SSSAJ Grow Your Career with SSSA
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online 28 September 2007
Published in Soil Sci Soc Am J 71:1740-1747 (2007)
DOI: 10.2136/sssaj2006.0177
© 2007 Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Weller, U.
Right arrow Articles by Wehrhan, M.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Weller, U.
Right arrow Articles by Wehrhan, M.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Weller, U.
Right arrow Articles by Wehrhan, M.
Related Collections
Right arrow Geostatistics
Right arrow Soil Methods/Instrumentation

SOIL & WATER MANAGEMENT & CONSERVATION

Mapping Clay Content across Boundaries at the Landscape Scale with Electromagnetic Induction

U. Wellera,*, M. Zipprichb, M. Sommerb, W. Zu Castellc and M. Wehrhand

a Helmholtz Centre for Environ. Research-UFZ, Dep. of Soil Physics, Theodor Lieser Str. 4, 06120 Halle (Saale), Germany
b ZALF-Leibniz Centre for Agric. Landscape Res., Institute of Soil Landscape Research, Eberswalder Str. 84, 15374 Müncheberg, Germany and Univ. of Potsdam, Institute of Geoecology, PO Box 601553, D-14415 Potsdam, Germany
c GSF-National Res. Centre for Environment and Health, Institute of Biomathematics and Biometry, Ingolstädter Landstrasse 1, 85764 Neuherberg, Germany
d ZALF-Leibniz Centre for Agric. Landscape Res., Institute of Soil Landscape Research, Eberswalder Str. 84, 15374 Müncheberg, Germany

* Corresponding author (ulrich.weller{at}ufz.de).


   ABSTRACT
TOP
 NOTES
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Detailed information on soil textural heterogeneity is essential for land management and conservation. It is well known that in individual fields, measurement of the soil's apparent electrical conductivity (ECa) offers an opportunity to map the clay content of soils with free drainage under a humid climate. At the catchment scale, however, units of different land management and differing sampling dates add variation to ECa and constrain the mapping across field boundaries. We analyzed their influence and compared three approaches for applying electromagnetic induction (EMv) to clay-content mapping at the landscape scale across the boundaries of individual fields and different sampling dates. In the study region, a separate calibration of the relation between clay and ECa for each field and sampling date (fieldwise calibration) yielded satisfactory clay-content predictions only if the costly precondition of sufficient calibration points for each field was fulfilled. We propose a method (nearest-neighbors ECa correction) for unifying ECa across boundaries based only on the ECa data themselves, and the assumption of continuity of textural properties at field boundaries, which was fulfilled in the landscape studied. Prediction is calibrated once for the entire landscape, which allows a reduced set of calibration points. The coefficient of determination for predicting clay content (here, including silt <4 µm) was improved from R2 = 0.66 (no correction for land use and sampling date) to R2 = 0.85 (n = 46). With the method developed, ECa offers a powerful and cheap method of clay-content mapping in agricultural landscapes.

Abbreviations: Corg, organic carbon content • ECa, apparent electrical conductivity • EMv, electromagnetic induction


   INTRODUCTION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Until recently, soil mapping has been based mostly on point information from soil profiles. Compared with the spatial variation of soil, sampling has been sparse and interpolation has had to be based either on expert knowledge or on geostatistics. This approach often does not provide information on soil heterogeneity as precisely and as cost effectively as needed, for example, for catchment management (Lathrop et al., 2000; Zhu and Mackay, 2001), land use planning, or precision farming (Moran et al., 1997; Stafford, 2000). This motivated the testing of several noninvasive methods that promise to overcome these problems associated with high-resolution mapping of soil properties (McBratney et al., 2003; Sommer et al., 2003).

Measurement of the soil's ECa offers a fast and fairly cheap way to obtain dense soil information (McNeill, 1992). The magnitude of the ECa signal is determined by several soil properties, including clay content and cation exchange capacity, as well as soil moisture, temperature, and salinity (Corwin and Lesch, 2005a; Durlesser, 1999; Rhoades et al., 1999; Auerswald et al., 2001).

The spatial variation in the ECa signal is controlled by clay content and mineralogy for soils in a humid climate, if they contain negligible amounts of salts and are not influenced by groundwater (Lesch et al., 2005; Durlesser, 1999; Auerswald et al., 2001). This is due to the electrical double layer on the surfaces of clay minerals that dominates the soil's electrical conductivity (Rhoades et al., 1976; Auerswald et al., 2001).

Several researchers have described the use of ECa for estimating the clay content of soil; Corwin and Lesch (2005a) gave a compilation of several studies. In individual fields, mapping clay content with ECa gave promising results (Durlesser, 1999; Dalgaard and Have, 2001). In this setting, the spatial variation of other factors influencing ECa—such as ionic composition of soil solutes, topsoil structure, bulk density, and organic carbon content (Corg)—is either reduced by the homogeneous cultivation inside the field boundaries or correlated to clay content.

Upon crossing the boundaries between individual fields, the application of ECa to map clay content is constrained. Each unit is characterized by unique land use (history) and agricultural management, both of which affect soil properties. At the landscape scale, this increases the variance of the ECa signal considerably. At boundaries between meadows and arable land, discontinuities in soil structure, water content at field capacity, and ionic composition are expected to be even larger. Furthermore, different measurement dates for different landscape units are likely to occur in large and diversely cultivated areas, adding variation to ECa (Durlesser, 1999; Sudduth et al., 2001). Some researchers have found that the ECa can be used for guiding further prospection but that there is no direct connection to specific soil properties on the regional level (Carroll and Oliver, 2005).

In this study, we investigated the mapping of clay given a spatial and temporal mosaic of several ECa measurements in an agricultural landscape. The methods applied were based only on the measurements themselves and on textural data at calibration points. All approaches were evaluated with respect to their accuracy of clay-content prediction at the landscape scale.


   MATERIALS AND METHODS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Research Area
The region investigated, the Klostergut Scheyern of the joint research project FAM (Schröder et al., 2002), is a farm in Bavaria at 11°26' E, 48°29.5' N. The soils are formed on Molasse, that is, Tertiary fluvial sediments of varied texture from clay to gravel with abrupt changes, which are partly covered by Pleistocene loess or Holocene colluvial deposits. The region is strongly undulating. There are no consolidated sediments in the area.

The soil temperature regime is mesic, with an udic moisture regime. The soil epipeda are predominantly ochric, and Eutrochrept is the predominant soil taxonomic suborder. Hapludalfs are common on gentle slopes with Pleistocene loess accumulation. Udorthents and Udipsamments occur on eroded hilltops and steeper slopes and are developed on Molasse. Areas with gleyic soils, that is, soils showing aquic conditions due to endosaturation, were not included in the study. They are present in low-lying areas, and can be easily mapped.

The development and comparison of the clay prediction methods took place in eight fields: five fields are in a cereal–potato (Solanum tuberosum L.)–grass rotation (A2, A3, A4, A5, and A7). Two fields, W2 and W3, are meadows, and F5 is fallow land taken out of use 10 yr ago (Fig. 1 ).


Figure 1
View larger version (44K):
[in this window]
[in a new window]

  		Fig. 1. Apparent electrical conductivity (EC) measurement points and fields investigated for method comparison.

 
Apparent Electrical Conductivity Measurement
The ECa was measured using a Geonics EM38 device (McNeill, 1980) in vertical mode. The resulting signal is referred to here as EMv. Location was determined by differential global positioning system with a mean error of <1 m and a maximal error up to 3 m (Ehrl et al., 2002). Measurements were taken by different people at various dates (Table 1 ). Soil moisture was near field capacity on all dates, to minimize the within-field spatial variance of soil water content not related to soil texture. Measurement between the end of October and April also guaranteed relatively small within-field variations of soil temperature and fertilizer residues.


View this table:
[in this window]
[in a new window]

  		Table 1. Correlation between electrical conductivity normalized to 25°C (EC25) and weighted clay content for homogeneous landscape units.

 
The density of observation points varied from <1 to 3 m in the track direction and from 3 to 15 m in between the tracks (Fig. 1). The measurements were made in the vertical mode. For computations based on individual fields only, we normalized raw EMv data to 25°C using the formula derived by Sheets and Hendrickx (1995):

Formula 1[1]
where ECT is the EMv at soil temperature T (°C). Soil temperature was measured at a depth of 50 cm. Temperature correction is not a precondition for the nearest neighbors EMv correction method (see below) and was therefore not applied to this. We used ordinary kriging to interpolate EMv between the tracks and for the prediction at the locations of the analyzed soil profiles. A spherical variogram model with nugget was used. Variogram calculations, fitting of a spherical variogram model, and predictions were accomplished with the geoR package for the statistic software R (Ribeiro and Diggle, 2001), while maps were generated with the aid of the software package ArcGIS 8.1 by ESRI (Redlands, CA).

Soil Calibration Data
Descriptive and analytical soil profile information for calibration was available for 46 points on a 50- by 50-m grid (Sinowski, 1995). Laboratory determinations of texture, organic C, and soil bulk density were made for each horizon. Soil bulk density was measured on 100-cm3 cylinders and corrected for stones >2 cm. It seemed adequate to integrate the fine silt fraction (2–4 µm) into the "clay fraction," since clay minerals form part of the mineral spectrum in this particle-size class (Allen and Hajek, 1989). To correlate the EMv signal with a soil quantity, q, a weighted sum Q was calculated from the density of the measuring signal (z) (McNeill, 1980):

Formula 2[2]
where R(dh) is the proportion of the signal measured below the lower boundary, d, of the hth horizon (d0 = 0, dN = 1.2 m), and qh is the considered soil quantity in the hth horizon. Only those calibration points that are inside the field boundaries and at a distance of <5 m from the nearest EMv measurement point were included in regression calculations. For correlating soil properties, Q, with interpolated EMv measurements, we assume a linear relationship:

Formula 3[3]
where {alpha}0 and {alpha}1 are the axis intercept and the slope, respectively.

Elimination of Land Use and Time Influences on Electromagnetic Induction Measurement
Method a: No Correction
First, we regarded the EC25 measurements, corrected for the influence of soil temperature only, as stable with respect to time and changes in cultivation. The first step in data analysis (Fig. 2a ) was then the spatial union of all different EC25 data sets, measured at different dates and on fields under different cultivation. Subsequently, we approximated EC25 values, at the locations of analyzed soil samples, by ordinary kriging. Finally, we performed regression analysis using the weighted clay contents. This approach provides a reference by which to evaluate the quality of the methods to correct EMv for land use and time.


Figure 2
View larger version (10K):
[in this window]
[in a new window]

  		Fig. 2. Workflow for electromagnetic induction measurement (EMv) to clay mapping: (a) no correction for influence of land use and time; (b) fieldwise calibration; and (c) nearest neighbors EMv correction.

 
Method b: Fieldwise Calibration
Assuming a linear influence of land use and time on ECa means that the coefficients {alpha}0 and {alpha}1 of Eq. [3] depend on land use and sampling date. Denoting each unique combination of management and sampling date (in practice, each field and measurement campaign) by an arbitrary indicator variable, n (n = 1, 2, ..., r), we rewrite Eq. [3] as

Formula 4[4]

The fieldwise calibration method, which pays attention to the influence of land use and sampling date on EMv, treats every EMv data set individually. Regression equations between weighted clay content and EMv were then calculated separately for each realization of the set of all unique space–time combinations, n (Fig. 2b). We then evaluated the quality of prediction at the landscape scale by comparing the predicted and the measured clay contents. We have also tested a simplified version of Eq. [4], where we assumed {alpha}n1 to be fixed to evaluate the type of influence that date and management has on ECa.

Method c: Nearest Neighbors Electromagnetic Induction Measurement Correction
To diminish the influence that different land management and different measurement dates have on the correlation between clay content and EMv, we have developed a two-step approach. In the first step, we tried to eliminate the differences between adjacent fields to form a fitted, more continuous signal that we call EMvadj. This step is based only on the EMv data and its location; no additional information on soil texture is included. In the second step, we then correlated this fitted signal against clay content of the reference data points (Fig. 2c). The first step is based on the assumption of continuity of clay content at field boundaries. Therefore, we assume the existence of a spatially autocorrelated variable, EMvadj, which is linearly correlated to the time- and field-dependent EC(x) values at position x. We can then write

Formula 5[5]
where the measurement at location xnj belongs to data set index n, and xmi to m, and {varepsilon} is a random variable whose variance depends on the sampling distance ||xmi – xnj|| with mean 0. We assumed that {varepsilon} is small if two points are in close proximity. We defined a set of points that are sufficiently close together and belonging to different fields. To minimize the numerical problems, we wanted to keep this data set small; therefore, we selected only those points that are the closest to each other. Thus, given two realizations of the EMv measurements, we have the set of coordinates Xm and Xn with corresponding EMv measurements EC(Xm) and EC(Xn) of Nm and Nn elements, respectively.

We define a set Pmn of EMv measurements at nearest neighbor points as

Formula 6[6]
with d(x, Xm)=mink=1Nm||x–xmk|| being the distance between Point x and the points in Set Xm. In other words, given two points belonging to different data sets (Point 1 to Set m and Point 2 to Set n), these two points are called nearest neighbors if, and only if, Point 1 is the closest point in Set m to Point 2, and Point 2 is the closest point in Set n to Point 1. In this study, we restricted the sets of pairs to those with a maximum separation distance of 15 m. Furthermore, if there were terrace steps between two nearest neighbor points, violating the assumption of spatial autocorrelation, we eliminated these pairs from the set of nearest neighbor points. These terrace steps violate the assumption of continuity of soil properties, and they were easy to detect in the landscape by using the digital elevation model that is available for the study area.

Given the union of all nearest neighbor sets Q={cup}m!=n,n=1rPmn, we have to minimize the sum of residuals from Eq. [5] as given by

Formula 7[7]
We also have the boundary condition that the overall mean should be preserved and the trivial solution of all coefficients being zero should be excluded. This is achieved by adding {Sigma}i=0rβ0i=0 and {Sigma}i=0rβ1i=r to the system of linear equations.

In matrix form, the equation system is given with

Formula 8[8]
with p1i and p2i as the first and second point of the ith pair, the matrix coefficients are given by

Formula 9[9]
The least square fit for the unknown variables Formula 9 was done by a standard iterative method (conjugate gradient, Fletcher and Reeves, 1964), which is implemented in the analytical software package R, Version 1.6.1 (Ribeiro and Diggle, 2001). The direct linear solution was too memory consuming for the larger data fields. The second step is then analogous to Method a, but based on the EMvadj values as defined in Eq. [5].

To analyze the correlation between the prediction error of the nearest neighbors method and the number of calibration points, k, we chose lk subsamples randomly from the n = 46 points with lk = min[1000,( )] and k = 4, 5, ..., n. For each subsample, we determined regression parameters and predicted the clay content for the n points. We then calculated the RMSE for predicted vs. measured weighted clay content and, finally, the median of the RMSE and the 25 and 75% quantiles for each k.


   RESULTS AND DISCUSSION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
CONCLUSIONS
 REFERENCES
 
To apply ECa to map clay content at the landscape scale, we had to verify the validity of two premises: first, we ascertained the relation of EMv to clay content locally; second, we established the continuity of this relation at the landscape scale.

Field Scale
Figure 3a shows the variograms for Fields A2 and F5. The differences in the local variation of the EC signal were high. None of the variograms show a significant nugget effect, thus the measurement error can be neglected. The coefficients of determination between weighted clay content and the interpolated EC25 values are listed separately in Table 1 for each field and sampling date. For the fields investigated, soil clay content controlled the spatial variation of the EMv signal. The influence of other soil properties on EMv was weak (Table 2 ).


Figure 3
View larger version (15K):
[in this window]
[in a new window]

  		Fig. 3. Semivariograms for electromagnetic induction (EMv) measurements: (a) variograms for two fields compared with regional semivariogram after electrical conductivity (EC) correction; and (b) regional semivariogram with and without EC correction.

 

View this table:
[in this window]
[in a new window]

  		Table 2. Correlations between electrical conductivity normalized to 25°C (EC25), weighted clay content, organic C content (Corg), and bulk density for homogeneous land units.

 
For three fields only (A3, A5, and F5), Corg or soil bulk density correlate with EMv. In these cases, there is usually strong interdependence between these variables and clay content. For prediction of clay content, the coefficients of determination, R2, ranged from 0.7 to 0.97 for fields with n ≥ 6 profiles with laboratory clay analysis (Table 1). This represents the upper limit of values reported by other researchers. Dalgaard and Have (2001) found similar coefficients of determination (R2 = 0.79) for soils developed on moraine, while Schmidhalter et al. (2001) reported values of R2 from 0.31 to 0.67. In the fields studied, several other conditions favored the application of EMv for clay prediction. The soil water status was near field capacity, and there was no influence of groundwater with a possibly differing ionic composition.

The F test for different statistical models showed a clear influence of the date and field of measurement. The performance of a model with a fixed slope but different intersects for the relation EMv vs. clay content performs only slightly better than the model with no correction (F = 1.5, P = 0.19). The influence of an adjusted slope for each date and field combination has a significant better correlation (F = 3.6, P = 0.005). The residual sums of squares were 0.064 for no correction, 0.052 for different intersects, and 0.024 for different intersects and slopes.

Influence of Time and Land Use
To unify multiple EMv measurements for clay prediction at the landscape scale, we analyzed the influence of management practices and time of measurement on EMv.

First, we analyzed the reproducibility of measurements made within 1 wk, but differing by track direction (Fig. 4 ). Regions where absolute and relative differences of interpolated EC25 exceeded 3 mS m–1 and 10%, respectively, were concentrated (Fig. 4b) between distant tracks as well as where there was a strong gradient in EC25, where small positioning errors have the greatest impact. The coefficient of determination between the two measurements with different track directions is R2 = 0.96, based on points within <1-m distance of the set in the other track direction (n = 163). Resulting EC25 patterns (Fig. 4a) are basically independent of track direction and can be reproduced by repeated measurement. The observed differences (RMSE = 2.6 mS m–1) are within the range reported by Sudduth et al. (2001) for short-term fluctuations in EMv signal due to drift of the EM38. For multiple EC25 measurements, made within a few days of each other on another field of Klostergut Scheyern, we do find a similarly good fit: R2 = 0.95, n = 185.


Figure 4
View larger version (67K):
[in this window]
[in a new window]

  		Fig. 4. Comparability of multiple electromagnetic induction (EMv) measurements: I. Reproducibility of electrical conductivity (EC) measurements.

 
Second, we compared measurements of the same field made at different times of the year and under different conditions in terms of land use. For all dates, the soil was near field capacity. Figure 5 compares two sampling dates in a meadow, and shows that patterns of EC25 are basically independent of time, but the EC25 values and their ranges differ significantly. Cultivation did not alter EC25 patterns either (Fig. 6 ). Sudduth et al. (2001) also found a strong linear correlation between multiple EMv measurements made on the same field in different years.


Figure 5
View larger version (51K):
[in this window]
[in a new window]

  		Fig. 5. Comparability of multiple electromagnetic induction (EMv) measurements: II. Influence of time of measurement on electrical conductivity (EC) measurements.

 

Figure 6
View larger version (21K):
[in this window]
[in a new window]

  		Fig. 6. Comparability of multiple electromagnetic induction (EMv) measurements: III. Influence of time and cultivation on electrical conductivity measurements normalized to 25°C (EC25). Each symbol represents one spatial point and its EC25 signal at Date 1 (x axis) vs. the measurement at Date 2 (y axis).

 
Correlation turns out to be weaker, though, if ECa values measured in summer are compared with spring measurements (Durlesser, 1999). During summer, the influence of spatially variant water tension and solute concentrations on ECa imply that ECa cannot always be optimally used for mapping soil clay content.

In summary, ECa is a function of time of measurement, management practice, land use, and soil surface topography (Corwin and Lesch, 2005b). The multiple measurements are linearly correlated with each other and most of the spatial variation of ECa is explained by soil clay content. Given the continuity of soil textural properties at field boundaries, this implies that the relative ECa patterns are constant across spatial and temporal boundaries at a first approximation.

Mapping at Landscape Scale
Method a: No Correction
To map clay content at the landscape scale, we used the entire multispatial and multitemporal set of EC25 data. Let us first disregard the influence of land use and sampling date (Fig. 2a). The correlation between interpolated EC25 and clay content then yields a lower coefficient of determination (R2 = 0.66; Fig. 7a ; Table 3 , Regression 1) compared with the single fields (Table 1).


Figure 7
View larger version (11K):
[in this window]
[in a new window]

  		Fig. 7. Comparison of methods to correct electromagnetic induction (EMv) measurement of electrical conductivity normalized to 25°C (EC25) for influences of time and cultivation with regard to prediction of soil clay content.

 

View this table:
[in this window]
[in a new window]

  		Table 3. Comparison of methods to correct electromagnetic induction (EMv) field values of electrical conductivity normalized to 25 °C (EC25) for influences of time and cultivation with regard to prediction of clay content (n = 46).

 
Different levels and ranges of EC25 measured on fields of differing land use and at different dates are shown by marked discontinuities of the predicted clay contents at field boundaries (Fig. 8a ). Thus, multiple ECa measurements cannot be evaluated jointly in terms of clay content without taking land management and time into account. Other investigators who looked at a set of fields, and who did not take into account the influence of different land use and management, also found fairly weak correlations between clay content and ECa. For measurements conducted on fields of the same research farm Klostergut Scheyern, Schmidhalter et al. (2001) and Durlesser (1999) reported coefficients of determination for the relation between clay content and ECa of R2 = 0.45 (here, clay + fine silt), and R2 = 0.51, respectively. The procedure (no correction) is similar to kriging with regression (Knotters et al., 1995).


Figure 8
View larger version (72K):
[in this window]
[in a new window]

  		Fig. 8. Discontinuities of soil clay content predicted by measurements of electrical conductivity (EC) on the boundaries between landscape units.

 
Method b: Fieldwise Calibration
One way of accounting for the influence of time and cultivation on ECa is to calibrate the relationship between EMv and clay for each data set separately (Fig. 2b). This gives an apparently strong prediction of clay content with R2 = 0.90 (Fig. 7b) and a RMSE of <4% clay content (Table 3, Regression 2). The correlation is, however, highly unstable, as in our example calibration points are sparse for some landscape units (Table 1). To achieve robust results, this method requires sufficient soil profiles for calibration in each individual field. This is a serious problem for mapping soil clay content at the catchment scale, where many, often small, land units are to be recognized. Obtaining calibration data is time consuming and expensive. In addition, artificial discontinuities of predicted clay content at field boundaries persist even for a relatively large number (n ≥ 6) of calibration points per field (Fig. 8b, Table 1).

The fieldwise calibration method represents a combination of kriging with regression and kriging with stratification (Stein et al., 1988). The stratification variables in this case are the fields and the sampling date of the data set.

Method c: Nearest Neighbors Correction
This method (Fig. 2c) takes into account spatial soil relationships and thereby reduces the variance of EMv by >50% (Fig. 3b). Correction of the influence of land use and time lowered the sill of the variogram from 265 mS2 m–2 for EC25 (no correction) to 125 mS2 m–2 for EMvadj. For the prediction of weighted clay content with EMvadj, a coefficient of determination of R2 = 0.85 was achieved with the nearest neighbors EMv correction method (Table 3, Regression 3; Fig. 7c). In contrast to the fieldwise calibration method, the correlation is more robust. The two parameters per field (for intersect and slope correction) are estimated from the large data set of EMv measurement point pairs, and the final correlation only costs two degrees of freedom.

The regression functions between EMv and clay content are built only once for all landscape units and sampling dates together. This is essential for the application of the method in the framework of mapping soil clay content at catchment and landscape scales. Compared with the fieldwise method, a significantly smaller set of calibration points is sufficient to achieve reliable prediction. Reducing the number of calibration points to eight, that is, only one profile per field, increases the median RMSE by <0.5% clay (Fig. 9 ).


Figure 9
View larger version (17K):
[in this window]
[in a new window]

  		Fig. 9. Dependence of RMSE for clay content prediction by measurements of electrical conductivity (EC) on the number of calibration points.

 
Besides improving clay prediction at point support, our method enables us to map patterns of clay content across the boundaries of individual fields (Fig. 8c). Furthermore, additional information in the form of multiple measurements of one field is incorporated into the optimization process. Other soil properties explaining ECa may easily be incorporated into the framework of the method, provided they are invariant to time and land use. Finally, the method implies no need to perform temperature correction of the ECa value explicitly: first, the formula for temperature correction strictly speaking demands local validation; and second, determination of the effective soil temperature of the measurement volume is tedious.

We need to stress, however, that the nearest neighbors EMv correction method is subject to two preconditions. First, a significant number of measurements should be placed near field boundaries, since the distance between neighboring EMv measurements should be small compared with the spatial range of correlation.

The second, more fundamental precondition is that there should be no sharp discontinuities in soil texture at the field boundaries considered. This assumption is fulfilled in our case: the parent materials in the Bavarian Molasse region mostly are highly variable fluvial sediments. On the one hand, this signifies that natural discontinuities in soil texture are unlikely to be linear, in contrast to, for example, a cuesta landscape. On the other hand, the size of single areas of the pattern is mostly small compared with the length of field boundaries. In its current form, the method uses an ordinary least squares fit. This could be generalized to take into account the distance between the data pairs. One way to do this would be to weight each equation with the spatial variance for the given point distance. Another way would be the use of a more general system to predict one data point from the other data set and to use these predictions in the linear equation system. We chose the nearest neighbor method due to the computational simplicity and as it proved to be sufficient with the given large number of data pairs.

The nearest neighbors EMv correction was applied to the entire farm Klostergut Scheyern to three contiguously measured subareas. Figure 10 shows the map of predicted clay content and its pattern.


Figure 10
View larger version (69K):
[in this window]
[in a new window]

  		Fig. 10. Map of estimated clay content of the Klostergut Scheyern using the nearest neighbors electromagnetic induction measurement (EMv) correction method.

 

   CONCLUSIONS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
REFERENCES
 
With our nearest neighbors EMv correction method, we can extend the application of ECa for clay-content mapping from the field to the landscape. This is possible without analyzing the signal variation due to land use, management practices, and sampling date in detail. The method developed adjusts for the unwanted influence of measurement conditions based merely on the signal, the spatial arrangement, and the stratification of the data. This nearest neighbors EMv correction provided the best results for the detection of continuous soil structures across field boundaries. In contrast, separately predicting clay content for each field and sampling date is not applicable to heterogeneous agricultural landscapes with numerous small fields, because a vast number of calibration points is necessary to attain unbiased results.

The easy measurement and simple calibration promotes the combination of the ECa technique and the nearest neighbors EMv correction as a standard tool in spatial soil investigation. The nearest neighbors EMv correction method can be easily applied to similar problems associated with mapping at the landscape scale, where land-use boundaries restrict the pedological interpretation of indirect techniques. It can be used as an independent preprocessing step for other methods, e.g., guided sampling (Lesch et al., 1995).


   ACKNOWLEDGMENTS
 
We thank U. Schmidhalter and H. Stanjek of the Technical University of Munich-Weihenstephan (TUM) for kindly making available the EM38 measuring device. We also thank K. Heil and E. Neudecker for joint EMi measurements of the fields A3, A4, A7, W2, and F5, and H.-P. Durlesser and C. Sperl for making their EMi data available. We wish to also thank M. Stephan for his assistance with field measurements during severe weather. This work was performed under the Research Network Agroecosystems Munich (FAM) and financed by the BMBF.


   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 May 3, 2006.


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





This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Weller, U.
Right arrow Articles by Wehrhan, M.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Weller, U.
Right arrow Articles by Wehrhan, M.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Weller, U.
Right arrow Articles by Wehrhan, M.
Related Collections
Right arrow Geostatistics
Right arrow Soil Methods/Instrumentation

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Crop Science
Journal of Natural Resources
and Life Sciences Education		 Vadose Zone Journal
Journal of Plant Registrations Journal of
Environmental Quality		 The Plant Genome