41 Simulating soil organic carbon in a rice-soybean

Milne et al., 2008
Simulating soil organic carbon in a rice-soybean-wheat-soybean chronosequence in
Prairie County, Arkansas using the Century model
Eleanor Milne1,2,*, Stephen Williams1, Kristofor R. Brye3, Mark Easter1, Kendrick Killian1,
and Keith Paustian1
each county of the US. However, many widely
grown crops and crop rotations are currently
excluded as the ability of Century, the model upon
which the system relies, to simulate them has yet to
be established.
One such crop is rice (Oryza sativa L.)
when grown in lowland, flooded conditions. Rice is
an important crop for several states in the US.
Although rice only accounts for ~ 1% of the total
cropland harvested in the US and just 1.5 to 2 % of
global production, rice is a major export crop for
the US, generating between 1 and 1.5 billion US
dollars per year (USDA, 2007b). United States rice
production is important for the states of Arkansas,
California, and Louisiana, which account for 80%
of US rice acreage with some production also
occurring in Texas, Mississippi, and Missouri
(USA Rice Federation, 2006). For these states,
inclusion of rotations that involve flooded rice in
tools such as COMET-VR would enhance carbon
accounting capabilities.
The Century Soil Organic Matter model
(Parton et al., 1988), similar to most other SOM
models, was designed to simulate decomposition
under aerobic soil conditions. This has led to few
instances of the model being used to simulate
agricultural rotations that include flooded rice. An
exception is a study by Bhattacharrya et al. (2007)
who tested the performance of Century against a
long-term fertilizer trial involving a jute (Cochorus
capsularis L.)-rice-wheat rotation in Barrackpore,
West Bengal, India. To address the study’s
objectives, the authors created new rice files for the
model with many parameters specific to the IndoGangetic Plains. The Century model was run using
these new rice files to test the model’s ability to
simulate the measured data set under five different
fertilizer regimes. Bhattacharyya et al. (2007)
reported that Century tended to over-estimate SOC
for all fertilizer regimes, but, in general, was able
to simulate trends in SOC over the 30-yr period
covered by the experiment.
Some other ecosystem models, such as
DeNitrification-DeComposition (DNDC, Li et al.,
2003), have subcomponents aimed at predicting
SOC turnover under anaerobic conditions. In the
case of DNDC, an ‘anaerobic balloon’ predicts the
soil aeration status in different soil layers and uses
this information mainly to predict changes in
nitrification and denitrification rates (Li et al.,
2003). DeNitrification-DeComposition has been
tested against long-term experiments involving
ABSTRACT
It is useful for ecosystem models, such as Century, to be able to
estimate soil organic carbon (SOC) turnover in anaerobic as
well as aerobic soils. This will allow the inclusion of cropping
systems involving flooded rice (Oryza sativa) in online C
inventory tools such as COMET-VR. The present study tested
the performance of the Century model against a cropped
chronosequence of fields planted to flooded rice once every
three years under a rice-soybean (Glycine max), wheat (Triticum
aestivum)-soybean rotation. Study sites were in Prairie County,
Arkansas and consisted of four fields which had been cropped
for varying lengths of time, up to a maximum of 44 years. Two
different chronosequence approaches were used for model
testing. Overall, Century was a suitable model for estimating
long-term SOC dynamics in rotations that involve submerged
rice once every three years, but other studies in the literature
have suggested that Century may be less suited to estimate
dynamics in rotations that are flooded every year. A comparison
of modelled crop yield with average county-wide crop yield
statistics for rice, wheat, and soybean during the study period
showed good agreement. The findings of this study have
important implications for the development of C inventory tools
that include flooded-rice rotations.
Key
words:
Arkansas,
Century
model,
chronosequence, flooded rice, soil organic C
1
The Natural Resource Ecology Laboratory,
Colorado State University, Fort Collins, Colorado,
USA; 2 The Macaulay Institute, Craigiebuckler,
Aberdeen, AB15 8QH, UK; 3 University of
Arkansas, Fayetteville, AR 72701, USA.
* Corresponding author ([email protected])
J. Integr. Biosci. 6(1):41-52.
INTRODUCTION
With the continued development of carbon (C)
markets in the United States (US), and the
increasing realization that the management of
agricultural land is an integral part of that market,
the need for soil organic matter (SOM) models to
simulate a wider variety of agricultural systems,
crops, and crop rotations is increasing. An example
of such a need is illustrated by the
US Department of Agricultures’ Voluntary
Reporting Carbon Management Tool (COMETVR), which uses data from the Carbon
Sequestration Rural Appraisal (CSRA) and
calculates annual carbon fluxes using a dynamic
Century model (Parton et al., 1988) simulation
(USDA, 2007a). This tool allows users to model
the dominant crops and crop rotations that occur in
41
29 December 2008 Electronic Journal of Integrative Biosciences 6(1):41-52.
Special Issue: Soil Quality for a Sustainable Environment (V.S. Green and K.R. Brye, co-editors)
1957
1968
Field
A
No cultivation
B
Cultivation 1986 onwards
C
Cult 1975 onwards
D
Cultivation 1957 onwards
1979
© 2008 by Arkansas State University
1990
2001
Sampled 1987 and 2001
Fig. 1. Cultivation schedule for the four fields in the Prairie County, Arkansas field trial.
Study site
The study site was in Prairie County,
Arkansas (34° 42’ N, 91° 31’ W). The sampled
area of the four fields was on a Crowley silt loam
(fine, smectitic, thermic Typic Albaqualf; Fielder et
al., 1981; USDA-NRCS-SSD, 2007) with sand, silt
and clay concentrations of 0.16, 0.67, and 0.17 kg
kg-1, respectively (Brye and Slaton, 2003). The
study site exists in an area of eastern Arkansas
known as the Grand Prairie (Brye and Pirani, 2005;
Brye et al., 2004b) where tallgrass prairie was the
dominant ecological community prior to settlement
and the influx of mechanised agriculture. Four
adjacent fields (A,B,C,D) were under managed
flooded rice in Japan (Shirato, 2005), but has not
yet been fully validated.
Although the Century model does not
have
a
subcomponent
specifically
for
decomposition under anaerobic conditions in the
same way as DNDC, the model does account for
the fact that anaerobic conditions (i.e., nearly
saturated soil water contents) will cause
decomposition rates to decrease. Century includes a
water-budget model that calculates saturated water
flow between soil layers and a soil drainage factor
that allows a soil to have differing degrees of
wetness. The aim of this study was two fold; firstly
to address whether Century works well enough for
Yr 1
Rice
Plant April
Harvest August
Yr 2
Soybean
Plant May
Harvest Oct.
Yr 3
Wheat
Plant Oct.
Harvest June
Soybean
Plant June,
Harvest Nov.
Fig. 2. Crop rotation used in the Prairie County, Arkansas four-field trial.
US crop rotations that include flooded rice, but are
not exclusively flooded, to allow these rotations to
be included in online tools such as COMET-VR
and secondly to consider some of the problems that
may arise when using chronosequence datasets for
model validation.
native prairie in 1956. One field remained as native
prairie from 1956 to 2001, while the other three
were put into cultivation at different times from
1956 to 2001 (Fig. 1). Soil samples were collected
from the fields in 1987 (Scott and Wood, 1989) and
again in 2001 (Brye and Slaton, 2003). When
sampled in 1987, the fields had been in cultivation
for 0 (A), 1 (B), 12 (C) and 30 (D) years. When
sampled again in 2001, the fields had been in
cultivation for 0 (A), 15 (B), 26 (C) and 44 (D)
years. The cultivated fields were managed under a
three year rice-soybean, wheat-soybean rotation
(Fig. 2). This meant that flooded rice was present in
one out of every three years.
MATERIALS AND METHODS
This study used published data from three past
studies: Brye and Slaton (2003), Scott and Wood
(1989) and Brye et al. (2004a). All three studies
reported on soil property changes in the same four
fields that were under managed native prairie in
1956. Three of the fields were subsequently put
into cultivation at different points in time.
Available data
The research articles from which data
were obtained for this study were not produced
with ecosystem modelling as their remit. Therefore,
42
Milne et al., 2008
they presented soil data only. Climate, fertilizer,
irrigation, yield, and productivity data were not
reported. For modelling purposes, the necessary
data were therefore obtained from other sources
and from consultation with some of the authors of
the original papers.
accumulated SOC mass in the 0-10cm depth
interval. In order to estimate SOC for 0-20 cm, the
relationship between the 0-10cm data for 1987 and
the 0-10 cm data for 2001 was established for each
field. For example, in the case of field A, the 1987
0-10 cm SOC data was ~ 37% lower than the 2001
0-10 cm data, the estimated 1987 0-20 cm SOC
value was therefore increased by 37% to give the
2001 0-20 cm SOC estimate.
1987 soil data
The 1987 soil data were obtained from
Scott and Wood (1989) who reported SOM at three
depths (0-5, 5-10 and 10-15cm), which was
estimated from soil C determined using the
Walkley-Black method (Hesse, 1971) along with
bulk density (BD) for the same depths. The SOM
data were used to calculate accumulated SOC mass
with profile depth down to 15 cm and it was
assumed that SOC is 58 % of SOM (Rowell, 1994;
Landon, 1991). The Century model estimates SOC
for the top 20cm of soil. In order to compare
modelled and measured data, it was therefore
necessary to extrapolate measured data to the 020cm depth interval. To extend the 0-15cm data to
a total for the 0-20cm depth, a regression was
performed for each of the fields (A,B,C and D)
using measured data from the three different depths
(0-5, 5-10 and 10-15cm) and accounting for the
fact that 0 g SOC m-2 would occur at 0 cm depth.
Regression relationships were determined and these
were then used to calculate SOC in the 0-20cm
depth interval (Fig. 3).
1987 and 2001 combined soil data
The 1987 and 2001 data were then
combined to form the chronosequence data set
(Fig. 4) against which the modelled data were
compared. For the native prairie (0 years under
cultivation, Field A), the 1987 data were used.
Climate data
Maximum and minimum air temperatures
and monthly precipitation were obtained from the
Global Historical Climate Network (GHCN)
weather station in Brinkley, Arkansas. This station
was located approximately 48km to the northeast of
the study site at 34° 88’ N, 91° 18’ W. Complete
data were available from 1945 to 2000.
Cultivation methods
Rice received 168 kg ha-1 N, ~67 kg ha-1
K and ~33 kg ha-1 P and was flood irrigated with a
~18 cm flood depth maintained. Soybeans were
furrow irrigated six times in a growing season.
Fertiliser inputs for wheat were obtained from the
University of Arkansas Cooperative Extension
Service (ACES, 2007). Tillage timing and
implements used for all crops were obtained from
2001 soil data
In 2001, soil samples were collected for
only the top (0-10 cm) (Brye and Slaton, 2003).
Measured bulk densities were used to determine
6000
5000
Priarie
yr1
yr12
yr30
Power
Power
Power
Power
-2
SOC (g m )
4000
0.8808
y = 269.7x
0.6597
y = 399.3x
2
R = 0.9999
2
R = 0.9947
3000
0.8036
y = 211.75x
2000
2
R =1
(yr1)
(Priarie)
(yr12)
(yr30)
0.9505
y = 125.7x
1000
2
R = 0.9998
0
0
5
10
15
20
25
Soil Depth (cm)
Fig. 3. Polynomial relationships between accumulated soil organic carbon (SOC) and soil depth used to
adjust data to a 0-20 cm depth interval.
43
29 December 2008 Electronic Journal of Integrative Biosciences 6(1):41-52.
Special Issue: Soil Quality for a Sustainable Environment (V.S. Green and K.R. Brye, co-editors)
© 2008 by Arkansas State University
-2
SOC (g m )
4500
4000
Measured SOC
3500
Modelled SOC
(a)
3000
2500
2000
1500
1000
500
0
0
10
20
30
40
50
4500
Measured
4000
Modelled (actual month)
Modelled (first month)
3000
-2
SOC (g m )
3500
(b)
2500
2000
1500
1000
500
0
0
10
20
30
40
50
Years Under Cultivation
Fig. 4. Century-modelled versus measured soil organic carbon (SOC) using the single-history approach (a)
and the multiple-history approach (b). For the multiple-history approach, Century output for the
first month of the year and the measurement month of the year are shown.
crop budgets from the University of Arkansas
Cooperative Extension Service (ACES, 2007).
Century testing methods
Site files and weather files
As stated previously, climate data were
obtained from the Brinkley, AR, weather station.
This was used to create the Century weather (.wth)
file. A site file for Prairie County was created using
site information (e.g., latitude and longitude, soil
texture, soil depth etc.) from the published
literature (Scott and Wood, 1989; Brye and Slaton,
2003; Brye et al. 2004a). Average weather
information, which forms part of the site file, was
generated from the Brinkley weather file. Sand,
Yield data
Actual crop yields from the three rowcrop agricultural fields were not available.
Therefore, average annual yield data for Prairie
County, AR were obtained from the National
Agricultural Statistics Service (NASS, 2007). Rice
yields were available from 1957 to 2001. Earlyseason soybean yields were available from 1978 to
2001. Wheat yields were available from 1961 to
2001 and late-season soybean from 1984 to 2001.
44
Milne et al., 2008
1835 modelled a time when there were a few
settlers who increased grazing in the area. The
second block, 1837 to 1886, modelled a time
period when there was a marked increase in the
human population. Grazing was reduced, but native
grasslands were harvested for hay each year in
October. Blocks 1 and 2 both used mean weather
from the site file. The third time block, 1887 to
1956 was the same as Block 2, but used actual
weather from the weather (.wth) file.
silt, and clay percentages were obtained from Brye
and Slaton (2003) for the native prairie field (A).
An average BD for the three measured depth
intervals (0-5, 5-10 and 10-15) was obtained from
Scott and Wood (1989). Soil pH data were also
obtained from Scott and Wood (1989).
Anaerobic
conditions
cause
decomposition to decrease. The Century model
determines soil decomposition rates based on
temperature and moisture and an estimate of the
soil aerobic conditions. Poorly drained soils can
meet the conditions for anaerobic decomposition,
which decreases belowground decomposition by a
multiplier factor (ANERB). The Century model's
soil drainage factor (DRAIN, in the site.100 input
file) allows a soil to have differing degrees of
wetness (e.g., DRAIN=1 for a well-drained soil to
DRAIN=0 for a poorly drained soil). In this study,
drainage was set to moderate, modelling normally
aerobic decomposition. An update to Century
allows management events to change the drainage
parameter (DRAIN) or restore the value back to the
value defined at the start of the simulation. To
better simulate the slower decomposition
conditions of flooded-rice systems, the soil was
temporarily set to poorly drained in the event file
for the periods when the fields were under flooded
rice. This approach has been used previously to
simulate the reduced decomposition conditions in
flooded-rice systems (Bhattacharyya et al., 2007).
In tests, it was determined that the ANERB
multiplier reduced decomposition rates by, on
average, 50% during a simulated rice growing
season. Testing with the anaerobic conditions
during rice growth produced a better fit to
measured data than simulations where the
parameter was held constant at a more easily
drained setting. While the fit to measured data was
improved, the difference was not large, reflecting
the short duration of flooding during the multi-crop
rotation.
Chronosequence approach
Between 1957 and 2001, general cropping
practices (i.e., fertilizer inputs, cultivars, irrigation
practices etc.) changed. It was therefore likely that
these changes were also seen in the four fields
considered here. The initial idea was to carry out
one model run accounting for changes in cropping
practices through time and to compare SOC output
with measured data from the fields for
corresponding points in time. However, it became
apparent that this would be inappropriate as this
approach would not compare similar outputs. For
example, measured SOC for fields cultivated for
one year using modern inputs would be compared
with modelled SOC based on 1957 inputs (Table
1).
Table 1. Measured verses modelled time periods
using a standard chronosequence approach.
Sample
year
Years
cultivated
Actual
cultivation
period
Simulated
cultivation
period
A
1987
0
0
0*
A1
2001
0
0
B
1987
1
1986-1987
1957-1958
B1
2001
15
1986-2001
1957-1972
C
1987
12
1975-1987
1957-1969
C1
2001
26
1975-2001
1957-1983
Field
code
Equilibrium and base
The Century equilibrium file simulates a
grassland growing a 75% warm-season mixed grass
with low-intensity grazing which has a moderate
(linear) effect on production. Fire frequency for
this tallgrass prairie was set at once every three
years. The equilibrium file was set to run for
10,000 years to ensure dynamic, steady-state
conditions were achieved prior to the time period
of interest.
The base period from initial settlement to
1956 was created using information from a Prairie
County
historical
documents
website
(Couchgenweb, 2007). This website gave a
description of Prairie County, AR in the 1880’s to
early 1900’s from a resident of the county at the
time. Using this information, the base period was
divided into three blocks. The first block, 1810 to
D
1987
30
1957-1987
1957-1987
D1
2001
44
1957-2001
1957-2001
* 1987 A represents 1957 with no cultivation.
Two options were identified; 1) to model
the experimental period using a single history with
modern fertilizer inputs throughout and to compare
this modelled output with measured results or 2) to
model each field individually using changing
fertiliser inputs over time and to take model output
from the years 1987 and 2001 for each of the four
model runs to compare with the measured data.
Both approaches were taken and are described
below in more detail.
45
Electronic Journal of Integrative Biosciences 6(1):41-52.
Special Issue: Soil Quality for a Sustainable Environment (V.S. Green and K.R. Brye, co-editors)
29 December 2008 © 2008 by Arkansas State University
throughout the entire experimental period to reflect
the actual situation.
Model runs for Field D were carried out
and choice of crop cultivar changed to match crop
productivity with average productivity from NASS
statistics for Prairie County (NASS, 2007). The
soybean cultivar option chosen for Field D was
SYBN1, an older soybean option in Century with a
low HI for the first block, SYBN2, a soybean with
a higher HI, for the second block, and a
combination of SYBN2 and SYBN, a high yielding
modern soybean, for the last block. Wheat files
used were W1 (low yielding) for the first block and
W3 (high yielding) for the second and third blocks.
For rice, only two options were available in
Century, upland and lowland rice where the
lowland rice option (RICL) was the appropriate
option for flooded rice. However, using RICL,
modelled rice yields were much greater than those
given by NASS statistics. Upon examination of the
modelled HI results for rice, it was determined that
the maximum HI (0.57) was being maintained
throughout the model period. The literature
suggests that the first tall and traditional rice
cultivars had a harvest index of around 0.3 (Ottis,
2004). This was increased to around 0.4 in 1969
with the introduction of short-statured, highyielding varieties (Khush, 2004). Therefore a new
rice cultivar option was created to represent a
slightly older rice cultivar with a lower maximum
HI (0.4) called RICL1. When this was used in the
event file for Field D, rice productivity better
matched the Prairie County NASS statistics
(NASS, 2007).
Fields C and B were cultivated from 1974
and 1986 onwards, respectively. Information from
the event file for Field D was used to create event
files for Fields B and C. With this approach,
fertilizer inputs, tillage practices, and choice of
crop were tailored to reflect historical changes.
Single-history chronosequence approach – SOC
and yield
For the single-history chronosequence
approach, one event file was created, which used
the same fertilizer inputs and tillage operations
throughout the entire cropped period from 1956 to
2001. The year before cultivation started, land
preparation with a mouldboard plough was
simulated using a heavy tillage event (i.e., cult K
from the cult.100 file). The crop rotation described
in Figure 2 was then simulated. Rice received two
nitrogen fertilizer applications per year, one before
flooding (10 g m-2) and one at mid-season (6.6 g m2
). Rice grown was the standard lowland rice option
from the crop.100 file in Century, which has a
maximum harvest index (HI) of 0.57. Tillage
information from the Arkansas Cooperative
Extension Service (ACES, 2007) crop budgets was
used to determine Century’s equivalent tillage
types for all crops. Soybean received no fertiliser
inputs. A modern, high-yielding soybean variety
(Century’s SYBN) was used for both early- and
late-season soybean crops. Soybean was row
irrigated six times during the growing season.
Information on fertilizer inputs to wheat was
obtained from ACES (2007). The wheat variety
grown was Century’s W3 option, a high-harvest
index, modern wheat. Wheat received two inputs of
nitrogen fertiliser of 6.6 g m-2 each in February and
March.
One model run was then carried out and
the SOMTC (i.e., total soil C including belowground structural and metabolic C) output was
taken from Century for the years corresponding to
the measured data (e.g., 0, 1, 12, 15, 26, 30 and 44
years after cultivation).
Multiple-history approach – SOC
For the multiple-history approach, four
Century event files were created and four model
runs were carried out, one for each field (A,B,C
and D). Field A was continuously managed prairie
throughout the entire experimental period. From
1957 to 1990, the prairie was modelled as the same
grazing and harvesting regime as the base period,
with a three-year fire interval. According to Brye et
al. (2004a), the prairie management regime
changed between 1991 and 2001 from a three-year
to a one-year burn interval. This was therefore
modelled in the second time block (1991 to 2001)
of the event file for Field A. Field D was
continuously cultivated throughout the whole of the
experimental period (1957 to 2001). For modelling
purposes, the experimental period was split into
three blocks, 1958 to 1969, 1970 to 1984, and 1985
to 2001, each with different fertilizer inputs, tillage
practices, and cultivar choices that reflected likely
farming practices during each time interval.
However, the same rotation was maintained
Statistical analysis (Modeval)
A quantitative comparison of measured
and modelled SOC, using the SOMTC output from
Century, was constructed using Modeval (Smith et
al., 1996). The sample correlation coefficient (r),
the root mean square of error (RMSE), the mean
difference between measured and simulated results
(M), and the coefficient of residual mass (CRM)
were determined.
RESULTS
Trends in the measured data
The seven measured data points (Fig. 4)
have to be considered in terms of four different
fields, which had been under cultivation for
different durations when measured at the two
sampling times. The seven measured data points do
not represent a typical chronosequence. However,
46
Milne et al., 2008
Table 2. Quantitative statistical analysis of modelled
versus measured soil organic carbon (SOC)
data for a rice-soybean-wheat-soybean
rotation in Prairie County, Arkansas.
in lieu of suitable long-term rice experiments, it
was useful to combine them to consider cultivation
effects on SOC after different lengths of
cultivation. One year after cultivation, SOC
increased considerably. Some increase would be
expected with the initial disturbance caused by
plough-out of the native prairie and the resultant
large one-time input of C that would come from
dead root and shoot material. However, the
increase appeared larger than expected (30 %) and
might, therefore, be partially attributable to
different sampling schemes used to obtain the 1987
and 2001 data. Soil organic carbon then declined
between 1 and 12 years after cultivation. Between
12 and 30 years of cultivation, SOC remained
relatively stable. The year 44 value showed a slight
increase from the year 30 value. This could be due
to inter-annual variability or may be attributable to
differences in analysis. Although the same method
(Walkley Black) was used in both instances to
measure SOC, there was a 14 year gap between
sampling times (Fig. 4).
Single
history
(1st month)
0.94
12
229
0.09
Multiple
history
(1st month)
0.92
13
193
0.07
Multiple
history
(measurement
month)
0.93
15
233
0.09
Statistic*
r
RMSE
M
CRM
* Sample correlation coefficient (r), the root mean square of error
(RMSE), the mean difference between measurements and
simulation (M) and the coefficient of residual mass (CRM).
showed a tendency to under-estimate the measured
data (Table 2).
As mentioned previously, no data for crop
yield were available from the actual fields in this
study. Therefore, average crop yields for Prairie
County, AR were obtained from NASS (2007). For
the single-history approach, it made sense to
compare modelled yield data from 1987 and 2001
(i.e., the sample years) with NASS data from 1987
and 2001 only. Fertilizer inputs and tillage
practices were not representative of those used in
the years prior to this. Table 3 shows modelled
grain-C (i.e., carbon content of the grain) and
grain-C calculated from NASS statistics for Prairie
County for 1987 and 2001.
Measured versus modelled SOC: Single-history
approach
Using the single-history approach, taking
output data from the first month of each year,
Century was able to simulate SOC in the four fields
quite well when SOMTC was plotted against
measured data (Fig. 4a). Century simulated the
initial rise in SOC in the first year immediately
following cultivation, although a smaller increase
was calculated by Century than in the actual field
data. Modelled SOC then declined to around 75%
of the initial prairie value and remained at around
this level for the rest of the model run. This
modelled data followed the pattern of the measured
data well for the first 30 years, after which the
measured data increased slightly, while modelled
SOC declined slightly (Fig. 4a).
Statistical evaluation of the simulation
using MODEVAL (Smith et al., 1996) revealed
acceptable model performance for the singlehistory approach (Table 2). A sample correlation
coefficient (r) of 0.94 showed good association
between simulated and measured data. This was the
highest correlation coefficient for either of the
approaches used. Root mean square error (RMSE)
gives a percentage term for the total difference
between predicted and observed values with 0
indicating no difference. For the single-history
approach, the RMSE was 12.03, which was
relatively low. MODEVAL also calculated the
coefficient of residual mass (CRM), which
indicates if the model is consistently under- or
over-estimating the measured data. In this case, the
CRM was positive (0.09) indicating that the model
Table 3. Modelled grain carbon and grain carbon
calculated from NASS (2007) Prairie
County average statistics for 1987 and
2001.
Crop
Rice
1st Soybean
Wheat
2nd Soybean
Grain-C (g m-2)
1987
2001
NASS Modelled*
NASS
Modelled*
215
75
105
54
322
129
125
55
275
96
143
70
267
126
130
62
* Modelled grain-C is for the year nearest to 1987 or 2001
(dependent on the position of the crop in the rotation).
Measured versus modelled SOC: Multiple-history
approach
Based on visual inspection of Figure 4, it
appears that the multiple-history approach did a
poorer job of simulating SOC data from the fields
than the single-history approach. The modelled
data increased at year 12 and then maintained the
same level until year 26, unlike the measured data
which declined during this time. The pattern was
the same irrespective of whether data were taken
from the first or the actual month of measurement.
This slightly poorer fit was confirmed by the
quantitative statistical evaluation of the data (Table
47
29 December 2008 Electronic Journal of Integrative Biosciences 6(1):41-52.
© 2008 by Arkansas State University
Special Issue: Soil Quality for a Sustainable Environment (V.S. Green and K.R. Brye, co-editors)
350
250
-2
C grain (g m )
300
200
150
Average from NASS
Modelled (HI = 0.57)
100
Modelled (HI = 0.4)
50
0
1950
1960
1970
1980
Year
1990
2000
2010
Fig. 5. Century-modelled versus calculated rice grain carbon (C) based on county-wide
estimates from NASS (2007).
This study aimed to test Century’s ability
to model long-term SOC dynamics in a system that
was flooded some of the time. The chronosequence
used in this study was comprised of a rotation that
included flooded rice every third year. This means
that within any 3-yr period, the soil is only flooded,
therefore under anaerobic conditions, for a total of
approximately five months out of the year. If total
decomposition for the 3-yr period is considered, the
amount of decomposition occurring under
anaerobic conditions is likely to be less than 20%.
Under these circumstances, Century appears to
simulate long-term changes in SOC turnover
relatively well. The biggest disagreement between
modelled and measured data came in the first year
immediately following conversion from grassland
to cropland when it is typical to see a large, but
temporary increase in SOC as a result of
decomposing roots and shoots that are ploughed
into the soil (Coleman, 1997). This increase is
difficult to sample for and measure as the increase
depends on the root mass, litter incorporation, root
turnover, the mechanical details of the soil
movement and the resulting soil decomposition.
Therefore, there is a possibility that inopportune
sampling could have led to an over-estimation in
the measured data. If the year 1 data point is
excluded from the analysis, the RMSE improves to
~ 9 for both the single- and multiple-history
approaches, suggesting the model is encountering
more of a problem modelling the initial plough-out
pulse of C following land conversion than it is in
modelling SOC turnover under the intermittently
anaerobic conditions produced by the flooded rice.
A recent modification to Century now
allows the user to change the drainage status for
part of the cropping season, which is particularly
2). The correlation coefficients were comparable (r
= 0.92 for the first month data and 0.93 for the
actual month data). The RMSE was greater (13.3
and 15) indicating a greater percentage difference
between modelled and measured values. The CRM
also indicated that overall Century showed a
tendency to under-estimate the measured data.
Grain-C comparison: Multiple-history approach
Measured grain-C calculated from the
NASS county yield statistics were compared with
modelled grain-C taken from the Field D (i.e.,
continuous cultivation) model run. Figure 5 shows
rice grain-C modelled using the default RICL rice
file (HI = 0.57) and the adapted lower HI RICL1
rice file (HI = 0.4). The adapted rice cultivar file
with the lower HI resulted in a better fit with the
county average yield data.
Figures 6 shows modelled versus actual
average grain-C for Prairie County for early-season
soybean, wheat, and late-season soybean. As
mentioned previously, yield data were only
available for some of the experimental period for
certain crops. From the data that were available,
Century modelled grain-C for wheat and soybean
reasonably well (Fig. 6). However, it has to be kept
in mind that the crop data presented was the
average available data for Prairie County and,
therefore, may have differed somewhat from actual
yields on the four fields.
DISCUSSION
Century model performance in simulating SOC
turnover in US crop rotations including flooded
rice
48
Milne et al., 2008
useful for rotations including flooded rice. Olk et
al. (1996) studied the changes in chemical
properties of SOM in four fields under rotations
including flooded rice, which were flooded for
120
(a)
100
80
60
40
Average from
NASS
Modelled
20
0
1960
1970
1980
1990
2000
2010
160
(b)
140
Grain –C (g m-2)
120
100
80
60
40
Average from
NASS
Modelled
20
0
1960
1970
1980
1990
2000
2010
80
70
(c)
60
50
40
30
Average from NASS
20
Modelled
10
0
1960
1970
1980
1990
2000
2010
Year
Fig. 6. Century-modelled versus calculated grain carbon (C) based on county-wide estimates from NASS 2007
for early-season soybean (a), wheat (b) and late-season soybean (c).
49
Electronic Journal of Integrative Biosciences 6(1):41-52.
Special Issue: Soil Quality for a Sustainable Environment (V.S. Green and K.R. Brye, co-editors)
29 December 2008 © 2008 by Arkansas State University
reported here, where the RMSE was 13 for the
single-history approach (Table 2).
In the same study, Bhattacharyya et al.
(2007) modelled a second experiment in the semi
arid area of the IGP. For this trial, rice was grown
in rotation with wheat and rice was irrigated as
needed to prevent the soil surface from being
without overlying water for more than two days
(Singh et al., 2004), implying that conditions may
have been predominantly anaerobic with
intermittent aerobic conditions during this time. For
this trial, Bhattacharyya et al. (2007) reported
Century was better able to simulate SOC change
than in the humid trial with RMSE values ranging
from 7 to 11. Unlike the study reported here, this
trial had been under cultivation before the longterm experiment began and therefore did not have
to simulate a pulse of SOC following conversion
from pasture. Taking this into account, the RMSE
value of 9 obtained for our study when year 1 was
omitted is in agreement with the second study
modelled by Bhattacharyya et al. (2007).
It appears that Century may be suitable for
estimating long-term SOC dynamics in rotations
that involve flooding once every three years,
however the situation becomes more complicated
for rotations that are flooded annually. More work
needs to be done to determine a threshold, above
which the total time spent in flooding precludes the
use of the Century model for the prediction of longterm SOC dynamics.
different portions of the cropping season. The
fields included a non-flooded rotation and a single-,
double- and triple-flooded rice rotation. Olk et al.
(1996) reported increasing phenolic content of
SOM, particularly the humic acid fraction, with an
increase in the period of time the fields were in
flooded rice. Olk et al. (1996) attributed this
increase to slower decomposition of lignin under
anaerobic conditions and pointed out that the
increased phenolic character of the SOM was
probably affecting N cycling. In Century, the
lignin-to-N ratio is used to partition plant and
animal residues into structural and metabolic pools
with different decay rates. The findings of Olk et
al. (1996) suggested that these decay rates will be
different under anaerobic conditions and may not
be so tightly linked to C:N ratios as they are under
aerobic conditions.
Shibu et al. (2006) provided a
comprehensive review of SOC dynamics and
models with reference to rice-based cropping
systems. They pointed out that in most flooded-rice
fields the soil can be divided into two zones, a
flooded top layer (0-15 cm), which is fully or
partially reduced, and a largely reduced lower
layer. They also state that the single-compartment
approach used by Century (i.e., treating the soil as
a uniform 0-20 cm layer) makes Century less
suitable for modelling flooded systems than models
such as DNDC, which treat the soil as 10 layers
with each layer 5 cm thick. However, Shibu et al.
(2006) also acknowledged that experimental
studies rarely have the information needed for all
the soil layers required by DNDC, especially
reliable data for the top 5 cm.
The issues cited above must be kept in
mind when using Century for rotations including
flooded rice, particularly for rotations which are
flooded more frequently than in the study
considered here. However, Century appeared able
to simulate the 1-rice-year-in-3 rotation considered
in this study reasonably well. This poses the
question of how frequently a predominantly
aerobic soil has to be flooded before long-term
SOC dynamics are changed enough to preclude use
of the Century model for the prediction of longterm SOC dynamics? Bhattacharyya et al. (2007)
evaluated Century against a long-term fertilizer
trial in a humid area of the Indo-Gangetic Plains
(IGP) that involved a rice-wheat-jute rotation, in
which rice was grown under submerged conditions
for more than five months a year every year for 30
years. In contrast to the study reported here, they
reported that Century tended to consistently overestimate SOC in this situation. A quantitative
analysis of the measure of coincidence between
measured and modelled data gave a RMSE of the
model of 19, 36, and 22 for the three trials
considered, indicating a poorer fit than that shown
for modelled and measured data in the study
Using a chronosequence approach for model
validation
Long-term studies that can provide
detailed datasets are the ideal when it comes to the
validation of ecosystem models such as Century.
However, such datasets are rare and all are
inevitably designed and carried out with purposes
other than model validation in mind. In lieu of
long-term studies, several researchers have used
chronosequence data to validate models (BondLamberty et al., 2006). Most commonly, this
approach has been used for studies where native
vegetation, such as forest, has been converted to
managed land, such as pasture, plantations, or
cropland (Cerri et al., 2004; Cerri et al., 2007;
Richards et al., 2007). In such cases, the native
vegetation is assumed to be unmanaged and
therefore problems do not arise from differences in
management practices between the different sites
over time. For example, Cerri et al. (2007) tested
the Century model against 11 different forest-topasture chronosequences. The sites used in each of
these were carefully chosen for similar
management histories in addition to similar soil
properties and proximity. Chronosequences were
comprised of either only well-managed pasture or
only degraded pasture. The oldest sites had been
converted to pasture approximately 26 years
50
Milne et al., 2008
threshold, above which the total time spent in
flooding precludes the use of the Century model for
the prediction of long-term SOC dynamics.
Regarding the use of a chronosequence of
this type for model evaluation, it appears that
extracting data from multiple model runs could be
an advantageous approach when sufficient details
of crop cultivars and management practices are
available. In this study, only county averages were
available and, in terms of SOC, the multiple-history
approach did not perform as well as a singlehistory model run, based solely on modern-day
crop
management
practices.
Further
parameterization of Century for the cultivars and
crop management practices corresponding to the
1950s and 1960s could enhance model
performance and is recommended before a
multiple-history approach is taken.
previous and the authors were fairly confident that
during this time pasture management practices had
changed very little.
Model evaluations, such as the one
reported here, using chronosequences of cropped
sites are less common. As mentioned previously,
problems can arise when long time periods are
considered using this approach as cropping
practices (i.e., cultivars, fertilizer inputs, tillage,
residue returns, etc.) have changed substantially in
the past 50 years. For example, the study described
here reports on the period 1957 to 2001 in the state
of Arkansas. During this time, the use of N
fertilizer increased considerably for both rice and
wheat, new cultivars with higher grain production
and lower straw production were introduced
(Khush, 2004), and tillage practices changed. In
addition, the situation was further complicated by
the fact that the four plots in this study were all
sampled twice (1987 and 2001) and both sets of
measurements were combined to produce a
‘chronosequence’ of fields of varying lengths of
cultivation.
In this study, these factors were accounted
for in the multiple-history approach. However, this
involved extracting data from four different model
runs to piece together a virtual model run and
therefore
did
not
use
a
conventional
chronosequence approach. The SOC output from
the multiple-history approach showed reasonable
agreement with the measured SOC data, but,
surprisingly, it was not as good as the approach
which used modern-day agricultural practices for
the entire time period. The advantage of using the
multiple-history approach is that crop outputs, such
as grain- and biomass-C among others, can also be
extracted from the model runs to evaluate model
performance for the simulation of plant growth.
Figures 5 and 6 showed that the grain-C values
simulated by Century were in reasonable
agreement with average grain-C for Prairie County
over this time period. A comparison with actual
data from the fields, which were unavailable for
this particular study, may have given a different
result.
ACKNOWLEDGEMENTS
The authors would like to acknowledge the Natural
Resource Conservation Service (NRCS) who
provide financial support for the development of
the COMET-VR carbon reporting tool and to the
reviewers of this paper for their valuable
comments.
REFERENCES
Arkansas Cooperative Extension Service (ACES), 2007.
[Online] available at http://www.uaex.edu (verified
16 January, 2007).
Bhattacharyya, T., D.K. Pal, M. Easter, S. Williams, K.
Paustian, E. Milne, P. Chandran, S.K. Ray, C.
Mandal, K. Coleman, P. Falloon, D.S. Powlson, and
K.S. Gajbhiye. 2007. Evaluating the Century model
using long-term fertiliser trials in the Indo-Gangetic
Plains, India. In: E. Milne, D.S. Powlson, and C.E.P.
Cerri (Eds.) Soil carbon stocks at regional scales.
Agric. Ecosys. Environ. 122:73-83.
Bond-Lamberty, B., S.T. Gower, M.L. Goulden, and A.
McMillan. 2006. Simulation of boreal black spruce
chronosequences: Comparison to field measurements
and model evaluation. J. Geophys. Res. 111: Art. No.
G02014.
Brye, K.R., and N.A. Slaton. 2003. Carbon and nitrogen storage
in a Typic Albaqualf as affected by assessment
method. Comm. Soil Sci. Plant Anal. 34:1637-1655.
Brye, K.R., E.E. Gbur, and D.M. Miller. 2004a. Relationships
among soil carbon and physiochemical properties of
a Typic Albaqualf as affected by years under
cultivation. Comm. Soil Sci. Plant Anal. 35:177-192.
Brye, K.R., C.P. West, and E.E. Gbur. 2004b. Soil quality
differences under native tallgrass prarie across a
climosequence in Arkansas. Am. Midl. Nat. 152:214230.
Brye, K.R., and A.L. Pirani. 2005. native soil quality and the
effects of tillage in the Grand Prairie region of
eastern Arkansas. Am. Midl. Nat. 154:28-41.
Cerri, C.E.P., K.P. Paustian, M. Bernoux, R.L. Victoria, J.M.
Mellilo, and C.C. Cerri. 2004. Modelling changes in
soil organic matter in Amazon forest to pasture
conversion using the Century model. Glob. Change
Biol. 10:815-832.
Cerri, C.E.P., M.E. Easter, K.P. Paustian, K. Killian, K.
Coleman, M. Bernoux, P. Falloon, D.S. Powlson, N.
Batjes, E. Milne, and C.C Cerri. 2007. Simulating
CONCLUSIONS
This study demonstrated Century to be a suitable
model for estimating long-term SOC dynamics in
rotations that involve submerged rice once every
three years, under the conditions present in
Arkansas, USA. Therefore, it can be recommended
that similar crop rotations be included in C
inventory tools, such as COMET-VR, which use
the Century model to estimate SOC dynamics. The
literature suggests that the situation is more
complicated when Century is used to estimate
dynamics in rotations that are flooded every year.
Therefore, further study is needed determine a
51
Electronic Journal of Integrative Biosciences 6(1):41-52.
Special Issue: Soil Quality for a Sustainable Environment (V.S. Green and K.R. Brye, co-editors)
SOC changes in 11 land use change chronosequnces
from the Brazilian Amazon with RothC and Century
models. Agric. Ecosys. Environ. 122:46-57.
Coleman, K., D.S. Jenkinson, G.J. Crocker, P.R. Grace, J. Klir,
M. Korschens, P.R. Poulton, P.R., and D.D. Richter.
1997. Simulating trends in soil organic carbon in
long-term experiments using RothC-26.3. Geoderma
81: 29-44.
Fielder, R.T., K.J., Crader, M.A. Simon, and C.R. Wilson. 1981.
Soil survey of Lonoke and Prairie counties,
Arkansas. United States Department of Agriculture,
Soil Conservation Service, Washington D.C.
Hesse, P.R. 1971. A textbook of soil chemical analysis. John
Murray, London. p245.
Landon, J.R. 1991. Booker Tropical Soil Manual, A handbook
for soil survey and agricultural land evaluation in the
tropics and subtropics. Longman Scientific and
Technical, Essex, UK. p139.
Li, C., Y. Zhuang, S. Frolking, J. Galloway, R. Harriss, B.
Moore III, D. Schimel, and X. Wang. 2003.
Modeling soil organic carbon change in croplands of
China. Ecol. Appl. 13:327-336.
National Agricultural Statistics Service (NASS). 2007. [Online]
available at http://www.nass.usda.gov (verified 16
January, 2007).
Olk, D.C., K.G. Cassman, E.W. Randall, P. Kinchesh, L.J.
Sangar, and J.M. Anderson. 1996. Changes in
chemical properties of soil organic matter with
intensified rice cropping in tropical lowland soils.
Eur. J. Soil Sci. 47:293-303.
Ottis, B.V., R.E. Talbert, and A.T. Ellis. 2004. Reducing
seeding rates with modern rice cultivars as a function
of Barnyardgrass control [Online]. Available at:
http://www.arkrice.org/research_results/2004_PDFs/
529_3a.pdf (verified 22 November, 2007).
Parton, W.J., J.W.B. Stewart, and C.V. Cole. 1988. Dynamics of
C, N, P and S in grassland soils: a model.
Biogeochem. 5:109-131.
Richards, A.E., R.C. Dalal, and S. Schmidt, 2007. Soil carbon
turnover and sequestration in native subtropical tree
plantations. Soil Biol. Biochem. 39:2078 – 2090.
Rowell, D.L. 1994. Soil Science Methods and Applications.
29 December 2008 © 2008 by Arkansas State University
Longman Ltd, Edinburgh, UK. p50.
Scott, H.D., and L.S. Wood. 1989. Impact of crop production on
the physical status of a Typic Albaqualf. Soil Sci.
Soc. Am. J. 53:1819-1825.
Shibu, M.E., P.A. Leffelaar, H. van Keulen, and P.K. Aggrawal.
2006. Quantitative description of soil organic matter
dynamics – A review of approaches with reference to
rice based cropping systems. Geoderma 137:1-18.
Shirato, Y. 2005. Testing the suitability of the DNDC model for
simulating long-term soil organic carbon dynamics in
Japanese paddy soils. Soil Sci. Plant Nutr. 51:183192.
Singh, Y., B. Singh, J.K. Ladha, C.S. Khind, R.K. Gupta, O.P.
Meelu, E. Pasuquin. 2004. Long-term effects of
organic inputs on yield and soil fertility in the ricewheat rotation. Soil Sci. Soc. Am. J. 68: 845-853.
Smith, P., J.U. Smith, and T.M. Addiscott, 1996 Quantitative
methods to evaluate and compare Soil Organic
Matter (SOM) Models. Pp. 183-202. In: D.S.
Powlson, P. Smith, and J.U. Smith (Eds.) Evaluation
of Soil Organic Matter Models. NATO ASI Series,
Vol. 138.
United States Department of Agriculture (USDA), 2007a. The
Voluntary Reporting of Greenhouse Gases-CarbOn
Management Evaluation Tool (COMET-VR).
[Online]. Available at
http://www.cometvr.colostate.edu/ (verified 22
November, 2007).
United States Department of Agriculture (USDA), 2007b.
Briefing Rooms/Rice. [Online]. Available at
http://www.ers.usda.gov/Briefing/Rice/ (verified 22
November, 2007].
United States Department of Agriculture (USDA), Natural
Resource Conservation Service (NRCS), Soil Survey
Division (SSD). 2007. Soil series name search
[Online]. Available at
http://ortho.ftw.nrcs.usda.gov/cgibin/osd/osdnamequery.cgi (verified 10 December,
2007).
USA Rice Federation 2006. [Online] available at
http://www.usarice.de/usarice_en/rice_basics/rice_re
gions (verified 16 January, 2007).
52