Geoscience Reference
In-Depth Information
A user-friendly software package (ESAP) developed by Lesch et al. (2000), which uses a
response-surface sampling design, has proven to be particularly effective in delineating spatial dis-
tributions of soil properties from EC
a
survey data (Corwin and Lesch, 2003; Corwin et al., 2003a,
2003b, 2006). The ESAP software package identifies the optimal locations for soil sample sites
from the EC
a
survey data. These sites are selected based on spatial statistics to reflect the observed
spatial variability in EC
a
survey measurements. Generally, six to twenty sites are selected depend-
ing on the level of variability of the EC
a
measurements for a site. The optimal locations of a minimal
subset of EC
a
survey sites are identified to obtain soil samples.
In a detailed study of the utility of EC
a
mapping, Farahani and Buchleiter (2004) used two differ-
ent soil sampling designs for comparison. Measured EC
a
data from three center-pivot fields, two near
the town of Wiggins and one near Yuma in eastern Colorado, were used to identify sample locations
using ESAP and a combination of cluster analysis and random soil sampling within clusters. For
the latter sampling design, three to five sample locations were randomly selected from each of five
delineated EC
a
zones (according to cluster analysis). For the sandy and nonsaline fields in eastern
Colorado, both sampling methods effectively captured the spatial variability. The random sampling
from within EC
a
clusters was found to be simple and subsequently used in a number of dryland and
irrigated fields to characterize EC
a
delineated zones for the purposes of site-specific management.
4.3.2 c
a l i b R a t i o n
of f
ec
a
t o
s
of i l
P
R of P e R t i e s
Apparent soil electrical conductivity can be calibrated to any soil property that significantly influences
the EC
a
measurement, such as salinity, θ
g
, clay content, SP, ρ
b
, and OM. As indicated in Table 4.1,
there are numerous studies that document the relationships between soil electrical conductivity and
various soil physical and chemical properties. All of the data analysis and interpretation presented in
these papers can be classified into two data modeling categories: deterministic and stochastic.
In general, stochastic models are based on some form of objective sampling methodology used
in conjunction with various statistical calibration techniques. The most common types of calibration
equations are geostatistical models (generalized universal kriging models and cokriging models)
and spatially referenced regression models.
Traditionally, universal kriging models have been viewed as an extension of the ordinary krig-
ing technique and used primarily to account for large-scale (nonstationary) trends in spatial data.
However, this modeling technique can be easily generalized to model ancillary survey data (such
EMI or ER data) when this data correlates well with some spatially varying soil property of interest
(e.g., soil salinity). This generalization is commonly referred to as a “spatial linear model” or “spa-
tial random field model” in the statistical literature. This modeling approach requires the estimation
of a regression equation with a spatially correlated error structure. This type of model probably rep-
resents the most versatile and accurate statistical calibration approach, provided enough calibration
sample sites are collected (
n
> 50) to ensure a good estimate of the correlated error structure.
Regardless of their versatility, spatial linear models are typically used in regional situations.
Such an approach is rarely used for field-scale survey work, due to the large number of required
calibration soil samples, which makes this approach economically impractical. Instead, most cali-
bration equations of soil properties are spatially referenced regression models. A spatially refer-
enced regression model is just an ordinary regression equation that includes the soil property being
calibrated with EC
a
and trend surface parameters. The model assumes an independent error struc-
ture that can usually be achieved through carefully designed sampling plans, such as the response-
surface sampling design. In practice, these are the only models that can be reasonably estimated
with a limited number of soil samples (
n
< 15).
Deterministic conductivity data modeling and interpretation have been carried out either from
a geophysical or a soil science approach. In the geophysical approach, mathematically sophisticated
inversion algorithms are generally employed. These approaches, which rely heavily on geophysical
theory, have met with limited success for the interpretation of near-surface EC
a
data. Part of the
