Sunteți pe pagina 1din 6

Bulk Density of Soils in Relation to Soil Physical and Chemical Properties

L. A. Manrique* and C. A. Jones


ABSTRACT
Soil scientists and engineers commonly require bulk density (/>
b
)
data as parameter input of simulation models predicting a wide range
of soil processes. Field methods exist to measure D
w
but are labor
intensive and time consuming. Much effort has been recently devoted
to evaluate alternative procedures to predict D
b
from soil physical
and chemical data. Using data of about 12 000 soil pedons from the
continental USA, Hawaii, Puerto Rico, and some foreign countries,
multiple regression relationships were developed to predict D
b
at
3 3 kPa moisture content. The data base was partitioned in small
homogeneous data sets according to their Soil Taxonomy classifi-
cation. Regression relationships between D
b
and soil properties in-
dicated that D
b
can be accurately predicted (R
1
>0.60) from organic-
C content for Inceptisols and Spodosols. Regression relationships
with organic C and clay contents, water content at 1500 kPa, and
the -1500 kPa water to clay ratio gave R
1
values from 0.53 to 0.74
for Ultisols, Alfisols, Vertisols, Oxisols, and Inceptisols. The pre-
dictive capability of the regression models was greatly improved
when the data were partitioned by suborders for all but Aridisols,
Entisols, and Mollisols. The results of this study show that, for soils
grouped according to their taxonomic classification, mnltiple-regres-
L.A. Manrique, 1290-D Maunakea St. 349, Honolulu, HI 96817;
and C.A. Jones, Blackland Res. Ctr., Texas Agric. Exp. Stn., Temple,
TX 76502. Received 26 Feb. 1990. *Corresponding author.
Published in Soil Sci. Soc. Am. J. 55:476-481 (1991).
sion models based on soil properties can provide a relatively accurate
alternative for predicting D
b
.
S
IMULATION MODELS such as the Erosion-Produc-
tivity Impact Calculator (EPIC) (Williams et al.,
1989) require large amounts of soil physical and chem-
ical data. Bulk density (D
b
) is one key property used
in many of the model components of EPIC. Although
Z?
b
measurements are normally available in most soil
data bases, often D
b
values are missing or have been
measured using different procedures. In order to ob-
tain accurate predictions of D
b
, labor-intensive and
time-consuming field measurements are required. It
has become desirable, therefore, to develop alternative
predictive models of D
b
as a function of existing soil
physical and chemical properties.
Bulk density almost invariably increases with soil
depth. This may be partially attributed to the higher
organic-matter content in the surface layer and to til-
lage practices that cause relatively loose structure in
the surface layer and compaction in the subsoil. Over-
all, differences in D
b
among soils are primarily attrib-
uted to differences in particle-size distribution. Several
procedures have been developed to predict Z>
b
based
MAN RIQU E & JONES: BULK DENSITY OF SOILS 477
on soil textural components. Van Wambeke (1974)
predicted D
b
for Oxisols based on sand-size particles.
Shaffer (1988) predicted D
b
as a function of clay con-
tent for soils of Minnesota. Jones (1983) used silt plus
clay contents to predict D
b
for soils with fragipans.
Mbagwu et al. (1983) and Scott and Wood (1989) pre-
dicted D
b
for Ultisols and Alfisols based on organic-
C content, while Alexander (1980) and Huntington et
al. (1989) predicted D
b
for California soils as a function
of the square root or logarithm term of organic-C con-
tent.
The objectives of this study were twofold: (i) to de-
velop statistical procedures for predicting D
b
of soils
as a function of soil physical and chemical properties,
and (ii) to examine the effects of grouping soils ac-
cording to their taxpnomic classification in reducing
the variability associated with D
b
predictions.
MATERIALS AND METHODS
The computerized attribute soils data base assembled at
the Blackland Research Center in Temple, TX (Manrique
et al., 1991), was used to develop predictive models of Z)
b
.
The data base contains information of approximately 12 000
pedons from the continental U.S.A., Hawaii, Puerto Rico,
and some foreign countries. Data contained in the data base
were gathered primarily from the USDA-SCS Soil Survey
Laboratory at Lincoln, NE. In addition to providing the spe-
cific data such as pedon number, date of sampling, soil series
name, location, county, state, and taxonomic classification,
the data base houses morphological, chemical, and physical
data. The laboratory procedures used to measure the soil
physical and chemical properties were described in Soil Con-
servation Service (1972).
Data for Histosols were not included in the statistical pro-
cedures. Histosols are organic soils that have unique soil
properties and behave differently from mineral soils (Soil
Survey Staff, 1975). Using IF-THEN rules (SAS Institute,
1982a), a quality control of data base was performed by
setting arbitrary limits for each soil property. The intent of
these limits was to eliminate incorrect values or those that
were outside the range normally found in agricultural soils.
Two statistical procedures were used to develop the mul-
tiple regression relationships between D
b
at 33 kPa soil
moisture content and soil physical and chemical properties.
The procedure REG (SAS Institute, 1982b) was used for
model analysis to relate D
b
with one or two variables, that
is, organic carbon (OC) and clay (CLAY) contents (Table 1).
The procedure STEPWISE using the STEPWISE routine
(SAS Institute, 1982b) was used for model analysis to relate
Z)
b
to variables such as OC, CLAY, silt (SILT), sand (SAND),
clay + silt (CS), water content at -1500 kPa soil water
potential (WC15), and the ratio WC15/CLAY (WC15C).
Variables such as CS, WC15, and WC15C were included in
Table 1. Variables considered for possible inclusion in the simple
and multiple regression models used in the prediction of bulk den-
sity.
Model Variablesf
1 OC
1A OC, OC
2
, OC
3
, OC"
2
2 CLAY
3 CLAY, SILT, CS
4 CLAY, OC
5 CLAY, SILT, CS, OC
6 OC, OC
2
, OC
3
, OC
1
'
2
, CLAY, CLAY
2
, CLAY
3
, SAND,
SAND
2
, SAND
3
, SILT, SILT
2
, SILT
3
, WC15, WC15
2
,
_________WC15
3
, WC15C, WC15C"
2
_____________
t OC = Organic C content, percent; CLAY = Clay content, percent; SAND
= Sand content, percent; SILT = Silt content, percent; CS = Clay plus
silt contents, percent; WC15 = Water content at -1500 kPa soil water
potential, percent by weight; WC15C = WC15/CLAY.
the regression models because earlier studies have shown
highly significant relationships between these variables and
D
b
(Alexander, 1980; Jones, 1983). In using the stepwise
procedure, only equations with variables that met the 0.05
significance level for entry into a given model were consid-
ered in the D
b
predictions. Also only variables that contrib-
uted more than 5% to the improvement of the overall K*
were included in the regression equations. Variables consid-
ered in each of the models are shown in Table 1.
RESULTS
The soils included in this study varied considerably
in their D
b
(Table 2). Inceptisols, Oxisols, and Spo-
dosols had mean D
b
values of 1.3 Mg nr
3
or smaller.
These soils had high OC content (generally higher than
1%). Clay contents were lower than 27% in all but
Oxisols and Vertisols. Because of the differences in
CLAY, OC, and WC15 among soil orders, the data
were partitioned by orders and suborders (Soil Survey
Staff, 1975). This procedure was tested to see if it
would improve the predictive capabilities of the mul-
tiple regression models.
Modeling Bulk Density
The examination of the multiple regression rela-
tionships between D
b
and soil physical and chemical
properties (Table 3) revealed that D
b
was significantly
(inversely) related to OC or its square root term
(Models 1 and 1 A). The R
2
of the regression equations
indicate that the square root term of OC accounted for
up to 53, 45, and 20% of the variation in D
b
for the A,
B, and C horizons, respectively, when all soils were
pooled together. It is interesting to note that the vari-
Table 2. The means and standard deviations of selected soil prop-
erties.
Propertyt
OC

Mean
SD
CLAY
n
Mean
SD
SILT
n
Mean
SD
SAND
n
Mean
SD
WC15
n
Mean
SD
At
11204
0.6
1.0
11927
24.7
14.8
11936
42.0
21.5
11732
32.1
24.9
10640
11.3
6.2
D
5540
0.5
0.5
5617
20.2
13.3
5616
31.3
17.6
5484
47.3
23.4
5602
10.2
5.7
E
4277
0.7
1.2
4341
16.4
16.0
4336
32.7
23.9
3537
40.4
29.5
4125
8.8
7.2
I
5472
1.5
2.3
5604
17.1
15.7
5601
40.5
17.6
5477
41.1
23.8
5294
11.7
8.6
M
16146
0.9
1.1
16603
26.3
13.2
16605
43.7
18.6
16279
28.7
23.4
16077
12.7
6.0
O
299
1.1
1.4
289
47.8
21.6
289
22.3
12.0
289
29.9
20.6
276
22.8
7.5
S
2133
2.6
2.9
2187
6.1
5.8
2186
32.5
19.8
1882
55.9
19.2
1900
9.5
8.6
U
5382
0.6
1.2
5687
26.4
16.8
5694
31.7
19.7
5533
40.3
23.9
4870
12.0
7.6
V
1416
0.8
0.7
1441
51.4
12.8
1440
35.3
10.2
1438
13.1
13.4
1249
22.0
6.2
n
Mean
SD
4541
1.5
0.2
2095
1.4
0.2
1289
1.4
0.2
2192
1.3
0.4
5178
1.4
0.2
173
1.2
0.2
595
1.3
0.4
2206
1.5
0.2
711
1.3
0.1
t OC = Organic C content, percent; CLAY = Clay content, percent; SAND =
Sand content, percent; SILT = Silt content, percent; WC15 = Water content
at -1500 kPa soil water potential, percent by weight; A, = bulk density, Mg
nr
3
.
t A = Alfisols, D = Aridisols, E = Entisols, I = Inceptisols, M = Mollisols,
O = Oxisols, S = Spodosols, U = Ultisols, V = Vertisols.
n = number of observations.
478 SOIL SCI. SOC. AM. J., VOL. 55, MARCH-APRIL 1991
Table 3. General regression equations expressing the relationship between bulk density and soil physical and chemical properties.
Model
1
1A
6
1
1A
6
1
1A
6
1A
6
bOf
1.510
1.660
1.740
1.509
1.731
, 1.768
1.520
1.702
1.773
1.675
1.767
CLAY OC WC15
All Soils
-1 -0.113
0.005 -0.023
Horizon A
-0.098
-0.011
Horizon B
-0.153
0.005 -0.022
Horizon C
0.005 -0.024
OC"
2
-0.318
-0.218
-0.327
-0.253
-0.409
-0.280
-0.396
-0.263
R
2
0.36
0.41
0.58
0.50
0.53
0.59
0.38
0.45
0.61
0.20
0.40
S y r f
0.20
0.19
0.16
0.18
0.17
0.16
0.19
0.17
0.14
0.23
0.20

19651
19651
19226
5113
5113
5018
10349
10349
10143
4189
4065
t Intercept.
$ Standard error of estimate.
N umber of observations.
11 Blanks in the table represent either variables not considered in the model (see Table 1) or variables that did not meet the criteria for stepwise inclusion in the
model.
Table 4. Regression equations expressing the relationship between bulk density and soil physical and chemical properties. Soil orders.
Model
1A
4
6
5
6
1A
5
6
1A
6
1A
6
1A
5
6
1
1A
1A
4
6
4
6
bO
1.676
1.704
1.832
1.617
1.692
1.587
1.576
1.639
1.720
1.872
1.574
1.677
1.396
1.393
1.876
1.582
1.861
1.678
1.721
1.749
1.661
1.773
CLAY OC WC15 OC"
2
bXf
Alfisols
-H ^0.271
-0.005 -0.127
0.005 -0.027 -0.224
Aridisols
-0.094 -0.003(CS)
0.005 -0.024 -0.199
Entisols
-0.259
-0.087 -0.002(CS)
0.006 -0.022 -0.169
Inceptisols
-0.403
0.007 0.125 -0.022 -0.596
Mollisols
-0.221
0.005 -0.022 -0.174
Oxisols
-0.185
-0.058 -0.004(SILT)
-0.011 -0.119 -0.361(WC15C"
2
)
Spodosols
-0.131
-0.452
Ultisols
-0.288
-0.005 -0.123
-0.245 -0.0005(WC15
2
)
Vertisols
-0.006 -0.061
-0.049 -0.019
/f
2
0.27
0.39
0.57
0.23
0.40
0.27
0.32
0.45
0.60
0.74
0.26
0.43
0.24
0.30
0.63
0.64
0.76
0.29
0.40
0.53
0.31
0.62
S y st
0.17
0.16
0.13
0.17
0.15
0.17
0.17
0.15
0.23
0.19
0.17
0.15
0.17
0.17
0.12
0.25
0.20
0.19
0.18
0.16
0.12
0.08

4296
4258
4213
2015
2007
1258
1238
1239
2145
2068
4970
4915
173
168
156
571
571
2079
2073
2058
654
654
t Regression coefficient for the variable in parentheses.
t Standard error of estimate.
Number of observations.
H Blanks in the table represent either variables not considered in the model (see Table 1) or variables that did not meet the criteria for stepwise inclusion in the
model.
ation in E^ attributed to OC decreased as soil depth
increased. It is likely that texture or other properties
played a more significant role in controlling D
b
as soil
depth increased where OC was a minor component.
Bulk density was poorly related to either CLAY or
CS in Models 2 to 5 (data not shown). In Model 6,
CLAY, WC15, and the square root of OC accounted
for up to 59, 61, and 40% of the variation in Z)
b
in the
A, B, and C horizons, respectively.
Multiple regression relationships between D
b
and soil
MANRIQUE & JONES: BULK DENSITY OF SOILS 479
Table 5. Regression equations (Models 1A and 6) expressing the relationship between bulk density and soil physical and chemical properties.
Soil suborders.
Suborder
Orthox
Ustox
Aquods
Humods
Orthods
Aquults
Humults
Model bO OC OC"
2
WC15 WC15
2
bXf
6
1A
6
6
1A
1A
6
6
1.429
1.471
1.829
1.850
1.660
1.859
1.753
1.783
1.255
-0.120
-0.402
-0.324
-0.455
-0.295
-0.227
-0.011
-0.004
0.0005
-0.466(WC15C"
2
)
0.003(CLAY)
-0.00001(WC15
3
)
-0.329(WC15C)
-0.00001(WC15
3
)
0.003(CLAY)
0.90
0.66
0.62
0.71
0.77
0.93
0.77
0.49
0.72
Aqualfs
Boralfs
U dalfs
Ustalfs
Xeralfs
Argids
Orthids
Aquents
Fluvents
O rthents
Psamment
Andepts
Aquepts
Ochrepts
Tropepts
Umbrepts
Albolls
Aquolls
Borolls
Udolls
Ustolls
Xerolls
Humox
6
1A
6
6
6
1A
6
6
6
1A
6
6
6
6
1A
6
1A
6
1A
6
1A
6
1A
6
6
6
6
6
6
6
1A
6
1.759
1.767
2.137
1.723
1.708
1.681
2.060
1.673
1.689
1.589
1.615
1.608
1.610
1.705
1.238
1.274
1.757
1.939
1.786
1.942
1.424
1.591
1.475
1.736
1.635
1.796
1.619
1.964
1.635
1.713
1.511
2.076
-H -0.217
-0.337
-0.248
-0.212
-0.163
-0.33
-0.228
-0.121
-0.163
-0.291
-0.241
-0.234
-0.234
-0.234
-0.128
-0.389
-0.241
-0.419
0.165 -0.686
-0.239
-0.204
-0.271
0.074 -0.434
-0.110
-0.140
-0.189
-0.149
-0.091
-0.193
-0.270
-0.017
-0.018
-0.016
-0.024
-0.012
-0.019
-0.029
-0.010
-0.022
-0.016
-0.020
-0.010
-0.013
-0.015
-0.063
-0.0006
-0.0006
-0.0004
-0.0003
-0.0004
-0.0003
-0.002
-0.001
0.002
-0.385(WC15C"
2
)
-0.298(WC15C"
2
)
0.00009(CLAY
2
)
-0.226(WC15C"
2
)
O.OOOSfCLAY
2
)
0.009(CLAY)
0.007(CLAY)
0.0001(CLAY
2
)
O.OOS(CLAY)
-0.395(WC15C"
2
)
-0.009(OC
2
)
0.53
0.48
0.70
0.41
0.60
0.42
0.71
0.42
0.35
0.52
0.65
0.21
0.41
0.26
0.50
0.63
0.51
0.71
0.55
0.65
0.47
0.63
0.58
0.79
0.73
0.59
0.37
0.41
0.43
0.49
0.48
0.13
0.14
0.13
0.14
0.11
0.13
0.14
0.14
0.15
0.12
0.12
0.12
0.16
0.18
0.12
0.19
0.14
0.18
0.14
0.17
0.15
0.16
0,13
0.14
0.11
0.13
0.18
0.13
0.13
0.16
0.18
745
601
586
1717
614
551
543
1190
817
183
182
415
524
115
314
277
338
334
1235
1213
140
127
109
108
67
608
984
873
1431
100
30
0.10
0.11
0.09
0.18
0.19
0.05
0.20
0.16
0.11
30
so
41
57
57
18
493
209
209
Udults
Ustults
Xerults
Torrerts
Uderts
listens
Xererts
6
6
1A
6
6
6
6
6
1.756
1.781
1.567
1.859
1.767
1.099
1.820
1.802
-0.246
-0.128
-0.265
-0.237
-0.095
-0.015
-0.013(OC
2
)
-0.0005
-0.020
-0.024
-0.020 -0.039(OC
2
)
0.074 -0.004 0.00006(WC15
3
)
-0.019
-0.020
0.43
0.40
0.60
0.50
0.73
0.59
0.78
0.64
0.62
0.18
0.15
0.16
0.17
0.13
0.10
0.06
0.08
0.10
157
1538
104
52
52
62
81
391
120
t Regression coefficient of the variable in parentheses,
jStandard error of estimate.
Number of observations.
1 Blanks in the table represent either variables not considered in the model (see Table 1) or variables that did not meet the criteria for stepwise inclusion in the
model.
properties for soil orders are provided in Table 4. Or-
ganic C was the best predictor of D
b
for Inceptisols and
Spodosols (Models 1 and 1 A). The R
2
values for Model
1A for these two soil orders were 0.60 and 0.76, re-
spectively. Organic C, however, did not contribute sig-
nificantly to the prediction of D
b
for the other soil orders
(R
2
<0.30).
For soil orders, D
b
was poorly related to CLAY and
CS in Models 2 and 3 (data not shown). Furthermore,
the inclusion of both CLAY and OC in Model 4 (Table
4) did not improve the prediction of D
b
for all but
Alfisols, Ultisols, and Vertisols. The R
2
values of the
regression equations for these three soil orders, however
were <0.40.
For soil orders, the best predictive model for Z)
b
was
a function of CLAY, OC, WC15, and WC15C (Model
6) (Table 4). The R
2
values of the regression equations
suggest that these four variables accounted for > 53%
and up to 74% of the variation in Z>
b
for Ultisols, Al-
fisols, Vertisols, Oxisols, and Inceptisols.
The prediction of D
b
was greatly improved when the
data were partitioned by soil suborders (Table 5). The
best predictive models for D
b
were from Models 1A
and 6. Unless otherwise stated, only regression equa-
tions for these two models are referred to in the fol-
lowing sections.
When D
b
was regressed against OC (Model 1 A), more
than 42% and up to 77% of the variation in D
b
could
be explained by the regression equations for 13 of the
37 suborders included in Table 5. The slopes of the
regressions equations for Aquods, Ochrepts, and Or-
thods indicated that the square root of each 1 g of OC
480 SOIL SCI. SOC. AM. J., VOL. 55, MARCH-APRIL 1991
per 100 g soil resulted in a decrease in D
b
of 0.402,
0.419, and 0.455 Mg irr
3
, respectively. For the other
10 suborders, the square root of each 1 g of OC per 100
g soil resulted in an average decrease in D
b
of about
0.292 Mg nr
3
. Overall, the slopes of the linear regres-
sions between D
b
and OC were different among subor-
ders although mean OC contents were similar for all
but Inceptisols and Spodosols (Table 2). These results
suggest that the kind of organic matter rather than the
total organic matter was more important in the inter-
action of organic matter with the mineral portion to
determine soil structure.
The main contributors to the R
2
values in the regres-
sion equations for Model 6 were WC15, the square root
of OC, and the quadratic term of WC15. These vari-
ables were inversely related to D
b
. The large magnitude
of the regression coefficients for OC suggests that OC
was the main predictor of D
b
for all suborders of In-
ceptisols and Spodosols. Water content at 1500 kPa,
in turn, was the largest contributor to the prediction of
D
b
for Humox and Uderts. The -1500 kPa water to
clay ratio (WC15C), which is indicative of the presence
of amorphous materials in soils (Soil Survey, Staff,
1975), also largely contributed to the prediction of D
b
for Boralfs, UdoUs, Humox, and Ustox. Bulk density
for these soils decreased in proportion to the square
root of WC15C as this ratio increased.
DISCUSSION
The high R
2
values of the regression equations for
Models 1 and 1A suggest that D
b
can be relatively ac-
curately predicted from OC for Inceptisols and Spo-
dosols. These results were consistent with published
relationships between D
b
and OC for Inceptisols and
Spodosols (Curtis and Post, 1964; Soil Survey Staff,
1975; De Coninck, 1980; Huntington et al. 1989). Bulk
density, however, was poorly predicted from OC or its
square root term for the remaining soil orders although
the R
2
of the equations indicated that OC alone ac-
counted for up to 58% of the variation for 10 of 37
suborders included in Table 5. These results differed
from other studies that have shown that OC or its
square root or logarithm term accounted for more than
40% of the variation in D
b
(Mbagwu et al., 1983; Alex-
ander, 1980; Scott and Wood, 1989).
Several studies have shown that clay and clay plus
silt contents contributed to a large extent to the pre-
diction of D
b
(Van Wambeke, 1974; Jones, 1983; Rawls,
1983; Shaffer, 1988). In this study, however, the R
2
values of the regression equations for Models 2 and 3
(data not shown) suggest that the contributions of
CLAY and CS were marginal at best to the prediction
of A> - We do not have a good explanation for these
poor contributions. But a plausible explanation could
be that most published relationships between soil par-
ticle components and D
b
were obtained using small and
homogeneous data sets, while the predictions of D
b
in
this study included large samples with several hundred
of observations in most cases. The large standard de-
viations of the mean D
b
for soil orders (Table 2) clearly
illustrate the difficulties in obtaining meaningful rela-
tionships between D
b
and soil properties. Nevertheless,
the results of this study show that by adding variables
such as water content at 1500 kPa and the 1500
kPa water to clay ration to regression equations con-
taining clay and OC improved significantly the predic-
tive inferences of D
b
(Model 6).
For discussion purposes, the soils included in this
study were grouped according to the R
2
values of their
corresponding regression equations for Model 6 (Table
4). Three groups were arbitrarily established at the order
category: Group 1 (R
2
<0.50), which included Aridi-
sols, Entisols, and Mollisols, Group 2 (R
2
0.50-0.70),
which included Alfisols, Oxisols, Ultisols, and Verti-
spls, and Group 3 (R
2
> 0.70), which included Incep-
tisols and Spodosols. Groups 1, 2, and 3 comprised 33,
44, and 23% of the data base, respectively. When similar
grouping was undertaken at the suborder category (Ta-
ble 5), 12,14, and 11 suborders were included in Groups
1, 2, and 3, respectively, and comprised 32, 38, and
30% of the data base.
The above results indicate that D
b
can be accurately
predicted (R
2
> 0.70) from soil properties for 30% of
the soils when the data were partitioned by suborders
as compared with 23% when the data were partitioned
by orders. A further examination of the results indicate
that D
b
was poorly predicted from soil properties for
Aridisols, Entisols, and Mollisols. Clay, silt, sand, and
OC contents exhibited large variability within Aridi-
sols, Entisols, and Mollisols (Table 2). Presumably the
poor relationships between D
b
and soil properties for
these orders were due in part to the high spatial vari-
ability in both vertical and horizontal dimensions of
soil properties. The partitioning of data by suborders
proved futile for these soil orders, as A> was poorly
predicted from soil properties for most suborders of
Entisols and Mollisols, and for all Aridisols, illustrating
the extensive heterogeneity of these soils.
Overall, the results of this study suggest that, for six
of the nine soil orders included in this study, the regres-
sion relationships developed constitute useful predic-
tive indices for predicting D
b
from existing soil physical
and chemical data and taxonomic classification.
ROLSTON ET AL.: MICROPENETROMETER FOR MEASU RIN G SOIL STRENGTH 481

S-ar putea să vă placă și