysis

The interpolated -ray and ECa data were clustered using FKM analysis. The FKM method is described elsewhere (e.g. McBratney et al., 1992; Triantafilis et al. 2001). Briefly, the similarity between an individual i and a cluster c is measured to determine how much they are alike in multi-variate space (Bezdek, 1981); whereby the objective function J(M,C) is minimised;

n k  2 J(M, C)    mic dic (xi ,cc ) ; (1) i  1c  1

where M = mic is a n × k matrix of membership values (n denoting the number of ancillary data points), C = (ccv) is a k p matrix of class centres (p denotes the number of

T ancillary data variables), ccv is the value of the centre of class c for variable v, xi = (xi1,…,xip)

T is the vector representing individual i, cc = (cc1,…,ccp) is the vector representing the center of

2 class c, and dic (xi ,cc ) is the square distance between xi and cc according to a distance

2 measure ( dic ).Given the local knowledge of the geology of the area, Mahalanobis is chosen

(Bezdek, 1981).

The fuzziness exponent () determines the degree of fuzziness. When  = 1 this is equivalent to the hard partition. As  increases, memberships tend to become uniform. The fuzziness performance index (FPI) and the normalized classification entropy (NCE) were then used to identify values for  and k. This is because the FPI is a measure of continuity between classes;

kF 1 FPI 1 (2) k 1

where F is the partition coefficient,

96

n k 1 2 F  (mic ) . (3) n i1 c1

Whereby a value of 1 suggests a very fuzzy classification, while a value approaching 0 indicates a hard one. The NCE is a measure of disorganization in data partitioning,

H NCE  ; logk (4)

where H is the entropy function;

1 n k H   mic log(mic ) (5) n i1 c1

Whereby, values approaching 0 indicate that the classes are well structured, whilst values approaching 1 suggest the classes are disorganized. Triantafilis et al. (2013) suggest that values around 0.5 provide a balance between continuity and structure. A more quantitative measure can be discerned for  and k using the derivative of J(M,C) with respect to 

(Bezdek, 1981):

n k dJ (M, C)  2    mic logmic  dic (6) d i 1c 1

For this study, the FuzMe software (Minasny and McBratney, 2002) is used and described further in Triantafilis et al. (2003). The number of classes is important in FKM clustering method. In order to classify the ancillary data, the data is clustered into k = 2 to 6 classes. To define a suitable degree of overlap between k, the fuzziness exponents () = 1.1–

1.7 are considered. The plot of . change in the FKM objective function (–J/) is also

97 considered, which assists in determining a suitable value of  (see McBratney and Moore,

1985).

4.2.3 Soil sampling and laboratory analysis To test the validity of the FKM classes, soil data were collected at 86 sampling points.

Soil samples were collected from the topsoil (0–0.3 m) and subsoil (0.9–1.2 m) with a gouge auger (highlighted in black and red dots in Fig. 4.2.1.1C). Each of the samples was stored in an individual water-tight plastic bag, taken back to the laboratory and stored in a cool room.

Ancillary data readings were also recorded at these locations to facilitate clustering of the ancillary data and assignation of a class to the sampling locations and allow determination of root mean square prediction of physical and chemical properties.

The samples were oven dried (60 ºC), homogenized and ground to 2 mm. The particle size distribution was determined from a subsample by first determining the various particle size fractions using the pipette method. The results were reported in terms of percent sand, silt and clay as defined by Coventry and Fett (1979). The cation exchange capacity (CEC) was determined by the method presented by Rayment and Higginson (1992). Owing to the expense of CEC analysis a smaller subset of 50 locations was only available (highlighted in black dots in Fig. 4.2.1.1C).

A saturated soil paste was prepared and the electrical conductivity of the extract (ECe) was measured using a SmartCHEM conductivity instrument. The results for clay content, ECe, and pH (at 86 sites) and CEC (at 50 sites) are reported in terms of differences among the classes obtained from the FKM analysis, to aid pedological interpretation and consider these soil properties for evaluating the use of the proximal and remote ancillary data for prediction of soil landscape units.

98

4.2.4 Linear mixed model (LMM) A linear mixed model (LMM) was used for the soil sample data of the following form because the samples were not chosen randomly:

y = X +  + , (7)

where y is a n×1 vector of values of the target soil variable, X is a n×p design matrix, is a p×1 vector of fixed effects coefficients, is a n×1 vector the elements of which are a realization of a spatially correlated random variable and , is a n×1 vector the elements of which are a realization of an independent and identically distributed random variable. The elements of the design matrix are the predictor covariables and the fixed effects coefficients correspond to these.

The correlated random variable  is assumed to be; normal, have zero mean and

 variance parameters ( ) and a distance parameter for a selected variogram function. The

 error variable  also has zero mean and a variance  . Models of the form in Equation (7) were fitted for target soil properties and with the fixed effects from the class of maximum membership for the FKM clustering of the ancillary data of k = 2 to 6. The fitting was done using the LME procedure from the NLME library for the R platform (Pinheiro et al., 2013; R

Development Core Team, 2010).

Using this approach, variance parameters for the random effects are first estimated by residual maximum likelihood (REML) and the fixed effects coefficients are estimated by weighted least squares. The null hypothesis that all class means are equal (where the fixed effects are classes) is tested by the Wald statistic. These methods are described by Lark et al.

(2006).

99

4.2.5 Computation of the prediction error variance for class means In this study, the expected value of the mean squared prediction error is calculated for t classification into k classes formed by the FKM algorithm for some sample size N following

Huang et al. (2014b) and as follows:

 2    p,C (N | k) =  k +  k)(1+k/N). (8)

The sum of the variances of the random effects in the model for k classes as the

  random effects,  k +  k, was treated as the expected value of the variance for the random variable. This approach is used elsewhere to compute values for the variances of design- based sample estimates from the results of model-based analyses (Cochran, 1977; Lark,

2011).

4.3 Results & discussion

4.3.1 Preliminary data analysis of ancillary and laboratory measured data Table 4.3.1.1 shows the pearson correlation coefficient (R2) between remote and proximally sensed ancillary data. The correlation coefficients for the survey radioelement data were largest between K (0.90) and Th (0.87) with TC, with a similarly large coefficient

(0.85) achieved between EM38h and EM38v. Similar coefficients were calculated when considering only the ancillary data recorded at the 86 sampling sites.

Table 4.3.1.2 shows the summary statistics of soil physical and chemical properties collected at the sampling points. It shows that clay content is high, on average for both the topsoil (47.4 %) and subsoil (52.7 %). This is consistent with the predominantly clay alluvial nature of the landscape. The reactivity of the soil overall, and as evident in topsoil (26.23 cmol(+)/kg) and subsoil CEC (28.23 cmol(+)/kg), is moderate. The salinity on average is

100 slightly saline in the topsoil (ECe = 3.79 dS/m) but moderately saline (5.98 dS/m) in the subsoil. The pH is neutral on average.

Table 4.3.1.1 Pearson correlation coefficient between remote and proximally sensed ancillary data along the survey transects and sampling points. Note: values shown for potassium (K-cps), uranium (U-cps), thorium (Th-cps), total count (TC-cps) and soil apparent electrical conductivity (ECa – mS/m) for EM38h and EM38v.

Survey data K U Th TC EM38h EM38v K 1.00 0.29 0.79 0.90 0.28 0.27 U 1.00 0.18 0.39 -0.24 -0.16 Th 1.00 0.87 0.42 0.34 TC 1.00 0.28 0.24 EM38h 1.00 0.85 EM38v 1.00 Sampling points data K 1.00 0.41 0.77 0.85 0.09 0.04 U 1.00 0.40 0.53 -0.11 -0.10 Th 1.00 0.83 0.19 0.13 TC 1.00 0.04 -0.01 EM38h 1.00 0.95 EM38v 1.00

Table 4.3.1.2 Summary statistics of soil physical and chemical properties collected at sampling points. . Note: values shown for Clay content (%), CEC (cmol(+)/kg), ECe (dS/m), pH, respectively. All values rounded to nearest integer.

Topsoil (0-0.3m) Mean Median Min. Max. Skewness Kurtosis Clay 47.4 53.00 7.6 70.2 -1.05 0.12 CEC 26.23 28.78 6.30 35.69 -1.25 1.24 ECe 3.79 2.01 0.28 21.68 2.18 6.03 pH 7.01 7.00 5.93 8.23 0.18 -0.23 Subsoil (0.9-1.2m) Clay 52.7 57.1 12.5 70.07 -1.37 1.52 CEC 28.23 30.39 9.24 39.68 -0.94 0.41 ECe 5.98 5.67 0.30 15.97 0.49 -0.23 pH 7.02 6.95 5.93 8.23 0.18 -0.23

4.3.2 Spatial distribution of proximally sensed data Fig. 4.3.2.1A shows the spatial variation of radioelement K (cps). Intermediate to intermediate-small counts (< 80 cps) are found in a small pocket in the north-west and north- east corners, associated with the wind deflated areas of the Nb4 soil landscape unit. This is also the case with a small contiguous area in the central and western; wherein all three locations coincide approximately were the red sand dune forming undulating plains (Qd).

101

Conversely, intermediate-large (>= 100 cps) K generally coincides with the expansive floodplains of clayey silt, sand, and gravel (Qrs) and the CC19 soil landscape in the north and the II1 landscape to the south.

Similar patterns are evident in U (Fig. 4.3.2.1B), Th (Fig. 4.3.2.1C) and TC (Fig.

4.3.2.2A), although there are differences. In the northwest corner U content is intermediate, whilst for the other channels, U is generally small. It is also noticed the Th is generally larger

(>= 45 ppm) in the northeast and southeast compared with the southwest (40-45 ppm). In addition, there are contiguous areas near the Darling River which are larger in terms of U (>=

35 cps), Th (>= 45) and TC (>= 1,600).

Fig. 4.3.2.2B shows the kriged distribution of the EM38h. For the most part small ECa

(<150 mS/m) characterises the wind deflated (Qd) areas characterised by Nb4 and My1 soil landscapes. The converse was true for the floodplain in the north associated with CC19 where

EM38h was intermediate-large (>=250 mS/m). The floodplain associated with II1 was generally intermediate (150 to 250 mS/m) in the southwest but intermediate-small (50–150 mS/m) southeast of the River. Fig. 4.3.2.2C shows the kriged distribution of the EM38v.

Similar patterns were evident, with only subtle differences. For example, the floodplain associated with II1 was intermediate-small (50–150 mS/m) to the southwest and southeast of the Darling River.

102

Fig. 4.3.2.1 Spatial distribution of -ray spectrometry data including; (A) potassium (K– counts per second), (B) uranium (U– counts per second) and (C) thorium (Th– counts per second).

103

Fig. 4.3.2.2 Spatial distribution of -ray spectrometry data including; (A) total count (TC – counts per second) and EM38 electrical conductivity including; (B) EM38h (ECa – mS/m) and (C) EM38v (ECa – mS/m).

104

4.3.3 FKM analysis Fig. 4.3.3.1A shows FPI results for clustering of the remote and proximal data for different values of k and . It is evident FPI is at a minimum when k = 4 and  > 1.4. When 

= 1.3, k = 6 is the minimum, whereas when  < 1.3, k = 5 is the minimum. Near identical results are shown in Fig. 4.3.3.1B when considering NCE. Fig. 4.3.3.1C shows the plot of

versus –J(M,C)/. In all cases (i.e. k = 2-6) a maximum occurs when 1.4. Given these results an exponent of 1.4 is selected to be suitable for two reasons. With respect to k, this is because McBratney and Moore (1985) indicated that is optimal when –

J(M,C)/is a maximum. With respect to , Triantafilis et al. (2009b) indicated when FPI and NCE are around 0.5 a good balance is achieved between continuity and organisation, respectively.

105

Fig. 4.3.3.1 Plot of; (A)fuzziness performance index (FPI), (B)normalized classification entropy (NCE) versus classes (k = 2 to 6) and (C) fuzziness exponent () versus – J(M,C)/.

106

4.3.4 Spatial distribution of the FKM classes Fig. 4.3.4.1A shows the result for k = 3. The three classes approximately coincide with the location of the three soil landscape units identified by Northcote (1966). It is clear that the wind deflated areas (Qd) are represented by class 3A. Table 4.3.4.1 shows the centroid values of the ancillary data is characterised by intermediate-small radioelement K

(72 cps), U (34 cps) and Th (31 cps) as well as small ECa (e.g. EM38h = 101 mS/m). Class

3B represents the southern part of the floodplain (Qrs) and more or less the II1 soil landscape.

It is defined by a centroid with the largest radioelement counts (i.e. TC = 1,523 cps) but only intermediate ECa (e.g. EM38h = 144 mS/m). By contrast, class 3D, which represents the

CC19 soil landscape and defines the northern part of the floodplain (Qrs), is characterised by intermediate radioelements (e.g., K (88 cps), U (42 cps) and Th (31 cps)) as well as very large ECa (e.g. EM38h = 279 mS/m).

Fig. 4.3.4.1B shows the spatial distribution for k = 4. The difference is that 3B essentially divides into 4B and 4C, whereby class 4C for the most part represents areas adjacent to the Darling River and 4B defines the area to the southeast of the Darling River and south of the isolated area of red sand forming undulating plain (Qd). Fig. 4.3.4.1C shows similar patterns for k = 5, where class 5E broadly corresponds to with the location of the

Darling River floodplain.

Table 4.3.4.1 Mahalanobis centroid values of remotely and proximally sensed ancillary data clustered using FKM and for classes k = 3, 4 and 5. Note: centroids shown for potassium (K-cps), uranium (U-cps), thorium (Th-cps), total count (TC-cps) and soil apparent electrical conductivity (ECa – mS/m) for EM38h (h) and EM38v (v).

k = 3 k = 4 k = 5 Class Centroid Centroid Centroid K Th U TC h v K Th U TC h v K Th U TC h v A 72 34 31 1231101 85 71 34 31 1218 95 79 71 34 30 1209 90 75 B 98 44 34 1523 144 110 102 43 33 1512 147 110 104 44 33 1519 136 102 C N/A N/A N/A N/A N/A N/A 89 44 35 1485 163 132 88 45 34 1476 155 126 D 88 42 31 1419 279 222 88 42 31 1408 296 236 88 42 30 1408 301 240 E 91 38 33 1409 202 90

107

Fig. 4.3.4.1 Spatial distribution of fuzzy k-means (FKM) derived classes for digital soil maps of k = (A) 3, (B) 4, and (C) 5.

108

4.3.5 Mean squared prediction error of digital soil maps To determine the efficacy of the various DSM of k = 2-6 the mean square prediction

2 error ( p,C) of these maps is determined and relative to the various laboratory measured soil physical (i.e. clay) and chemical (i.e. CEC, ECe and pH) properties. Table 4.3.5.1 shows the

2 results. In terms of topsoil (0–0.3 m) properties, the minimum  p,C was achieved when k = 4 is considered for clay content (159.7), CEC (21.9), ECe (13.5) and pH (0.22). Table 4.3.5.1

2 also shows the minimum  p,C was also k = 4 and considering most of the subsoil properties

(i.e. 0.9–1.2 m), and including clay content (80.8), CEC (31.2) and ECe (16.6).

These results indicate that when k = 4 seven of the 8 topsoil and subsoil properties are

2 well accounted, excepting subsoil pH (0.21) which was minimised in terms of  p,C when k =

6. This number of classes is also consistent with the number of soil landscape units previously identified by Northcote (1966), however in that case, only two soil landscape units characterised the clay plain (CC19 and II1), whilst two units were identified in the aeolian areas (Nb4 and My1).

To compare DSM of k = 4 using DSM methods with the traditional soil landscape unit

2 2 map (Fig. 4.2.1.1A) the  p,C for all properties is calculated. In all cases, the  p,C for the traditional map was larger for all properties except for subsoil ECe for (k = 2 and 5) and clay content (k = 2).

109

2 Table 4.3.5.1 Mean Squared Prediction Errors (MSPE) ( p,C ) for topsoil (0-0.3 m) and subsoil (0.9-1.2 m) for the measured soil properties used in the REML analysis. Different rows represent the different numbers of classes as categorized by the fuzzy k-means analyses from k = 2-6. Bold highlighted values indicate the lowest values of each row (i.e. lowest MSPE for each measured soil property).

0-0.3 m Clayk CEC ECe pH No. of lowest MSPE 204.42 35.9 15.2 0.23 0 170.93 23.5 13.8 0.23 0 159.74 21.9 13.5 0.22 4 194.35 22.0 14.1 0.23 0 179.46 25.0 15.5 0.23 0 Traditional 282.84 63.1 19.3 0.24 0 0.9-1.2 m Clayk CEC ECe pH No. of lowest MSPE 388.82 48.9 18.3 0.24 0 91.43 37.1 16.9 0.24 0 80.84 31.2 16.6 0.23 3 81.55 33.8 17.4 0.23 0 81.36 32.5 16.7 0.21 1 Traditional 279.54 60.2 17.3 0.26 0

4.3.6 REML analysis of FKM class map = 4 Here the mean values of the various laboratory-measured soil properties relative to the classes obtained by FKM analysis of the remote and proximal data when k = 4 are interpreted.

This is because not only is the spatial distribution of k = 4 is most closely aligned with the

2 previously identified soil landscape units (Northcote, 1966), but because the  p,C was a minimum for k = 4 for most of the soil properties and also that k = 4 was a minimum when considering the FPI and NCE when  = 1.4. Fig. 4.3.4.1 shows the mean values of topsoil and subsoil physical and chemical properties and the standard error for each class obtained from the LMM estimated by REML. In all cases, the Wald statistic allows rejection of the null hypothesis of no difference among class means.

The clay content and CEC appear most revealing about the partitioning of the k = 4 classes and in terms of their land-use and management. Fig. 4.3.6.1A shows that class 4A has lowest topsoil (0–0.3 m) clay content (29.7 %). This is similarly the case with regard to subsoil

(0.9–1.2 m) clay content (24.8 %) (Fig. 4.3.6.1B). This result is consistent with the soil texture of the wind deflated aeolian sand of Nb4, which was typically loamy (~25 %). The average

110

CEC of both topsoil (15.26 cmol(+)/kg) and subsoil (17.26 cmol(+)/kg) CEC also suggests the soil in this area will have poor shrink-swell potential (Fig. 4.3.6.1C and D). As a result, the land has not been cleared or used extensively for agricultural purposes. Recent development has occurred and to take advantage of the good drainage and for irrigated horticultural production.

111

Fig. 4.3.6.1 Plot of mean and standard deviation of: (A) topsoil clay content (%), (B) subsoil clay content (%), (C) topsoil CEC (cmol(+)/kg), (D) subsoil CEC (cmol(+)/kg), (E) topsoil ECe (dS/m), (F) subsoil ECe (dS/m), (G) topsoil pH and (H) subsoil pH (the results for clay content, ECe, and pH at 86 sites and CEC at 50 sites are reported).

112

Conversely, class 4D has the highest topsoil (57.35 %) and subsoil (59.18 %) clay.

This is also the case for topsoil (29.93 cmol(+)/kg) and subsoil (31.54 cmol(+)/kg) CEC. This result is not unexpected given the CC19 soil landscape unit which coincides with the location of this class, is characterised by uniform and deep grey self-mulching clays (Ug5.24 and

Ug5.25, Northcote, 1979). Herein, and owing to the generally more fertile nature of the clay, the soil was originally cleared extensively in the late 1860’s for wool production. More recently, and owing to the moderate shrink-swell nature of these clays, the area was extensively developed for irrigation and for cotton production.

Owing to the success of irrigation and high profitability of cotton, the area ascribed by class 4B and in the southwest corner, was subsequently extensively developed. This most likely was because the soil was recognised as being similar to class 4D. This is reinforced herein, whereby topsoil (54.42) and subsoil (56.22) clay, as well as topsoil (26.83 cmol(+)/kg) and subsoil (32.32 cmol(+)/kg) CEC is more or less equivalent to that of 4D. Class 4C, which also shows similar clay content and CEC values, has not however been extensively developed for irrigation. Whilst the potential is there the location of the class relative to the current

Darling River floodplain would predispose the area to excessive flooding and would require a large investment in flood mitigation works.

It is interesting to note that the difference between these three classes, and based on their centroids, was because 4D has much higher ECa. For example, and as shown in Table

4.3.4.1, soil ECa in class 4D (e.g. EM38h = 296 mS/m) was twice as large as compared with

4B (147 mS/m). This result is consistent with subsoil salinity where 4D (ECe = 8.30 dS/m) was double that of 4B (3.76 dS/m). It is similarly noticed that soil ECa was intermediate for 4C

(e.g. EM38h = 163 mS/m) as was subsoil ECe (6.03 dS/m) (Fig. 4.3.6.1F).

113

In terms of land-use implications, the lower topsoil ECe of 4B suggests that soil to the south is more freely draining. With respect to 4D, and given both topsoil (7.19 dS/m) and subsoil (8.30 dS/m) ECe were at a level (i.e. > 7.7 dS/m) (Fig. 4.3.6.1E) where even tolerant crops such as cotton are susceptible to 50 % reduction in germination (see Rhoades et al.,

1989), there may be problems with soil salinity. The reason is because of the presence of shallow water tables (Buchanan and Triantafilis, 2009), and because various salinity hazard factors have been realised (Buchanan et al., 2016) in the area delineated by class 4D.

Another subtle difference between the classes is shown in Fig. 4.3.6.1G and Fig.

4.3.6.1H, which shows the standard deviation of pH for topsoil and subsoil pH, respectively.

The results are nearly identical at both depths. It is also noteworthy that pH for 4A (7.14), which represents the aeolian areas, is not significantly different to 4B (7.04) and 4D (7.01) which are associated with the clay alluvial plain. Nevertheless, the neutral to slightly alkaline reaction trend of the classes reflects the prevailing climate of the study area which is semi-arid.

In class 4C the slightly acidic nature of the soil is most likely a reflection of the younger age of the soil associated with the current floodplain and therefore the smaller amount of time for salt accumulation (e.g. calcium carbonate).

4.3.7 Fuzzy canonical analysis of FKM class map = 4 To understand the distribution of the remote and proximal data in character space on a small number of axes, a fuzzy canonical analysis (see FCA-Triantaiflis et al., 2003) was conducted. The biplot in Fig. 4.3.7.1 shows the plot of Canonical Axes 1 and 2 of the FCA and the ellipses indicate where 95 % of each class falls when m  0.5. Along axis 1, the centroids are well spaced for 4A (–0.95), 4B (0.28) and 4D (–0.43), with 4C (–0.27) close to 4B. The rays of the ancillary data are also shown. With respect to the radioelements, Th contributes most to axis 1 (0.26) with K most to 2 (–0.51). The TC contributes equally (0.16, 0.18) whilst

114

U contributes more to 2 (–0.29). The EM38h has a similar contibution to both axes, with a slightly higher contribution to axis 1 (0.82), while the EM38v contributes most to axis 1 (–

0.47).

The directions of the rays are also informative with respect to class discrimination. The direction of the K and U rays and location of 4B and 4C indicate these were discriminated most by these radioelements. In the case of 4B, and as shown in Fig. 4.3.2.1A, this was because of large K (>100 cps). With respect to 4C, and as shown in Fig. 4.3.2.1B, this was because of large U (> 35 cps). These results are consistent with where K and U could be expected to be large and in lower lying areas where radioelement rich clays (Pickup and

Marks, 2001) accumulate. This is also the case with Th which represents silty soil associated with active floodplains (Bierwirth, 1996).

Fig. 4.3.7.1 Plot of first two fuzzy canonical axes for k = 4 classes and fuzziness exponent () = 1.4. Note: Fuzzy canonical rays for each ancillary data sources including radioelement potassium (K), uranium (U), thorium (Th) and total count (TC) and EM38h and EM38v and centroids for each of the k = 4 classes (i.e., 4A, 4B, 4C and 4D) with location of 95 % density ellipses for each class.

115

A larger discriminator of 4D is the EM38h and as evidenced by the ray being greatest in the direction of 4D which has the largest EM38h (296 mS/m) centroid. This is also the reason why 4B and 4C are discriminated and because the latter had larger centroids of EM38h

(163 mS/m). The converse is true for 4A which is orthogonal to all rays. These results are consistent with the soil property which is most commonly known to influence ECa. Namely, class 4C has moderately saline topsoil (4.34 dS/m) and subsoil (6.03 dS/m) ECe, which is larger than the only “slightly” saline topsoil (2.13 dS/m) and subsoil (3.76 dS/m) ECe of 4B but less that of 4D which had “highly” saline amounts of topsoil (7.19 dS/m) and subsoil (8.30 dS/m) ECe.

In addition, we note that the ancillary data used in the FKM analysis did not contribute equally to the various management classes. For example, EM38h and EM38v and K played more important roles in discriminating the managemen classes than U, Th and TC. In order to under fully explore the effect of ancillary data on predicting soil properties (e.g. soil salinity), a conditional simulation approach was used in Chapter 5 along with a linear mixed model fitted by dropping different sources of ancillary data in a systematic way.

4.4 Conclusions The use of various FKM indices (e.g. FPI, NCE and -J(M,C)/) suggested  = 1.4 as a suitable fuzziness exponent to identify soil landscape units that could be determined solely from four remote- (i.e. K, U, Th and TC) and two proximal-sensed (i.e. EM38h and

EM38v) ancillary data. Using LMM and REML analysis, and considering physical (i.e. clay) and chemical (i.e. CEC, ECe and pH) properties of the topsoil (0–0.3 m), the mean squared

2 prediction error (i.e.  p,C) was minimised when k = 4 for all four soil properties. The same was true for three of the four subsoil properties. This was not the case for pH, however. The

116 same analysis revealed that the k = 4 classes derived from DSM of the remote- and proximal-

2 sensed data was superior in terms of minimising  p,C as compared with the results achieved with the traditional soil landscape unit map of k = 4 (Northcote, 1966).

It is concluded the scope to identify soil differences objectively and at the district level with remote -ray spectrometer (K, U, Th, and TC) and proximal ECa (EM38h and

EM38v) data shows potential because the k = 4 were also broadly representative of soil landscape units previously identified (Northcote, 1966) using traditional methods. For example, 4A represented the Nb4 and Qd unit associated with the wind deflated red sand forming undulating plains. The DSM was unable to identify some soil mapping (i.e. My1) or geomorphological units (i.e. Qrd), however, these were not widely identified across the study area. Conversely, the expansive Qrs unit which characterized the clayey floodplains of silt, sand, and gravel was defined by three soil landscape units. The results indicate that CC19 which is found in the northern half of the clay plain coincides with the spatial extent of 4D.

Interestingly, II1 is not as homogenous as indicated by the fact this area is characterized by

4C which was mostly associated with the current Darling River floodplain; whereas 4B defined areas on either side of the Darling River.

117

4.5 References Altdorff, D., Dietrich, P., 2012. Combination of electromagnetic induction and gamma spectrometry using K-means clustering: A study for evaluation of site partitioning. J. Plant Nutr. Soil Sci. 175 (3), 345–354.

Anderson-Cook, C.M., Alley, M.M., Roygard, J.K.F., Khosla, R., Noble, R.B., Doolittle, J.A., 2002. Differentiating soil types using electromagnetic conductivity and crop yield maps. Soil Sci. Soc. Am. J. 66 (5), 1562–1570.

Muckenhausen, E., 1980. Avery, B.W.: Soil Classification for England and Wales (Higher Categories). Soil Survey Technical Monograph No. 14. Rothamsted Experimental Station, Harpenden, England.

Bezdek, J.C., 1981. Pattern recognition with fuzzy objective function algorithms. Plenum Press, New York.

Bierwirth, P.N., 1996. Investigation of airborne gamma-ray images as a rapid mapping tool for soil and land degradation - Wagga Wagga, NSW. Australian Geological Survey Organisation Record 1996/22.

Bramley, R.G.V., Ouzman, J., Boss, P.K., 2011. Variation in vine vigour, grape yield and vineyard soils and topography as indicators of variation in the chemical composition of grapes, wine, and wine sensory attributes. Aust. J. Grape Wine R. 17 (2), 217–229.

Brunker R.L., 1971, Bourke 1:250 000 Geological Sheet SH/55-10, 1st edition, Geological Survey of New South Wales, Sydney

Buchanan, S., Triantafilis, J., 2009. Mapping water table depth using geophysical and environmental variables. Ground Water 47 (1), 80–96.

Buchanan, S.M., Triantafilis, J., Odeh, I.O.A., Subasinghe, R., 2012. Digital soil mapping of compositional particle-size fractions using proximal and remotely sensed ancillary data. Geophysics 77 (4), 201–211.

Buchanan, S.M., Huang, J., Triantafilis, J., 2016. Salinity risk assessment using fuzzy multiple criteria evaluation. Soil Use Manage., minor revision.

Cochran, W.G., 1977. Sampling Techniques-3. Wiley, New York.

Coventry, R. J., Fett, D.E.R., 1979. A pipette and sieve method of particle-size analysis and some observations on its efficacy. Division of Soils, CSIRO.

Fraisse, C.W., Sudduth, K. A., & Kitchen, N. R., 2001. Delineation of site-specific management zones by unsupervised classification of topographic attributes and soil electrical conductivity. T. ASAE, 44 (1), 155.

Gooley, L., Huang, J., Page, D., Triantafilis, J., 2014. Digital soil mapping of available water content using proximal and remotely sensed data. Soil Use Manage. 30 (1), 139–151.

Coventry, R.J., Fett, D.E.R., 1979. A pipette and sieve method of particle-size analysis and some observations on its efficacy. Division of Soils, CSIRO.

118

Guo, Y., Shi, Z., Li, H.Y., Triantafilis, J., 2013. Application of digital soil mapping methods for identify salinity management classes in coastal lands of central China. Soil Use Manage. 29 (3), 445–456.

Hansen, M. K., Brown, D. J., Dennison, P. E., Graves, S. A., Bricklemyer, R. S., 2009. Inductively mapping expert-derived soil-landscape units within dambo wetland catenae using multispectral and topographic data. Geoderma, 150(1), 72-84.

Huang, J., Lark, R.M., Robinson, D.A. Lebron, I., Keith, A.M., Rawlins, B., Tye, A., Kuras, O., Raines, M., Triantafilis, J., 2014a. Scope to predict soil properties at within-field scale from small samples using proximally sensed -ray spectrometer and EM induction data. Geoderma 232, 69–80.

Huang, J., Nhan, T., Wong, V., Johnston, S., Lark, R.M., Triantafilis, J., 2014b. Digital soil mapping of a coastal acid sulfate soil landscape. Soil Res. 52 (4), 327–339.

Huang, J., Taghizadeh-Mehrjardi, R., Minasny, B., Triantafilis, J., 2015a. Modelling soil salinity along a hill slope in Iran by inversion of EM38 data. Soil Sci. Soc. Am. J. 79 (4), 1142–1153.

Huang, J., Scudiero, E., Clary, W., Corwin, D.L., Triantafilis, J., 2016. Time-lapse monitoring of soil water content using electromagnetic conductivity imaging. Soil Use Manage., in press.

Lark, R. M., Cullis, B. R., Welham, S. J., 2006. On spatial prediction of soil properties in the presence of a spatial trend: the empirical best linear unbiased predictor (E-BLUP) with REML. Eur. J. Soil Sci. 57 (6), 787–799.

Lark, R. M., 2011. Spatially nested sampling schemes for spatial variance components: Scope for their optimization. Comput. Geosci-UK. 37 (10), 1633–1641.

McBratney, A.B., Moore, A.W., 1985. Application of fuzzy-sets to climatic classification. Agri. Forest Meteorol. 35 (1), 165–185.

McBratney, A.B., De Gruijter, J.J., Brus, D.J., 1992. Spatial prediction and mapping of continuous soil classes. Geoderma 54 (1-4), 39–64.

Minasny, B., McBratney, A.B., Whelan, B.M., 1999. VESPER version 1.6, Precision Agriculture Laboratory, Sydney, Australia.

Minasny, B., McBratney, A.B., 2002. FuzME version 3.0, Precision Agriculture Laboratory, The University of Sydney, Australia.

Moore, I.D.,Gessler, P.E., Nielsen, G.A., and Petersen, G.A., 1993. Soil attribute prediction using terrain analysis. Soil Sci. Am. J. 57 (2), 443–452.

Northcote, K.H., 1966. Atlas of Australian Soils, Sheet 3, Sydney Canberra-Bourke-Armidale Area, with Explanatory Data.

Northcote, K.H., 1979. A Factual Key for the Recognition of Australian Soils. 4th edn., Rellim Technical Publishers, Glenside, SA.

Pickup, G., Marks, A., 2001. Regional-scale sediment process models from airborne gamma ray remote sensing and digital elevation data. Earth Surf. Proc. Land. 26 (3), 273–293.

Pinheiro, J., Bates, D., DebRoy, S., Sarkar, D., The R Development Core Team, 2013. nlme: Linear and Nonlinear Mixed Effects Models. R package version 3.1–110.

119

R Development Core Team 2010. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

Rayment, G.E., Higginson, F.R., 1992. Australian laboratory handbook of soil and water chemical methods. Inkata Press Pty Ltd.

Rhoades, J.D., Bingham, F.T., Letey, J., Hoffman, G.J., Dedrick, A.R., Pinter, P.J., Replogle, J.A., 1989. Use of saline drainage water for irrigation: Imperial Valley study. Agricultural Water Management, 16 (1), 25–36.

State of Environment (SoE) Reports. 2012. Chapter 3.1, Management of soil and land. NSW Environmet Protection Authority (EPA). Accessed on 4 August 2015 at http://www.epa.nsw.gov.au/soe/soe2012/chapter3/chp_3.1.htm#3.1.12

Saey, T., Simpson, D., Vermeersch, H., Cockx, L., Van Meirvenne, M., 2009. Comparing the EM38DD and DUALEM-21S sensors for depth-to-clay mapping. Soil Sci. Soc. Am. J. 73 (1), 7– 12.

Triantafilis, J., Huckel, I.A., Odeh I.O.A., 2001. Comparison of statistical prediction methods for estimating field-scale clay content using different combinations of ancillary variables. Soil Science 166 (6), 415–427.

Triantafilis, J., Odeh, I.O.A., Minasny, B., McBratney, A.B., 2003. Elucidation of physiographic and hydrogeological units using fuzzy k-means classification of EM34 data in the lower Namoi valley. Environ. Mod. Software 18, 667–680.

Triantafilis, J., Odeh, I.O.A., Jarman, A.L., Short, M., and Kokkoris, E., 2004. Estimating and mapping deep drainage risk at the district level in the lower Gwydir and Macquarie valleys, Australia. Aust. J. Exp. Agric. 44 (9), 893–912.

Triantafilis, J., Lesch, S. M., La Lau, K. & Buchanan, S. M. 2009a. Field level digital soil mapping of cation exchange capacity using electromagnetic induction and a hierarchical spatial regression model. Aust. J. Soil Res. 47 (7), 651–663.

Triantafilis, J., Kerridge, B., Buchanan, S.M., 2009b. Digital soil-class mapping from proximal and remotely sensed data at the field level. Agron. J. 101 (4), 841–853.

Triantafilis, J., Gibbs, I.D., Earl, N.Y., 2013. Digital soil pattern recognition in the lower Namoi valley using numerical clustering of gamma-ray spectrometry data. Geoderma 192, 407–421.

Van Meirvenne, M., Islam, M. M., De Smedt, P., Meerschman, E., Van De Vijver, E., Saey, T., 2013. Key variables for the identification of soil management classes in the aeolian landscapes of north- west Europe. Geoderma 199, 99–105.

Weller, U., Zipprich, M., Sommer, M., Castell, W. Z., & Wehrhan, M., 2007. Mapping clay content across boundaries at the landscape scale with electromagnetic induction. Soil Sci. Soc. Am. J. 71 (6), 1740–1747.

Woodforth, A., Triantafilis, J., Cupitt J., Malik, R.S., Geering, H., 2012. Mapping estimated deep drainage in the lower Namoi Valley using a chloride mass balance model and EM34 data. Geophysics 77 (4), 245–256.

Zare, E., Huang, J., Santos, F. A., & Triantafilis, J., 2015. Mapping salinity in three dimensions using a DUALEM-421 and electromagnetic inversion software. Soil Sci. Soc. Am. J. 79 (6), 1729– 1740.

120

5 An error budget for soil salinity mapping using different ancillary data

5.1 Introduction Secondary soil salinization occurs as a function of human interaction with the landscape and results from the mobilisation of primary salts (i.e. cyclical, aeolian or connate) into the rootzone. Salinization of soil is problematic because increasing salinity affects soil physical, chemical, and biological processes (Karlen et al., 2008) is a major constraint on crop yield (Mass and Hoffman, 1977; Mass, 1991). The standard world-wide measure of soil salinity is the electrical conductivity of a saturated soil paste extract (ECe – dS/m). Various values define the level of salinity in soil and specifically, non- (0-2 dS/m), slightly- (2-4 dS/m), moderately- (4-8 dS/m), highly- (8-16 dS/m) and severely (> 16 dS/m) saline (Barrett-

Lennard et al., 2008). To determine appropriate management for individual crops, farmers need to map the spatial variation of soil ECe. This is problematic because different crops have different cut-off values. For example, the nitrogen-fixing Dolichus lablab requires non-saline conditions (Land and Water Australia, 2006), whilst wheat (Triticum sp.) can tolerate moderate levels of ECe (i.e. 6 dS/m; Natural Resources Conservation Service, 2011). The ECe method is time-consuming and an expensive property to map (Triantafilis et al., 2001;

Corwin, 2008).

To speed-up the process of ECe mapping of soil, pedometric methods (McBratney et al.,

2003) are being employed. In the first instance, remotely sensed data such as digital elevation model data have been used as ancillary information because in arid-landscapes the soil forming factor of elevation leads to the movement of salts into lower-lying areas (Bilgili,

2013) or can account for saline conditions in estuarine environments (Huang et al., 2014a, b).

More recently, airborne -ray data (i.e. radioelements of K, U and Th) have been shown to be useful in mapping dryland salinity (Wilford, 2001) because gamma ray data are freely available nationwide (Minty et al., 2009) and it is relatively cheap to acquire (i.e., less than

121

$1 per ha: Spies and Woodgate, 2004) and is of potential use to characterise parent materials and soil types. More commonly, proximally sensed electromagnetic induction (EMI) data have been used over the past 40 years (Doolittle and Brevik, 2014), because apparent electrical conductivity (ECa - mS/m), as measured by EM38 and EM34 instruments, is used to calibrate and then subsequently map ECe in the rootzone at field (Amezketa and de Valle de Lersundi, 2008; Li et al., 2015), farm (Triantafilis et al., 2001) and district (Herrero et al.,

2003; Triantafilis and Buchanan, 2010) scales. This is because a large number of georeferenced ECa measurements can be made quickly; accounting for variability of salinity

(Doolittle and Brevik, 2014).

However, various sources of errors may be introduced and propagated during the pedometric process, which leads to uncertainty in the final prediction results. From the perspectives of both map-makers and end-users, it is of importance to quantify the associated errors using uncertainty analysis (Malone et al., 2011). According to Nelson et al. (2011), four main sources of error occur in digital maps, including model, covariate, analytical and positional errors. McBratney et al. (2006) used bootstrapping to estimate the model uncertainty of pedo-transfer functions. In terms of the covariate error, Bishop et al. (2006) discussed the error in elevation data and its propagation into the calculation of slope and used as a covariate to predict clay. Similarly, Ma et al. (2011) discussed accuracy issue of ECa associated with temperature. Furthermore, Brown and Heuvelink (2006) and Heuvelink et al.

(2007) developed a probabilistic framework for simulating ancillary data, considering their inherent positional and analytical errors propagation during mapping. More recently, Nelson et al. (2011) quantified and compared sources of errors by conducting a conditional simulation.

122

Although researchers (e.g., Guo et al., 2013; Scudiero et al., 2013, 2014) are increasingly combining different sources of ancillary data to map soil ECe, no studies have evaluated the uncertainty caused by using different sources of ancillary data. In this study, an error budget procedure is developed to quantify the relative contributions made by model, input (for all the ancillary data) and particularly the individual covariate (for each of the ancillary data) error when combining remotely sensed digital elevation model (DEM) or -ray spectrometry data and/or using ECa data collected using EM38 or EM34 data to map soil ECe.

The spatial distribution of different sources of errors will also be studied by using restricted maximum likelihood (REML) analysis across the predominantly irrigated cotton growing area of the Bourke district of Darling River Valley in New South Wales, Australia, where point-source (i.e. short scale) secondary soil salinization is problematic and as a function of leakage from various water storages and remobilisation of salts from a Cretaceous marine mudstone (Triantafilis and Monteiro Santos, 2011).

5.2 Materials and methods

5.2.1 Ancillary data There are three sources of remote and proximal sensed data. The remote sensed data include digital elevation model (DEM) and -ray spectrometry. The DEM data was collected using an 11 Channel RMS GR33A chart recorder flying at an altitude of 60 m. Data acquisition was collected concurrently with -ray data but the sampling interval is 7 m. Raw point-based elevation data was extracted using the buffering Tool with a width of 1 km using

ArcMap 10.2 (ESRI, 2012). Across the whole area, 141,572 measurements were obtained. Due to the memory limitation of the R software and a large amount of time caused by simulating the elevation dataset, the elevation dataset was resampled by taking every one point out of the ten in the order of Eastings. The resampled elevation dataset includes 14,158 measurements

123

(Fig. 5.2.1.1C). These measurements were spaced 70 m apart from west to east and 250 m apart from south to north.

The-ray spectrometry measures the natural radioactive emissions of -rays from the decay of potassium (K), uranium (U), thorium (Th) and across the whole spectrum (total count

– TC). The relative abundance provides information on the upper 0.30 m of soil (Minty, 1997).

The -ray data was obtained from the Geoscience Australia Data Delivery System. It was collected by New South Wales Department of Mineral Resources between May and July of

1995. The detector consisted of sodium iodide (NaI) treated with thallium. The crystal detector has a total volume of 33 L. The survey lines were approximately from west to east. Line spacing was generally 250 m, an interval of 60 m and at a height of 60 m. To avoid edge effects, raw point-based -ray spectrometry data were extracted using the buffering tool with a width of 1 km using ArcMap 10.2 (ESRI, 2012). Prior to modelling soil ECe, -ray spectrometry data were processed by removing the extremely small values (< mean - 3 × standard deviation) because these small values are mostly located within the water storage reservoirs (Fig. 5.2.1.1B). The reduced -ray spectrometry dataset has 13,946 measurements

(Fig. 5.2.1.1C).

Proximal measurements include ECa (mS/m) acquired using an EM38 (Geonics Ltd.,

Mississauga, Ontario, Canada), which was placed on the ground with measurements of soil electrical conductivity (ECa made in horizontal (EM38h) and vertical (EM38v) modes.

Practically ECa measures the solum (0-0.75 m) and rootzone (0-1.5 m), respectively (McNeill,

1990). EM34 (Geonics Ltd., Mississauga, Ontario, Canada) ECa measurements were also acquired at 10-(EM34-10), 20-(EM34-20) and 40-m (EM34-40) coil spacing in horizontal mode. This gives practical ECa to depths of 0 to 7, 0 to 15 and 0 to 30 m, respectively

(McNeill 1980). In both cases, ECa was measured on an approximate 0.5 km grid in irrigated and 1 km in dryland areas. In all, over 556 sites were visited and used in this study, with their

124 location shown in Fig. 5.2.1.1C. A Magellan NavPRO5000 was used for geo-referencing

(Geocentric Datum of Australia 1994).

125

Fig.5.2.1.1 A) Air-photo and B) infrastructure of the Bourke study area and C) locations of the EM38/34, -ray spectrometry and DEM surveys and soil sampling sites.

126

5.2.2 Soil sampling data The rationale for selecting the calibration sites is described in detail in Chapter 4.

Briefly, a total of 50 cores were selected and drilled in accordance with four classes derived from a fuzzy k-means (FKM) clustering of the initial 1,236 x 3 EM34 signal data matrix

(Triantafilis et al., 2003). This is done to allow for the calibration over the full range of EM34 signal data response. In addition, samples were obtained across the study area and along an approximate east-west running transect (Triantafilis et al., 2011). The samples collected include soil obtained at 0.30 m depth increments to a depth of 1.8 m. The samples were oven dried (60 ºC) and passed through a 2-mm sieve and analysed for electrical conductivity of a saturated soil paste extract (ECe - dS/m) using the method described by the U.S. Salinity

Laboratory Staff (1954). The average soil ECe values between 0-1.8m are used.

5.2.3 Digital mapping Because it is less expensive to obtain more densely observed ancillary data than to collect soil samples and conduct chemical or physical analysis, the predictive model is used to predict the target soil property (i.e. ECe) at unsampled locations where ancillary data is available. The process of model estimation and prediction are conducted using linear mixed models (LMMs). A LMM includes the fixed effect component (퐗흉), which is a relationship between the soil property of interest (i.e. ECe) and the ancillary data, a form of spatial correlation model (i.e. variogram) used to model the spatial dependence, the random effect, and an error component (휺). Generally, a LMM has the following form:

퐲 = 퐗흉 + 퐙퐮 + 휺 (1),

where 퐲 is a vector of the observed response (i.e. soil ECe). 퐗 is a matrix of predicting ancillary data at the observation points, and the vector 흉 contains the coefficients that

127 describe the fixed effects relationship. The vectors 퐮 and 휺 contain random errors which are spatially correlated such that

퐮 0 휉휎2퐆 ퟎ [ ] ~훮 ([ ] , [ ]) (2) 휺 0 ퟎ 휎2퐈

where 퐆 is the correlation matrix where the correlation depends only on the relative location of the observations. 퐈 is the identity matrix and 휎2 is the variance of the independent error and 휉 is the ratio of the variance of u to 휎2 (Lark et al., 2006). It should be noted the random terms of the explicit assumption that is being made are jointly Gaussian. The term 휺 represents both independent measurement errors and variation that arises from processes that are spatially dependent over shorter distances than separate samples; this is the nugget in geostatistical terms.

Empirical best linear unbiased prediction (E-BLUP) requires the kriging estimate using the LMM. Full details of restricted maximum likelihood (REML) estimation and E-BLUP can be found in Lark and Cullis (2004) and Lark et al. (2006). The geoR package (Ribeiro and Diggle, 2001) in R was used to fit the model parameters in Equations (1) and (2) while the gstat package (Pebesma, 2004) was used for prediction and simulation.

5.2.4 Predicting ECe

Fig. 5.2.4.1A shows the used approach. Briefly, ECe was predicted across the study area by first developing a 200-m prediction grid (5,918 locations). A regular three-step kriging approach is followed, whereby E-BLUP is used. In brief, the following steps are carried out (See Fig. 5.2.4.1A and B):

(i) Interpolate the ancillary (i.e. elevation, -ray spectrometry and ECa) data, 퐦 onto the prediction grid (퐦푝) and at the sampled locations (퐦표) using ordinary kriging (OK);

128

(ii) Fit a LMM between the measured ECe and 퐦표.

(iii) Use the 퐦푝 and the new LMM to interpolate ECe onto the prediction grid using E-

BLUP;

In step (ii) LMM selection was performed on the observed ECe prior to the error budget calculation. The fixed effects were determined by first fitting a full model with all of the ancillary data, including elevation, -ray (K, U, Th, and TC), ECa (EM38h, EM 38v, EM 34-

10, EM34-20 and EM34-40) and coordinates as predictors. The backward elimination was performed until all predictors had a probability less than 0.05. Leave-one-out cross-validation of the selected LMM model was performed to assess the model accuracy (root-mean-square- error, RMSE) and bias (mean error, ME). Other statistical indices were calculated to evaluate the goodness of fit, including Pearson’s r, Lin’s concordance (Lin, 1989) and standardised squared prediction error (ratio of prediction error squared to kriging variance, SSPE) (Bishop et al., 2015).

129

Fig. 5.2.4.1 Flowchart of E-BLUP and error budget calculation and its component (ancillary data (퐦 ), prediction grid(퐦푠), prediction grid (퐦푝) and sampled locations (퐦표) ). Note: ‘–’ means subtraction.

130

5.2.5 Error budget calculation To calculate an error budget on the different sources of error associated with the final digital map of ECe and as described above using E-BLUP (i.e. using LMM and REML), the procedure described by Nelson et al. (2011) is followed. However, and given Nelson et al.

(2011) found that the positional and analytical error only account for a relatively small proportion of the error budget, the discussion is limited to the following four different sources of error and described each one in turn and in order of the steps undertaken. That is model, input, combined and individual covariate sources of error. It should be noted that Nelson et al. (2011) only considered a spatial model with one covariate where here three sets of covariates are considered. The flowchart of the error budget calculation can be found in Fig.

5.2.4.1C and D.

The first step requires the use of 퐦푝 (i.e., 퐦푝−푒푙푒푣푎푡푖표푛, 퐦푝−푔푎푚푚푎, 퐦푝−푒푚) and 퐦표 to simulate ECe onto the prediction grid using 100 conditional simulations. The mean standard deviation of the conditional simulated ECe across the prediction grid is called the model-associated error (Fig. 5.2.4.1C).

Afterward, simulations of the ancillary data at the prediction locations are required. In brief, the following steps are carried out:

(i) Simulate the ancillary (i.e. elevation, -ray spectrometry and ECa) data, 퐦 onto

the prediction grid (퐦푠) using 100 conditional simulations (Fig. 5.2.4.1A);

this was essentially generated via sequential Gaussian simulation on the grid

(Pebesma, 2017);

(ii) Use the 퐦푠 (i.e., 퐦푠−푒푙푒푣푎푡푖표푛 , 퐦푠−푔푎푚푚푎 , 퐦푠−푒푚 ) and the new LMM

calculated during step (ii) of the digital mapping to interpolate ECe onto the

prediction grid using E-BLUP;

131

The mean standard deviation of the simulated ECe maps across the prediction grid is the input-associated error (Fig. 5.2.4.1C).

The next step requires calculation of the combined error. This requires the use of 퐦푠 and 퐦표 to simulate ECe onto the prediction grid using 100 conditional simulations. The mean standard deviation of the 100 simulated ECe is the combined error (Fig. 5.2.4.1 c), which includes both model error and input error.

It was assumed that the combined error comprises of errors due to the use of a spatial model and input ancillary data. Now that the combined error that accounted for the model- associated and input-associated errors are calculated, the model and input error can be calculated (Fig. 5.2.4.1C), respectively, by subtracting the input-associated and model- associated errors from the combined error.

To calculate the specific source of covariate error from a given ancillary data source, the exact same steps used to calculate the combined error is used but instead and for example considering elevation, 퐦푝−푒푙푒푣, 퐦푠−푔푎푚푚푎, 퐦푠−퐸푀, and 퐦표 are used to simulate ECe onto the prediction grid using 100 conditional simulations. The mean standard deviation of the 100 simulated ECe is called elevation-associated error. The same approach is used to determine the associated error for -ray and EM as required. By subtracting the elevation-associated error, -ray-associated and EM-associated error (Fig. 5.2.4.1D) from the combined error, the elevation error, -ray error and EM error, are obtained respectively.

5.3 Results and discussion

5.3.1 Exploratory data analysis

The summary statistics of the observed soil ECe (0-1.8 m) and ancillary data are shown in Table 5.3.1.1. The target soil property of ECe has a mean of 6.78 dS/m and standard

132 deviation of 3.61 dS/m. The large coefficient of variation (CV) of ECe (i.e. 53.30 %) indicates a strong variation across the study area. This reflects the short scale variation in soil salinity and as a function of salinization processes acting in close proximity to various large water storages (i.e. point source salinisation). Nevertheless, as the skewness value (0.48) is less than 1 (Oliver and Webster, 2014) ECe is considered to be normally distributed. The CV of elevation is small (1.9 %) in comparison and the data is not skewed (-0.1). This is similarly the case for the -ray data (e.g. Th: skewness = 0.1). However, the strongly skewed (i.e. > 1) nature of the ECa measurements of the EM38h (2.1) and EM38v (4.0) means that it is required to transform these data by log transformation. This reduced the skewness accordingly and to acceptable levels (e.g. EM38v = -0.5) with both log-transformed ECa data were used in subsequent analysis.

Table 5.3.1.1 Summary statistics of soil samples and reduced ancillary data.

EC (d EM38 EM38- logEM38- logEM38- EM34- EM34- EM34- e elevation K U Th TC S/m) -h v h v 10 20 40

N 50 14,158 13,946 13,946 13,946 13,946 556 556 556 556 556 556 556

Min 0.55 92.2 9.2 2.9 5.4 751.8 14 9 2.2 2.6 28 48 78

Mean 6.78 103.2 88.3 32.3 40.8 1422.7 190.8 148.8 4.7 5.0 145.4 161.2 186.2

Max 16.46 113.3 170 66.7 82.9 1890.5 1,118 1,409 7.3 7.0 300 278 284

Medium 7.08 103.1 88.5 31.9 40.7 1444.7 170 124 4.8 5.1 148 160 184

Std. Dev. 3.61 1.9 20.2 8.2 9.5 165.1 136.8 140.8 0.8 0.8 49.8 43.6 33.1

Skewness 0.48 -0.1 -0.0 0.2 0.1 -0.6 2.1 4.0 -0.2 -0.5 0.0 -0.0 0.0

Kurtosis 0.29 1.7 0.0 0.1 0.0 0.4 8.0 25.9 0.0 0.2 -0.4 -0.4 0.4

CV (%) 53.30 1.9 22.9 25.3 23.2 11.6 71.7 94.6 17.2 15.0 34.3 27.1 17.8

133

5.3.2 Variograms of ancillary data Fig. 5.3.2.1 shows the sample variograms and empirical variograms calculated for the raw ancillary data. Fig. 5.3.2.1A shows the variograms for raw elevation data. The fitted empirical variogram is a sum of a nugget variogram (C0 = 1.05) and an exponential variogram (C1 = 2.70, range = 2387.85 m). Fig. 5.3.2.1B shows the variograms for raw Th data. The fitted empirical variogram is a sum of a nugget variogram (C0 = 63.45) and an exponential variogram (C1 = 26.11, range = 1293.34 m). Fig. 5.3.2.1C shows the variograms for raw logEM38v data. The fitted empirical variogram is a sum of a nugget variogram (C0 =

0.13) and a spherical variogram (C1 = 0.49, range = 4887.97 m). The ratio of partial sill ( the difference between sill and nugget) to sill for elevation, Th and logEM38v are 0.72, 0.29 and

0.79, respectively. This value will be 1 for a variogram with no nugget variance (where the curve passes through the origin); conversely, it will be 0 where there is no spatially dependent variation at the range specified. Webster and Oliver (2007) suggested that the sample size and variability of the data affected the structure of the semi-variograms. Herein, the sampling density of elevation and -ray data are similar while the partial sill/ sill of the two ancillary data are quite different. In addition, although the size of -ray data is 24 times larger than EM data, the partial sill/ sill of EM data is larger than that of -ray data. Therefore, it is expected to see the large -ray error, intermediate elevation error and small EM-error for predicting soil ECe.

134

Fig. 5.3.2.1 Variograms of the raw ancillary data and including; a) elevation; b) thorium (Th) and c) logEM38v.

135

5.3.3 Spatial distribution of kriged ancillary data Fig. 5.3.3.1 shows the spatial distribution of some of the OK ancillary data. Fig.

5.3.3.1A shows elevation. The highest points (> 104 m) are associated with the elevated aeolian dune in the northwest, northeast and central-western part of the study area.

Conversely, the lowest points (< 102 m) identified through the centre and from the northeast to southwest, represent parts of the playa basin, the irrigated lands and the alluvial floodplain and adjacent to the Darling River. Between these two values, intermediate elevation (102-104 m) defines the irrigated lands which are characterised by Vertosols.

Fig. 5.3.3.1B shows the spatial distribution of Th. Intermediate-high (40-45 ppm) to high (> 45 ppm) Th predominantly characterises the south-eastern half of the study area.

Conversely, intermediate-low (30-35 ppm) to low (< 30 ppm) Th is identified in the north- western half of the study area and defining the aeolian dunes, with intermediate (35-40 ppm)

Th found in between and associated with the irrigated lands in the centre and the plains on the southern side of the Darling River.

Fig. 5.3.3.1C shows the mapped logEM38v. Intermediate-high (4.5-5.0 mS/m) to high (> 5.0 mS/m) values are found in the central west, southwest, and northeast of the study area. Interestingly the largest values are associated with the many large water storages where secondary salinization is evident. For example, this is the case to the north of the large circular and semi-circular water storages in the centre and also the irregularly shaped dual- cell storage at the southern end of the study area. Conversely, intermediate-low (3.5-

4.0mS/m) values adjacent to the Darling River, indicate where salts have not accumulated and owing to the relative age of the soil, whilst the lowest values (< 3.5 mS/m) characterise the aeolian dunes in the northwest and northeast.

136

Fig. 5.3.3.1 Spatial distribution of kriged A) elevation (m), B) thorium (Th) (ppm) and C) logEM38v (mS/m) across the Bourke study area.

137

5.3.4 Model selection

Table 5.3.4.1 shows the summary statistics of the OK ancillary data (i.e. 푚표) at the

50 ECe sampling points and their Pearson’s correlation coefficient (r) with soil ECe. In general, the OK ancillary data statistics are equivalent to the raw data (Table 5.3.1.1). Of greater significance for the simulation modelling and interpolation using E-BLUP is the correlations between the ancillary data and the target variable of ECe. Here, the highest correlation is achieved with two sources of ancillary data and specifically Th (r = 0.457) and

TC (r = 0.353). This is closely followed by logEM38v (r = 0.351) and logEM38h (r = 0.328).

Interestingly, the trend surface of Northing (r = 0.351) and EM34-10 (r = 0.284) showed significant correlation with ECe, whilst the correlation with elevation (r = -0.046) was not significant.

To determine which OK ancillary data is useful for LMM a backwards elimination is undertaken until all predictors had a probability less than 0.05. Table 5.3.4.2a shows the summary statistics of the fixed effects of the LMM model comprising the three statistically significant ancillary data (i.e. P < 0.05) including elevation (0.003), Th (0.001) and logEM38v (0.05). It should be noted that elevation is selected via the backward elimination although it is not significant with soil ECe. This is because of elevation is strongly correlated with logEM38v (Pearson’s r = -0.43) and Th (Pearson’s r = -0.55). In this study, we retained elevation to see its role in the DSM and uncertainty analysis. Then, a spherical model (see

Table 5.3.4.2b) is fitted, because of the slightly larger log-likelihood (-115.1) as compared with an exponential (-115.3) and non-spatial model (-115.3).

138

Table 5.3.4.1 Summary statistics of kriged ancillary data and Pearson’s r between soil ECe and kriged ancillary data.

Eastisg Northing logEMlogEM EM34- EM34 EM34 elevation K U Th TC s s 38h 38v 10 -20 -40

N 50 50 50 50 50 50 50 50 50 50 50 50

37649 Min 6665791 99.8 64.8 29.2 30.2 1048 3.4 3.7 88.4 111.2 150.2 0

38633 Mean 6674031 103.1 86.5 31.8 41.2 1419 4.8 5.1 153.8 169.3 192.8 6

39404 Max 6679701 106.8 105.4 35.2 47.7 1612 5.7 5.9 210.5 234.1 247.9 5 38614 Medium 6674946 102.9 88.6 31.8 42.5 1427 4.9 5.2 155.7 165.8 192.6 5

Std. Dev. 4774 3741 1.5 9.4 1.5 4.4 125.9 0.6 0.6 33.2 32.6 24.7

Skewness -0.1 -0.7 0.7 -0.7 0.4 -0.9 -1.2 -0.7 -0.9 -0.2 0.1 0.2

Kurtosis -1.0 -0.4 0.7 -0.0 -0.6 0.2 1.3 -0.3 0.2 -1.0 -0.9 -0.8

CV 1.2 0.1 1.5 10.8 4.8 10.6 8.9 12.4 11.5 21.6 19.3 12.8

Pearson’s 0.18 0.35* -0.05 0.27 -0.02 0.46** 0.35* 0.33* 0.35* 0.28* 0.27 0.26 r

*: P < 0.05; **: P < 0.01.

139

Table 5.3.4.2 Summary statistics of selected LMM comprising elevation, -ray spectrometry and ECa data.

a) Fixed effects of the LMM

Estimate Standard Fixed effects t value Prob > |t| s Error

intercept -153.100 45.481 -3.366 0.002

elevation 1.255 0.397 3.163 0.003

Th 0.493 0.134 3.678 0.001

logEM38-v 1.990 0.997 1.997 0.050 b) Variogram of the LMM Variogram Spherical 2 (partial sill) 8.818 range (m) 966.9 nugget 0.000 Log- -115.1 likelihood c) Model precision and bias Cross-validation Pearson’s r 0.536

Lin’s Estimate 0.476 concordance Lower 0.270 correlation coefficient Upper 0.641

Mean 0.982 SSPE Median 0.223

ME (dS/m) -0.03

RMSE (dS/m) 3.03

5.3.5 Spatial distribution of predicted ECe using E-BLUP

Fig. 5.3.5.1A shows the map of predicted ECe (dS/m) using E-BLUP and LMM of

퐦표 the OK ancillary data 퐦푝. In brief, highly saline soil (ECe > 8 dS/m) characterises the western and northern parts of the alluvial plain and the playa basin. This is particularly the

140 case with the irrigated lands. This is the case in areas adjacent to several large water storage reservoirs, where point source salinization is evident. This is particularly the case in fields directly to the north of the circular storage where soil salinization is so severe that the field is no longer used for irrigated agricultural production (i.e., Easting: 383000, Northing:

6673000).

This is similarly the case to the south of the western corner of the semi-circular storage (i.e., Easting: 387500, Northing: 6672000) as well as the irregularly shaped dual-cell storages (i.e., Easting: 383000, Northing: 6667500) where shallow saline water tables occur

(Buchanan and Triantafilis, 2009). By contrast, slightly saline (ECe: 2-4 dS/m) to non-saline soil (ECe < 2 dS/m) conditions characterise in the alluvial plain in the southwest corner of the study area and also adjacent to the Darling River. The elevated areas associated with the aeolian dunes are also non-saline for the most part.

Fig. 5.3.5.1B shows the plot of measured versus predicted ECe (dS/m) using leave- one-cross validation results of the E-BLUP. Table 5.3.4.2c shows that the performance of the fitted model is moderate, with the results reasonably precise (RMSE = 3.03 dS/m) given

RMSE is less than 20 % of the range of ECe (15.91 dS/m), with the results not biased (ME = -

0.03 dS/m). Lin’s concordance is relatively small (0.476) and SSPE has a mean of 0.982 and median of 0.223. With regard to SSPE, a mean value of 1.00 and median of 0.45 indicates that the prediction variance accurately reflects the actual errors (Lark, 2002). Therefore, this model of spatial variation is acceptable.

Whilst Pearson r (0.536) and Lin’s concordance (0.476) are moderate, Fig. 5.3.5.1B shows that for the most part the measured and observed ECe are equivalent and fall close to the 1:1 line. For example, the two measured non-saline (38 and 43) and slightly saline (48) sites located on the alluvial plain in the southern part of the study area, are predicted to be

141 non-saline (i.e. < 2 dS/m) and slightly saline (2-4 dS/m), respectively. Conversely, two of the most highly saline (i.e. 2 and 9) sites located in the playa basin at the northern end are similarly predicted to be highly saline (i.e. 8-16 dS/m). In between these extremes, it is possible to predict ECe fairly well at moderately saline ECe (4-8 dS/m) in irrigated areas and on the alluvial plain in the southern (i.e. 40 and 49), central (i.e. 12, 14 and 16) and northern

(i.e. 4 and 30) areas.

However, in some cases, it was not possible to predict ECe well. The best example is the extremely saline site (45) located at the western edge of the study area where ECe is 17 dS/m. The reason for this is mostly to do with the fact that salinity here is isolated and a function of the site's proximity to a large conveyance channel which is known to leak and where point source salinization is apparent. Another reason is because the site is located at the edge of the study area and there is insufficient data (e.g. ECa) to enable a more accurate prediction. This is similarly the case for sites where predicted ECe is overestimated (i.e. sites

25 and 50). In this case, these sites are located in close proximity to the Darling River, and adjacent to areas where measured ECa is intermediate-large, and because there is actually no salinity issues the interpolation leads to an overestimation of ECe.

It would be interesting to know if a similar model can be obtained by excluding the relatively expensive ground-based ECa data (logEM38v). Table 5.3.5.1 shows the leave-one- out cross-validation results using only air-borne -ray spectrometry data (i.e. Th) and elevation data. It is found that the model performs almost as well as the model comprising logEM38v data, given the RMSE (3.066 dS/m) and ME (-0.063 dS/m) are slightly greater with Lin’s concordance (0.435) and Pearson’s r (0.516) slightly smaller. This suggests that the selected LMM comprising of elevation, EM and -ray data may be over-fitted. However,

142 for the purpose of understanding the error propagation of different ancillary daya in the modelling process, we kept all the ancillary data for the following interpretation.

Similarly, air-borne -ray spectrometry data for LMM analysis is excluded. The results indicate that this data model performs much worse than the current LMM (RMSE =

3.42 dS/m, ME = -0.21 dS/m, Lin’s = 0.242). This is an interesting result because although the -ray spectrometry data (i.e. Th) has a large error, it still provides necessary information for the LMM and should be retained for predicting the distribution of soil salinity.

143

Fig. 5.3.5.1 A) Spatial distribution of predicted ECe (dS/m) using E-BLUP; B) predicted ECe (dS/m) vs. measured ECe (dS/m) using leave-one-out cross-validation.

144

Table 5.3.5.1 Summary statistics of selected LMM comprising elevation and -ray spectrometry data. a) Fixed effects of the LMM Standard Fixed effects Estimates t value Prob > |t| Error

intercept -113.448 42.430 -2.674 0.010

elevation 0.580 0.132 4.395 < 0.001

Th 0.933 0.376 2.483 0.017 b) Variogram of the LMM

Variogram Spherical

2 (partial 9.41 sill) range (m) 1058.1

nugget 0.000

Log- -119 likelihood c) Model precision and bias

Cross-validation

Pearson’s r 0.516

Estimat Lin’s 0.435 e concordance correlation Lower 0.234 coefficient Upper 0.600 Mean 0.969 SSPE Median 0.205 ME (dS/m) -0.06

RMSE 3.07 (dS/m)

145

5.3.6 Error budget evaluation: model, input and combined

Fig. 5.3.6.1A shows the standard deviation of predicted ECe at each prediction point due to model error and calculated using simulated ancillary data (i.e. elevation, Th, and logEM38v), measured ECe and E-BLUP that yielded the input-associated error. The input- associated error was subtracted from the combined error. It is evident that the model error variance map is random. This is unsurprising given the model error is caused by the interpolation process of E-BLUP, and includes fixed effect components(퐗흉)(i.e. elevation,

Th and logEM38v), the spatially correlated component (i.e. spherical variogram) and the random error (휺). A combination of all these components makes the model error randomly distributed. Table 5.3.6.1 shows that the model error is small (0.02 dS/m).

Table 5.3.6.1 Calculated error budget for different sources of error

Source of error Main causes Error (dS/m) Combined Kriging error + covariate error 4.44 Model Kriging error 0.02 Input Covariate error 1.38 Elevation Elevation error 0.24 -ray -ray spectrometry error 2.74 EM ECa error 0.14

Fig. 5.3.6.1B shows the standard deviation of predicted ECe at each prediction point due to input error using OK of ancillary data (i.e. elevation, -ray and ECa) and 100 conditional simulations with measured ECe that yielded the model-associated error. The model-associated error was subtracted from the combined error. It is evident that the input error is unevenly distributed across the field and as shown in Fig. 5.3.5.1B. Specifically, intermediate-large (1.5-1.7 dS/m) to large (> 1.7 dS/m) input error is evident in the central part of the study area encompassing the playa basin in the northeast and extending down into the clay plain in the centre and southwest and also adjacent to the River.

146

The larger error is in part attributable to the lack of ancillary data and with regard to

ECa and as evident in Fig. 5.3.6.1C. Specifically, this is the case in areas adjacent to the

Darling River and also on the aeolian dune in the central-southern part of the study area and southwest of the large circular storage. In these locations, the paucity of ECa data available is due to difficulties in obtaining easy access. This was because the land associated with the aeolian dune is not suitable for agriculture owing to its clay to loamy sand texture

(Buchanan et al., 2012) and its susceptibility to wind erosion. This is similarly the case with the area adjacent to the Darling River, but here agriculture is unsuitable because of the presence of large levees and channels associated with previous water courses of the Darling

River. In both cases, the land has been left for the most part uncleared. As such, ECa surveying was limited to data collection along either side of earthen roadways (see Fig.

5.3.6.1C).

Conversely, the input error is intermediate-small (1.1-1.3 dS/m) to small (< 1.1 dS/m) in areas southeast of the Darling River. Of interest, and in terms of predicting highly saline

ECe (8-12 dS/m), the input error is generally small in the known areas of secondary soil salinity and adjacent to many of the large storages. The two best examples are immediate to the north of the circular storage and the southwest corner of the semi-circular storage. In both cases, the variance is small (< 1.1 dS/m) and as compared with the moderately (6-8 dS/m) to highly (8-12 dS/m) saline nature of the predicted ECe (see Fig. 5.3.6.1C). Table 5.3.6.1 shows that the input error is, however, larger (1.38 dS/m) than the model error described previously.

Fig. 5.3.6.1C shows the standard deviation of predicted ECe at each prediction point due to combined error using simulated ancillary data (i.e. elevation, -ray and ECa), measured

ECe and 100 conditional simulations for each set of the 100 simulated ancillary data. This

147 map summarises 10,000 maps of predicted ECe. Again, and as with the model error, it is evident that the variance due to combined error is randomly distributed. Table 5.3.6.1 shows that the combined error is larger (4.44 dS/m) the standard deviation (3.61 dS/m) of measured

ECe.

148

Fig.5.3.6.1 Spatial distributions of standard deviation of predicted ECe (dS/m) due to A) model error, B) input error and C) combined error across Bourke study area.

149

5.3.7 Error budget evaluation: covariate In order to further understand the role of the different ancillary data made in the input error, the specific sources of covariate error are calculated. Fig. 5.3.7.1A shows the standard deviation of predicted ECe at each prediction point due to elevation input using only the OK elevation data, and simulated -ray and ECa from 100 conditional simulations with measured

ECe. This yields the elevations-associated error, which was subtracted from the combined error. Fig. 5.3.7.1B and C show the same result but for the -ray and ECa data.

The maps are equivalent to Fig. 5.3.5.1A in that the standard deviation is similarly random. Despite this, Table 5.3.6.1 shows that of these three sources of ancillary data, the elevation error (0.24 dS/m) is more than 10 times smaller than the -ray error (2.74 dS/m).

This is despite the fact that an equivalent amount of data is available. One reason for the larger error might be attributable to the larger CV of the -ray Th (10.6 %) data as compared to the elevation (1.5 %). It is more likely that the -ray Th has a relatively large proportion of error of all the ancillary data, given the nugget of the variogram for Th is the largest (63.45).

This is consistent with Bierwirth and Brodie (2008), who reported counting errors and noise associated with -ray collection can be around 10% for a single measurement point.

With regard to the EM error (0.14 dS/m) it is smaller than both the elevation and -ray error. One reason might be because the signature-to-noise ratio (quality of the useful information compared to the environmental noise) of the elevation and -ray data derived from the airborne survey is larger than the ECa data collected during the ground-based EM surveying. In terms of DEM, various sources of error occur due to data collection (Chaplot et al., 2006), interpolation methods (Carrara et al., 1997) and characteristics of terrain surface

(Skidmore, 1989). With regard to the ECa measurements, although it is affected by environmental noises such as the temperature (Sudduth et al., 2001), the routine calibration

150 process of the EM induction instrument should minimise the measurement error. In this study, the smallest EM error illustrates us ECa data is more reliable and can be used without further calibration for soil mapping while some pre-processing techniques may be required to improve the quality of the air-borne elevation and -ray data.

151

Fig.5.3.7.1 Spatial distributions of the standard deviation of predicted ECe (dS/m) due to A) elevation error, B) -ray error and C) EM error across Bourke study area.

152

5.4 Conclusions

In this study, an error budget procedure for digital mapping of soil ECe at the district scale using a combination of elevation, -ray spectrometry and EM data is developed. The budget considers the contributions of the model, input, combined and covariate errors using the standard deviation of the predicted ECe of the prediction grid. The combined error is approximately 4.44 dS/m, which is relatively large compared to the standard deviation of measured ECe. Of this error, most of it is attributable to the input error (1.38 dS/m) which is larger than the model error (0.02 dS/m). In terms of the input error, it is determined that the larger standard deviation is attributable to the lack of ancillary data, namely the ECa in areas adjacent to the Darling River and also on the aeolian dune where data collection was difficult owing to dense native vegetation.

Given the relatively large error caused by the use of -ray spectrometry data (61.7%), it is suggested more advanced pre-processing techniques (e.g., filtering) should be applied to the -ray spectrometry data or incorporate other available ancillary data such as satellite imagery (Lobell et al., 2010; Guo et al., 2013) to improve the prediction. In addition to this, it is also necessary to improve the robustness of the interpolation model given the current LMM may be over-fitted. This may be achieved by adopting non-linear prediction algorithms, such as Geoadditive model (Kammann and Wand, 2003), support vector machine (Cai et al., 2010), decision trees (Evans, 1998) and artificial neural network (Patel et al., 2002).

In future, it is worth further exploring the meaningfulness of the error produced using the digital soil mapping approach. This requires collecting information about the cost of the soil sampling, EM surveys, as well as the cost of the various management practices to remediate ths salt affected soils. Along with the estimated error in the final maps, an implicit loss function can be used to estimate the potential loss to the land owners or farmers when

153 management practices are carried out based on the inaccurate soil maps (Lark and Knights,

2015).

5.5 References Amezketa, E, del Valle de Lersundi, J, 2008. Soil classification and salinity mapping for determining restoration potential of cropped riparian areas. Land Degradation and Development 19, 153-164.

Barrett-Lennard, EG, Bennett, SJ, Colmer TD, 2008. Standardising the terminology for describing the level of salinity in soils. 2nd International Salinity Forum, Adelaide, 31 March – 3 April, 4 pp.

Bierwirth, PN, Brodie, RS, 2008. Gamma-ray remote sensing of aeolian salt sources in the Murray– Darling Basin, Australia. Remote Sensing of Environment 112, 550-559.

Bilgili, AV, 2013. Spatial assessment of soil salinity in the Harran Plain using multiple kriging techniques. Environmental monitoring and assessment 185, 777-795.

Bishop, TFA, Horta, A, Karunaratne, SB, 2015. Validation of digital soil maps at different spatial supports. Geoderma 241, 238-249.

Bishop, TFA, Minasny, B, McBratney, AB, 2006. Uncertainty analysis for soil-terrain models. International Journal of Geographical Information Science 20, 117-134.

Brown, JD, Heuvelink, GDM, 2006. The Data Uncertainty Engine (DUE): a software tool for assessing and simulating uncertain environmental variables. Computers and Geosciences 33, 172-190.

Buchanan, SM, Triantafilis, J, 2009. Mapping water table depth using geophysical and environmental variables. Ground Water 46, 80-96.

Buchanan, SM, Triantafilis, J, Odeh, IOA, 2012. Digital soil mapping of compositional particle-size fractions using proximal and remotely sensed ancillary data. Geophysics 77, WB201-211.

Cai, S, Zhang, R, Liu, L, Zhou, D, 2010. A method of salt-affected soil information extraction based on a support vector machine with texture features. Mathematical and Computer Modelling 51, 1319-1325.

Cambardella, CA, Moorman, TB, Parkin, TB, Karlen, DL, Novak, JM, Turco, RF, Konopka, AE, 1994. Field-scale variability of soil properties in central Iowa soils. Soil Science Society of America Journal 58, 1501-1511.

Carrara, A, Bitelli, G, Carla, R, 1997. Comparison of techniques for generating digital terrain models from contour lines. International Journal of Geographical Information Science 11, 451-473.

Chaplot, V, Darboux, F, Bourennane, H, Leguédois, S, Silvera, N, Phachomphon, K, 2006. Accuracy of interpolation techniques for the derivation of digital elevation models in relation to landform types and data density. Geomorphology 77, 126-141.

Corwin, D.L., 2008. Past, present, and future trends in soil electrical conductivity measurements using geophysical methods. In: Allred, B.J., Daniels, J.J., Ehsani, M.R. (Eds.), Handbook of Agricultural Geophysics. CRC Press, Taylor and Francis Group, Boca Raton, Florida, pp. 17-44. Corwin, D.L., Rhoades, JD (1982) An improved technique for determining soil electrical

154

conductivity — depth relations from above ground electromagnetic induction measurements. Soil Science Society of America Journal 46, 517-520.

Doolittle, JA, Brevik, EC, 2014. The use of electromagnetic induction techniques in soils studies. Geoderma 223-225, 33-45.

Evans, F, 1998. An investigation into the use of maximum likelihood classifiers, decision trees, neural networks and conditional probabilitistic networks for mapping and predicting salinity. MSc thesis, School of Computer Science, Curtin University of Technology, Perth, WA.

Guo, Y, Shi, Z, Li, HY, Triantafilis, J, 2013. Application of digital soil mapping methods for identifying salinity management classes based on a study on coastal central China. Soil Use and Management 29, 445-456.

Herrero, J, Ba, AA, Aragüés, R, 2003. Soil salinity and its distribution determined by soil sampling and electromagnetic techniques. Soil Use and Management 19, 119-126.

Heuvelink, GMB, Brown, D, van Loon, EE, 2007. A probabilistic framework for representing and simulating uncertain environmental variables. International Journal of Geographical Information Science 21, 497–513.

Huang, J, Nhan, T, Wong, V, Johnston, S, Lark, RM, Triantafilis, J, 2014a. Digital soil mapping of a coastal acid sulfate soil landscape. Soil Research 52, 327-339.

Huang, J, Wong, V, Triantafilis, J, 2014b. Mapping soil salinity and pH across an estuarine and alluvial plain using electromagnetic and digital elevation model data. Soil Use and Management 30, 394-402.

Kammann, EE, Wand, MP, 2003. Geoadditive models. Journal of the Royal Statistical Society. Series C: Applied Statistics 52, 1-18.

Karlen, DL, Tomer, MD, Neppel, J, Cambardella, CA, 2008. A preliminary watershed scale soil quality assessment in north central Iowa, USA. Soil and Tillage Research 99, 291-299.

Land and Water Australia, 2006. Lablab fact sheet for Grain and Graze - May 2006. Accessed from http://lwa.gov.au/files/products/grain-and-graze/pn20436/pn20436.pdf.

Lark, RM, 2002. Robust estimation of the pseudo cross‐variogram for cokriging soil properties. European Journal of Soil Science 53, 253-270.

Lark, RM, Cullis, BR, Welham, SJ, 2006. On spatial prediction of soil properties in the presence of a spatial trend: the empirical best linear unbiased predictor (E-BLUP) with REML. European Journal of Soil Science 57, 787-799.

Lark, RM, Cullis, BR, 2004. Model-based analysis using REML for inference from systematically sampled data on soil. European Journal of Soil Science 55, 799–813.

Lark, R. M., & Knights, K. V., 2015. The implicit loss function for errors in soil information. Geoderma, 251, 24-32.

Li, HY, Webster, R, Shi, Z, 2015. Mapping soil salinity in the Yangtze delta: REML and universal kriging (E-BLUP) revisited. Geoderma 237-238, 71-77.

155

Lobell, DB, Lesch, SM, Corwin, DL, Ulmer, MG, Anderson, KA, Potts, DJ, Doolittle, JA, Baltes, MJ, 2010. Regional-scale assessment of soil salinity in the Red River Valley using multi-year MODIS EVI and NDVI. Journal of Environmental Quality 39, 35-41.

Ma, R, McBratney, A, Whelan, B, Minasny, B, Short, M, 2011. Comparing temperature correction models for soil electrical conductivity measurement. Precision Agriculture 12, 55-66.

Malone, BP, de Gruijter, JJ, McBratney, AB, Minasny, B, Brus, DJ, 2011. Using additional criteria for measuring the quality of predictions and their uncertainties in a digital soil mapping framework. Soil Science Society of America Journal 75, 1032-1043.

Mass, EV, 1991. In Agricultural Assessment and Management, ASCE Manuals and Report on engineering Practice No.71. (K.K. Tanji, Ed.) American Society of Civil Engineers 262-304.

Mass, EV, Hoffman, GJ, 1977. Crop Salt Tolerance - Current Assessment. Journal of the irrigation and drainage division 103, 115-134.

Natural Resources Conservation Service, 2011. Crop tolerance and yield potential of selected crops as influenced by irrigation water salinity (ECw) or soil salinity (ECe). Accessed from http://www.nrcs.usda.gov/Internet/FSE_PLANTMATERIALS/publications/azpmstn10485.pdf.

McBratney, AB, Santos, MLM, Minasny, B, 2003. On digital soil mapping. Geoderma 117, 3-52.

McBratney, AB, Minasny, B, Viscarra Rossel, RA, 2006. Spectral soil analysis and inference systems: a powerful combination for solving the soil data crisis. Geoderma 136, 272-278.

McNeill, JD, 1980. Electromagnetic terrain conductivity measurement at low induction numbers: Technical Note TN6, Geonics Limited, Mississauga, Ontario, Canada.

McNeill, JD, 1990. Geonics EM38 Ground Conductivity Meter: EM38 Operating Manual. Geonics Limited, Ont., Canada.

Minty, B, Franklin, R, Milligan, P, Richardson, M, Wilford, J, 2009. The Radiometric Map of Australia. Exploration Geophysics 40, 325-333.

Nelson, MA, Bishop, TFA, Triantafilis, J, Odeh, IOA, 2011. An error budget for different sources of error in digital soil mapping. European Journal of Soil Science 62, 417-430.

Northern Floodplains Regional Planning Committee, 2004. Vegetation Communities of the Northern Floodplains Western New South Wales Books 1–3: Walgett. Brewarrinna and Bourke Shires, Northern Floodplains Regional Planning Committee.

Patel, RM, Prasher, SO, God, PK, Bassi, R, 2002. Soil salinity prediction using artificial neural networks. Journal of the American Water Resources Association 38, 91-100.

Pebesma, EJ, 2004. Multivariable geostatistics in S: the gstat package. Computers and Geosciences 30, 683-691.

Pebesma, E., 2017. Package ‘gstat’. Accessed on 05/02/2017 from https://cran.r- project.org/web/packages/gstat/gstat.pdf

Pickup, G, Marks, A, 2000. Identifying large-scale erosion and deposition processes from airborne gamma radiometrics and digital elevation models in a weathered landscape. Earth Surface Processes and Landforms 25, 535-557.

Ribeiro, PJ Jr, Diggle, PJ, 2001. geoR: a package for geostatistical analysis. R-News 1, 14-18.

156

Scudiero, E, Skaggs, TH, Corwin, DL, 2014. Regional scale soil salinity evaluation using Landsat 7, western San Joaquin Valley, California, USA. Geoderma Regional 2-3, 82-90.

Scudiero, E, Teatini, P, Corwin, DL, Deiana, R, Berti, A, Morari, F, 2013. Delineation of site-specific management units in a saline region at the Venice Lagoon margin, Italy, using soil reflectance and apparent electrical conductivity. Computers and Electronics in Agriculture 99, 54-64.

Skidmore, AK, 1989. A comparison of techniques for calculating gradient and aspect from a gridded digital elevation model. International Journal of Geographical Information System 3, 323-334.

Spies, B, Woodgate, P, 2003. Salinity mapping methods in the Australian context. Technical Report. Programs Committee of the Natural Resource Management Ministerial Council and the National Dryland Salinity Program.

Sudduth, KA, Drummond, ST, Kitchen, NR, 2001. Accuracy issues in electromagnetic induction sensing of soil electrical conductivity for precision agriculture. Computers and Electronics in Agriculture 31, 239-264.

Triantafilis, J, Laslett, GM, McBratney, AB, 2000. Calibrating an electromagnetic induction instrument to measure salinity in soil under irrigated cotton. Soil Science Society of America Journal 64, 1008-1017.

Triantafilis, J, Odeh, IOA, Minasny, B, McBratney, AB, 2003. Elucidation of physiographic and hydrogeological units using fuzzy k-means classification of EM34 data in the lower Namoi valley. Environmental Modelling and Software 18, 667-680.

Triantafilis, J, Buchanan. SM, 2010. Mapping the spatial distribution of subsurface saline material in the Darling River valley. Journal of Applied Geophysics 70, 144-160.

Triantafilis, J, Monteiro Santos, FA, 2011. Hydrostratigraphic analysis of the Darling River valley using a joint inversion of EM38 and EM34 data and a 1D spatially constrained algorithm for quasi-3D conductivity imaging. Hydrogeology Journal 19, 1053-1063.

Triantafilis, J, Odeh IOA, McBratney, AB, 2001. Five geostatistical methods to predict soil salinity from electromagnetic induction data across irrigated cotton. Soil Science Society of America Journal 65, 869-978.

United States Salinity Laboratory Staff., 1954. Diagnosis and improvement of saline and alkali soils. USDA Agricultural Handbook No. 60US Government Printing Office, Washington, DC.

Webster, R., & Oliver, M. A. 2007. Geostatistics for environmental scientists. John Wiley & Sons.

Wilford, JR, Dent, DL, Dowling, T, Braaten, R,2001. Rapid mapping of soils and salt stores: Using airbone radiometrics and digital elevation models. AGSO Research Newsletter 33-40.

157

6 Conclusion Digital soil mapping is the process whereby easier and cheaper to acquire ancillary data is used along with spatial and non-spatial mathematical and statistical methods to generate maps of soil types or soil properties. In this thesis, DSM was used for both purposes and across the

Bourke Irrigation District.

With respect to digital soil mapping to generate a map of soil types, it was shown in this thesis that the non-spatial clustering algorithm of FKM analysis (FuzMe) was capable of identifying soil landscape units, from the clustering of proximal (i.e. EM induction) and remote (i.e. airborne -ray spectrometry) sensed data. The results were consistent units previously identified using traditional soil survey and extrapolation by air-photo interpretation. Herein, the units were also shown to be statistically significant with respect to various soil physical and chemical soil properties which are highly relevant to soil use and management in this highly productive agricultural landscape. Specifically, the DSM units could be discriminated based on soil texture where the elevated aeolian Dunes were sandy, whereas the clay alluvial plain was characterised by heavy clay textures. Chemically they differed and as a function of salt content and relative to various thresholds of slightly-, moderate and strongly saline topsoil and subsoil values. Future work in this regard could explore the use of other ancillary data covariates (i.e. inclusion of digital elevation model data) or the expansion of the work across larger areas.

With respect to digital soil mapping to generate a map of a soil property, it was shown in this thesis that spatial REML analysis was capable of developing a model suitable for mapping soil salinity, from proximal (i.e. EM induction) and remote (i.e. airborne -ray spectrometry) sensed data. In this regard, the results provided a high-resolution map of soil salinity, which enabled its spatial extent to be discerned. It was clear that the patterns of salinity were consistent with the location of large earthen water storages and also the major

158 conveyance infrastructure which channels the water from the Darling River to the storages.

The errors of the final map could also be ascertained by the use of the error budget approach which gave insights into what measures in terms of ancillary data generation are required to improve prediction of salinity in prediction. In this regard, the collection of more detailed EM data would improve salinity prediction, particularly in the areas where errors were largest near the River and where the sampling spacing was ~ 1 km.

In both cases, that is for DSM of soil type and soil properties, future work to enhance applications for precision agriculture is to develop a mobile sensing system using a ground- based -ray spectrometer and EM induction instruments for more precise and cost-effective acquisition of the ancillary. Moreover, in order to better understand the 3-D nature of the conductivity distribution from the topsoil and into the subsoil, a collection of multiple-coil array ECa (e.g. DUALEM-421) might be appropriate. In this regard, there is a growing body of literature where this type of data is being used along with inversion algorithms which allow the change in conductivity values with depth to be determined and which helps to better understanding of the variation of different soil properties with depth and reduces the uncertainty in the predictions made in the subsoil.

159