HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Cockx, L. Articles by De Vos, B. Search for Related Content PubMed Articles by Cockx, L. Articles by De Vos, B. GeoRef GeoRef Citation Agricola Articles by Cockx, L. Articles by De Vos, B. Related Collections Electromagnetic Induction, EMI Spatial Variability Forest Soils

PEDOLOGY

Using the EM38DD Soil Sensor to Delineate Clay Lenses in a Sandy Forest Soil

^a Dep. of Soil Management and Soil Care, Ghent Univ., Coupure 653, 9000 Gent, Belgium
^b Research Institute for Nature and Forest, Gaverstraat 4, 9500 Geraardsbergen, Belgium

^* Corresponding author (liesbet.cockx{at}ugent.be).

	ABSTRACT

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
The objective of this study was to locate clay lenses in a sandy forest soil. The study site was a Scots pine (Pinus sylvestris L.) plantation in the Campine region of Belgium, which has been selected as a European Union-funded Level II Research Site for soil monitoring. Typically, the soils in the area have homogeneous sandy soil profiles, but clay lenses occur locally within a depth of 2 m of the soil surface. Locating these clay lenses is necessary because they can have a substantial impact on soil processes. Therefore, we used the EM38DD soil sensor to measure the soil apparent electrical conductivity (ECa) simultaneously in two orientations. Apparent electrical conductivity maps were generated and it was found that the variation in ECa was mainly driven by the spatial variability of soil texture across the study site. The ratio of the two orientations (profile ratio or PR) clearly revealed a circular pattern with decreased PR values (<1), which was identified as a clay lens. To delineate the extent of this lens, two numerical methods were used: (i) a fuzzy-k-means classification of the PR map focusing on the lowest centroid class, and (ii) a probability approach through indicator kriging. Using a validation image obtained from directed auguring, cell-by-cell comparisons were made for these two methods complemented with spatial accuracy measures. Although both methods tended to underestimate the spatial extent of the clay lens, the indicator kriging method was the most accurate, with an overall accuracy of 0.838, a proportion of error due to locational errors (-loc) of 0.864, and an average similarity of 0.841.

Abbreviations: AUC, area under the curve • CEC, cation exchange capacity • ECa, apparent electrical conductivity • EC_h, electrical conductivity measurement in the horizontal orientation • EC_v, electrical conductivity measurement in the vertical orientation • EMI, electromagnetic induction • FPI, fuzziness performance index • GPR, ground penetrating radar • IK, indicator kriging • NCE, normalized classification entropy • OK, ordinary point kriging • PR, profile ratio • ROC, receiver operating characteristic • RNE, relative nugget effect

	INTRODUCTION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
The spatial variability of forest soils is an important factor for forest site planning, quality, and productivity (Schoenholtz et al., 2000). Forest soils serve multiple production and environmental functions and maintaining these functions is crucial for sustainable forest management.

Accurate characterization of the soil is a preliminary step in site establishment, since both lateral and vertical soil heterogeneity may have an impact on tree growth patterns. Soil texture is a fundamental qualitative soil physical property, influencing many other properties and processes including soil formation, water movement, erosion potential, and cation exchange capacity (CEC). It strongly affects soil moisture and controls the pool of nutrients available for plant uptake (Mc Bride et al., 1990). Soil discontinuities may be defined by significant changes in soil texture, and their identification and spatial delineation is important in land use decision making (Ogg et al., 2000). The presence of a clay layer in a sandy soil constitutes a discontinuity that restricts infiltration and influences the lateral movement of soil water and agrochemicals (Doolittle et al., 1994). Concerning forest productivity, Woolery et al. (2002) selected clay content in the B horizon as an important parameter for estimating species productivity. Also, Bravo and Montero (2001) found texture to be an important factor for forest productivity. Generally, soil texture is known to determine the impact of soil disturbances on tree growth (Gomez et al., 2002). A potential consequence of soil compaction caused by harvesting or site preparation is the significant loss of site productivity. It should be mentioned, however, that forest productivity relies on the interplay of soil physical, chemical, and biological properties and processes, which can be complex and varying among forest ecosystems (Schoenholtz et al., 2000). Consequently, future forest management practices should be site specific and account for the spatial variability of soil properties. Therefore, knowledge of the spatial variability of soil texture may be important for interpreting tree productivity and planning forest management strategies.

New techniques have evolved that allow soil spatial variability to be identified using noninvasive, geophysical soil sensors. Sensors based on electromagnetic induction (EMI) have already proved their utility for soil characterization in agricultural soils (Corwin and Lesch, 2005). In this sensing technique, the bulk or apparent soil electrical conductivity (ECa) is measured, which is used as an indirect indicator of some soil properties (McNeill, 1980). In nonsaline soils, the variation in ECa is primarily determined by soil texture, moisture content, and CEC, all of which are important to plant biomass production. The inverse of ECa, the electrical resistivity, measured through direct current, can also be considered as a surrogate for the variability of soil physical properties (Samouëlian et al., 2005). Its depth of investigation depends on the distance between the electrodes, which is limited in densely populated forests by the inter-tree distance. On the other hand, ground-penetrating radar (GPR) has proved to be an effective tool for exploring subsurface horizons, also in forests (Butnor et al., 2003). Kung and Donohue (1991) showed that GPR was able to locate soil layers with textural discontinuities, whereas Boll et al. (1996) predicted the depth to textural interfaces using GPR. Both techniques detect changes in electromagnetic soil properties. The use of GPR is limited, however, since only electrically resistive soils are amenable to study (Butnor et al., 2003). As a response to this disadvantage, EMI has been used as a precursory tool to guide the more costly, complex, and time-consuming GPR measurements (Gish et al., 2002; Inman et al., 2002). Generally, it is recognized that EMI allows the detection of gradual lateral changes in textural properties (Doolittle and Collins, 1998), whereas GPR is better suited to vertical exploration of textural discontinuities. Nevertheless, several researchers (Brus et al., 1992; Bork et al., 1998; Mueller et al., 2003) have reported the use of EMI to find textural discontinuities within the soil profile.

This study performed both lateral and vertical soil explorations using an EMI sensor. The sandy soil of the study area is characterized by the presence of a clay lens at variable depth. We aimed to locate this textural discontinuity with the EM38DD sensor (Geonics Ltd., Mississauga, ON, Canada). This sensor has the advantage of measuring the ECa simultaneously in two orientations. The ratio of these two orientations provides an indirect measure of the degree of soil profile heterogeneity. Additionally, we investigated different data analysis techniques to delineate the spatial extent of the clay lens.

	MATERIALS AND METHODS

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Study Site
The study site is an even-aged, 75-yr-old Scots pine plantation of 2 ha in Brasschaat, Belgium (central coordinates: 51°18'33'' N, 4°32'14'' E). The plantation is part of a 150-ha mixed coniferous–deciduous state forest, called "De Inslag." Since 1988, the Flemish Research Institute for Nature and Forest has used this site as a research area and in 1992 it was integrated as a Level II Research Site into a European research program for monitoring forest ecosystems. Several studies concerning tree physiology, nutrient behavior, CO₂ and water cycles, forest vitality, and air pollution monitoring were conducted or are ongoing at this site.

The topography of the area is flat with a maximum height variation of 33 cm. The soil is a moderately wet sandy soil, characterized by an anthropologically disturbed spodic horizon, called a "post-podzol" in the Belgian soil classification (symbolized as Zegb on the Belgian soil map) or an Anthreptic Haplorthod according to the USDA Soil Taxonomy (Soil Survey Staff, 2003). The soil parent material is a homogeneous aeolian coarse sand, but at variable depths (usually from 1.5 to 2 m) clay with a thickness of at least 0.20 m can be found (>40lay, with clay defined as particles with a diameter <2 µm). The origin of the sedimented clay is unclear. It is hypothesized to be the remains of a bed of small fens covered by sand in the Pleistocene epoch. The mineral soil is characterized by very low pH values (pH in H₂O between 3.6 and 4.1) and a very low CEC; the forest floor is of the Mor type and varies in thickness across the site (1–13 cm). At some locations, a thick moss layer (3–7 cm) covering the litter layer is present. In the northwestern corner, traces of Second World War activities were found.

Apparent Electrical Conductivity Survey
We conducted a soil sensing survey using the EM38DD electromagnetic induction sensor. The EM38DD sensor consists of two EM38 units fixed perpendicular to each other. In each unit a transmitter coil is energized with an alternating current, inducing a primary electromagnetic field. This induces very small currents in the soil, which generate a secondary electromagnetic field that is sensed by a receiver coil located 1 m from the transmitter coil. The ratio of the secondary to the primary fields provides a measure of the ECa of the soil.

With the EM38DD, the soil ECa is measured simultaneously in two orientations, each having its own depth response profile. The sensitivity of the sensor can reach a depth of 2 m in low-conductive soils. The vertical orientation (EC_v) receives its major influence from deeper soil layers, while the horizontal orientation (EC_h) receives a dominant influence of the near surface soil. Combining the ECa measured in the two orientations in a so-called profile ratio (PR) provides an indication of the heterogeneity of the soil profile (Corwin et al., 2003): PR = EC_h/EC_v. A PR close to 1 indicates a uniform profile, a PR < 1 indicates a more conductive subsoil compared with the topsoil and a PR > 1 indicates decreasing conductivity with depth.

In total, 156 ECa measurements were taken by putting the EM38DD manually on the soil surface. Fifty-four locations were selected on the basis of a predefined grid with a 20- by 20-m spacing; the other locations were taken between these grid nodes to complement the regular sampling design with shorter distance measurements.

Soil Sampling
To provide an interpretation of the ECa measurements, soil samples at 23 locations were taken at 50-cm intervals down to 2 m. Texture was analyzed using the conventional pipette method. Additionally, in the area where we expected to find a clay lens, the presence of heavy clay within a depth of 2 m was checked by hand auguring at 60 locations. Where a clay lens was encountered, the depth to its top was registered. These observations were used as validation data.

Interpolation Techniques
The ECa measurements were interpolated using ordinary point kriging (OK). This is a geostatistical method that provides an estimate of a variable Z at any unsampled location x₀ using a linear combination of observations within a predefined neighborhood around x₀ (Goovaerts, 1997). The spatial structure of Z is represented by a variogram model, which is used to assign a weight to the n(x₀) neighbors Z(x), yielding the OK estimator:

[1]

The sum of these weights must equal unity to guarantee that the predictor is unbiased.

Indicator kriging (IK) was used to obtain a map of the probability of occurrence of the clay lens in terms of a critical PR and to interpolate the categorical (binary) presence/absence data to a validation map showing the presence of heavy clay within 2-m depth. To obtain a probability map of PR, alternative methods like stochastic simulations or disjunctive kriging could be used as well; however, we used IK because it is conceptually quite simple and its performance was evaluated to be similar (Lark and Ferguson, 2004). The principles of IK have been discussed in detail by Goovaerts (1997), Deutsch and Journel (1998), and others. Indicator kriging obtains the probability of a certain critical threshold z_c being exceeded by building the conditional cumulative distribution function (ccdf) at each point based on the behavior and correlation structure of indicator-transformed data. The ccdf F[x₀; z|(n)] signifies a probabilistic model for the uncertainty around the unknown value at x₀, where |(n) represents the conditioning to local information:

[2]

Therefore, original data have to be transformed into indicators i(x; z_k) in respect to a series of k threshold values z_k, selected across the range of data:

[3]

Generally, the quantiles of the ccdf are taken as a series of k threshold values, covering the range of variation of z (Van Meirvenne and Goovaerts, 2001). For each threshold, an indicator variogram is modeled and ordinary IK estimates of the indicators are used to approximate the ccdf at the every point x₀. After the ccdf is built, it must be post-processed to assess the probability that the estimation is larger, or smaller, than the critical threshold z_c.

Classification and Delineation Methods
Two numerical methods were used for the delineation of the clay lens from ECa measurements: (i) a fuzzy-k-means classification and (ii) a classification based on the probability of exceeding a critical threshold.

The fuzzy-k-means classification is an unsupervised classification used to identify "natural clusters" in a data set X (with elements x_ig; with i = 1, ..., n; g = 1, ..., p) having n individuals with p attributes (Bezdek, 1981). Each individual is allocated a membership [0,1] to each of the k clusters through an iterative algorithm starting with a random set of cluster means. Each individual is then assigned to the closest of these means and new means are recalculated based on the distance in attribute space between the individual and the cluster mean. This is repeated until a specified convergence criterion is met. The aim is to identify cluster centroids that minimize the generalized objective function, which is defined as follows:

[4]

where M is the matrix with membership values m_ij (with i = 1, ..., n; j = 1, ..., k), C is the matrix of class centers c_jg (with j = 1, ..., k; g = 1, ..., p), d²(x_i,c_j) is the squared distance between each data point x_i and its cluster centroid c_j, and is the fuzzy exponent defining the degree of fuzziness of the solution. When = 1 the solution is a hard partition; as approaches infinity, the solution approaches its maximum degree of fuzziness. In this study, the Euclidean distance was taken as the distance measure because only one input variable (p = 1) was used. The optimum number of clusters is determined by minimizing two indices: the fuzziness performance index (FPI), which measures the degree of fuzziness, and the normalized classification entropy (NCE), which indicates the degree of fuzzification. The FPI is defined as follows (Roubens, 1982):

[5]

where F is the partition coefficient:

[6]

The NCE is defined as:

[7]

where H is the entropy function:

[8]

One of the resulting clusters was retained as optimal, indicating the presence of the clay lens.

The second method is based on a map of the probability of the clay lens occurring. This probability map was obtained through IK; the centroid of the optimal cluster from the fuzzy-k-means classification was taken as the critical threshold z_c. The resulting probability map was classified into a Boolean map showing predicted absence or presence of the clay lens.

Accuracy Measurements
The results of the delineation methods were compared with a validation image showing the experimentally observed presence of the clay lens. Categorical comparisons are generally based on a confusion matrix containing categorical similarities obtained from a pixel-by-pixel comparison. Table 1 shows a two by two confusion matrix in which the elements are the number of grid cells that fall into each categorical combination. Diagonal elements represent an agreement between the reference and the prediction, and off-diagonal elements represent misclassifications (Congalton and Green, 1999).

View this table: [in this window] [in a new window]   Table 2. Formulae to calculate the kappa statistic () and its variants from a two by two confusion matrix.

To validate the probability map a receiver operating characteristic (ROC) curve can be constructed (Pontius and Schneider, 2001). The ROC curve relates relative proportions of correctly classified cells [a/(a + c)] and incorrectly classified cells [b/(b + d)] across a continuous range of probability thresholds. The area under this curve (AUC) is a measure of the ability to correctly discriminate between the absence and the presence of an event of interest-in our case, the clay lens. An AUC of 0.5 (a diagonal line on the curve) indicates a classification performance no better than chance, while a classification with perfect discrimination has an AUC of 1. Manel et al. (2001) showed that the AUC statistic is prevalence independent.

Besides for accuracy purposes, information from the contingency matrix was also used to determine the optimal probability threshold through the F measure. The F measure is defined as the harmonic mean of precision (P) and recall (R) where precision is the proportion of presences that are real presences [P = a/(a + b)] and recall is the proportion of correctly classified presences [R = a/(a + c)]. A weighted version of the F measure (van Rijsbergen, 1979) was used:

[9]

with ß [0,+] as a weighting factor controlling the relative importance of precision vs. recall. With ß = 1, precision and recall have equal weights; a smaller ß emphasizes precision, a larger ß emphasizes recall. In this study, ß was set to 0.5 and F_0.5 was maximized to select the optimal probability threshold.

	RESULTS AND DISCUSSION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Apparent Electrical Conductivity Measurements
The locations of the 156 ECa measurements are shown in Fig. 1a. The descriptive statistics of the EM38DD measurements (Table 3) show that the ECa values of this sandy soil were very low (3–9 mS m^–1) with relatively small CVs (22 and 29 0.000000or EC_h and EC_v, respectively). The EC_h and EC_v measurements had similar mean values within the same data range. Their distribution was symmetrical but platykurtic, whereas the PR measurements showed a skewed, leptokurtic distribution. In most studies, the correlation between EC_h and EC_v has been reported to be very strong (e.g., Triantafilis and Lesch, 2005; Vitharana et al., 2006), but in this study the correlation coefficient was only 0.74, indicating some deviation from a linear relationship.

View larger version (31K): [in this window] [in a new window]   Fig. 1. Point ordinary kriging maps of electrical conductivity (a) using the horizontal orientation (EC_h, mS m–1) with measurement locations of EM38DD (dots) and (b) using the vertical orientation (EC_v, mS m–1) with locations of texture samples (squares), and (c) the profile ratio (PR). (Metric Lambert coordinates on x and y axes.)

[10]

where (h) is the semivariance at lag distances h, C₁ is the sill, and a the range. The variograms of EC_h and EC_v had a similar spatial structure with a strong spatial relation (low relative nugget effect, RNE, of 10 and 6%, respectively), whereas the PR measurements had a larger RNE (30%; Table 3). The RNE measures the proportion of random or short-distance variability (RNE = [C₀/(C₀ + C₁)]100). The range of the variograms was on the order of 80 m for all three variables. Although the data range was rather small, maps of EC_h and EC_h, obtained by point OK with a pixel resolution of 1 by 1 m showed clear and rather similar patterns (Fig. 1a and 1b). The highest ECa values occurred in the east of the field, the lowest in the west. The PR map is given in Fig. 1c. It indicates that most of the study site is characterized by a rather homogeneous soil (PR values around 1), but in the northwest corner increased PR values were found (1.2–1.3) whereas in the southeast a circular phenomenon with decreased PR values (0.7–0.8) can be seen.

Clay Content and Apparent Electrical Conductivity Variability
Based on the clay content of the 23 sampled locations (Fig. 1b), two types of textural profiles could be distinguished: one homogeneous with low clay content down to 2 m (18 locations), the other one with a marked increase in clay content from a depth varying between 1 and 2 m downward (five locations) (Fig. 2). The standard deviation of the clay content of the homogeneous soil profile was quite uniform for all layers, on average 2.6%. The top 1 m of the heterogeneous soil profile had a similar standard deviation, but it increased strongly below 1 m to 6.50n average. This increase in standard deviation resulted from the variable depth at which the clay lens occurred. The heterogeneous soil profiles were all characterized by an increased EC_v, resulting in a decreased PR. The average PR of the heterogeneous profiles was 0.81, compared with an average PR value of 1.03 for the homogeneous soil profiles.

View larger version (14K): [in this window] [in a new window]   Fig. 2. Mean clay content of the heterogeneous and homogeneous soil profiles, shown in layers of 0.5 m (error bars represent the standard deviation).

Table 4 shows the Pearson correlation coefficients (r) between the mean clay content in the different soil layers (0.5-m intervals) and EC_h and EC_v. Up to 1.5-m depth, the correlations for both EC_h and EC_v were rather weak (r < 0.50) because of the textural homogeneity. In the 1.5- to 2-m layer, the correlation increased. In particular, EC_v was well related with the more variable clay content in that layer (r = 0.67); with EC_h, the correlation was considerably less (r = 0.41). Looking at the correlation across the whole soil profile, averaged for the top 1.5 and 2 m, the clay content in the 1.5- to 2-m layer seemed to increase the correlation, especially for EC_v (r increased from 0.44 to 0.56).

A fuzzy-k-means classification of the interpolated PR values was performed with a fuzzy exponent () of 1.3, which is in the middle of the range 1.12 to 1.5 as suggested for soil data by Odeh et al. (1990). Minimizing the NCE and the FPI resulted in the least fuzzy and least disorganized number of classes (Fig. 3a). The study site was optimally classified into five classes: two classes with a PR centroid around one, two classes with a PR centroid larger than one, and one class with a PR centroid smaller than one (Fig. 3b).

View larger version (18K): [in this window] [in a new window]   Fig. 3. (a) Fuzziness performance index (FPI) and normalized classification entropy (NCE) as a function of the number of classes, and (b) map of fuzzy classification of the profile ratio into five classes.

View larger version (35K): [in this window] [in a new window]   Fig. 4. Prediction of the clay lens according to (a) the fuzzy-k-means reclassification and (b) the indicator kriging probability map showing the probability that the profile ratio is 0.81.

Validation
All auger observations (those performed for the textural analyses and the binary observations used for the validation) were binary coded: 1 if a clay lens occurred within a depth of 2 m, 0 if it did not occur. These values were interpolated using IK with a spherical variogram model having a RNE of 11% and a range of 90 m (Fig. 5a), resulting in an indicator map of the presence of a clay lens. This map was categorized by a threshold of 0.5: whenever the estimated indicator was at least 0.5, a clay lens was expected. This map was taken as the validation image (Fig. 5b). Figure 5b also shows the sampled locations as white dots when the clay lens was observed within 2 m, or as black dots if no clay lens was encountered.

View larger version (16K): [in this window] [in a new window]   Fig. 5. (a) Indicator variogram model of the validation data (dots with number of data pairs) and (b) the validation image.

View larger version (13K): [in this window] [in a new window]   Fig. 6. (a) The measure of the harmonic mean of precision and recall weighted at 0.5 (F0.5) in function of probability thresholds and (b) classification of the indicator kriging probability map (with a probability threshold of 0.50).

To quantify the accuracy of the two methods, the Boolean maps were compared with the validation image on a pixel-by-pixel basis. Table 5 shows the resulting accuracy measures. The A_o reached high values for both methods (0.819–0.838), but it should be noted that there was a prevalence effect of 21%; the proportion of pixels without an increase in subsoil (<2 m) clay is larger than with presence of the clay lens. Predicting the absence of the clay lens in the whole study site would still give an A_o of 0.79. The values were 0.341 and 0.349 for the two methods, respectively, indicating a fair agreement (Landis and Koch, 1977). After correction for the prevalence effect, a substantial strength of agreement was reached, indicated by * values of 0.634 and 0.676. In terms of -loc, the IK method scored best with a -loc of 0.86. The spatial context was taken into account with S_a. Fuzziness of category was not considered and fuzziness of location was set with an exponential decay function (with a halving distance of two cells and a neighborhood defined by a radius of two cells). This measure was, for both methods, higher (0.819 and 0.841) than the indices calculated on a pixel basis. So we can conclude that the delineation was good in terms of spatial agreement. The IK-based method was found to be the most accurate, but it is also a more elaborate method based on a PR threshold value obtained from the fuzzy-k-means method. Nevertheless, the IK-based method was preferred to delineate the clay lens.

Depth of the Clay Lens
The depth to the upper boundary of a clay layer (D_c) was reported to be an important factor for biomass development in pine stands (Usoltsev and Vanclay, 1995). In our study site, a significant relationship between D_c (measured at 42 locations) and EC_v was found, with a Pearson correlation coefficient of –0.64 (Fig. 7a). The closer the textural discontinuity was to the surface, the more it contributed to the response of the sensor. Following Doolittle et al. (1994), we fitted an exponential regression model, which appeared to be the best model to predict D_c (Fig. 7b):

[11]

View larger version (17K): [in this window] [in a new window]   Fig. 7. (a) Scatterplot of electrical conductivity in the vertical orientation (EC_v) vs. the depth of the clay lens (Dc), and (b) the depth of the textural discontinuity (clay lens in meters below the surface and metric Lambert coordinates on x and y axes).

	CONCLUSIONS

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Analyzing the EM38DD data resulted in an accurate identification of the location of the clay lens, but tended to underestimate its spatial extent. Indicator kriging appeared to be the most appropriate processing technique through the use of a probability map showing the probability of the clay lens to occur. The fuzzy-k-means algorithm was used as an initial step to determine the optimal (and objective) PR threshold. In delineation studies, the spatial context of similarities should be taken into account; both -loc and S_a were found to give good results.

This sensor appeared to be a suitable instrument for detecting lateral and vertical soil textural variability important for forest management applications. To determine a pedological discontinuity, both orientations of the EM38DD sensor were essential because only the PR value was useful for analyzing the vertical heterogeneity of the soil profile. We concluded that the dual dipole version of the sensor allows extra opportunities for EMI applications. One field survey with the EM38DD is sufficient to characterize the spatial textural variability, to locate the presence of a clay lens, and to map its depth approximately.

	NOTES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Permission for printing and for reprinting the material contained herein has been obtained by the publisher.

Received for publication September 12, 2006.

	REFERENCES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 

• Bezdek, J.C. 1981. Pattern recognition with fuzzy objective function algorithms. Plenum Press, New York.
• Boll, J., R.P.G. van Rijn, K.W. Weiler, J.A. Ewen, J. Daliparthy, S.J. Herbert, and T.S. Steenhuis. 1996. Using ground-penetrating radar to detect layers in a sandy field soil. Geoderma 70:117–132.[CrossRef][Web of Science]
• Bork, E.W., N.E. West, J.A. Doolittle, and J.L. Boettinger. 1998. Soil depth assessment of sagebrush grazing treatments using electromagnetic induction. J. Range Manage. 51:469–474.
• Bravo, F., and G. Montero. 2001. Site index estimation in Scots pine (Pinus sylvestris L.) stands in the High Ebro Basin (northern Spain) using soil attributes. Forestry 74:395–406.[Abstract/Free Full Text]
• Brus, D.J., M. Knotters, W.A. van Dooremolen, P. van Kernebeek, and R.J.M. van Seeters. 1992. The use of electromagnetic measurements of apparent soil electrical conductivity to predict the boulder clay depth. Geoderma 55:79–93.[CrossRef][Web of Science]
• Butnor, J.R., J.A. Doolittle, K.H. Johnson, L. Samuelson, T. Stokes, and L. Kress. 2003. Utility of ground-penetrating radar as a root biomass survey tool in forest systems. Soil Sci. Soc. Am. J. 67:1607–1615.[Abstract/Free Full Text]
• Byrt, T., J. Bishop, and J.B. Carlin. 1993. Bias, prevalence, and kappa. J. Clin. Epidemiol. 46:423–429.[CrossRef][Web of Science][Medline]
• Cohen, J. 1960. A coefficient of agreement for nominal scales. Educ. Psychol. Meas. 20:37–46.[CrossRef][Web of Science]
• Congalton, R., and K. Green. 1999. Assessing the accuracy of remotely sensed data: Principles and practices. Lewis Publ., New York.
• Corwin, D.L., S.R. Kaffka, J.W. Hopmans, Y. Mori, J.W. van Groenigen, C. van Kessel, S.M. Lesch, and J.D. Oster. 2003. Assessment and field-scale mapping of soil quality properties of a saline sodic soil. Geoderma 114:231–259.[CrossRef][Web of Science]
• Corwin, D.L., and S.M. Lesch. 2005. Apparent soil electrical conductivity measurements in agriculture. Comput. Electron. Agric. 46:11–43.[CrossRef]
• Deutsch, C.V., and A.G. Journel. 1998. GSLIB: Geostatistical software library and user's guide. 2nd ed. Appl. Geostat. Ser. Oxford Univ. Press, Oxford, UK.
• Doolittle, J.A., and M.E. Collins. 1998. A comparison of EM induction and GPR methods in areas of karst. Geoderma 85:83–102.[CrossRef][Web of Science]
• Doolittle, J.A., K.A. Sudduth, N.R. Kitchen, and S.J. Indorante. 1994. Estimating depths to claypans using electromagnetic induction methods. J. Soil Water Conserv. 49:572–575.
• Fielding, A.H., and J.F. Bell. 1997. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ. Conserv. 24:38–49.[CrossRef][Web of Science]
• Foody, G.M. 1992. On the compensation for chance agreement in image classification accuracy assessment. Photogramm. Eng. Remote Sens. 58:1459–1460.
• Gish, T.J., W.P. Dulaney, K.-J.S. Kung, S.C.T. Daughtry, J.A. Doolittle, and P.T. Miller. 2002. Evaluating use of ground-penetrating radar for identifying subsurface flow pathways. Soil Sci. Soc. Am. J. 66:1620–1629.[Abstract/Free Full Text]
• Gomez, A., R.F. Powers, M.J. Singer, and W.R. Horwath. 2002. Soil compaction effects on growth of young ponderosa pine following litter removal in California's Sierra Nevada. Soil Sci. Soc. Am. J. 66:1334–1343.[Abstract/Free Full Text]
• Goovaerts, P. 1997. Geostatistics for natural resources evaluation. Oxford Univ. Press, Oxford, UK.
• Hagen, A. 2003. Fuzzy set approach to assessing similarity of categorical maps. Int. J. Geogr. Inf. Sci. 17:235–249.[CrossRef]
• Hagen-Zanker, A., B. Straatman, and I. Uljee. 2005. Further development of a fuzzy set map comparison approach. Int. J. Geogr. Inf. Sci. 19:769–785.[CrossRef]
• Inman, D.J., R.S. Freeland, J.T. Ammons, and R.E. Yoder. 2002. Soil investigations using electromagnetic induction and ground-penetrating radar in southwest Tennessee. Soil Sci. Soc. Am. J. 66:206–211.[Abstract/Free Full Text]
• Kung, K.-J.S., and S.V. Donohue. 1991. Improved solute-sampling protocol in a sandy vadose zone using ground-penetrating radar. Soil Sci. Soc. Am. J. 55:1543–1545.[Web of Science]
• Landis, J.R., and G.G. Koch. 1977. The measurement of observer agreement for categorical data. Biometrics 33:159–174.[CrossRef][Web of Science][Medline]
• Lark, R.M., and R.B. Ferguson. 2004. Mapping risk of soil nutrient deficiency or excess by disjunctive and indicator kriging. Geoderma 118:39–53.[CrossRef][Web of Science]
• Manel, S., H.C. Williams, and S.J. Ormerod. 2001. Evaluating presence–absence models in ecology: The need to account for prevalence. J. Appl. Ecol. 38:921–931.[CrossRef]
• McBride, R.A., A.M. Gordon, and S.C. Shrive. 1990. Estimating forest soil quality from terrain measurements of apparent electrical conductivity. Soil Sci. Soc. Am. J. 54:290–293.[Web of Science]
• McNeill, J.D. 1980. Electromagnetic terrain conductivity measurement at low induction numbers. Tech. Note TN-6. Geonics Ltd., Mississauga, ON, Canada.
• Mueller, T.G., N.J. Hartsock, T.S. Stombaugh, S.A. Shearer, P.L. Cornelius, and R.I. Barnhisel. 2003. Soil electrical conductivity map variability in limestone soils overlain by loess. Agron. J. 95:469–507.
• Odeh, I.A.O., A.B. McBratney, and D.J. Chittleborough. 1990. Design of optimal sampling spacings for mapping soil using fuzzy-k-means and regionalized variable theory. Geoderma 47:93–122.[CrossRef][Web of Science]
• Ogg, C.M., W.J. Edmonds, and J.C. Baker. 2000. Statistical verification of soil discontinuities in Virginia. Soil Sci. 165:170–183.[CrossRef]
• Pontius, R.G., Jr. 2000. Quantification error versus location error in comparison of categorical maps. Photogramm. Eng. Remote Sens. 66:1011–1013.
• Pontius, R.G., Jr., and L.C. Schneider. 2001. Land-cover change model validation by an ROC method for the Ipswich watershed, Massachusetts, USA. Agric. Ecosyst. Environ. 85:239–248.[CrossRef]
• Roubens, M. 1982. Fuzzy clustering algorithms and their cluster validity. Eur. J. Oper. Res. 10:294–301.[CrossRef][Web of Science]
• Samouëlian, A., I. Cousin, A. Tabbagh, A. Bruand, and G. Richard. 2005. Electrical resistivity survey in soil science: A review. Soil Tillage Res. 83:173–193.[CrossRef]
• Schoenholtz, S.H., H. Van Miegroet, and J.A. Burger. 2000. A review of chemical and physical properties as indicators of forest soil quality: Challenges and properties. For. Ecol. Manage. 138:335–356.[CrossRef]
• Soil Survey Staff. 2003. Keys to Soil Taxonomy. 9th ed. NRCS, Washington, DC.
• Swets, J.A. 1988. Measuring the accuracy of diagnostic systems. Science 240:1285–1293.[Abstract/Free Full Text]
• Triantafilis, J., and S.M. Lesch. 2005. Mapping clay content using electromagnetic induction techniques. Comput. Electron. Agric. 46:203–237.[CrossRef]
• Usoltsev, V.A., and J.K. Vanclay. 1995. Stand biomass dynamics of pine plantations and natural forests on dry steppe in Kazakhstan. Scand. J. For. Res. 10:305–312.
• Van Meirvenne, M., and P. Goovaerts. 2001. Evaluating the probability of exceeding a site-specific soil cadmium contamination threshold. Geoderma 102:75–100.[CrossRef][Web of Science]
• van Rijsbergen, C.J. 1979. Information retrieval. Butterworths, London.
• Visser, H., and T. de Nijs. 2006. The map comparison kit. Environ. Modell. Softw. 21:346–358.[CrossRef]
• Vitharana, U.W.A., M. Van Meirvenne, L. Cockx, and J. Bourgeois. 2006. Identifying potential management zones in a layered soil using several sources of ancillary information. Soil Use Manage. 22:405–413.
• Woolery, M., K. Olson, J. Dawson, and G. Bolero. 2002. Using soil properties to predict forest productivity in southern Illinois. J. Soil Water Conserv. 57:37–45.

This article has been cited by other articles:

				L. Cockx, M. Van Meirvenne, U. W. A. Vitharana, L. P. C. Verbeke, D. Simpson, T. Saey, and F. M. B. Van Coillie Extracting Topsoil Information from EM38DD Sensor Data using a Neural Network Approach Soil Sci. Soc. Am. J., October 21, 2009; 73(6): 2051 - 2058. [Abstract] [Full Text] [PDF]

				T. Saey, D. Simpson, H. Vermeersch, L. Cockx, and M. Van Meirvenne Comparing the EM38DD and DUALEM-21S Sensors for Depth-to-Clay Mapping Soil Sci. Soc. Am. J., January 21, 2009; 73(1): 7 - 12. [Abstract] [Full Text] [PDF]

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Cockx, L. Articles by De Vos, B. Search for Related Content PubMed Articles by Cockx, L. Articles by De Vos, B. GeoRef GeoRef Citation Agricola Articles by Cockx, L. Articles by De Vos, B. Related Collections Electromagnetic Induction, EMI Spatial Variability Forest Soils