Agriculture Reference
In-Depth Information
Table 5.5
Accuracies of on-the-go and full spectrum carbon sensing via reflectance for defined
point estimations depending on the variables considered in a stepwise regression
Corr.coeff. squared (r
2
),
without dimension
Root mean squared
error in 10 g/kg
Variable
Carbon only
0.70
0.189
Carbon + topography
0.81
0.156
Carbon + topography + water
0.83
0.144
Carbon + topography + water + clay
0.88
0.127
Extract from Huang et al. (
2007
)
Legend: field size = 50 ha. 85 geo-referenced and defined points for traditional soil analyses. 3,700
points for mean reflectance readings. Carbon negatively correlated to topography and positively
correlated to water or clay
Range of total carbon: 5.5-28.9 g/kg. Range of elevation: 290-303 m
accuracies is not possible since this would imply the same spatial resolution for
records from laboratories and from on-the-go sensing in fields. This prerequisite
does not hold. The technique shown in Figs.
5.25
and
5.26
easily can provide a
spectrum for every 8 cm of travel. Present practice is to use 50 readings respectively
for an average spectrum. Thus for every 4 m of travel, a signal is available. And with
a swath-width of 20 m, about 125 readings per ha result. Such a high resolution is
completely beyond any real possibilities when the recording is based on analyses of
samples in laboratories. Hence from the
spatial resolution
that can be obtained,
online and on-the-go sensing inherently is much better suited for site-specific farm-
ing than analyzing in laboratories. It could have a lower
accuracy per signal
than
laboratory techniques yet still provide a benefit as a result of the much better spatial
coverage
But disregarding any spatial resolution, online and on-the-go reflectance sens-
ing can be compared with traditional analysing on the basis how well for defined
points within a field it supplies soil property data that agree with those that are
obtained with conventional state of the art laboratory methods. Such a
defined
point estimation
relies on the assumption that conventional laboratory analyses
are accurate. This assumption probably is reasonable because of the long experi-
ences with these conventional laboratory methods and the well developed proce-
dures when using them.
Among the various natural soil properties, much interest goes to organic mat-
ter or the
carbon
content, particularly since water is a special case as a result of
its transient character and clay can be roughly detected by electrical conductivity
- except for arid regions. However, the natural soil properties are interdependent.
A question is how much can be gained in accuracy for the
sensing of carbon
if
the
dependence on other soil properties
is taken into account.
The results in Table
5.5
are based on defined point estimations of total carbon
sensing by the on-the-go technique of Figs.
5.25
and
5.26
in a sandy loam of glacial
origin in Michigan, USA, hence in a humid, moderate climate. The reflectance sig-
nals were recorded within a range of 900-1,700 nm with a spectral resolution of
6 nm. A multiple soil property sensing approach and a sophisticated processing of
