Modelling Soil Carbon - Soil Carbon: Science, Management and Policy for Multiple Benefits - page 207

over time in a one-step process assuming a
linear rate of change over a given period
(the default being 20 years). It has been used
in regional-scale assessments, showing most
success when incorporating detailed regional
data (Grace et al ., 2004). Other studies have
used regression approaches to estimate SOC
stock changes at the regional scale, extrapo-
lating several long-term experimental data
sets into the future (Gupta and Rao, 1994;
Smith et al ., 2000). Others still have used
more complicated regressions based on spa-
tially explicit soil databases (Kern and John-
son, 1993; Kotto-Same et al ., 1997).
As pointed out by Falloon et al . (2002),
such regression approaches fail to capture
the dynamic nature of SOC change. This is
particularly relevant when looking at SOC
change following a change in land use, as
SOC tends to deplete rapidly at first, then
more slowly. In order to capture these dy-
namics, studies have linked dynamic SOC
models (originally developed at the plot
scale) to large-scale data sets using geo-
graphic information systems (GIS) to make
estimates at the regional to subnational
scale. This allows many of the site-scale ob-
servations of SOC dynamics to inform larger-
scale assessments (Paustian et al ., 1997).
Paustian et al . (1995) outlined the approach
needed to link spatial data sets of soils, cli-
mate and land-use information to the CEN-
TURY model, and this was subsequently
used to make state and regional assessments
of SOC stock change in the USA (Paustian
et al ., 2001, 2002). Other examples of this
approach include Falloon et al . (1998), who
linked the RothC model to GIS layers of
soils, climate and land-use data for a 25,000
km 2 area of central Hungary. The methods
developed by Paustian et al . (1995) were
used to develop the GEFSOC Modelling
System, a tool that allows the user to make
large-scale assessments of SOC stocks and
changes (Easter et al ., 2007; Milne et al .,
2007). The tools have been used to make
national-scale (Al-Adamat et al ., 2007; Kamoni
et al ., 2007) and subnational-scale (Bhattacha-
rryya et al ., 2007; Cerri et al ., 2007) assess-
ments, but could be applied at any scale
providing data are available to parameterize
and run the models.

The scale that perhaps presents the
most challenges for SOC modelling, and
any estimation of SOC stock change, is the
landscape scale. Issues arise with data col-
lation and adequate representation of het-
erogeneous land covers/uses. Also, at this
scale, SOC stocks under fragmented land
covers such as hedgerows or urban green
spaces gain in significance (Viaud et al .,
2010), and these are difficult to model. An-
other area where SOC modelling is in need
of further development is in the horizontal
transport and deposition of C, the relevance
of which changes with scale. Some SOC
turnover models include estimates of C loss
from erosion, but few deal with deposition
to another area or loss of dissolved C (Tip-
ping et al ., 2007).

Uncertainties in Modelling SOC

Uncertainty in large-scale simulations has
two components: uncertainty due to inad-
equacies in the model (referred to by Beven,
2002, as structural errors), and uncertainty
due to reduced detail and precision in data
available at large scale (referred to here as
input errors). If a model is to be applied at
large scale, uncertainty in the simulations
is likely to be greater than at field scale, due
to the reduced detail of input data avail-
able. For example, detailed management
factors of crops, such as sowing date and
timing of fertilizer applications, cannot
usually be specified when the resolution of
the simulations is larger than the size of
the management unit; the resolution of the
simulation might be a 1 km 2 grid cell, whereas
the size of a management unit might be a
5 ha field, so there will be many different
values for the management factors within
each 1 km 2 cell. Uncertainty in simulations
at large scale is also greater than at field
scale, due to the reduced precision of the
input values; for example, the C content of
the soil in a 10 ha field can be measured
precisely, and the error in the measure-
ment defined using replicates, whereas for
applications at larger scale, the soil C con-
tent is often determined for 1 km 2 grid cells,