Precision Agric (2009) 10:490–507 DOI 10.1007/s11119-008-9103-z
Interpretation of electrical conductivity patterns by soil properties and geological maps for precision agriculture Ju¨rgen Ku¨hn Æ Alexander Brenning Æ Marc Wehrhan Æ Sylvia Koszinski Æ Michael Sommer
Published online: 21 December 2008 Ó Springer Science+Business Media, LLC 2008
Abstract Precision farming needs management rules to apply spatially differentiated treatments in agricultural fields. Digital soil mapping (DSM) tools, for example apparent soil electrical conductivity, corrected to 25°C (EC25), and digital elevation models, try to explain the spatial variation in soil type, soil properties (e.g. clay content), site and crop that are determined by landscape characteristics such as terrain, geology and geomorphology. We examined the use of EC25 maps to delineate management zones, and identified the main factors affecting the spatial pattern of EC25 at the regional scale in a study area in eastern Germany. Data of different types were compared: EC25 maps for 11 fields, soil properties measured in the laboratory, terrain attributes, geological maps and the description of 75 soil profiles. We identified the factors that influence EC25 in the presence of spatial autocorrelation and field-specific random effects with spatial linear mixed-effects models. The variation in EC25 could be explained to a large degree (R2 of up to 61%). Primarily, soil organic matter and CaCO3, and secondarily clay and the presence of gleyic horizons were significantly related to EC25. Terrain attributes, however, had no significant effect on EC25. The geological map unit showed a significant relationship to EC25, and it was possible to determine the most important soil properties affecting EC25 by interpreting the geological maps. Including information on geology in precision agriculture could improve understanding of EC25 maps. The EC25 maps of fields should not be assumed to
J. Ku¨hn M. Wehrhan S. Koszinski M. Sommer Leibniz-Centre for Agricultural Landscape Research (ZALF), Institute of Soil Landscape Research, Eberswalder Str. 84, 15374 Mu¨ncheberg, Germany J. Ku¨hn (&) ¨ berlinger Str. 39, 78628 Rottweil, Germany U e-mail:
[email protected] A. Brenning Department of Geography and Environmental Management, University of Waterloo, Waterloo, ON N2L 3G1, Canada M. Sommer Institute of Geoecology, University of Potsdam, Karl-Liebknecht-Str. 24-25, 14476 Potsdam, Germany
123
Precision Agric (2009) 10:490–507
491
represent a map of clay content to form a basis for deriving management zones because other factors appeared to have a more important effect on EC25. Keywords Soil apparent electrical conductivity Geological maps Digital terrain analysis Soil organic matter Soil carbonates Soil texture Kriging
Introduction Precision agriculture requires the delineation of site-specific management zones to describe the spatial variation of soil and soil properties at the field scale (McBratney et al. 2005). This enables cost-effective and environmentally sound use of resources such as soil, water, nutrients, pesticides and seeds (Carroll and Oliver 2005). Non-invasive methods of digital soil mapping (DSM) are being used increasingly in precision agriculture (Lagacherie et al. 2007). McBratney et al. (2000) give an overview of DSM techniques. Mapping the apparent electrical conductivity (ECa) of soil is a common DSM tool to define management zones (Sommer et al. 2003; McBratney et al. 2005; Carroll and Oliver 2005; Vitharana et al. 2008) as it is inexpensive and rapid. Spatial heterogeneity in the crop canopy can be used as indicators for soil properties if the climatic environment and local relief position are known (Auerswald et al. 1997; Maidl et al. 1999). The potential of yield maps has been examined intensively to indicate soil properties (Basso et al. 2001) because yield integrates all site effects (Moran et al. 1997). Consequently, it is difficult to infer single soil properties from yield maps, and their use is limited due to interannual variation of weather conditions, management measures and the occurrence of diseases (Vitharana et al. 2008). The leaf area index (LAI) of winter cereals, derived by remote sensing, might be a better indicator of soil properties than yield maps, because it is independent of the way the harvester is driven, has a greater spatial resolution of up to 1 m and does not require interpolation (Sommer et al. 2003). Digital elevation models (DEM) are used to characterize sites and to determine soil properties, e.g. soil moisture, nutrient supply, soil organic matter (SOM) content or thickness of the A horizon. Attempts to relate soil types and their spatial distribution to terrain attributes are widely used (e.g. Lagacherie and Voltz 2000; McBratney et al. 2003; Behrens and Scholten 2006), but the predictive accuracy of these models is limited. We analyzed the spatial pattern of soil electrical conductivity maps to determine relationships with soil properties. As ECa depends on soil temperature, a correction to 25°C (EC25) is necessary (Durlesser 1999; Domsch and Giebel 2004). Maps of EC25 have to be interpreted to derive management measures, as does all DSM-derived spatial information, because they contain only indirect information about soil properties (Lagacherie and McBratney 2007). Bronson et al. (2005) suggested that the interpretation of ECa data can be improved by using additional spatial information. Pedotransfer functions are frequently used to transform DSM measurements into single soil property maps, e.g. ECa values into clay content (e.g. Dalgaard and Have 2001; Sommer et al. 2003; Domsch and Giebel 2004; McBratney et al. 2005; Weller et al. 2007) or the depth of clay lenses (Cockx et al. 2006). Before drawing any conclusions from DSM-derived field maps (e.g. ECa maps) it is important to know which factors might affect the measured signal, and how they are distributed spatially in reality within the fields examined. McNeill (1980a, 1992) and Sudduth et al. (2001) mention the following factors that affect soil ECa: soil moisture, pore size and distribution, salt content, soil texture (especially clay content), cation exchange capacity (CEC), clay mineral
123
492
Precision Agric (2009) 10:490–507
composition, pore geometry and tortuosity, the concentration of dissolved electrolytes in soil water, amount and composition of colloids, temperature of the soil water and the content of electrically conductive rare soil minerals. Sudduth et al. (2001) also include soil temperature and bulk density, and McBratney et al. (2005) suggest that soil depth is another factor. The EC25 may vary not only because of soil texture differences, but also because of differences in soil organic matter and carbonate content (Domsch and Giebel 2004); this is the case in non-saline soil in central Europe in spring when soil is at field capacity and vertical and horizontal soil matric potential differences are negligible. A deeper ground water table possibly increases EC25 as the measurement depth of the EM38 reaches a maximum of 4 m (McNeill 1980b). Dissolved carbonates affect the composition of dissolved electrolytes in the soil water and these might be transported to the surface by ground water (Sommer and Schlichting 1997) and precipitate. The EC25 is usually mapped field by field because of differences in cultivation. Timedependent effects related to cultivation and weather might introduce systematic differences in EC25 measurements at the field boundaries even though the soil properties such as soil texture are continuous in space. To eliminate this effect, algorithms have been developed to adjust adjacent EC25 maps to obtain regional EC25 maps (Weller et al. 2007; Brenning et al. 2008). The effect of soil properties on EC25 might be related to and even dominated by relief and geology. The soil at all locations is affected by the mobilization and transport of solutes, colloids and sediments from the surrounding area. Transport processes are controlled by terrain and geology at the landscape scale, not at the field scale (Sommer 2006). Topography affects many of the properties that affect EC25. A better understanding of the factors that determine EC25 and their interrelations is required to improve predictions of the spatial distribution of relevant soil properties based on EC25 values. This will improve the use of EC25 maps to define management zones in precision agriculture. Our hypothesis was that terrain, geology and geomorphology, as well as the soil can, in general, explain the spatial distribution of EC25. To test this we performed an exploratory data analysis with a large variety of data to explain EC25. We compared the effects of soil properties measured by laboratory analysis (e.g. soil texture), of terrain attributes (e.g. flow accumulation), of geology and of soil factors, such as ground water, on EC25. Linear mixed-effects models with residual spatial autocorrelation and stepwise variable selection were applied to identify variables that control EC25, an important property for representing soil heterogeneity in the landscape.
Materials and methods Landscape characteristics of the research area The research area is around Wulfen near Ko¨then in Saxony-Anhalt, eastern Germany (51°520 3100 –51°450 0500 N, 11°510 3200 –11°560 3700 E) and consists of fields on the farm WIMEX, Baasdorf. Plains and fluvial terraces are the dominant geomorphological units in the landscape. The main parent materials of the soil on the higher terraces are Pleistocene glacial and periglacial sediments of the Weichsel and Saale glaciations, deposited on top of Mesozoic to Cainozoic sedimentary rocks (Fig. 1, Table 1). The soil parent material consists of glacial till, fluvial deposits (mainly sands), loess and clay stones. Towards the north, the study area extends into the wide valley of the Elbe River with Holocene and Weichselian alluvial sediments.
123
Precision Agric (2009) 10:490–507
493
Fig. 1 Research area near Wulfen, Germany: a geological map 1:25,000 (Ko¨niglich Preußische Geologische Landesanstalt 1913a, b; modified); b elevation (DEM 2 m) with point EC25 field maps; location of the 75 soil profiles on the 11 experimental fields is indicated in both maps; Units of the geological map: blue colors (a, b, c): Holocene calcareous sands, high ground water table; brown colors (d, e, f): Holocene carbonate free sands, high ground water table; yellow and orange colors (g, h, m): Holocene or Weichsel dry sands; green colors (i, j, k, l): Weichsel loess above different underlying materials; red color (n): Saale glacial till, loam; violet and purple colors (o, p): Oligocene clay stone; for a detailed explanation of units see in Table 1
Chernozems are the dominant soil type in the region of Wulfen. Other soil types include Gleysols, Cambisols and Stagnosols, using the soil classification World Reference Base (IUSS Working Group WRB 2006). Climate is temperate and semi-humid with precipitation in all months. The mean annual temperature is about 8.5°C in Ko¨then, the mean precipitation is 507 mm and drought is common in summer (Walter and Lieth 1964; Deutscher Wetterdienst 1999). Experimental setup The EC25 was measured in 11 agricultural fields (ranging in size from 24 to 68 ha, mean: 43 ha). To improve landscape knowledge a high-resolution LIDAR (light detection and ranging) DEM was obtained, and geological maps at the scale of 1:25,000 (Ko¨niglich Preußische Geologische Landesanstalt 1913a, b) were digitized (Fig. 1a). Soil maps at the same scale were not available, in spite of their importance for predicting soil properties in
123
494
Precision Agric (2009) 10:490–507
Table 1 Units of the geological map 1:25,000 (Ko¨niglich Preußische Geologische Landesanstalt 1913a, b) at the 75 soil profiles; 16 single units a–p were grouped by geological similarity to the variable ‘Geological unit’ with 6 units A–F Unit
Description
N
a
Calcareous-sandy organic layer, above calcareous sand (Holocene); high ground water table
4
b
Calcareous-sandy organic layer, above calcareous sand (Holocene), above glacial till (loam, marl; Weichsel); high ground water table
1
c
Calcareous-sandy organic layer, above a poorly permeable calcareous layer (meadow chalk), above calcareous sand (Holocene); high ground water table
2
d
Sandy organic layer, above sand (Holocene); high ground water table
1
e
Sandy organic layer, above sand (Holocene), above impermeable clay stone (Oligocene); high ground water table
1
f
Sandy organic layer, above impermeable clay, above sand (Holocene); high ground water table
5
g
Sand with low organic matter content, partially gravelly, above sand (Weichsel); dry
11
h
Organic layer, above sand with low organic matter content, sometime gravelly, above sand (Weichsel); dry
1
i
Organic layer, above loess, above sand (Weichsel)
8
j
Loess, above sand, above low permeable loam or marl (Weichsel)
2
k
Loess, above low permeable loam or marl (Weichsel)
8
l
Loess (Weichsel), above impermeable clay stone (Rotliegendes)
1
m
Organic sandy layer, above sand (Saale); dry
n
Loamy sand to loam, derived from low permeable glacial till (loam or marl; Saale)
6
o
Sand (Saale), above impermeable clay stone (Oligocene)
2
p
Calcareous clay, derived from calcareous clay stone (Oligocene)
3
A
Units a, b, c: calcareous Holocene sands, high ground water table
7
B
Units d, e, f: carbonate free Holocene sands, high ground water table
C
Units g, h, m: dry sands from Holocene or Weichsel
31
D
Units i, j, k, l: loess from Weichsel, above different underground
19
E
Unit n: glacial till from Saale, loam
6
F
Units o, p: clay stone from Oligocene
5
19
7
N = Number of soil profiles in the unit
the DSM (Walter et al. 2007). With these three sources of spatial information we selected locations for the soil profiles in the 11 fields to describe the variation in the properties mentioned above within the fields and throughout the landscape. Soil profile sites were placed in the centre of relatively homogeneous areas, which were characteristic and representative for the range of EC25, terrain elements and geology in the research area. For each location the values of EC25 and terrain attributes were computed, geological information was taken from the geological maps, and the soil profiles were described and named. Each soil horizon was sampled and the properties were determined subsequently by laboratory analysis. The effect of organic layers with a high SOM content up to half-bog horizons, of secondary carbonates and of ground water, expressed by the presence of gleyic horizons, was recorded for each soil profile. The unit of the geological map was compared to the geological parent material from the soil profile description to control the quality of the medium-scale maps.
123
Precision Agric (2009) 10:490–507
495
Soil sampling and laboratory procedures Soil samples were taken from 75 soil profiles in the 11 fields with a hydraulic auger of 2 m length and inner diameter of 8 cm. Depths of 1–2 m were reached by augering, depending on soil properties, soil genesis, geology and the current soil moisture. Bulk soil samples were taken for each soil horizon to determine gravel (wt%), organic (Corg, wt%) and carbonate carbon content (Ccarb, wt%) and soil texture of the fine earth fraction by the pipette-sieve method (pre-treatment involved removal of organic matter and carbonates; the methods are described in Schlichting et al. 1995). Organic C was multiplied by 2 to obtain the SOM (Schlichting et al. 1995), and Ccarb was multiplied by 8.33 to obtain the CaCO3 content. All carbonates were calcite. Core samples of 100 cm3 volume were taken for each soil horizon, except where the gravel content was too large. The core samples were dried at 105°C and weighed to measure soil bulk density. The electrical conductivity in water extract (ECWE; soil:water 1:5) was measured for soil samples from 24 soil profiles on six of the fields to detect salinity or gypsum (Normenausschuss Wasserwesen (NAW) im DIN Deutsches Institut fu¨r Normung e.V. 2002). Soil electrical conductivity measurements The apparent electrical conductivity of soil (ECa) was measured with the EM38 sensor (Geonics Ltd., Mississauga, Ontario, Canada) in the vertical mode in the 11 fields (Fig. 1). The sensor was pulled on a sledge behind an all-terrain vehicle about 5 cm above ground surface at a speed of 2–3 m s-1 along parallel tracks 10–30 m apart in agricultural fields with bare soil. Measurements were made in spring when the soil was near field capacity. The raw ECa measurements were corrected to 25°C (EC25) using the method of Durlesser (1999). Point EC25 measurements of all fields were interpolated by ordinary kriging to a region-wide grid of 5 9 5 m with the appropriate variogram model for each field. Regional digital elevation model A regional-scale airborne high-resolution LIDAR DEM was recorded on January 10, 2006 for the whole 17.5 9 7.5 km Wulfen research area. Flight height of the Cessna 404 aeroplane was 1000 m above ground and the speed was 65 m s-1. The strips were 728 m wide with a distance of 500 m between the flight lines, resulting in an overlap of 228 m. A point density of about 1 point per m2 was achieved at a pulse frequency of 50,000 Hz. Vertical accuracy of the measurements was about 3.2 cm in the z coordinate (mean difference). The DEM was initially interpolated to a resolution of 2 m and resampled to the same 5 m grid as the EC25 estimates. Terrain attributes, such as elevation, sine and cosine of aspect, slope, profile and plan curvature, log10 of flow accumulation and SAGA wetness index were calculated using SAGA GIS 1.2 (O. Conrad, University of Hamburg, Germany; see Table 2 below). The SAGA wetness index (SWI) differs from the topographic wetness index (TWI) in the calculation of the catchment area, which results in larger potential soil moisture values for SWI (refer to Bo¨hner and Selige 2006, for a definition of this index). Definition of the variables and data preparation The EC25 data and terrain attributes for all 75 soil profiles were retrieved from the grid nodes by SAGA GIS 1.2. There were 12 numerical predictor variables in total: eight terrain attributes and four soil properties. The contents (wt%) of SOM, carbonates, clay and silt for
123
496
Precision Agric (2009) 10:490–507
Table 2 Descriptive statistics of the dependent variable EC25, derived from the 5 9 5 m grid, and all numerical predictor variables from the 75 soil profiles Variable name (unit)
Mean
Standard deviation
Median
Quartile Lower
Minimum
Maximum
Upper
EC25 (mS m-1)a
44.41
22.07
40.24
25.51
62.11
13.66
92.01
SOM (kg m-2 m-1)b
23.73
9.85
22.17
16.98
29.08
6.83
61.09
CaCO3 (kg m-2 m-1)c,d
57.11
73.52
32.34
2.09
83.32
0.95
420.02
1.26
0.80
1.51
0.32
1.92
-0.02
2.62
Clay (kg m-2 m-1)
190.37
100.69
198.91
136.82
232.30
10.32
591.19
Silt (kg m-2 m-1)
489.80
270.02
459.65
257.63
767.63
43.37
962.06
Elevation (m asl)d
69.64
10.85
71.98
60.55
75.47
52.47
90.85
Sine aspectd
-0.10
0.68
-0.18
-0.75
0.51
-1.00
1.00
Cosine aspectd
0.25
0.69
0.49
-0.45
0.90
-0.99
1.00
Slope (°)
0.65
0.54
0.51
0.23
0.84
0.07
2.34
-0.03
0.27
0.01
-0.12
0.12
-1.09
0.75
0.05
0.22
0.02
-0.08
0.12
-0.32
1.32
12.63
3.15
12.44
10.35
15.83
6.35
17.71
2.13
0.63
2.10
1.68
2.48
1.00
4.27
Log10 CaCO3
Profile curvature (m-1)e Plan curvature (m-1)e SWIf Log10 Flow acc.g
Origin of the variables: EC25, Elevation, Aspect, Slope, Curvature, SWI and Flow accumulation: data retrieved from the 5 9 5 m grids for the positions of the 75 soil profiles; with variable SOM, CaCO3, Clay and Silt from laboratory analysis Variables without unit indication are dimensionless a
EC25: about 1.5 m soil depth is measured; the signal reaches at most to 4 m (vertical mode)
b
SOM, CaCO3, clay and silt: addition of the amounts of all soil horizons of a soil profile down to 1 m soil depth c
Only mentioned to allow for the comparison of CaCO3 to the other soil properties
d
Not used as potential explanatory variables in linear modeling
e
Multiplied by 100
f
SWI = SAGA wetness index
g
Log10 Flow acc. = log10 Flow accumulation
all soil horizons were converted to amounts (kg m-2 m-1) by multiplying with horizon thicknesses and bulk densities, and corrected for gravel content. The amount of each horizon was weighted according to the depth function of the EM38 signal (McNeill 1980b) for the horizon thickness between the upper and lower boundaries of the soil horizon. The amounts for the soil horizons were added to a depth of 1 m for each soil profile. These weighted amounts were used for modeling. Descriptive summary statistics for all numerical variables used are given in Table 2. Nearly all soil and topographic properties (except slope) show considerable variability. The EC25 values have a large range from 14 to 92 mS m-1, but with more sites with smaller values (median = 40.2 mS m-1). The categorical variable ‘Geological unit’ was based on the geological maps (Ko¨niglich Preußische Geologische Landesanstalt 1913a, b). The 16 geological units originally identified at the 75 soil profiles were grouped into six units based on their main characteristics related to genesis, grain size and organic matter (Table 1). This aggregation provided sufficient samples per geological unit for statistical analysis. The categorical
123
Precision Agric (2009) 10:490–507
497
variable ‘Gleyic horizons’ describes the effect of ground water and was defined by the soil profile observations, e.g. redox features such as iron concretions and iron staining of matrix or aggregate surfaces. In the model it is included as a dichotomous categorical variable (presence of gleyic horizons = 1, absence = 0). Statistical analyses The aim of the statistical analysis was to identify soil and soil-landscape variables that are related to EC25 (dependent variable) using spatial linear mixed-effects models. Details of model structure and fitting are given later in this section. Correlations between explanatory variables and the response are examined as part of the explanatory data analysis. Eleven predictor variables were considered as potential explanatory variables for linear modeling: four numerical soil variables, five terrain attributes (Table 2), and two categorical variables (‘Geological unit’ and ‘Gleyic horizons’). Elevation and aspect were not used for statistical modeling because of potential confounding with other field effects; however relative elevation was represented by flow accumulation. The variable representing the size of the contributing area (‘Flow accumulation’) and the amount of calcium carbonate (CaCO3) were transformed to common logarithms, log 10, to reduce skewness. In the linear model, each level of the geological unit variable was compared with unit C (Holocene and Weichselian dry sands; Table 1). This latter unit was chosen as the baseline because it had by far the most observations; it was also the unit with the smallest mean EC25 (Table 5 in the results section). Given the complex spatial distribution of the sample sites within the 11 fields, we decided to use a linear mixed-effects model that could take account of spatial autocorrelation in the data. In addition to the fixed effects corresponding to the explanatory variables, we considered random effects on EC25 for each field to account for differences that might be related to differences in cultivation history (Brenning et al. 2008). The spatial proximity of sample sites within each field suggested that the data—or more precisely the model residuals—may be spatially autocorrelated. We modeled this by applying a general linear model that estimated the structure of spatial dependence by fitting a variogram model to the residuals as outlined below (Cressie 1993; Pinheiro and Bates 2004). To obtain a parsimonious model that we could interpret, we used a combined stepwise forward and backward variable selection based on the Akaike Information Criterion (AIC), starting with the null model. The AIC, as opposed to R2, penalizes the number of explanatory variables in the model and so avoids overfitting to the data. Since the residual autocorrelation structure of large models will be different from that of the null model, the following approach for variable selection and autocorrelation modeling was chosen: (1) perform a stepwise variable selection assuming no spatial autocorrelation to obtain a temporary mixed model; (2) fit a temporary variogram to the residuals of this model; (3) perform the combined forward–backward variable selection starting with the temporary mixed model and using a fixed spatial autocorrelation structure based on the temporary variogram model; (4) re-fit the variogram to the selected mixed model and (5) re-estimate the selected mixed model with the final variogram of the residuals. All models were fitted by the maximum-likelihood method, and variogram parameters were estimated by weighted least squares (Cressie 1993). We analyzed the data in R 2.6.0 and used the component ‘nlme’ for regression modeling (Pinheiro and Bates 2004; R Development Core Team 2006). Since soil factors such as SOM, carbonates and soil texture were represented by both soil properties and the geological units, two separate models were fitted to the data. Thus, Model 1 was based on directly measured soil variables (SOM, log10 CaCO3, clay, silt),
123
498
Precision Agric (2009) 10:490–507
terrain attributes and the occurrence of gleyic horizons, whereas Model 2 was derived with terrain attributes and geological units only as potential explanatory variables in stepwise model selection. As the occurrence of gleyic horizons is also included in the unit of the geological map, the variable ‘Gleyic horizons’ was omitted in Model 2.
Results Pearson product moment (rp) and Spearman rank (rs) correlation coefficients indicate that EC25 is mainly related to SOM (rp = 0.59; rs = 0.65), CaCO3 (log10 CaCO3: 0.56; 0.56), and clay amount (0.49; 0.45) (Table 3). This exploratory analysis provided the first insight into which properties relate to EC25. Other variables with weaker correlations with EC25 are silt amount, slope angle, SWI, and log10 Flow accumulation, and for the other terrain attributes they are mainly \0.1 (Table 3). The stepwise selection of spatial linear mixed-effects models resulted in the two models described in Table 4, where the model parameters are indicated. To summarize the model selection process, in the construction of Model 1, the initial stepwise variable selection produced the following improvements in AIC (null model: AIC = 679) in the order of variable selection: log10 CaCO3 (DAIC = -35.01), SOM (DAIC = -20.12 relative to previous step), clay (-6.2), gleyic horizons (-9.66), silt (-1.82), flow accumulation (-1.42), SWI (-0.03). After incorporating the appropriate variogram function for the residuals, the additional stepwise selection step removed SWI (DAIC = -0.28). Regarding Model 2, the initial stepwise selection improved the AIC incrementally by including the geological units (DAIC = -56.4), flow accumulation (-1.8) and slope (-0.5) to give an AIC of 620.3. After including the spherical variogram model fitted by weighted least squares, this model’s AIC (621.8) could be improved be removing slope (DAIC = -1.16). The final Model 2 included only the geological unit and flow accumulation as fixed effects. Table 3 Correlation coefficients describing the relationship between each numerical variable and EC25 Variable (unit)
Pearson correlation rp
Spearman rank correlation rs
SOM (kg m-2 m-1)
0.59
0.65
Log10 CaCO3
0.56
0.56
Clay (kg m-2 m-1)
0.49
0.45
Silt (kg m-2 m-1)
0.16
0.24
Elevation (m asl)a
0.02
-0.07
-0.08
-0.16
Cosine aspecta
0.00
-0.02
Slope (°)
0.17
0.14
Profile curvature (m-1)
-0.16
-0.08
Plan curvature (m-1)
Sine aspecta
-0.07
-0.03
SWIb
0.13
0.14
Log10 Flow acc.c
0.10
0.16
Variables without unit indication are dimensionless a
Not used as potential explanatory variables in linear modeling
b
SWI = SAGA wetness index
c
Log10 Flow acc. = log10 Flow accumulation
123
Precision Agric (2009) 10:490–507
499
Table 4 Coefficients of the fixed effects in the final linear mixed–effects model for soil conductivity (EC25) Model 1
Model 2
Multiple R2 (%)
61.4
56.0
Adjusted R2 (%)
56.0
49.9
Range (m)
347
Nugget:sill ratio (%) Variable (unit)
Coefficient
Intercept
-6.56
SOM (kg m-2 m-1)
0.762
Standard error
p-value
7.15
0.362
0.192
\0.001*
2.24
\0.001*
0.072
0.018
\0.001*
Silt (kg m-2 m-1)
-0.018
0.011
0.098
Log10 Flow acc.a
4.50
2.29
0.054
Log10 CaCO3 Clay (kg m-2 m-1)
421
3.3
10.84
0 Coefficient
Standard error
p-value
26.35
6.36
\0.001*
3.41
2.54
0.204
Geological unit A
40.28
7.62
\0.001* \0.001*
Geological unit B
25.65
7.64
Geological unit D
2.34
6.48
0.714
Geological unit E
13.43
2.65
0.204
46.81
9.18
\0.001*
Geological unit F Gleyic horizons
15.26
4.63
0.002*
Variables without unit indication are dimensionless * Significant at the 5% level a
Log10 Flow acc. = log10 Flow accumulation
Model 1 shows which variables relate to EC25 when geology was excluded. All soil properties (SOM, log10 CaCO3, clay, silt and ‘Gleyic horizons’) are represented, but only SOM, log10 CaCO3, clay and ‘Gleyic horizons’ have a significant effect (t-tests, significance level 5%; Table 4). Of the terrain attributes log10 Flow accumulation only is included in the model, but the relation with EC25 is not significant (Table 4). A spherical model with a range of 347 m and a nugget:sill ratio of 3.3% was fitted to the residual spatial autocorrelation. To assess the relative importance of the different explanatory variables in the models, we examined the effects that two standard deviation increments of the variables have on EC25 values predicted by the models. The size of this effect was obtained by multiplying the variable’s standard deviation as given in Table 2 with the corresponding model coefficient in Table 4. Using this criterion, the effects of SOM, carbonate and clay amount have similar magnitudes. Large values of each of these variables correspond with large EC25 values. The presence of gleyic horizons also increased EC25 by about 15 mS m-1 compared to non-gleyic soil (Table 4), which was about the same as the effect of the other significant variables under a change of two standard deviations. The fixed-effects component of Model 1 resulted in a multiple R2 of 61.4% and an adjusted R2 of 56.0%. A scatter plot of the predicted versus observed EC25 values of Model 1 is shown in Fig. 2. Model 2 was based on terrain attributes and geological units. The parent material in the soil profiles was consistent in 65 cases with the corresponding geological map unit, but not in the other ten cases. As the geological map was tested as a variable, the values A–F of the variable ‘Geological unit’ were not changed in case of discrepancy.
123
500
Precision Agric (2009) 10:490–507
Fig. 2 Scatterplot of measured and predicted EC25 values (mS m-1) at 75 sample sites using Model 1. The R2 corresponding to the model’s fixed effects is 61.4%, the adjusted R2 is 56%. The line in the diagram is the 1:1 line
Flow accumulation and aggregated geological unit were the only fixed effects in Model 2. Its residual autocorrelation structure was fitted by a spherical model with a range of 421 m and no nugget effect. In Model 2, flow accumulation did not show a significant effect on EC25 (t-test, 5% level of significance). Three geological units (A, B and F) have EC25 values significantly different from the baseline unit C (F-test, p-value \ 0.001), the largest contrast was the increase in EC25 by 54.1 mS m-1 in unit F compared to unit C (Table 5). The multiple R2 of the model’s fixed-effects component, which serves only as a rough indicator because of the non-independent residuals, was 56.0% and the adjusted R2 was 49.9%. The soil electrical conductivity (ECWE) was measured in selected soil samples to determine possible effects of salinity or gypsum (Table 6). The maximum ECWE was 0.18 S m-1, so an effect from salt content cannot be excluded. The German threshold value of ECWE [ 0.04 S m-1 (Table 6) indicating a need to dissolve salts or gypsum before soil texture analysis of the fine earth fraction (Normenausschuss Wasserwesen Table 5 Summary statistics of EC25 within each level of the categorical variables Variable and level
Number of observations
EC25 (mS m-1) Median
Quartile Lower
Upper 81.4
Geological unit A
7
73.9
68.4
Geological unit B
7
59.2
42.6
75.7
Geological unit C
31
25.5
23.1
38.7
Geological unit D
19
37.6
29.2
45.8
Geological unit E
6
43.3
34.9
61.6
Geological unit F
5
79.6
72.4
83.1
Gleyic horizons: No
56
36.0
24.4
54.8
Gleyic horizons: Yes
19
55.2
32.8
73.0
123
Precision Agric (2009) 10:490–507
501
Table 6 Electrical conductivity of selected soil samples (ECWE) from 24 soil profiles in the research area Wulfen; comparison to literature threshold values for salinity (ECSE) N
Mean
Standard deviation
Median
Quartile Lower
Upper
Minimum
Maximum
ECWE (S m-1)a 143
0.0290
0.0299
0.0198
0.0113
0.0339
0.0030
0.1813
0.1495
0.0990
0.0565
0.1695
0.0150
0.9065
ECSE (S m-1)a 143
0.1450
Comparison to threshold values for saline horizons (ECSE)
N
1. Number of samples with ECSE [ 0.075 S m-1:
87
2. Number of samples with ECSE [ 0.2 S m-1:
27
3. Number of samples with ECSE [ 0.4 S m-1:
10
Citation of the threshold values: 1. Arbeitsgruppe Boden (2005): Definition of saline soil horizons, saturation extract (ECSE), soil:water 1:1: threshold value [0.075 S m-1 2. Normenausschuss Wasserwesen (NAW) im DIN Deutsches Institut fu¨r Normung e.V. (2002): Threshold value for salt or gypsum removal for soil texture analysis of the fine earth, water extract (ECWE), soil:water 1:5: [0.04 S m-1; converted to saturation extract (ECSE), soil:water 1:1: threshold value [0.2 S m-1 3. McBride et al. (1990); McNeill (1992): Definition of saline soils, saturation extract (ECSE), soil:water 1:1: threshold value [0.4 S m-1 N = Number of samples a Measured ECWE (electrical conductivity in water extract) was converted to ECSE (electrical conductivity in saturation extract) by multiplication by 5 to compare to threshold values
(NAW) im DIN Deutsches Institut fu¨r Normung e.V. 2002) was exceeded in 27 soil samples. Two sample soil profiles with similarly large EC25 values (Fig. 3) had different ECWE values: the mean ECWE of all soil horizons in profile 1 was 0.07 S m-1 and in soil profile 2 it was 0.03 S m-1. In spite of large differences in clay and silt contents both soil profiles had EC25 values of about 85 mS m-1.
Discussion The results of the spatial linear mixed-effects models indicate that SOM, calcium carbonate, clay and the aggregated geological unit are the key factors in explaining much of the variation in soil conductivity at the landscape scale. The occurrence of gleyic soil horizons also had a significant effect on EC25 when the geological map unit was not included in the regression models. Topographic characteristics derived from a high resolution DEM provided only limited additional information according to the correlation analysis, and they had no significant influence in the final models. An adjusted R2 of 56.0% in Model 1, which included direct measurements of soil properties, shows that a considerable proportion of the spatial variation in EC25 measurements in complex soil landscapes can be explained by detailed information from soil and landscape properties observed in situ or measured in the laboratory. The simpler approach of Model 2 with geological map units resulted in a slightly smaller adjusted R2 of 49.9%.
123
502
Precision Agric (2009) 10:490–507
Fig. 3 a Soil profile 1: gleyic-calcic Chernozem (siltic); EC25 87 mS m-1, organic matter 23.3 kg m-2 m-1, CaCO3 159 kg m-2 m-1, clay 177 kg m-2 m-1, silt 898 kg m-2 m-1; average ECWE of all soil horizons 0.07 S m-1; geological unit b & A (Table 1); white color: secondary carbonates; b Soil profile 2: vertic-calcic Stagnosol (clayic); EC25 83 mS m-1, organic matter 21.3 kg m-2 m-1, CaCO3 145 kg m-2 m-1, clay 560 kg m-2 m-1, silt 399 kg m-2 m-1; average ECWE of all soil horizons 0.03 S m-1; geological unit p & F (Table 1); white color on the left and the right side of the soil profile b: no secondary carbonates, but traces caused by a spade. Scale in 10 cm increments
Terrain attributes had the smallest effect on EC25 in the regression models. Reports in the literature concerning DEMs are contradictory. Sommer et al. (2003) showed that relief was more important for the spatial distribution of colluvium than geology in a loesscovered agricultural landscape in Bavaria, Germany. Terrain properties were also important for the delineation of management zones and high-yielding zones in an area of Belgium with similar geomorphology to that in our study area (Vitharana et al. 2008). Lagacherie and Voltz (2000) found that a DEM improved the mapping of soil properties only under distinct preconditions, however. Carre´ and McBratney (2005) could not improve the suitability of soil maps for management applications by adding terrain information. Our results indicate that the contribution of terrain attributes is small when geological and soil properties are included in the model to explain EC25. Overall, the geological units with calcareous materials, organic layers, Oligocene clay stones and relatively high ground water tables (geological units A, B, and F) were associated with large soil electrical conductivities. Soil developed from loess or sand had small EC25 values (geological units C, D, and E), especially if the sites were relatively dry because of a lower ground water table. This is consistent with the more detailed relationships between EC25 and soil properties identified by Model 1. This latter model refines the landscape-scale
123
Precision Agric (2009) 10:490–507
503
representation of Model 2, indicating that large SOM, carbonate and clay amounts, and gleyic horizons result equally in large soil conductivities. Therefore, we can indicate that interpretation of the geological map, which is not used by default in precision agriculture, enables a reliable prediction of factors that relate to EC25 at the landscape scale. In both models about 40–50% of the variation in EC25 remained unexplained. Some other studies had larger proportions of the variation in ECa or EC25 explained using a single explanatory variable. For example, Dalgaard and Have (2001) reported a relationship for ECa and clay content in a single field with an R2 of 79%, as did Weller et al. (2007) for different fields with an R2 of 85% after nearest neighbor correction of ECa. Domsch and Giebel (2004) had an R2 of 55% for clay content in 413 soil profiles in Brandenburg, Germany, and they recorded that silt content had a considerable effect on EC25. Heil and Schmidhalter (2003) also found a good relation between silt and ECa. Carroll and Oliver (2005) reported large correlation coefficients between ECa and each of sand, silt and clay in two fields in England. Several reasons might explain the smaller R2 in our study: 1. The ECa measurements in our study area were made at different times in and between years. Therefore the effect of soil moisture content and differences in management (e.g. ploughed, not ploughed) cannot be excluded because they might induce field-specific variation in the coefficients of the explanatory variables rather than random effects on the intercept as used here. A correction, such as that in Weller et al. (2007) or Brenning et al. (2008) might have been necessary. They proposed fieldwise linear transformations (or more general nonlinear transformations; Brenning et al. 2008) of soil conductivity to obtain some continuity in the data across the boundaries of adjacent fields. This was not done as our EC25 data were from isolated fields (see Fig. 1). 2. Other important factors affecting EC25 were not examined. Soil salinity might exert an added effect on EC25 in some areas (McNeill 1980a, 1992), but this was not expected in our study area at the beginning. McBratney et al. (2005) suggested calculating the ECa:clay ratio to identify saline soil. In terms of the statistical modeling of EC25, this would correspond to an interaction term between clay amount and soil salinity, e.g. the slope (or regression coefficient) of clay amount in the regression model would depend on soil salinity. Since summer drought is common in our study area, further research is required to assess the potential effect of soil salinity on EC25. Our first measurements at selected sites suggest that soil salinity might be responsible for the large EC25 values associated with some soil profiles. It is interesting that soil from the units A and F behave in the same way in relation to EC25, even though there are marked differences in the types of soil (Fig. 3), especially in their clay and silt amounts. Both types of soil would require different types of crop management. 3. The EC25 was measured in our study with the EM38 in the vertical mode, and SOM, CaCO3, clay, and silt amounts were calculated to a depth of 1 m (Model 1). About 55% only of the EM38 signal in the vertical mode comes from the top 1 m of the soil, and the signal reaches a depth of about 4 m below ground, if the soil is homogeneous or horizontally stratified (McNeill 1980b; Domsch and Giebel 2004). The usual depth of exploration of the EM38 in vertical mode is specified as 1.5 m (McNeill 1992). Therefore, sampling the soil to a greater depth might have improved the regression results of Model 1. 4. Model 2 could possibly be improved by using the geological units observed in the soil profiles rather than those indicated by the digitized geological maps. This uncertainty relates to the scale of 1:25,000 of the geological maps.
123
504
Precision Agric (2009) 10:490–507
We investigated four of the seven factors of the ‘scorpan’ model of McBratney et al. (2003) in our study. The factors integrated in our models were s = ‘soil’ (EC25 and its influencing factors), r = ‘relief’ (DEM), p = ‘parent material’ (Geological map) and n = ‘position’ in the sense of the topographic position in the landscape. The relevance of the scorpan approach for precision farming has been emphasized by the results of this study, and we suggest that landscape-scale factors should be included in the future interpretation of EC25 maps. Our exploratory data analysis at the landscape scale showed that EC25 is affected by a combination of several soil properties, and that SOM and CaCO3 amounts have a stronger influence on EC25 than clay. This result was unexpected in our research on EC25. Auerswald et al. (2001), Sudduth et al. (2005) and Siri-Prieto et al. (2006) reported that clay was more important for ECa than organic matter. McNeill (1992) mentions clay as a factor influencing ECa, but does not mention SOM and CaCO3. Nevertheless, our results are consistent with soil science knowledge. Korsaeth (2005) found that SOM had a larger effect on ECa than clay in a field trial. McBride et al. (1990) showed that ECa in non-saline forest soil was markedly affected by exchangeable Ca in the soil solution. Bronson et al. (2005) explained the variation of ECa by partial least squares regression models with an R2 of 61% for the variables clay, silt, Ca, Mg, Na, soluble salts content and cation exchange capacity. Dissolved CaCO3 might have increased EC25 in the soil of our study area. Soil organic matter also has an effect on EC25 because its content is strongly related to CEC (McNeill 1980a; Scheffer and Schachtschabel 2002). Shatar and McBratney (1999) reported that multivariate models are more suitable for explaining yield variation than univariate models. Similarly, EC25 measurements should not be used only to indicate the spatial variation in a single variable, e.g. clay content, unless the local calibration of such a univariate correspondence has been proved in a study area. Carroll and Oliver (2005) came to the same conclusion. Behrens and Scholten (2006) also recommended the use of DSM in combination with field survey, including soil sampling and analysis, to predict soil properties. The EC25 is, and remains, a mixed and indirect signal, influenced by several independent soil properties, that integrates the soil properties to a depth of 1.5 m (vertical mode of the EM38). The factors and soil properties relevant for EC25 can be assessed only by field work, soil sampling and laboratory analysis at characteristic sites in the fields. This remains necessary to interpret EC25 maps to define management zones.
Conclusion The EC25 values in the research area were strongly correlated with soil profile properties and geology as represented by units of the 1:25,000 geological map. Terrain attributes had only a weak, non-significant influence on EC25. Geological map units can be regarded as higher-level indicators that integrate information about SOM, CaCO3, texture (e.g. clay content) and ground water influence, amongst others. These individual variables and the geological units were the most important in the regression models. The amounts of SOM and CaCO3 were more important in relation to EC25 than clay. The EC25 reflects a combination of several factors, and focusing on one could be misleading. An effect from salt content cannot be excluded because this region in eastern Germany is relatively dry. Geological maps already imply most of the factors that affect EC25, but there is less spatial detail in them than in maps of EC25. Therefore geology, geomorphology and the soil can suggest suitable and characteristic locations for sampling sites to design an experimental setup that explains the variation in EC25 accurately and
123
Precision Agric (2009) 10:490–507
505
inexpensively. It is advantageous to consider landscape at the regional scale together with geology and geomorphology to explain the spatial variation in EC25 for its use as a DSM tool in precision agriculture, and in particular to identify management zones. The derivation of maps of clay content from those of EC25 might be inappropriate because there is no certainty in a given field which of the properties considered above will have the greatest effect on EC25. Our landscape-scale data analysis approach offers a better way to understand the spatial relationships between the geofactors that are the basis of site-specific agricultural production than the field-scale perspective. Acknowledgements Sincere thanks go to Ulrich Wagner from WIMEX (Baasdorf, Saxony-Anhalt, Germany) to use their fields as study sites. This work was carried out through sub-project ‘Development of an integrated site analysis using non-invasive methods’ of pre agro II, which is a collaborative research project funded by the German Federal Ministry of Education and Research (BMBF), under Grant Number 0339740/2. The constructive comments of the two reviewers and the editor are gratefully acknowledged.
References Arbeitsgruppe Boden (2005). Bodenkundliche Kartieranleitung (Soil Survey Instruction, 5th ed., 438 pp). Bundesanstalt fu¨r Geowissenschaften und Rohstoffe (Ed.), Hannover, Germany. Stuttgart, Germany: E. Schweizerbart’sche Verlagsbuchhandlung. Auerswald, K., Simon, S., & Stanjek, H. (2001). Influence of soil properties on electrical conductivity under humid water regimes. Soil Science, 166, 382–390. doi:10.1097/00010694-200106000-00003. Auerswald, K., Sippel, R., Kainz, M., Demmel, M., Scheinost, A. C., Sinowski, W., et al. (1997). The crop response to soil variability in an agroecosystem. Advances in Geoecology, 30, 39–53. Basso, B., Ritchie, J. T., Pierce, F. J., Braga, R. P., & Jones, J. W. (2001). Spatial validation of crop models for precision agriculture. Agricultural Systems, 68, 97–112. doi:10.1016/S0308-521X(00)00063-9. Behrens, T., & Scholten, T. (2006). Digital soil mapping in Germany—a review. Journal of Plant Nutrition and Soil Science, 169, 434–443. doi:10.1002/jpln.200521962. Bo¨hner, J., & Selige, T. (2006). Spatial prediction of soil attributes using terrain analysis and climate regionalisation. In J. Bo¨hner, K. R. McCloy, & J. Strobl (Eds.), SAGA—analyses and modelling applications. Go¨ttinger Geographische Abhandlungen, 115 (pp. 13–28 and 118–120). Go¨ttingen, Germany: Verlag Goltze. Brenning, A., Koszinski, S., & Sommer, M. (2008). Geostatistical homogenization of soil conductivity across field boundaries. Geoderma, 143, 254–260. doi:10.1016/j.geoderma.2007.11.007. Bronson, K. F., Booker, J. D., Officer, S. J., Lascano, R. J., Maas, S. J., Searcy, S. W., et al. (2005). Apparent electrical conductivity, soil properties and spatial covariance in the U.S. Southern high planes. Precision Agriculture, 6, 297–311. doi:10.1007/s11119-005-1388-6. Carre´, F., & McBratney, A. B. (2005). Digital terron mapping. Geoderma, 128, 340–353. doi:10.1016/ j.geoderma.2005.04.012. Carroll, Z. L., & Oliver, M. A. (2005). Exploring the spatial relations between soil physical properties and apparent electrical conductivity. Geoderma, 128, 354–373. doi:10.1016/j.geoderma.2005.03.008. Cockx, L., Van Meirvenne, M., & De Vos, B. (2006). Using the EM38DD soil sensor to delineate clay lenses in a sandy forest soil. Soil Science Society of America Journal, 71, 1314–1322. doi:10.2136/ sssaj2006.0323. Cressie, N. A. C. (1993). Statistics for spatial data. New York, USA: Wiley, 928 pp. Dalgaard, M., & Have, H. (2001). Soil clay mapping by measurement of electromagnetic conductivity. In A. Werner & A. Jarfe (Eds.), Programme book of the joint conference of ECPA-ECPLF (pp. 367–372). Wageningen, Netherlands: Academic Publishers. Deutscher Wetterdienst (Ed.). (1999). Klimaatlas Bundesrepublik Deutschland. Teil 1. Lufttemperatur, Niederschlagsho¨he, Sonnenscheindauer (Climate Atlas of Germany. Part 1. Air Temperature, Precipitation, Sunshine, 23 pp., 57 maps). Offenbach/M., Germany. Domsch, H., & Giebel, A. (2004). Estimation of soil textural features from soil electrical conductivity recorded using the EM38. Precision Agriculture, 5, 389–409. doi:10.1023/B:PRAG.0000040807. 18932.80. Durlesser, H. (1999). Bestimmung der Variation bodenphysikalischer Parameter in Raum und Zeit mit elektromagnetischen Induktionsverfahren (Determination of soil physical properties variation in space
123
506
Precision Agric (2009) 10:490–507
and time using electromagnetic induction methods). PhD. Thesis, Technical University of Munich, Munich, Germany. FAM-Bericht, 35. Aachen, Germany: Shaker Verlag, 120 pp. Heil, K., & Schmidhalter, U. (2003). Derivation of soil texture and soil water content from electromagnetic induction measurements. In A. Werner & A. Jarfe (Eds.), Programme book of the joint conference of ECPA-ECPLF (pp. 429–430). Wageningen, Netherlands: Academic Publishers. IUSS Working Group WRB. (2006). World reference base for soil resources 2006—a framework for international classification, correlation and communication. World Soil Resources Reports 103. Rome, Italy: FAO, 128 pp. Ko¨niglich Preußische Geologische Landesanstalt (Ed.). (1913a). Geologische Karte von Preußen und benachbarten Bundesstaaten 1:25.000, Blatt Wulfen 4137 (Geological map of Prussia and adjacent federal states 1:25,000, Sheet Wulfen 4137). Berlin, Germany. Ko¨niglich Preußische Geologische Landesanstalt (Ed.). (1913b). Geologische Karte von Preußen und benachbarten Bundesstaaten 1:25.000, Blatt Co¨then 4237 (Geological map of Prussia and adjacent federal states 1:25,000, Sheet Co¨then 4237). Berlin, Germany. Korsaeth, A. (2005). Soil apparent electrical conductivity (ECa) as a means of monitoring changes in soil organic N on heterogeneous morainic soils in SE Norway during two growing seasons. Nutrient Cycling in Agroecosystems, 72, 213–227. doi:10.1007/s10705-005-1668-6. Lagacherie, P., & McBratney, A. B. (2007). Spatial soil information systems and spatial soil inference systems: Perspectives for digital soil mapping. In P. Lagacherie, A. B. McBratney, & M. Voltz (Eds.), Digital soil mapping, an introductory perspective. Developments in Soil Science 31 (pp. 3–22). Amsterdam, Netherlands: Elsevier. Lagacherie, P., McBratney, A. B., & Voltz, M. (Eds.). (2007). Digital soil mapping. An introductory perspective (600 pp, 41 Color Plates). Developments in Soil Science 31. Amsterdam, Netherlands: Elsevier. Lagacherie, P., & Voltz, M. (2000). Predicting soil properties over a region using sample information from a mapped reference area and digital elevation data: A conditional probability approach. Geoderma, 97, 187–208. doi:10.1016/S0016-7061(00)00038-0. Maidl, F.-X., Brunner, R., Sticksel, E., & Fischbeck, G. (1999). Ursachen kleinra¨umiger Ertragsschwankungen im bayerischen Tertia¨rhu¨gelland und Folgerungen fu¨r eine teilschlagbezogene Du¨ngung (Site effects of small-scale yield variation in the Tertiary hills north of Munich (Germany) and conclusions for site specific farming). Journal of Plant Nutrition and Soil Science, 162, 337–342. doi:10.1002/ (SICI)1522-2624(199906)162:3\337::AID-JPLN337[3.0.CO;2-2. with English abstract. McBratney, A. B., Mendonc¸a Santos, M. L., & Minasny, B. (2003). On digital soil mapping. Geoderma, 117, 3–52. doi:10.1016/S0016-7061(03)00223-4. McBratney, A. B., Minasny, B., & Whelan, B. M. (2005). Obtaining ‘useful’ high-resolution soil data from proximally-sensed electrical conductivity/resistivity (PSEC/R) surveys. In J. V. Stafford (Ed.), Precision agriculture ‘05 (pp. 503–510). Wageningen, Netherlands: Wageningen Academic Publishers. McBratney, A. B., Odeh, I. O. A., Bishop, T. F. A., Dunbar, M. S., & Shatar, T. M. (2000). An overview of pedometric techniques for use in soil survey. Geoderma, 97, 293–327. doi:10.1016/S0016-7061 (00)00043-4. McBride, R. A., Gordon, A. M., & Shrive, S. C. (1990). Estimating forest soil quality from terrain measurements of apparent electrical conductivity. Soil Science Society of America Journal, 54, 290–293. McNeill, J. D. (1980a). Electrical conductivity of soils and rocks. Technical Note TN-5. Mississauga, Ontario, Canada: Geonics Limited, 22 pp. McNeill, J. D. (1980b). Electromagnetic terrain conductivity measurement at low induction numbers. Technical Note TN-6. Mississauga, Ontario, Canada: Geonics Limited, 17 pp. McNeill, J. D. (1992). Rapid, accurate mapping of soil salinity by electromagnetic ground conductivity meters. In G. C. Topp, W. D. Reynolds, & R. E. Green (Eds.), Advances in measurement of soil physical properties. Bringing theory into practice (pp. 209–229). SSSA Special Publication No. 30. Madison, Wisconsin, USA: Soil Science Society of America. Moran, M. S., Inoue, Y., & Barnes, E. M. (1997). Opportunities and limitations for image-based remote sensing in precision crop management. Remote Sensing of Environment, 61, 319–346. doi: 10.1016/S0034-4257(97)00045-X. Normenausschuss Wasserwesen (NAW) im DIN Deutsches Institut fu¨r Normung e.V. (2002). Bodenbeschaffenheit. Bestimmung der Partikelgro¨ßenverteilung in Mineralbo¨den. Verfahren mittels Siebung und Sedimentation (Soil texture. Determination of particle size distribution in mineral soil. Methods by sieving and sedimentation). DIN ISO 11277. Ref. Nr. DIN ISO 11277:2002–08. Berlin, Germany: DIN Deutsches Institut fu¨r Normung e.V., 25 pp. Pinheiro, J. C., & Bates, D. M. (2004). Mixed-effects models in S and S-PLUS. New York, USA: Springer, 528 pp.
123
Precision Agric (2009) 10:490–507
507
R Development Core Team. (2006). R. A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. ISBN 3-900051-07-0. Scheffer, F., & Schachtschabel, P. (2002). Lehrbuch der Bodenkunde (Textbook of soil science) (15th ed.). Heidelberg, Germany: Spektrum Verlag, 593 pp. Schlichting, E., Blume, H.-P., & Stahr, K. (1995). Bodenkundliches Praktikum. Eine Einfu¨hrung in pedologisches Arbeiten fu¨r O¨kologen, insbesondere Land– und Forstwirte, und fu¨r Geowissenschaftler (Practicing soil science. An introduction to pedological working for ecologists, especially agriculturists and foresters, and for earth scientists, 2nd ed.). Berlin, Germany: Blackwell Verlag, 295 pp. Shatar, T. M., & McBratney, A. B. (1999). Empirical modeling of relationships between sorghum yield and soil properties. Precision Agriculture, 1, 249–276. doi:10.1023/A:1009968907612. Siri-Prieto, G., Reeves, D. W., Shaw, J. N., & Mitchell, C. C. (2006). World’s oldest cotton experiment: Relationships between soil chemical and physical properties and apparent electrical conductivity. Communications in Soil Science and Plant Analysis, 37, 767–786. doi:10.1080/00103620600564018. Sommer, M. (2006). Influence of soil pattern on matter transport in and from terrestrial biogeosystems—a new concept for landscape pedology. Geoderma, 133, 107–123. doi:10.1016/j.geoderma.2006.03.040. Sommer, M., & Schlichting, E. (1997). Archetypes of catenas in respect to matter—a concept for structuring and grouping catenas. Geoderma, 76, 1–33. doi:10.1016/S0016-7061(96)00095-X. Sommer, M., Wehrhan, M., Zipprich, M., Weller, U., zu Castell, W., Ehrich, S., et al. (2003). Hierarchical data fusion for mapping soil units at field scale. Geoderma, 112, 179–196. doi:10.1016/S0016-7061 (02)00305-1. 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. doi:10.1016/S0168-1699(00)00185-X. Sudduth, K. A., Kitchen, N. R., Wiebold, W. J., Batchelor, W. D., Bollero, G. A., Bullock, D. G., et al. (2005). Relating apparent electrical conductivity to soil properties across the North-central USA. Computers and Electronics in Agriculture, 46, 263–283. doi:10.1016/j.compag.2004.11.010. Vitharana, U. W. A., Van Meirvenne, M., Simpson, D., Cockx, L., & De Baerdemaeker, J. (2008). Key soil and topographic properties to delineate potential management classes for precision agriculture in the European loess area. Geoderma, 143, 206–215. doi:10.1016/j.geoderma.2007.11.003. Walter, C., Lagacherie, P., & Follain, S. (2007). Integrating pedological knowledge into digital soil mapping. In P. Lagacherie, A. B. McBratney, & M. Voltz (Eds.), Digital soil mapping, an introductory perspective (pp. 281–300). Amsterdam, Netherlands: Elsevier. Developments in Soil Science 31. Walter, H., & Lieth, H. (1964). Klimadiagramm-Weltatlas, 2. Lieferung (World atlas of climate diagrams, second part). Jena, Germany: VEB Gustav Fischer. 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.
123