Upload
vineet-yadav
View
217
Download
3
Embed Size (px)
Citation preview
*
01
do
Spatially explicit historical land use land cover and soil
organic carbon transformations in Southern Illinois
Vineet Yadav *, George Malanson
Department of Geography, 316 Jessup Hall, The University of Iowa, Iowa City, IA 52242, USA
Received 12 April 2007; received in revised form 7 July 2007; accepted 19 July 2007
Available online 14 September 2007
www.elsevier.com/locate/agee
Agriculture, Ecosystems and Environment 123 (2008) 280–292
Abstract
Soil organic carbon (SOC) is a biophysical parameter, which is also directly linked to above ground land use and land cover (LULC).
Currently changes in LULC and variations in SOC often are studied and modeled separately. However, both are conjoined and should be seen
as part of a cascading ecosystem framework. This is not only true for SOC but also for other biophysical parameters which are governed by
human activities. At a watershed scale, this relationship is exceptionally important and the focus should be towards studying the impact of
LULC change on the levels of SOC in spatially explicit terms. To advance knowledge on this front, we studied transformations of LULC,
erosion and SOC from the start of settled agriculture in a moderate size basin of�9340 ha in Union and Pulaski Counties of Southern Illinois.
The primary objective of this research was to study the evolution of SOC at the regional scale, as a result of historical land use change and
erosion from 1851 to 2005. To model SOC, we used CENTURY 4.0 whereas LULC changes in the area were derived by visually classifying
aerial photographs. Long-term erosion associated with different LULC was computed through Revised Universal Soil Loss Equation
(RUSLE). To simplify the task of carrying out numerous simulations (>5000), the study area was divided into cells of 100 m � 100 m. Since
CENTURY is only vertically spatially explicit, each of these cells was designated as individual locations homogeneous with respect to
different input and output parameters like SOC, erosion and LULC. Validation in the study was performed by aggregating and comparing
CENTURY derived output with SOC estimates given in the Natural Resource Conservation Service (NRCS) Soil Survey Geographic
Database for the year 2000. Our results show a correlation of 0.63 between the simulated and observed SOC estimates. In the study area,
approximately 64% of SOC has been lost since the establishment of European–American settlements. Losses are considerable for the soil
types which had higher initial levels of SOC. However, from the beginning of 1980, the simulations indicate rising carbon sequestration due to
conservational management practices. This assessment is common across all LULC classes considered here. Though comparatively, rates of
recovery of lost SOC are higher for areas which converted from agriculture to forest.
# 2007 Elsevier B.V. All rights reserved.
Keywords: Land use land cover; Soil organic carbon; Erosion; CENTURY
1. Introduction
Modifications of land use and land cover (LULC) have an
important influence on the dynamics of soil organic carbon
(SOC). Insight into these dynamics can only be accom-
plished through modeling, as it is not possible to collect field
measurements of SOC at significant spatial and temporal
detail, to estimate the periodic effects of LULC change and
their associated management options over the landscape
Corresponding author. Tel.: +1 319 335 0151; fax: +1 319 335 2725.
E-mail address: [email protected] (V. Yadav).
67-8809/$ – see front matter # 2007 Elsevier B.V. All rights reserved.
i:10.1016/j.agee.2007.07.010
(Elliott et al., 1996). Temporally, variations in SOC levels
can also be used to understand changes in the fundamental
properties of the ecosystems that are governed by land use or
management practices (Tan et al., 2005). In the last two
decades, many biogeochemical models have been developed
to simulate SOC at various spatial scales. However, process
models based on first order kinetics have gained acceptance
and become the principal mode for studying the transforma-
tions in soil carbon cycling. Prime examples of these models
are CENTURY (Parton et al., 1987, 1988), ROTH C
(Jenkinson et al., 1990; Parshotam, 1996), DAISY (Mueller
et al., 1996; Jensen et al., 1997), CANDY (Franko, 1996;
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292 281
1 The United States is divided by United States Geological Survey in
successively smaller hydrologic units which are classified into four levels
i.e. regions, sub-regions, accounting units, and cataloging units. Each
hydrologic unit is identified by a unique hydrologic unit code (HUC)
consisting of 2–8 digits based on the four levels of classification in the
hydrologic unit system.
Franko et al., 1997), NCSOIL (Molina, 1996) and DNDC
(Li, 1996; Li et al., 1997).
Spatial and temporal variability of SOC has been
extensively studied by both experimental and modeling
approaches (Smith et al., 1997; Bernoux et al., 1998; Six
et al., 1999; Collins et al., 2000; Arrouays et al., 2001).
Though only few studies combine both these dimensions
(Papritz and Webster, 1995a,b) and the ones that do neglect
short and medium range spatial variability. Thus, the impact
of management strategies on soil at both field and landscape
scales, cannot be adequately evaluated (Nieder and Richter,
2000; Walter et al., 2003). The two major lacunae in the
current modeling based SOC studies at the landscape level
are (1) no explicit inclusion of LULC change and (2)
relegation of erosion as a factor of no importance in
determining SOC concentrations. In case of erosion, it is a
lacuna induced mainly by limitations of SOC models though
with respect to LULC change it has happened despite the
availability of explicit information from remote sensing
since 1930’s.
Erosion is the single most important means of horizontal
transfer of SOC and LULC is the primary governor of its
intensity (Yoo et al., 2006). Spatially explicit arrangement of
LULC alters the SOC budget over a landscape in comparison
to non-explicit budgeting of the same. However, changes in
the carbon budget do not result from the spatial representa-
tion of LULC, but by formulating processes responsible for
the flux of carbon in a spatially explicit framework.
Complete spatially explicit modeling of SOC of a basin is
only possible when both horizontal and vertical fluxes
between different LULC are modeled at a particular spatial
resolution. Site-based estimates do not provide any idea
about the horizontal input/output of SOC at any particular
location. This creates the need for merging SOC and erosion
models (Yadav and Malanson, 2007). In the current SOC
models, only CENTURY and EPIC (Williams, 1990)
account for erosion of which only CENTURY has the
capability to represent both the processes of erosion and
deposition. Modeling of the movement of SOC across
landscapes has not been effectively realized in any SOC
model and neither CENTURY nor EPIC are exceptions to
that. This leaves lot to be desired regarding spatially explicit
modeling of SOC over a landscape.
With these considerations in the background, this research
examines the transformations in SOC since the introduction of
agriculture (1851) to the present times (2005)’ in a moderate
size basin in Union and Pulaski Counties of Southern Illinois
(USA). The main objective of this research is to study the
evolution of SOC at the regional scale as a result of historical
land use change and erosion from 1851 to 2005. To fulfill this
objective, the study proposes combining of LULC change
information derived from remote sensing, erosion and SOC
models. The practical implications of this research are far
reaching as it exemplifies a methodology for timely
assessment of spatial SOC variations which has become
more feasible in the current scenario where remote sensing
information is available at more frequent intervals. In this
research, LULC change information was explicitly derived
from aerial photographs and the associated erosion was
computed through Revised Universal Soil Loss Equation
(RUSLE). All this information was entered into CENTURY
ecosystem model and parameterized accordingly to obtain a
spatially explicit time series of SOC changes. The CENTURY
model was chosen because it has been successfully applied to
simulate SOC levels in various ecosystems and it allows the
user to input different crop management practices and erosion
amounts on monthly basis. Validation in the study was
performed by comparing levels and measuring strength of
association between the estimated SOC of soil groups given in
the Soil Survey Geographic Database (SSURGO) and
CENTURY derived output.
2. Study area
The study area is located in the Cache River basin
(Hydrologic Unit Code no. 07140108)1 in Union and Pulaski
Counties of Southern Illinois (Fig. 1) and covers an area of
�9340 ha.
The long-term records (1931–2005) from the nearest
climatic station indicate that the mean annual precipitation
in the region is around 122 cm with average monthly
temperatures reaching �4 8C in January and 32 8C in July.
Elevation in the area ranges from 50 to 195 m and the slope
of the land is from northwest to southeast and varies from 0
to 26%. The region is drained by the Big Creek River which
remains a major contributor of the sediment load at the point
of its confluence with the Lower Cache River. In the last two
centuries, conversion of forested land to cropland has
considerably altered the hydrologic response of the basin.
Net erosion from the cultivated cropland in the Big Creek
basin is approximately 16.35 Mg ha�1 year�1 and nearly
2207 ha are classified as highly erodible land (Guetersloh,
2001). Such is the impact of erosion and land use change that
identical soil types found on same slopes vary in their SOC
content by �1–1.5 kg C m�2 an example of which is shown
in Fig. 2 (NRCS, 2005).
With respect to LULC, in 2005, 79% of the total study
area was either devoted to cropland or pastures. Another
18% was classified as deciduous forest and the remaining
3%, was devoted to fulfill urban and transportation
requirements (USDA – National Agricultural Statistics
Service, 2005). Spatially most of the eroded and severely
eroded soils lie in the upland areas because deforestation in
the 19th and early 20th century was more pronounced in
these regions (Duram et al., 2004). In contrast, bottomlands
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292282
Fig. 1. Location of the study area.
are fertile but flooding remains a major concern. Soils in the
bottomlands are classified as occasionally and frequently
flooded and as expected have more SOC than upland soils.
3. Data sources
Multiple sources are employed in this research to derive
parameters for CENTURY simulations. Climate data of
Fig. 2. Impact of erosion and LULC change on SOC of soils found on same
slopes.
Anna (National Climatic Data Center Station No. 110187,
latitude 378 280 and longitude �898150, State: Illinois,
County: Union) from 1939 to 2005 is used to estimate
monthly temperature and precipitation. The crop rotation
sequence is generated by utilizing yield and acreage
statistics of Union and Pulaski Counties, available from
the Decennial (1851–1920) and 4 or 5 yearly Agricultural
Censuses (1925 onwards) of USA (U.S. Census, 1850–
2002). The type, the timing and the frequency of tillage
operations i.e. primary tillage, planting, cultivation and
harvest were determined from an Environmental Protection
Agency (EPA) report (Donigian et al., 1994) and by
researching annual crop residue surveys of Conservation
Technology Information Center (CTIC, 1994–2003). Nat-
ural Resources Conservation Service (NRCS) SSURGO
forms the basis for computing soil parameters (Soil Survey
Staff, 2000a,b). A digital Elevation Model (DEM) of 1/3 arc
seconds (�10 m spatial resolution) available from the U.S.
Geological Survey (USGS) Seamless Data Distribution
System (USGS, 1999) is used to derive length–slope factor
of RUSLE. Atmospheric Nitrogen fixation variables after
1851 in CENTURY are initialized by using deposition data
from Dixon Springs agricultural center located in the Pope
County in Southern Illinois (latitude 378 260 and longitude
�888 400) and lastly, aerial photographs acquired by
Agricultural Adjustment Administration in 1938 (AAI,
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292 283
1938) and by United States Department of Agriculture
(USDA) under National Agriculture Imagery Program
(NAIP) in 2005 (USDA-FSA, 2005) are used to visually
identify type and location of LULC change.
4. Methodology
To simplify the task of carrying out numerous simulations
the study region was divided into cells of 100 m � 100 m
(1 ha) though at edges the area of individual cells was lesser
than this norm. Since CENTURY is only vertically spatially
explicit, each of these cells was designated as individual
locations homogeneous with respect to different parameters.
Two primary steps of (1) determining LULC state of each
1 ha cell from 1851–2005 and (2) erosion associated with
these LULC states were completed before simulating SOC
changes from CENTURY.
4.1. Land use change and agriculture rotation
sequences
European–American settlements in the region started
appearing by the end of the first decade of 19th century,
however the population and the farmland of the counties in
which the study area is located did not increase substantially
before 1850 (Fig. 3).
The area was under native vegetation i.e. Oak-Hickory
forest before 1851, which also became the initial land cover
state. No secondary information is available on LULC change
from 1851 to 1938 when aerial photographs were acquired by
AAI at the scales ranging from 1:21,000 to 1:22,000. Ten of
these photos were combined together to create an uncon-
trolled photo mosaic. These aerial photographs became the
determinants of second LULC state. State of the scanned
aerial photographs prevented creation of a controlled or semi
controlled photo mosaic within a researchable time-period,
Fig. 3. Change in population (absolute numbers) and percent farmland (of
total land area) of Union and Pulaski Counties from 1840 to 2000.
making visual classification a necessity to derive a LULC
map. Lastly, NAIP aerial photographs of 2005 with a ground
sample distance of 2 m became the determinants of the final
LULC state. As the aerial photographs had better spatial
resolution than 1 ha cells, a majority rule was used to classify
LULC at both time states of 1938 and 2005. Three LULC
classes were identifiable from the aerial photographs of 1938
these were forest, agriculture and urban and transportation.
However, as urban and transportation LULC were not in
majority within 1 ha cells, they became part of forested or
agricultural classification groups. Moreover, it was assumed
that cells which were agricultural or forested in 1938,
remained in their respective land classes since 1851. Cropland
was not separately determinable from pasture in 1938 and
agricultural trends in the region shows that Conservation
Reserve Program (introduced in 1985) or pasture became
major land uses after 1990’s. Pasture and crops sequester
different amounts of carbon with higher levels occurring
under the former due to minimal erosion and year-round
supply of litter for decomposition. The difference in the levels
of SOC under both these land uses however becomes
statistically significant only after more than a decade of
continuous practice (Paustian et al., 1997). Hence, for
simulating SOC, cropland and pasture categories were
merged together in a single class of agriculture until the
end of the study time period. In 2005, other than agriculture
and forest there were few areas at 1 ha resolution which had
urban and transportation as a majority LULC, rather than
excluding these cells they were grouped into agriculture or
forested categories on the basis of nearest neighbor rule. Thus,
we ended up with four LULC classes from 1851 to 2005.
These four classes were (1) areas which were agricultural
from 1851 to 2005 (2) areas which were forested from 1851 to
2005 (3) areas which remained agricultural until 1938 and
forested thereafter and (4) areas which were forested until
1938 and agricultural afterwards. For the purpose of
discussion, these classes from hereafter will be called as
A–A (agricultural from 1851 to 2005), F–F (forested from
1851 to 2005), A–F (agricultural until 1938 and forested from
1938 to 2005) and F–A (forested until 1938 and agricultural
from 1938 to 2005). During the total simulation time period of
154 years, nearly 64.29% of the study region was covered by
A–A class, 11.45% was covered by F–F class, 20.17% was
covered by F–A class and the remaining 4.07% belonged to
the A–F class. Spatial distribution of these classes is shown in
Fig. 4.
Forested conditions in this research were simulated by
using parameters of hardwood forests given in CENTURY.
To model rotation sequence for agricultural conditions, we
substituted space for time and used crop acreage of Union
and Pulaski Counties available from Decennial and
Agricultural Censuses (U.S. Census, 1850–2002) to derive
time periods of different crop types. The proportion of time
devoted to a crop changed according to its acreage reported
in the Census. Statistics from the Census were made
representative of the number of years after which the Census
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292284
Fig. 4. Spatial distribution of the simulation LULC classes.
was conducted. Thus, from 1851 to 1920 the proportion of
time of different crop types changed after every 10 years and
from 1925 this was 4 or 5 years depending on the frequency
of Agricultural Censuses. This process resulted in two tables
of crop time periods for Union and Pulaski Counties. There
was no significant difference between the two tables,
considering which only one single table i.e. Table 1 of Union
County (larger proportion of study area lies in the Union
County) was utilized for constructing the crop rotation
sequence from 1851 to 2002. For the last 3 years of the
simulation time period i.e. from 2003 to 2005, we once more
repeated the rotation sequence of 1999–2002. This was done
as no Agriculture Census has been conducted after 2002 at
the time of this research.
Similar strategy was also used to extract information
regarding yield, manure application, fertilizers, herbicides
and insecticides from the Census. Yield statistics became the
source for determining changes in the crop varieties from
1851 to 2005. Tillage information for part of the rotation
sequence before 1990 remained same as given in the EPA
report (1994). After 1990’s, however, CTIC (1994–2003)
annual crop residue management surveys were utilized for the
same purpose. An important thing to note is that the amount of
organic matter added to the soil in long term is governed by
yield, management practices and accurate representation of
the time periods of different crops and not so much on the
exact combinations in which they are sown. This however
remains true when erosion is not considered, because erosion
in any particular year is determined by the previous year crop
cover type and thus SOC content of the agricultural soils do
get affected by crop combinations. In our research, it is highly
likely that our generated rotation sequence might not exactly
match what was practiced at any particular location but it still
is the best possible average representation for such a long
time-period for crop combinations and erosion computation.
4.2. Erosion
CENTURY does not have an inbuilt functionality to
compute erosion. It however has a facility where a user can
input monthly erosion estimates which are taken into
account while simulating SOC levels. Hence, there is no
direct interaction between CENTURYand an erosion model.
In this study, we used RUSLE to get estimates of erosion
associated with different LULC over 154 years study time
period. In the first step, length–slope (LS) factor was
calculated by using a DEM of the study area. The technique
proposed by Moore and Burch (1986a,b) was employed to
estimate LS factor based on flow accumulation and slope
steepness. In their equation it is given as:
LS ¼�
As
22:31
�m
��
sin b
0:0896
�n
(1)
where m = 0.4–0.6 (0.4 was used) and n = 1.2–1.3 (1.3 was
used), LS = calculated LS factor, As = specific catchment
area and b = slope angle in degrees. Soil erodibility (K)
estimates for different soil types were taken from SSURGO
County database. A mean value of rainfall–runoff erosivity
(R) factor was computed on the basis of annual average
precipitation of Anna from five different methods as pro-
posed by Arnoldous (1977, 1980), Lo et al. (1985), Renard
and Freimund (1994) and Yu and Rosewell (1996). Cover
(C) and Support Practice (P) factors varied according to crop
types, tillage history and rotation sequence and were deter-
mined by using the information and methods described in
the Agricultural Handbook 537 (Wischmeier and Smith,
1978). The enrichment ratio (ER) controls the amount of
organic matter in the eroded sediments. A constant ER of 2
was used in this research based on the results obtained by
Massey and Jackson (1952) who report the ER of organic
matter for soils of silt loam texture (most of the soils are of
this texture in our study area) to be between, 1.8 and 2.3.
Process models of sediment movement are available
which can predict both erosion and deposition but they are
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292 285
Table 1
Time periods of different crops for the Union County (1851–2002)
Year Winter
wheat
Wheat Oats Corn Rye Barley Hay Hay
seed
Soybean 0rchards Sorghums Cotton Small
fruits
Potatoes Fallow Idle Total
1850 0.00 1.40 1.07 5.45 0.00 0.00 0.09 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2.00 10.00
1860 0.00 3.40 0.18 4.06 0.01 0.00 0.35 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2.00 10.00
1870 0.00 2.60 1.03 3.87 0.03 0.00 0.47 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2.00 10.00
1880 0.00 3.67 0.61 3.16 0.00 0.00 0.56 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2.00 10.00
1890 0.00 2.93 0.96 2.74 0.01 0.00 1.33 0.00 0.00 0.00 0.02 0.00 0.00 0.00 2.00 10.00
1900 0.00 3.39 0.45 2.82 0.01 0.00 1.33 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2.00 10.00
1910 0.00 2.26 0.27 3.35 0.01 0.00 2.00 0.00 0.00 0.00 0.00 0.00 0.00 0.10 2.00 10.00
1920 0.00 1.84 0.26 2.96 0.03 0.00 2.83 0.00 0.00 0.00 0.00 0.00 0.00 0.07 2.00 10.00
1925 0.53 0.00 0.14 1.41 0.00 0.00 1.70 0.00 0.00 0.00 0.00 0.05 0.00 0.00 1.16 0.00 5.00
1930 0.47 0.00 0.13 1.40 0.00 0.00 1.22 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.78 0.00 5.00
1935 0.59 0.00 0.09 1.72 0.00 0.00 1.26 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.35 0.00 5.00
1940 0.49 0.00 0.08 1.42 0.00 0.00 1.29 0.00 0.05 0.41 0.00 0.00 0.04 0.00 1.21 0.00 5.00
1945 0.20 0.00 0.11 1.90 0.00 0.00 1.06 0.00 0.41 0.38 0.00 0.00 0.02 0.00 0.92 0.00 5.00
1950 0.39 0.00 0.11 1.56 0.01 0.03 0.95 0.06 0.57 0.29 0.00 0.00 0.00 0.00 1.04 0.00 5.00
1954 0.00 0.30 0.11 1.49 0.02 0.04 0.83 0.10 0.60 0.16 0.00 0.00 0.00 0.00 0.35 0.00 4.00
1959 0.00 0.42 0.04 1.98 0.01 0.05 0.86 0.06 0.75 0.25 0.00 0.00 0.00 0.00 0.57 0.00 5.00
1964 0.00 0.58 0.01 1.80 0.01 0.01 0.94 0.00 0.76 0.21 0.01 0.00 0.00 0.00 0.68 0.00 5.00
1969 0.00 0.57 0.03 1.60 0.01 0.00 0.64 0.00 0.96 0.19 0.09 0.00 0.00 0.00 0.37 0.54 5.00
1974 0.00 0.88 0.01 0.78 0.01 0.00 0.81 0.00 1.80 0.12 0.18 0.00 0.00 0.00 0.05 0.38 5.00
1978 0.00 0.37 0.00 0.71 0.00 0.00 0.72 0.00 1.79 0.10 0.11 0.00 0.00 0.00 0.04 0.15 4.00
1982 0.00 0.77 0.01 0.77 0.00 0.00 0.64 0.00 1.60 0.10 0.00 0.00 0.00 0.00 0.01 0.10 4.00
1987 0.00 0.37 0.00 0.87 0.00 0.00 1.03 0.00 1.83 0.17 0.07 0.00 0.00 0.00 0.11 0.55 5.00
1992 0.00 0.33 0.01 0.98 0.00 0.00 1.13 0.00 1.78 0.14 0.06 0.00 0.00 0.00 0.07 0.50 5.00
1997 0.00 0.24 0.00 0.78 0.00 0.00 1.34 0.00 1.90 0.10 0.00 0.00 0.00 0.00 0.03 0.60 5.00
2002 0.11 0.00 0.00 0.93 0.00 0.00 0.97 0.00 2.45 0.00 0.02 0.00 0.00 0.00 0.01 0.50 5.00
Total 2.67 26.33 5.72 49.61 0.17 0.14 25.37 0.22 14.79 2.61 0.53 0.05 0.06 0.17 25.73 2.83 157.00
not designed for moderate to large basin scale computations.
This became the primary reason for not considering them as
a viable alternative to RUSLE in this study.
4.3. Model simulation framework
To compute initial SOC levels for the year 1851, Oak-
Hickory forest was simulated for 7000 years in CENTURY.
After 7000 years, SOC stabilized and its levels changed little
in time. Erosion for these undisturbed forested conditions was
again based on RUSLE and due to its spatial variability and
effects of drainage the net primary productivity (NPP) ranged
from 300 to 520 g C m�2. Results of these simulations were
also used to define the initial SOC levels of individual
locations for further simulations. After 1851, depending on
the individual site’s LULC history, it was assigned to one of
the four different simulation paths which were A–A, F–F, A–
F, and F–A. Nitrogen fixation was increased for this time
period keeping into consideration the release of atmospheric
nitrogen due to industrial revolution. C factor for erosion
computation was also altered for forested conditions after
1851 to represent disturbance effects.
4.4. Verification and analysis framework
Validation in the study was performed by comparing the
amount of SOC predicted by CENTURY (year 2000) with
those available from SSURGO for the year 2000. As
SSURGO only gives estimates across various soil groups,
results from CENTURY were grouped in these soil
categories. Soil groups for which lesser than 30 individual
pixels were present in the study area were left out of the
validation process. Thus, 27 soil groups which covered
�97% of the total area were utilized for comparative
evaluation. As CENTURY only simulates SOC to a depth of
20 cm, SSURGO estimates of soil organic matter (SOM)
were considered to that depth. To convert SOM into SOC it
was multiplied by 0.58 (Nelson and Sommers, 1982).
Verification by contrasting and correlating simulated output
with the field appraisal of SOC at individual locations was
not deemed suitable in this study because of the discordance
between the simulated and measurement scales. Moreover,
there is enough variability within 1 ha, that it is extremely
difficult to accomplish a decent evaluation of the output on
the basis of model parameterization realized from average
climate, soil, agricultural rotation and three land change
states, as carried out in this study. Furthermore, even if
CENTURY is parameterized to fit the pattern of sampled
SOC locations, it might not accurately represent scenarios at
other individual locations, because of which it was decided
to run all the simulations in the study area from a same set of
initial average parameters and compare it to SSURGO
output.
Other than examining CENTURY’s output for temporal
and spatial changes in SOC, two different frameworks were
built for analysis. In the first one, a comparative evaluation
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292286
Fig. 5. Soil group aggregated scatterplot of the observed (mean SSURGO
SOC) and simulated SOC for the year 2000.
of the simulated and observed SOC levels was performed.
As part of this framework, two best possible linear
regression models were built to determine the strength of
relationship of the observed and simulated SOC values in
different soil groups (for the year 2000) with elevation,
slope, Inverse Topographic Wetness Index (ITWI)2, drai-
nage, erosion class, sand, silt, clay and bulk density.
Elevation, slope and ITWI were computed by combining
their respective values at individual locations under various
soil groups. Drainage, erosion class, sand, silt and clay were
all derived from SSURGO data and bulk density was
calculated on the basis of texture.
In the second analytical framework, we explored the
relationship of simulated SOC with elevation, slope, ITWI,
clay, drainage3 and slope2 for individual locations at every 5-
year interval from 1851 to 2005. These variables were
selected after careful research which led to the elimination
of all other non-contributing topographic and soil attributes.
Due to the heteroskedastic nature of the SOC output, we
examined the influence of both global and local effects with
the help of robust4 linear (White, 1980) and spatial
autoregressive error models (Anselin, 1988). However, as
there was no appreciable difference in the explained
variance by adopting any of the two statistical approaches
we went for simplicity and used robust global ordinary least
squares model for our analysis.
5. Results and discussion
CENTURY is a monthly time step simulation model. In
this research, however, we assessed its output for time series
of changes at 5-year intervals from 1851 to 2005. This also
formed the basis for assessing performance of the regression
models in the second analytical framework.
5.1. Comparison of CENTURY output and SSURGO
data
SSURGO reports a range (low to high) for SOM content
of various soil groups in a County. The average of this range
was compared against CENTURY’s output grouped into soil
categories. The mean SOC according to CENTURY in the
2 ITWI is defined as slope/specific catchment area. It is related to Wetness
Index defined as ln (a/tan b) where a is the local upslope area draining
through a certain point per unit contour length and tan b is the local slope.
The inverse is used to avoid dividing by 0 when slope is 0.3 SSURGO classifies soil types in the study region into four categories
which are rarely flooded, not flooded, occasionally flooded and frequently
flooded. These were numerically coded as 1, 2, 3 and 4, respectively. In this
coding scheme, each successive integer point increase was thought to be
representative of decreasing drainage quality. Adoption of this approach
eliminated the need for creating dummy variables for representing drainage
in the regression model.4 Regression model resistant to the problem of heteroskedasticity was
used for analysis. In this model, assumption of homoskedasticity was
relaxed by applying White’s standard error estimates.
year 2000 was �2.56 kg C m�2 whereas according to
SSURGO this was between �1.0 and �3.0 kg C m�2. The
scatter plot of the mean SSURGO and CENTURY derived
output shown in Fig. 5 clearly exemplifies the nature and
strength of this relationship. Spatial representation of the
mean SSURGO and CENTURY simulated output for the
year 2000 is shown in Fig. 6b and c. Fig. 6a shows DEM of
the study region which if seen together with Fig. 6b and c
indicates the importance of topography in controlling the
levels of SOC.
Statistically, the r between the observed and the
simulated output was 0.63 and root mean square error
(RMSE) stood at �1.0 kg C m�2. The r however increases
to 0.75 and RMSE decreases to 0.59 kg C m�2 if we exclude
an outlier i.e. soil class 1843A. This soil class with 49
observations is listed in Table 2, which gives estimates of
simulated and observed SOC in soil groups. Even after the
elimination of this outlier, the proportional representation of
the study area stands at 96 .5%. Hence, the region of soil type
1843A represents error in our analysis which might have
occurred due to not knowing LULC history of these cells at
much more frequent intervals. Additionally, it is also likely
that better sample representation i.e. inclusion of more than
49 observations may further reduce this error. Soil class
1843A is included in all further analysis as it has more than
30 individual observations and thereby fulfills all our sample
size requirements.
SSURGO categorizes soil types on the basis of erosion. In
our study area, soils were classified as eroded, severely
eroded or were not assigned any erosion designation in
which case they were assumed to be not eroded. As erosion
was specifically modeled in our study scheme, we tested for
the locational difference in the simulated SOC values in the
three erosion classes through one-way ANOVA. The
difference in the locational SOC values for the year 2000
in these erosion regions visible from the categorical boxplot
shown in Fig. 7 was found to be highly significant at
P < 0.001 and the R2 for the same stood at 0.16.
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292 287
Fig. 6. DEM (a) and spatial variations in the observed (Mean SSURGO, b) and estimated SOC (Simulated SOC, c) for the year 2000.
Table 2
Simulated and mean SSURGO SOC for the year 2000 by soil groups
Soil type No. of
observationsa
Simulated
SOC 2000
(C in kg m�2)
Mean SSURGO
SOC 2000
(C in kg m�2)
79D3 1435.00 1.67 1.37
79C3 974.00 1.83 1.37
8333A 835.00 3.36 2.98
214D3 559.00 1.71 1.32
8334A 479.00 4.22 3.13
8331A 474.00 2.84 3.27
214C3 469.00 1.83 1.32
5079C3 430.00 1.96 1.38
477C2 399.00 2.13 1.76
79E2 386.00 2.66 1.77
694F 359.00 2.51 3.33
214B 280.00 2.0 2.2
79D2 277.00 2.47 1.77
79B 273.00 1.97 1.96
79E3 230.00 1.88 1.37
834F 224.00 2.71 2.52
5079B2 155.00 2.34 1.77
79C2 154.00 2.21 1.77
477B 139.00 2.45 1.94
214D2 134.00 2.41 1.69
5079D3 133.00 2.67 1.38
79F 69.00 2.53 1.94
694D2 62.00 2.82 3.24
1843A 49.00 7.8 3.29
8475B 43.00 2.69 2.4
214C2 35.00 2.17 1.69
216D2 30.00 1.51 0.71
a No. of observations indicate number of individual locations i.e. cells in
a particular soil group
Assessment of the relationship of simulated and observed
SOC values with topographic and soil attributes showed that
in case of CENTURY the best determinants were elevation,
drainage, sand, silt and bulk density though elevation was
not significant at P < 0.05. On the other hand, these
determinants for SSURGO were erosion class, drainage,
clay, sand, silt and bulk density however constant and clay
were not significant at P < 0.05. With CENTURY’s best
predictors SSURGO’s R2 stood at 0.83 but elevation was not
significant whereas with SSURGO’s best predictors CEN-
TURY’s R2 stood at 0.90 but constant, clay and erosion class
were not significant. Multicollinearity remains a problem in
Fig. 7. Boxplot of the simulated SOC for the year 2000 by erosion class
(differences in means are significant at P < 0.001).
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292288
Fig. 8. (a and b) LULC class aggregated time series of simulated SOC changes. Individual locations i.e. each 100 m � 100 m cell was grouped and aggregated
by LULC class and their mean was computed at every 5-year interval from 1851 to 2004 to derive these curves.
these regression frameworks (clay, sand silt and bulk density
are correlated) though heteroskedasticity is not but our aim
here was to determine the basic controllers of SOC and not
build the best possible regression model and this analysis
clearly shows that same sort of variables govern the
observed and predicted output. Furthermore, comparative
evaluation and performance of regression models substanti-
ate our approach to compute SOC at a medium scale and also
gives credence to the generated time series of SOC changes
in the study area.
5.2. Time series of SOC changes
Mean SOC in the study area decreased from �6.88 to
�2.51 kg C m�2 from 1851 to 2005. These reductions were
Fig. 9. (a–c) Spatially explicit simulat
�6.95 to �2.03 kg C m�2 for A–A class, �5.61 to
�5.19 kg C m�2 for F–F class, �6.6 to �2.31 kg C m�2
for A–F class and �6.76 to �3.40 kg C m�2 for F–A class.
The continuous time series of these changes are shown
in Fig. 8a and the spatial nature of these changes can be
gleaned from Fig. 9. Certain important facets of the curves in
Fig. 8a are (1) SOC of both mean and A–A land cover
class decreases until the end of 1950’s and stabilizes
thereafter to start increasing around 1980 (2) in F–F class
SOC remains approximately constant and only reduces by
�0.42 kg C m�2 due to the disturbance effects, which leads
to greater erosion in comparison to earlier time periods and
(3) the respective SOC content of A–F and F–A class
increases and reduces substantially due to LULC changes
from 1938 onwards.
ed time series of SOC changes.
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292 289
Fig. 10. Losses and gains of SOC by individual locations. (*) Initial state
SOC for the year 1851 by individual locations i.e. for each 100 m � 100 m
cell is plotted against the difference between the simulated SOC for the year
1939 and 1851. Difference was taken from the year 1939 because firstly this
was the year when LULC changed and secondly because around this time
significant shifts occurred in the agricultural rotation and management
practices in the study area. The curvature in A–A and A–F individual
observations is mainly due to the interaction between drainage and soil type.
(**) SOC estimates by individual locations i.e. for each 100 m � 100 m cell
for the year 1939 is plotted against the difference between the simulated
SOC for the year 2005 and 1939. Again, the curvature in A–A and A–F
individual observations is mainly due to the interaction between drainage
and soil type.
Fig. 11. Timely changes in the variance accounted by topographic and soil
attributes of the simulated SOC. The curves show amount of explanation
gained by adding each variable in the regression model. Thus, for the year
1851 elevation, slope, ITWI, clay, drainage and slope2 together account for
90% of the variance.
The time trajectories of the mean SOC and A–A category
in Fig. 8a are similar because of the higher proportion of the
latter in the study area. Spatially, initial SOC levels in 1851
shown in Fig. 9a were higher in lowlands than in uplands due
to poor drainage and lesser erosion. In 1851, the study area
was covered by F–F class and elevation shown in Fig. 6a and
slope accounted for 76% variance seen in Fig. 9a.
By 1939, substantial reductions in SOC took place
everywhere but were more pronounced for higher SOC
regions on lower slopes under agricultural LULC. Thus,
soils which had higher levels of SOC lost more in
comparison to the soils which had low SOC content. This
relationship becomes apparent once an association is made
between Figs. 6a, 9b and 10a in which the difference
between the levels of SOC in 1939 and 1851 are plotted
against the baseline levels for the year 1851. In F–A and F–F
class losses by 1939 were more for upland regions, as they
suffered from higher erosion, whereas basin areas either
increased or maintained their SOC levels due to increased
availability of atmospheric nitrogen after 1851 and lower
potential for erosion. From 1939 to 2005, a comparison of
Figs. 6a, 9c and 10b reveals that cells with higher initial SOC
levels in A–F class recovered much faster in comparison to
their lower concentration counterparts while regions in the
F–F and A–A class maintained the same pattern as seen from
1851 to 1939, except for the fact that for A–A class the gains
and the losses hovered around 0.
5.3. Non-erosion scenario
The non-erosion scenario represents changes occurring in
SOC levels only due to LULC transformations from 1851 to
2005. A comparison between Fig. 8a and b clearly depicts
that total SOC losses which can be attributed to erosion are
substantial in the region. The mean SOC, if erosion is not
taken into account, decreased from�6.88 to�4.6 kg C m�2
from 1851 to 2005. Levels of SOC also declined for A–A,
A–F and F–A land classes. These levels reduced from�6.95
to �3.84 kg C m�2 for A–A class, �6.6 to �5.6 kg C m�2
for A–F class and �6.33 to �4.49 kg C m�2 for F–A class.
In F–F class, after removing the effects of erosion, levels of
SOC increased from 5.61 to 6.27 kg C m�2 due to the impact
of increased nitrogen fixation in this period. Comparatively,
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292290
Fig. 12. (a–d) Relationship of the topographic and soil attributes with simulated SOC: scatterplot of the simulated and fitted SOC by LULC class for the year
2005.
5 There were only two land classes till 1939 i.e. A and F. However, due to
physical and locational factors the difference in the means of regions or
individual pixel observations which followed the four LULC trajectories
was substantial from 1851 to 1939. Separating LULC classes for building
regression models is extremely useful as it removes the confounding caused
by them.
the mean level of SOC in the year 2005 in case of non-
erosion scenario was higher by 83%. Similarly these levels
were higher by 89% for A–A class, by 21% for F–F class, by
142% for A–F class and by 42% for F–A class.
5.4. Time series of SOC changes and performance of
predictors at individual site locations
Examining the controllers of SOC such as topographic
and soil attributes on the landscape for one time state does
not give any information about the changes in their time
varying explanatory performance. This is important to
understand the transformations in the proportion of variance
accounted by SOC regression models under complex LULC
regimes. The curve of % cumulative sequential sums of
squares (SSE) shown in Fig. 11 clearly depicts that with
timely progression the explained variance of simulated SOC
declines from 90 to 31%.
Why does the explained variance declines with succes-
sive time states? The answer to this question is simple, but
has profound consequences for statistical model building to
predict SOC at any scale. The reduction in regression
coefficients is primarily the result of different SOC time
trajectories of the four LULC classes and as time progresses
confounding increases thereby reducing the explanatory
performance of regression models. This is substantiated by
the ANOVA curve shown in Fig. 11 where it is evident that
the explained SOC variance of LULC class increases until
1939 when the differences in the mean SOC of different
LULC classes is highest as can be seen from Fig. 8a or b.
After 1939, the explained variance in the timely SOC output
accounted by LULC classes declines due to reduction in the
differences in their means5. Both these results imply that
statistical models whether global or local will perform much
better if LULC class is included as a factor and not as a
covariate. Such is the impact that if we analyze the
associations by LULC as shown in Fig. 12 for the year 2005,
than the R2 increases to 0.93 for A–A class, 0.93 for A–F
class, 0.98 for F–A class and 0.97 for F–F class.
Simplified assumptions and model parameterization
make the explained variance here much higher than what
would be observed if associations are developed between
individual site field estimates of SOC and the independent
variables used here. But we expect that similar conclusions
can also be reached from this latter approach to analysis. In
this research LULC was assumed to be constant from 1851
to 1938 and once again from 1938 to 2005. Population of the
study area subsuming counties grew till mid 1930’s and
decreased thereafter as visible from Fig. 3. Similarly, the
total land area devoted to farms stabilized after 1930’s. Thus,
errors in SOC estimation might result from not knowing the
exact time period for LULC change from 1850 to 1938. But
we expect that long time period of simulation from 1851 to
2005 will allow CENTURY to equilibrate removing any
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292 291
minor variations. Still errors are present in the output as
mentioned in Section 5.1 which cannot be removed until
much better information about LULC is available from 1851
to 1938. After 1938, however, we think that our LULC
parameterization is more or less accurate due minor
variations in the population and % total farmland of the
Counties where the study area is located.
6. Conclusion
In this research, a novel approach to examine historical
progression of SOC changes was adopted. Primary
determinants of SOC like erosion and LULC transforma-
tions were combined to obtain its present levels in a spatially
explicit framework. Associations between CENTURY
output and SSURGO estimates of SOC in the study area
substantiate our modeling methodology. The results
obtained in the study suggest that:
1. H
istorical LULC practices have led to considerablereduction in SOC until about 1980. Since then however
the levels of SOC are increasing due to the adoption of
conservational management practices.
2. T
he rate of SOC loss is considerably higher for soils andregions which had higher initial levels of SOC. Though
after 1980, the rate of gain is also higher for these soils
and regions.
3. A
ll the four LULC classes show different trajectories oftemporal SOC transformations due to the effects of
associated erosion, litter availability and decomposition
mechanics.
4. R
egression models positively indicate that topographicand soil attributes control the levels of SOC on the
landscape and statistical models of SOC will perform
much better if different LULC classes are modeled
separately.
5. E
rosion has played a major role in governing the levels ofSOC in the study area and its contribution is only �20%
lesser than SOC losses caused by historical LULC
changes.
Despite the fact that impoverishment of SOC caused by
soil loss can increase the potential for sequestering more
carbon in the eroded soil profiles (Van Oost et al., 2004),
overall in the study area with the current rate of SOC gain it
would take decades of conservational management practices
to make up for the considerable losses of SOC induced since
the introduction of agriculture. Methodologically, in this
study, erosion was separately computed and entered into
CENTURY, which is a cumbersome procedure to couple an
erosion and a SOC model. This also leads to aggregation of
erosion estimates at the time step of the SOC model, creating
hindrances in studying the impacts of extreme events on soil
and SOC loss that are of much shorter duration. Furthermore,
huge disparity has been observed between the presumed
erosion and the measured downstream sediment yield
(Trimble and Crosson, 2000) which means that large volumes
of sediment and by implication SOC remains stored in the
watershed (Smith et al., 2005; van Wesemael et al., 2006).
This increases the uncertainty associated with spatially
explicit modeled estimates of SOC. Hence, the accuracy
achieved in this research though acceptable, can be further
improved by tightly coupling SOC models with process
erosion models that considers both erosion and deposition and
by including LULC information at much more frequent
intervals to account for timely biomass changes.
Acknowledgement
This research is an outcome of the financial assistance
provided under a subcontract to George Malanson, from
NSF grant 0410187 to Christopher Lant.
References
AAI, 1938. 1938–1941 Illinois Historical Aerial Photographs. Illinois
Natural Resources Geospatial Data Clearinghouse.
Anselin, L., 1988. Spatial Econometrics: Methods and Models. Kluwer
Academic Publisher, Dordrecht, Netherlands.
Arnoldous, H.M.J., 1977. Methodology Used to Determine the Maximum
Potential Average Annual Soil Loss Due to Sheet and Rill Erosion in
Morocco. FAO Soils Bulletin No. 34. Food and Agricultural Organiza-
tion, Rome.
Arnoldous, H.M.J., 1980. An approximation of the rainfall factor in the
universal soil loss equation. In: De Boodt, M., Gabriels, D. (Eds.),
Assessment of Erosion. John Wiley and Sons, Chichester, pp. 127–132.
Arrouays, D., Deslais, W., Badeau, V., 2001. The carbon content of topsoil
and its geographical distribution in France. Soil Use Manage. 17, 7–11.
Bernoux, M., Cerri, C.C., Neill, C., de Moraes, J.F.L., 1998. The use of
stable carbon isotopes for estimating soil organic matter turnover rates.
Geoderma 82, 43–58.
Collins, H.P., Elliott, E.T., Paustian, K., Bundy, L.C., Dick, W.A., Huggins,
D.R., Smucker, A.J.M., Paul, E.A., 2000. Soil carbon pools and fluxes in
long-term corn belt agroecosystems. Soil Biol. Biochem. 32, 157–168.
CTIC, 1994–2003. National Crop Residue Management Survey. West
Lafayette, IN.
Donigian Jr., A.S., Barnwell Jr., T.O., Jackson, R.B., Patwardhan, A.S.,
Weinrich, K.B., Rowell, A.L., Chinnaswamy, R.V., Cole, C.V., 1994.
Assessment of Alternative management Practices and Policies Affecting
Soil Carbon in Agroecosystems of the Central United States. United
States Environment Protection Agency, Washington D.C., p.
194 + appendix.
Duram, L.A., Bathgate, J., Ray, C., 2004. A local example of land-use
change: Southern Illinois—1807, 1938, and 1993. Prof. Geogr. 56, 127–
140.
Elliott, E.T., Paustian, K., Frey, S.D., 1996. Modeling the measurable or
measuring the modelable: a hierarchical approach to isolating mean-
ingful soil organic matter fractionations. In: Powlson, D.S., Smith, P.,
Smith, J.U. (Eds.), Evaluation of Soil Organic Matter Models: Using
Existing Long-term Datasets. Springer-Verlag, Berlin and Heidelberg,
pp. 161–179.
Franko, U., 1996. Modelling approaches of soil organic matter turnover
within the CANDY system. In: Powlson, D.S., Smith, P., Smith, J.U.
(Eds.), Evaluation of Soil Organic Matter Models Using Existing Long-
term Datasets. Springer-Verlag, Berlin and Heidelberg, pp. 247–254.
V. Yadav, G. Malanson / Agriculture, Ecosystems and Environment 123 (2008) 280–292292
Franko, U., Crocker, G.J., Grace, P.R., Klir, J., Korschens, M., Poulton, P.R.,
Richter, D.D., 1997. Simulating trends in soil organic carbon in long-
term experiments using the CANDY model. Geoderma 81, 109–120.
Guetersloh, M., 2001. Big Creek Watershed Restoration Plan: A Component
of the Cache River Watershed Resource Plan. Illinois Department of
Natural Resources, Springfield, IL, p. 52.
Jenkinson, D.S., Andrew, S.P.S., Lynch, J.M., Goss, M.J., Tinker, P.B., 1990.
The turnover of organic carbon and nitrogen in soil and discussion.
Philos. Trans. R. Soc. B 329, 361–368.
Jensen, L.S., Mueller, T., Nielsen, N.E., Hansen, S., Crocker, G.J., Grace,
P.R., Klir, J., Korschens, M., Poulton, P.R., 1997. Simulating trends in
soil organic carbon in long-term experiments using the soil–plant–
atmosphere model DAISY. Geoderma 81, 5–28.
Li, C., 1996. The DNDC model. In: Powlson, D.S., Smith, P., Smith, J.U.
(Eds.), Evaluation of Soil Organic Matter Models Using Existing Long-
term Datasets. Springer-Verlag, Berlin and Heidelberg, pp. 263–267.
Li, C., Frolking, S., Crocker, G.J., Grace, P.R., Klir, J., Korchens, M.,
Poulton, P.R., 1997. Simulating trends in soil organic carbon in long-
term experiments using the DNDC model. Geoderma 81, 45–60.
Lo, A., El-Swaify, S.A., Dangler, E.W., Shinshiro, L., 1985. Effectiveness of
EI30 as an erosivity index in Hawaii. In: El- Swaify, S.A., Molden-
hauerand, W.C., Lo, A. (Eds.), International Conference on Soil erosion
and conservation, Malama Aina 83, Soil Conservation Society of
America, University of Hawaii, Department of Agronomy and Soil
Science, Honolulu, HI, pp. 384–392.
Massey, H.F., Jackson, M.L., 1952. Selective erosion of soil fertility
constituents. Soil. Sci. Soc. Am. Pro. 16, 353–356.
Molina, J.A.E., 1996. Description of the model NCSOIL. In: Powlson,
D.S., Smith, P., Smith, J.U. (Eds.), Evaluation of Soil Organic Matter
Models Using Existing Long-term Datasets. Springer-Verlag, Berlin
and Heidelberg, pp. 269–274.
Moore, I.D., Burch, G.J., 1986a. Physical basis of the length-slope factor in
the universal soil loss equation. Soil Sci. Soc. Am. J. 50, 1294–1298.
Moore, I.D., Burch, G.J., 1986b. Sediment transport capacity of sheet and
rill flow—application of unit stream power theory. Water Resour. Res.
22, 1350–1360.
Mueller, T., Jensen, L.S., Hansen, S., Nielsen, N.E., 1996. Simulating soil
carbon and nitrogen dynamics with the soil–plant–atmosphere system
model DAISY. In: Powlson, D.S., Smith, P., Smith, J.U. (Eds.), Eva-
luation of Soil Organic Matter Models Using Existing Long-term
Datasets. Springer-Verlag, Berlin and Heidelberg, pp. 275–281.
Nelson, D.W., Sommers, L.E., 1982. Total carbon, organic carbon and
organic matter. In: Page, A.L. (Ed.), Methods of Soil Analysis. Part 2.
American Society of Agronomy,, Madison, Wisconsin, USA, pp. 539–
579.
Nieder, R., Richter, J., 2000. C and N accumulation in arable soils of West
Germany and its influence on the environment—developments 1970 to
1998. J. Plant Nutr. Soil. Sc. 163, 65–72.
NRCS, 2005. Soil Survey of Union County Illinois. United States Depart-
ment of Agriculture, Springfield, Illinois, p. 428.
Papritz, A., Webster, R., 1995a. Estimating temporal change in soil mon-
itoring 1. Statistical-theory. Eur. J. Soil Sci. 46, 1–12.
Papritz, A., Webster, R., 1995b. Estimating temporal change in soil mon-
itoring. 2. Sampling from simulated fields. Eur. J. Soil Sci. 46, 13–27.
Parshotam, A., 1996. The Rothamsted soil–carbon turnover model—dis-
crete to continuous form. Ecol. Model. 86, 283–289.
Parton, W.J., Schimel, D.S., Cole, C.V., Ojima, D.S., 1987. Analysis of
factors controlling soil organic-matter levels in great-plains grasslands.
Soil Sci. Soc. Am. J. 51, 1173–1179.
Parton, W.J., Stewart, J.W.B., Cole, C.V., 1988. Dynamics of C, N, P and S
in grassland soils: a model. Biogeochemistry 5, 109–131.
Paustian, K., Collins, H.P., Paul, E.A., 1997. Management controls on soil
carbon. In: Paul, E.A. (Ed.), Soil Organic Matter in Temperate Agroe-
cosystems. CRC Press, Boca Raton, pp. 15–49.
Renard, K.G., Freimund, J.R., 1994. Using monthly precipitation data to
estimate the R-factor in the revised use. J. Hydrol. 157, 287–306.
Six, J., Elliott, E.T., Paustian, K., 1999. Aggregate and soil organic matter
dynamics under conventional and no-tillage systems. Soil Sci. Soc. Am.
J. 63, 1350–1358.
Smith, P., Powlson, D.S., Glendining, M.J., Smith, J.U., 1997. Potential for
carbon sequestration in European soils: preliminary estimates for five
scenarios using results from long-term experiments. Global Change
Biol. 3, 67–79.
Smith, S.V., Sleezer, R.O., Renwick, W.H., Buddemeier, R., 2005. Fates of
eroded soil organic carbon: Mississippi basin case study. Ecol. Appl. 15,
1929–1940.
Soil Survey Staff, 2000a. N.R.C.S., United States Department of Agricul-
ture. Soil Survey Geographic (SSURGO) Database for Pulaski County,
Illinois.
Soil Survey Staff, 2000b. N.R.C.S., United States Department of Agricul-
ture. Soil Survey Geographic (SSURGO) Database for Union County,
Illinois.
Tan, Z.X., Liu, S.G., Johnston, C.A., Loveland, T.R., Tieszen, L.L., Liu, J.X.,
Kurtz, R., 2005. Soil organic carbon dynamics as related to land use
history in the northwestern Great Plains. Global Biogeochem. Cycle 19.
Trimble, S.W., Crosson, P., 2000. Land use – US soil erosion rates – myth
and reality. Science 289, 248–250.
U.S. Census, 1850–2002. Decadal and Agricultural Censuses of United
States. Bureau of Census, Washington, D.C..
USDA-National Agricultural Statistics Service, 2005. Cropland Data Layer
of Illinois, Year: 2005. Available From Illinois Department of Agri-
culture.
USDA-FSA, 2005. Natural Color Orthophoto Mosaics of Union and Pulaski
County. USDA Geospatial Data Gateway.
U.S. Geological Survey (USGS)—EROS Data Center, 1999. Digital Eleva-
tion Model Extracted from National Elevation Dataset (1/3 arc second).
U.S. Geological Survey, Sioux Falls, SD.
Van Oost, K., Govers, G., Quine, T.A., Heckrath, G., 2004. Comment on
‘‘Managing soil carbon’’ (I). Science 305, 1567.
van Wesemael, B., Rambaud, X., Poesen, J., Muligan, M., Cammeraat, E.,
Stevens, A., 2006. Spatial patterns of land degradation and their impacts
on the water balance of rainfed tree crops: a case study in southeast
Spain. Geoderma 133, 43–56.
Walter, C., Rossel, R.A.V., McBratney, A.B., 2003. Spatio-temporal simu-
lation of the field-scale evolution of organic carbon over the landscape.
Soil Sci. Soc. Am. J. 67, 1477–1486.
White, H., 1980. A heteroskedasticity-consistent covariance matrix esti-
mator and a direct test for heteroskedasticity. Econometrica 48, 817–
838.
Williams, J.R., 1990. The Erosion–Productivity Impact Calculator (EPIC)
Model: a case history. Philos. Trans. R. Soc. B 329, 421–428.
Wischmeier, W.H., Smith, D.D., 1978. Predicting Rainfall Erosion
Losses—A Guide to Conservation Planning. Agriculture Handbook
No. 537. U.S. Department of Agriculture, Washington D. C.
Yadav, V., Malanson, G., 2007. Progress in soil organic matter research:
litter decomposition, modelling, monitoring and sequestration. Prog.
Phys. Geogr. 31, 131–154.
Yoo, K., Amundson, R., Heimsath, A.M., Dietrich, W.E., 2006. Spatial
patterns of soil organic carbon on hillslopes: integrating geomorphic
processes and the biological C cycle. Geoderma 130, 47–65.
Yu, B., Rosewell, C.J., 1996. A robust estimator of the R-factor for the
universal soil loss equation. Trans. ASAE 39, 559–561.