Soil-Core Break Method to Estimate Pine Root Distribution

J. A. Escamilla, N. B. Comerford*, and D. G. Neary

ABSTRACT obtained by one of two methods. Either (i) roots visible
Root-distribution studies are tedious and time consuming; there- at each soil cross section! are counted or (ii) roots are
fore, root systems of forest ecosystems are one of the least studied washed from the soil sample. Root data from the form-
components. Yet they are essential in order to assess belowground er are expressed as the number of roots per unit soil
effects of forest-management activities.This study used the soil core- area (N), while the latter are expressed as either root
break method to measure the number of roots crossing a horizontal weight or root length per unit of soil volume (Lv).
unit area (N, expressed in no. cm'2) as an alternative to measuring Comparing the measurement of Lv and N, Lv is the
the more time consuming root-length density (Lv, expressed as cm more precise procedure since roots are totally removed
root length cm~3 soil volume). Soil core samples were taken from the from the soil and measured directly. This procedure
surface horizon of a 7-yr-old slash pine (Pinus elliottii Engelm.) is the most common approach, but it is also one of
stand growing on a Pomona sand (sandy, siliceous, hyperthermic the most time consuming. In comparison, Vepraskas
Ultic Haplaquod). Soil cores, 15.2-cm diameter by 30 cm long, were and Hoyt (1988) showed that measuring N required
extracted from two microsite locations within the check and com- only one-fifth the time required to measure Lv. In
plete-weed-control plots. Root-length density and roots per square addition, N has been used to calculate Lv through ap-
centimeter were measured at 2-cm depth increments. The roots-per- plication of geometrical probability theory leading to
square centimeter measure was a useful predictor of root-length den- the following equation (Melhuish and Lang, 1968):
sity only when treatment means were used, and even then the data
did not fit the theoretical relationship of Lv = 2N. An L\/N slope LY = 2N [1]
of ~1 suggested that roots have a preferential vertical orientation in The coefficient of 2 presumes roots to be randomly
the A horizon. However, depth trends of root-length density and oriented in three-dimensional space (i.e., no prefer-
roots per square centimeter were similar regardless of management ence for vertical or horizontal orientation). Equation
practices or microsite. Roots measured as roots per square centimeter [1] is appropriate to anisotropic root distributions if
was also a less variable measure of root quantity than was root- root density is evaluated in three mutually perpendic-
length density. Although roots per square centimeter can only be ular planes. In the case of nonrandom or preferential
used to calculate root-length density on an empirical basis, the more root orientation, the coefficient will be different than
easily obtained value of roots per square centimeter can be used to 2. Bennie et al. (1977) documented a coefficient of 2
depict root response to belowground competition. for the taproot system of the dicotyledonous cotton
(Gossypium hirsutum L.) and sunflower (Helianthus
annuus L.), while a value of 1.5 for monocotyledonous
R OOT-DISTRIBUTION STUDIES provide information
on the extent, density development, and frame-
work of plant root systems. Yet the procedures used
sorghum (Sorghum bicolor [L.] Moench) indicated a
more vertical orientation of the roots. A similar study
found that Eq. [1] overestimated root-length density
are generally tedious and time consuming. As a result, by factors of 1.31 to 1.89, demonstrating that roots in
root systems are one of the least understood compo- those soils were also more vertically aligned (Drew and
nents of forest ecosystems. Most root investigations Saker, 1980). A recent study documented that roots of
of forest trees in the U.S. Southeast have employed cotton were approximately randomly oriented (Lv/N
some version of the core-break method for sampling = 2.08), while roots of wheat (Triticum aestiyum L.)
the root system (Gholz et al., 1986; Van Rees and and sorghum were more horizontal than vertical (Lv/
Comerford, 1986). W = 3.9 and 7.6, respectively; Bland, 1989). There are
The core-break method consists of retrieving a soil no similar data for the genus Pinus. However, the
core and breaking it at the depth of interest (Bohm, sandy, single-grain texture and light color matrix of
1979). Once the soil core is retrieved, root data are the soils of the Coastal Plain make them ideal for the
J.A. Escamilla and N.B. Comerford, Soil Science Dep.; and D.G. use of the core-break method. If the relationship be-
Neary, Southeast. For. Exp. Stn., U.S. Forest Service, and Soil Sci- tween N and Lv could be documented, root-distri-
ence Dep., Univ. of FL, Gainesville, FL 32611. Contribution of the bution studies in forest ecosystems could proceed at
Florida Agric. Exp. Stn., Journal Series no. R-00962. Received 17 a faster, more productive rate. The objectives of this
Aug. 1990. 'Corresponding author.
study were to: (i) document the relationship between
Published in Soil Sci. Soc. Am. J. 55:1722-1726 (1991). ./V and Lv for slash pine in the surface horizon of a
 ESCAMILLA ET AL.: ESTIMATING PINE ROOT DISTRIBUTION 1723

Spodosol, and (ii) document if N and Lv both give Means for each depth of each treatment were calculated
similar statistical interpretations for root distribution. and N was regressed against Lv for these data. Means rep-
resented (i) pine in the check plot, (ii) pine in the weed-
MATERIALS AND METHODS control plot, and (iii) nonpine in the check plot.
The study area was at the Gator Nationals Forest ap-
proximately 10 km northeast of Gainesville, FL. This study Analyses of Variance for Transformed Root-Length Density
area and treatments are the same as described in Escamilla and Number of Roots Crossing a Horizontal Unit Area
et al. (1991) and Swindel et al. (1988). Briefly, the study Initial analyses showed that Lv and N were not normally
design for this area was three replicates of a split-plot ran- distributed. Root-density values were then transformed into
domized, complete-block design. The data are based on 24 the form: log(Lv + 0.001) and log(Ar + 0.001). Transfor-
soil cores collected from the check and complete-weed-con- mations resulted in normal distributions.
trol treatments (Escamilla et al., 1991) during the seventh Statistical comparisons were done separately for log(Lv +
growing season. 0.001) and log,(N + 0.001) in order to test if each had similar
Soil cores were the same as those described in Escamilla statistical inferences. We used a randomi/ed complete-block
et al. (1991), and N and Lv were determined as described design with three blocks. Treatments were the two manage-
in that reference with the following exceptions. Since the ment practices (check and weed control). Computations were
objective was to correlate Lv with N, and since Lv repre- obtained with the SAS system using a repeated-measures
sented a 2-cm-deep soil volume, the value for N was deter- analysis as a split-split-plot design (SAS Institute, 1985). The
mined by taking the average of the N values between two whole-plot factor was treatment (check or weed control). The
measured depths. For example, the value of N for the 8- to subplot factor was within-plot location (within-bed or in-
10-cm depth (denned by the midpoint depth, 9 cm) was terbed) and the sub-subplot factor was depth. Depth was also
determined by averaging the N values for the 8- and 10-cm the subject factor in the repeated-measures analysis. Com-
faces. For this reason, it was not possible to have an lvalue parisons were run for (i) root category (pine vs. nonpine roots
for the 0- to 2-cm depth since there was no N value at the from the check plots), and (ii) pine roots only (check vs.
0-cm location. weed control). The H-F (Huynh-Feldt) condition was tested
and found to be valid, therefore the univariate test was used
Statistical Analysis (Freund et al., 1986). For ease of interpretation, P values of
Linear Relationship between Root-Length Density relevant interactions with depth are reported (Tables 1 and
and the Number of Roots Crossing a Horizontal Unit Area 2). The within-depth portions of the tables test the hypoth-
esis that response variables change at each depth increment.
Simple linear regressions were made between Lv and N The between-depths portions of the tables test the hypothesis
from all 2-cm segments of each core for (i) all roots in the that sample depth had no effect on the response variables,
check and weed-control plots, (ii) pine roots alone in the thereby allowing the inclusion of the within-depth effects in
check and weed-control plots, and (iii) other roots in the the design. All means presented in the figures are untrans-
check plots. Similar
2
analyses were made using only the 0- formed (antilog means) and based on the analysis of the log
to 0.3 roots cm- range for A''for pine'roots and the 0- to 1.0 values. Antilog means were calculated by the following equa-
roots cm"2 range for N. for nonpine roots and for all roots. tion:
Table 1. Regression results for the model: Lv = ft, + &7V, where Lv is root density in cm cm'3 and N is root density as number of roots cnr2.
Regression nt SE of ffi P <F
Check and Weed Control
All roots
N 264 0.154 0.491 0.060 0.209 0.0001
N (0 to 1 roots cm-2) 258 0.069 0.999 0.093 0.309 0.0001
Pine roots
AT 264 0.059 0.585 0.090 0.138 0.0001
N (0 to 0.3 roots cm-2) 253 0.048 0.797 0.117 0.156 0.0001
Check
Other roots
N 132 0.242 0.233 0.075 0.069 0.0023
N (0 to 1 roots cm-2) 128 0.168 0.582 0.131 0.136 0.0001

Table 2. Results of repeated-measures analysis of variance for root categoriest in the check plots.
Within depths Between depths
Source D X J\l)t D X L(2)§ D X T X L(3fl 7X1) L(2) T X L(2) BX L(T) (3)
n ^ r>

log(Lv# + 0.001) 0.0001 0.0499 0.0066 0.1059 0.0871 0.0549 0.2877 0.5181
logoff + 0.001) 0.0001 0.9665 0.3575 0.1346 0.0815 0.0152 0.1925 0.4361
t T = treatment (check, weed control); L - location (within- and interbed); D - depth (4, 6, 8, 10, 12, 14, 16, 20, 22, 24 cm); B = block.
j(1) Depth X block X treatment as an error term.
§ (2) Depth X block X location (treatment) as an error term.
H (3) Residual as an error term.
# Lv = root-length density.
ft N = the number of roots crossing a horizontal unit area.
 1724 SOIL SCI. SOC. AM. J., VOL. 55, NOVEMBER-DECEMBER 1991

+ O.So-2) [2] with data where TV" was <0.3, the slope increased to
where M = mean value, y = log value of the variable, x = 0.80 (Table 1).
antilog, and a2 = variance of the log values. Mean values There was also a poor relationship between Lv and
of Lv and N were plotted at depth midpoints. N for the nonpine roots in the check plots (Table 1
Fig. Ic). The value of TV accounted for only 7% of the
RESULTS variability in Lv. Even when the high root-density val-
ues were removed from the regression (Table 1), N
Linear Relationship between Root-Length Density accounted for only 14% of the variability in Lv, with
and Number of Roots Crossing a slope of 0.58.
a Horizontal Unit Area Using treatment means instead of individual ob-
When all observations from all plots for all root servations gave a much better predictive relationship
categories were used, there was a poor relationship between N and Lv. Using all data, the linear relation-
between Lv and AT (Table 1). While the regression was ship had a slope of 0.51 (r2 = 0.69). However, it was
significant, only 21% of the variability of Lv using all apparent that the several highest observations were
roots in the check and weed-control plots could be controlling the relationship. Using data where N <
explained by N (Fig. la). The slope of the relationship 0.8 roots cm"2, the relationship looked quite good, with
was 0.49, indicating a predominantly vertical root ori- a slope of 1.1 (SE = 0.07) and r3 = 0.86 (Fig. 2).
entation. The slope appeared to be controlled by those
samples with the highest root densities. If the data Comparative Trends of Root-Distribution Patterns
where N <l.O was used, the linear Lv/Nrelationship
had a slope of about 1.0; however, r2 was only 0.31 Check Plots: Pine Roots vs. Nonpine Roots
(Table 1).
\ Considering just the pine roots from all plots, A7 was Root-distribution trends, as suggested by analysis of
variance, were comparable between the two root-den-
still a poor predictor of Lv (Table 1 and Fig. Ib). It sity measures. Analyses suggested similar inferences
explained just 14% of the variability of pine root Lv. for the pine-root distribution, in that there were depth
The slope was 0.59, still suggesting a more vertical X location interactions (Table 2a) and location dif-
orientation of roots. By evaluating the relationship
ferences (Table 2b). Figure 3 indicates that the mag-
nitude of the differences; are also similar.
When the nonpine-root category is considered, N
provided a slightly different interpretation than did
Lv. Where Lv suggested that depth distribution of pine
and nonpine roots differed, analyses based on N sug-

0.70

0.60

0.50

"g 0.40
u

•^•0.30« Non-pine
> o
Pine, check
D
0.20
Pine, weed control

0.10

0.00
1 2 3 0.2 0.4 0.6 0.8
N (no.cm"2) N (no. cm ~2)
Fig. 1. Root density expressed as root-length density (Lv) regressed Fig. 2. Root density expressed as root-length density (Lv) regressed
against root density expressed as the number of roots crossing a against root density expressed as the number of roots crossing a
horizontal unit area (N) for (a) all roots in the check and weed- horizontal unit area (N) for pine roots in the check treatment,
control plots, (b) pine roots in the check and weed-control plots, pine roots in the weed-control treatment, and nonpine roots in
and (c) nonpine roots in the check plots. Each plotted point is an the check treatment. Each plotted point is a mean for a soil depth
observation for a depth, root-species category in a soil core. of all soil cores in a treatment.
 ESCAMILLA ET AL.: ESTIMATING PINE ROOT DISTRIBUTION 1725

Table 3. Results of repeated-measures analysis of variancet for pine roots in check and weed-control plots.
Within depths Between depths
Source D X 7XW D X £(2)§ DXTX £(3)1! 71(1) L(2) T X L(2) B X L(T) (3)

log(Lv# + 0.001) 0.0001 0.3764 0.0010 0.1481 0.4255 0.0122 0.9932 0.3775
log(Nft + 0.001) 0.0001 0.8377 0.0011 0.2938 0.0718 0.0135 0.11381 0.3856
t 71 = treatment (check, weed control); L = location (within- and interbred); D = depth (4, 6, 8, 10, 12, 14, 16, 20, 22, 24 cm); B = block.
t (1) Depth X block X treatment as an error term.
§ (2) Depth X block X location (treatment) as an error term.
1 (3) Residual as an error term.
# Lv = root-length density.
tt N = the number of roots crossing a horizontal unit area.

gested no such interaction (Table 2a), yet the patterns through the faces of the rectangular box (Melhuish and
of Lv and N with depth were similar (Fig. 3). Lang, 1968). This simplification assumes that, given
a soil volume, a random root measured on one face
Pine Roots: Check Plots and Weed-Control Plots will go through the soil volume and appear on another
Root density decreased with depth regardless of the soil core face. Or at least, if the root stops in the soil
volume, another root will begin in the soil volume and
expression of root density (Table 3a). Likewise, Wand randomly occur on another face. While this approach
Lv gave similar inferences, in that there were depth seems to be appropriate for roots of some agronomic
X location interactions (Table 3a) and location dif- plants (Bennie et al. 1977; Bland, 1989), it does not
ferences (Table 3b). Figure 4 shows that both N and hold for roots of this forest ecosystem.
Lv gave similar depth-distribution patterns for each Equation [1] might still be appropriate if N were
location. Within-bed root density was greater than that measured in two or three mutually perpendicular
in the interbed, with maximum differences in the 6- planes. While this study shows that Lv does not equal
to 18-cm-depth range. 2N when N is measured in a horizontal plane, there
still is the potential that Lv = 2N if both horizontal
Variability of Root-Length Density and Number and vertical planes were analyzed and averaged. This
of Roots Crossing a Horizontal Unit Area approach comes from the same theoretical base, but
The variability of N was less than that of Lv. In could not be tested with these data.
order to determine differences in root response as low While Lv did not equal 2N, there was still a good
as 0.02 cm cm-3 with a P < 0.05, 16 cores would be correlation between the value of N and that of Lv when
necessary. In comparison, only seven cores would be mean data were used. Under these conditions, for this
required for an equivalent response in N. soil, Lv was approximately equal to N, regardless of
the root category tested. Neither weed-control treat-
DISCUSSION ment nor depth in horizon significantly affected this
relationship. While this is a much more empirical ap-
The use of Eq. [1] to predict Lv from N using the proach than that based on probability theory, it does
core-break method was not possible. Equation [1] is suggest that N can be a useful measure of Lv under
based on geometrical probability theory, which im- field circumstances.
plies that, if a rectangular box of soil is small, then, The use of N is attractive as long as N identifies the
by simplification, a random straight line (a root) passes same root responses seen with Lv, and if N is a more
Lv (cm cm ~3) or N (no. cm-*)
0 0.2 0.4 0.6 0.8 1
Lv (cm cm"3) or N (no. cm~2)
2 0 0.1 0.2 0.3 0.4

4
6 6
8 8
•-10 •g-10

t 12
£14 £ 14
LU
016 Q 16
18 18
20 20
22 22
24 24
Fig. 3. Root-length density (Lv) and the number of roots crossing Fig. 4. Pine-root root-length density (Lv) and number of roots cross-
a horizontal unit area (N) depth distributions for pine and nonpine ing a horizontal unit area N depth distribution for interbed and
roots by 2-cm increments in the A horizon. Plotted points are within-bed locations across both treatments. Plotted points are
means of 12 observations. means of 12 observations.
1726 SOIL SCI. SOC. AM. J., VOL. 55, NOVEMBER-DECEMBER 1991

efficient field method. These data showed that statis-

tical inferences for pine roots were similar when either
N or Lv was used as the response variable. Likewise,
depth trends were also similar when all root species
or just pine roots were evaluated. These data also con-
firmed that TV was a less variable measure of root quan-
tity and required fewer samples to get similar
precision, thus making it more time efficient. In ad-
dition, Vespraskas and Hoyt (1988) have already
shown that measuring TV requires about one-fifth the
time required to measure Lv. Therefore, it seems jus-
tifiable to conclude that the faster, potentially more
precise and less laborious TV measurements could be
used as successfully as Lv in evaluating root responses
to treatments. This is particularly interesting given the
result that mean N correlated well with mean Lv.