Modelling Soil Carbon - Soil Carbon: Science, Management and Policy for Multiple Benefits - page 204

or humified organic matter with CO 2 evolved
during the process. In addition, there is a
small pool of inert organic matter with a
turnover time too slow to detect. Soil tem-
perature and moisture modify the rates of
decomposition. The partition of C to either
CO 2 , biomass or stabilized humus material
is adjusted for soil texture (Coleman and
Jenkinson, 1996).
This multicompartmental approach has
been used in the majority of models that
deal with the long-term turnover of C in
soils, the main difference between models
being the number of pools and associated
rate constants. Short-term turnover rates
can be determined by decomposition stud-
ies. For longer-term turnover, techniques
have been employed that use natural label-
ling of SOM using stable 13 C tracers, labelling
of SOM with 'bomb' 14 C and/or 14 C-dating
techniques (von Lutzow et al ., 2007). In
reality, these pools are hypothetical, with
decay being a continuous process. How-
ever, representing this in a model involves
developing different equations for numerous
individual types of plant material (Agren
and Bossata, 1987) and has therefore proved
less popular.
RothC requires the user to provide data
on plant material inputs to the soil. In the
1980s and 1990s, several models were de-
veloped that linked plant growth models to
SOM models, the purpose being to describe
nutrient turnover both above and below
ground. One example is CENTURY, devel-
oped using data from long-term experiments
in the American Great Plains. CENTURY has
a three-pool SOM model, two litter pools
(often counted as SOM in conventional
measurements) - metabolic and structural,
which are roughly analogous to the DPM
and RPM in RothC, three plant productivity
submodels and a water budget model (Par-
ton et al ., 1988). Throughout the 1980s and
1990s, many models were developed using
different data sets in different climate and
soil regions. In 1996 (updated in 2001),
model descriptions and metadata were
brought together in a report by the SOM net-
work (Smith et al ., 1996a). In addition,
Smith et al . (1997) carried out a comparison
of nine of the most frequently used SOM

models by testing them using data sets from
a range of land uses. They found six of the
models (RothC, CANDY, DNDC, CENTURY,
DAISY and NCSOIL) performed better than
the rest across a range of conditions.
The majority of SOC models were devel-
oped primarily to describe the dynamics of
C in aerobic conditions in mineral soils.
Fewer have been developed specifically for
flooded soils and peatlands. By the 2000s,
the need for models that could describe C
dynamics in the large global C stock repre-
sented by peatlands had been recognized.
The ECOSSE model was developed in re-
sponse to this need, taking a multicompart-
mental approach in the same way as RothC
(on which it is partially based), but including
calculations that account for the domination
of CH 4 evolution under anaerobic conditions
(Smith et al ., 2010). In recent years, the SOC
modelling community has started to explore
the use of different approaches such as bio-
climatic envelope models to estimate C
stocks and changes in peatlands at the re-
gional to global scale with models such as
PEATSTASH (Gallego-Sala et al ., 2010).

Application to Different Climates
and Land Covers/Uses

The application of SOM models to different
environments, soils, land uses and climates
is dependent on the data available for their
development and evaluation. Models can be
used to interpret observations at a particular
site, to predict what might happen at another
site or to predict SOM turnover in the future.
If a model is used only to interpret what is
occurring in a particular laboratory or field
experiment, it can be run using just the data
available in that experiment. However, if the
application is predictive, simulations of the
inherently long-term processes of SOM turn-
over must be tested against independent
data. Such applications cannot be developed
adequately by testing simulations against
laboratory or short-term field data alone. Easy
access to long-term experimental data has
therefore controlled the development of such
models, as long-term experiments (of at least