Geoscience Reference
In-Depth Information
4.3 GUIdelIneS foR CondUCtInG A fIeld-SCAle eC
a
-dIReCted
SoIl SAMplInG SURvey foR AGRICUltURe
The basic steps of a field-scale EC
a
survey for characterizing spatial variability include (1) EC
a
sur-
vey design, (2) geo-referenced EC
a
data collection, (3) soil sample design based on geo-referenced
EC
a
data, (4) soil sample collection, (5) physicochemical analysis of pertinent soil properties,
(6) stochastic or deterministic calibration of EC
a
to soil properties, (7) determination of the soil
properties influencing the EC
a
measurements at the study site, and (8) GIS development. Details on
EC
a
-directed soil sampling protocols are presented by Corwin and Lesch (2005b, 2005c). Outlined
protocols are provided in Table 4.2. Of the eight basic steps, EC
a
-directed soil sample design, sto-
chastic or deterministic calibration of EC
a
, and determination of the soil properties influencing the
geospatial EC
a
measurements are the least understood and yet are crucial for correctly understand-
ing and interpreting spatial EC
a
data. Ideally, efforts must be directed toward mapping EC
a
when
the soil property of interest is expected to have its greatest influence on EC
a
values. This maximizes
the likelihood of inferring the spatial patterns of the soil property of interest from the EC
a
map. For
instance, the effect of texture (or clay content) on EC
a
is more pronounced at higher water contents
(Dalgaard et al., 2001), suggesting EC
a
field mapping when the soil is wet rather than dry.
4.3.1 ec
a
-d
i R e c t e d
s
of i l
s
a M P l e
d
e s i g n
An EC
a
survey of a field is most often conducted with either mobile ER or EMI equipment that
has been coupled to a GPS. Depending on the level of detail desired, from 100 to several thousand
spatial measurements of EC
a
are taken generally in regularly spaced traverses across the field of
interest. The use of mobile EMI equipment has one slight advantage over the use of mobile ER
equipment due to the fact that EMI is noninvasive, which is the ability to take measurements on dry
and stony soils.
Once a geo-referenced EC
a
survey is conducted, the data are used to establish the locations of
the soil core sample sites for (1) calibration of EC
a
to a correlated soil sample property (e.g., salin-
ity, water content, and clay content) and (2) delineation of the spatial distribution of soil properties
correlated to EC
a
within the field surveyed. Currently, two different sampling schemes are used to
establish the locations where soil cores are to be taken: design-based and model-based sampling
schemes. Design-based sampling schemes have historically been the most commonly used and,
hence, are more familiar to most research scientists. Design-based methods include simple random
sampling, stratified random sampling, multistage sampling, cluster sampling, and network sampling
schemes. The use of unsupervised classification by Fraisse et al. (2001) and Johnson et al. (2001) is
an example of design-based sampling. Model-based sampling schemes are less common. Specific
model-based sampling approaches that have direct application to agricultural and environmental
survey work are described by McBratney and Webster (1983), Russo (1984), and Lesch et al. (1995a,
1995b, 2005).
The sampling approach introduced by Lesch et al. (1995a, 1995b, 2005) is specifically designed
for use with ground-based soil EC
a
data. This sampling approach attempts to optimize the estima-
tion of a regression model (i.e., minimize the mean square prediction error produced by the calibra-
tion function), while simultaneously insuring that the independent regression model residual error
assumption remains approximately valid. This, in turn, allows an ordinary regression model to be
used to predict soil property levels at all remaining (i.e., nonsampled) conductivity survey sites.
There are two main advantages to the response-surface approach. First, a substantial reduction
in the number of samples required for effectively estimating a calibration function can be achieved,
in comparison to more traditional design-based sampling schemes. Second, this approach lends
itself naturally to the analysis of remotely sensed EC
a
data. Many types of ground-, airborne-, and
satellite-based remotely sensed data are often collected specifically because one expects this data
