Abstract
The responsible usage of water by facilities that rely on wet log
storage in the southern United States has become an issue of great
importance as restrictions on water usage have grown in recent years. In
order to learn about the dynamics of moisture content in wet-stored logs
over time, it is necessary to conduct continuous monitoring of log
piles. Time domain reflectometry (TDR) is a method that current research
has shown to have potential for use in this area. In this study, TDR
probes of three lengths (75, 100, and 125 mm) were systematically
inserted into 39 saturated bolts of Pinus taeda L., and both TDR and
moisture content measurements were taken nine times over a period of 16
days as the bolts air dried. The samples were then oven dried, and
measurements were taken three more times during that process. TDR
readings from the 125-mm probes had the strongest relationship
([R.sup.2] = 0.9426) with moisture content measurements. This result
indicates TDR readings are sufficiently correlated with moisture content
to accurately predict moisture variation over time and can be used to
learn how water application and other factors affect the moisture
content of wet-stored logs.
**********
In the wood products sector, timely management of available
resources is an ongoing concern. Because of the seasonal nature of
harvesting, in the southern United States it is common practice to place
logs in wet storage during periods of increased wood supply. Wet storage
helps to ensure that mill facilities will have adequate wood available
to allow operation during times when weather and other seasonal
difficulties slow or prevent harvesting activities. As opposed to dry
storage, wet storage maintains the wood under a system of sprinklers,
allowing it to be stored for long periods of time without experiencing
high levels of decay and damage by insects.
Wet storage of wood requires large amounts of water to maintain
high moisture content in the logs. At present, most facilities operate
under the assumption that more water is better, with high levels of
water being continuously applied to wet-stored logs. While this has been
shown to be an effective method of maintaining wood quality (Syme and
Saucier 1995), increasing concerns in the southern United States
regarding high levels of water consumption due to increasing
urbanization and recent drought make reduced water use desirable.
Elowsson and Liukko (1995) have shown that alternative regimes of water
delivery may also be effective.
In order to learn about the effectiveness of different rates of
water application to wet-stored logs, it would be advantageous to study
the moisture content of the logs throughout the period of storage.
Because of problems of accessibility (piles can be from 6 to 8 m high)
and the high moisture content of stored logs, resistance-type and
capacitance/power-loss moisture meters are unsuitable, as they cannot
measure moisture content above 30 percent (Haygreen and Bowyer 1996),
and at present there are no rapid assessment techniques available for
this application. However, a literature review suggested two methods
that could possibly be adapted for this purpose: near infrared (NIR)
spectroscopy and time domain reflectometry (TDR).
NIR spectroscopy has proven useful for measuring moisture content
of trees using core and disk samples. Of the two, it was determined that
NIR spectroscopy was not a viable method to use on a standing log pile.
Typically, NIR spectroscopy could be used to measure the moisture
content of a core, or a probe could be inserted in a log for continuous
measurements. It is not possible to obtain core samples from or insert
probes into logs while they are in wet storage, as most logs cannot be
accessed. Additionally, the fiber-optic probes that could be inserted
into a log are unable to withstand the hostile environment inside a
standing log pile for long periods of time and cannot be reliably left
inserted in the logs from one measurement to the next.
TDR is a standard method for determining soil moisture content. TDR
can be used for this purpose because the dielectric constant of porous
materials for frequencies between 1 MHz and 1 GHz is strongly dependent
on volumetric water content and largely independent of bulk density
(Constantz and Murphy 1990). For measurement of soil moisture, probes of
known length are inserted into the soil, and a pulse is transmitted
through the probes. As the dielectric constant of the soil is strongly
influenced by its moisture content, the apparent length of the probe
rods differs depending on the moisture content of the soil (Topp et al.
1980, Nadler et al. 2003).
The same principal can be applied to wood, and several studies have
provided encouraging evidence indicating that TDR may be a reliable
method of measuring the moisture content of wet-stored logs over time.
Constantz and Murphy (1990) demonstrated that TDR could reasonably
measure changes in moisture content in several species of living trees
over time, though they showed that the calibration curve for determining
moisture content of trees was different from the general calibration
curve used for determining moisture content of soil. The study
additionally demonstrated that individual tree species would likely need
individual calibration curves to ensure accuracy, though no further
studies were done to confirm this. Later, Wullschleger et al. (1996)
applied similar techniques to several hardwood species and concluded
that TDR was a sufficient method for determining seasonal variations in
the moisture content of standing trees and confirmed that the
calibration curve for standing trees was different from that for
measuring the moisture content of soils. Their experiment, however,
indicated that it may be possible to have one universal calibration
curve for various tree species. One further study (Nadler et al. 2003)
of 5-year-old lemon (Citrus limon (L.) Burman f.) trees used TDR to
simultaneously measure differences in soil and stem moisture content
under various irrigation regimes and concluded that TDR could be used
successfully to measure changes in tree water status as they respond to
water stress. While each of these studies showed potential for TDR as a
means of measuring moisture content in trees, it should be noted that in
each study, only one probe length was used (probe length varied from
study to study, however), two of them (Constantz and Murphy 1990, Nadler
et al. 2003) did not attempt unique calibrations for the species
involved in the studies, and all of these studies assumed that the same
part of the waveform reading that results from TDR measurement used in
measuring soil moisture (the dielectric constant) would be most
appropriate for measuring tree moisture content.
Therefore, in order to implement TDR as a basis for monitoring
moisture content of wet-stored logs, it was first necessary to carry out
a calibration study. The main goal of this study was to determine if TDR
is a reasonable method for monitoring moisture content in logs. To
accomplish this, several secondary goals were identified, including the
following:
* determination of which part of the waveform reading is most
highly correlated to moisture content of wood;
* determination of what probe rod length is the most conducive for
predicting moisture content;
* determination of the nature of the statistical relationship
between the apparent length of the probes and moisture content; and
* development of a mathematical model for use in further studies
should a reasonable relationship be found.
Materials and Methods
TDR measurements
In order to adapt TDR technology to measure the moisture content of
logs, probes were designed and built. Each probe consisted of two
3-mm-diameter stainless steel rods brazed to a length of copper coaxial
cable. The brazed ends of the probes and the cable were cast inside a 30
by 30 by 60-mm plastic block so that the probe rods were spaced 25 mm
apart, enabling them to be systematically inserted into logs or bolts.
Once inserted, a pulse of energy was sent down the cable into the probes
and then reflected back to the TDR instrument, where the apparent length
of the probe rods was read as a waveform trace on an oscilloscope
display. A Tektronix TDR cable tester was used to determine the apparent
length of the probe rods inserted in short wood bolts.
Calibration samples
The main set of calibration wood samples was initially made up of
30 bolts, each 152.5 mm long, cut transversely from logs of Pinus taeda
L. The diameter of the test bolts ranged from 124 to 229 mm with the
bark removed. Because of a malfunctioning probe, one of the bolts was
excluded from the experiment, and a total of 39 bolts were used. The
main calibration set (29 bolts) had 75-mm probes inserted. Two
additional sets of bolts, with five samples in each set, were tested
with 100- and 125-mm probe rods inserted, respectively. These wood
samples ranged from 124 to 220 mm in diameter for both sets of five.
Data collection
Wood samples were hydrated in a saturation tank for 2 weeks prior
to the start of the experiment. Following saturation, probe weight, wood
sample weight, and gravimetric weight were measured. Two 3-mm holes, 25
mm apart, were then drilled in the samples using a guide. The depth of
holes was consistent with the length of the probe rods (i.e., 75, 100,
or 125 mm) to avoid problems with inserting longer probes, and probe
orientation was parallel to the grain. Once the holes were drilled, the
probes were inserted and the first waveform readings taken.
Samples were air dried during the next 16 days and measured nine
more times during this period. At each measurement, the samples were
weighed without removing the probes (probe weights were later subtracted
to obtain sample weights) and TDR waveform readings collected. After
these initial 10 measurements, the wood samples were oven dried for 1
day at 50[degrees]C without removing the probes, and then weights and
waveforms were recorded. For the 12th and final measurement interval,
the probes were removed, and the wood samples were oven dried for 3 days
at 103[degrees]C, and then ovendry weights were recorded. This final
weighing provided a reference weight and percent moisture content on an
ovendry basis for all previous readings.
The primary purpose of this calibration study was to use the
apparent length of the probe rods as measured on the TDR display to
predict wood moisture content. Note that apparent length is not the
actual length of the rods but rather the point at which wave reflection
signals appear. This apparent length is a function of the conductivity
of the medium; wood with higher moisture content has higher conductivity
and carries the signal farther before reflection Occurs.
Figure 1 shows the waveform. In this figure, the starting point for
measurement of the waveform is point X; this corresponds to the point
where the probe rods emerge from the plastic block. The position of X is
determined for each probe by short-circuiting the rods at the probe base
while the probe is connected to the TDR.
Initially, lengths from point X to points A, B, C, and D (Fig. 1)
were recorded. Ideally, we would have saved each waveform for later
analysis but this was not possible using the Tektronix TDR cable tester.
Previous TDR studies of soil moisture content measurement over the past
20 years have used points B, C, and D. Point D was abandoned during the
second measurement interval because of difficulties in consistently
determining the location of this point on the waveform. Point E,
referred to also as the dropoff point of the signal, showed a more
stable response and was recorded during the next 10 measurement
intervals in place of point D. Point C, also referred to as the
inflection point of the rising curve, was seemingly the most consistent
feature on the waveform, as it was not difficult to locate and
measurements were unambiguous. On visual assessment at the end of the
experiment period, both point C and point E showed a steady shortening
of the apparent length of the waveform over time as the samples dried;
points A and B had considerable instability. Only points C and E were
used in subsequent analyses.
[FIGURE 1 OMITTED]
Statistical analysis
The form of the relationship between apparent distances of the
waveform and the moisture content of the wood samples was unknown at the
start of the analysis. However, soil moisture content is often related
to the waveform reading through the Topp equation (Topp et al. 1980).
This equation is a third-order polynomial regression model, suggesting
that a polynomial regression may also be appropriate for relating the
TDR waveform to wood moisture content. The analysis therefore began with
a visual assessment of plots of apparent distances against moisture
content for each probe length to determine the validity of this
conjecture. Following this, further analyses were conducted to determine
the values of model parameters and evaluate the validity of these
models.
Note that probe and sample effects were not accounted for, even
though repeated measurements were taken on each wood sample over time.
Because the ultimate goal of this study was to create a prediction
equation that would work with other probes and wood samples rather than
an equation descriptive of this particular data set, removing variance
through the inclusion of random effects would be ineffectual.
Additionally, autocorrelation within wood samples due to repeated
measures was ignored. Note that repeated measures also have the
potential to lead to an inflated [R.sup.2] estimate. This was because
future application of these methods cannot take autocorrelation into
account since actual moisture content will not be measured.
Results
The analysis of the TDR calibration data began with an examination
of separate plots of the data for each probe length and apparent
distance since the form of the relationship between apparent distance
and percent moisture content was unknown. From this visual examination,
a linear or curvilinear relationship between the apparent distance to
the inflection point of the rising curve (point C) and the percent
moisture content in the wood sample appeared reasonable. The
relationship between the apparent distance to the dropoff point of the
signal (point D) and the percent moisture content in the wood sample was
clearly not linear but could possibly be modeled with a higher-order
polynomial.
The analysis then continued with further examination of linear and
second-order relationships between apparent distance to the inflection
point of the rising curve and percent moisture content. Regression
statistics indicated that apparent distance to the inflection point and
the square of that distance were highly significant in prediction of
percent moisture content for each probe length (P < 0.0001 in all
cases for distance to inflection point, and P = 0.0006, 0.0018, and
0.0102 for square of distance to inflection point for 75-, 100-, and
125-mm probe lengths, respectively). The degree to which apparent
distance was able to predict moisture content varied among probe
lengths; the 125- and 100-mm probes were the most effective, and the
75-mm probes were the least effective ([R.sup.2] = 0.7920 for 75-mm
probes, [R.sup.2] = 0.9031 for 100-mm probes, and [R.sup.2] = 0.9281 for
125-mm probes). Figure 2 shows plots of the apparent distance to the
inflection point and percent moisture content of each sample at each
measurement time, separated by probe length; the line resulting from the
second-order regression is shown in each case. Figure 3 shows the
residual plots associated with each of these regressions. The regression
plots in Figures 2 and 3 indicate that a second-order regression was
appropriate in each case. Analysis of a third-order model verified this,
as the cube of the apparent distance proved insignificant in each case.
However, the residual plots from Figure 3 indicate a problem with the
variance of the residuals. This issue is most obvious in the data from
the 75-mm probes, where the variance of the residuals was clearly larger
for high predicted moisture content than low predicted moisture content.
There could be several reasons for this, including influence on the
apparent distance to the inflection point from a source other than
moisture, differences between individual probes, or naturally occurring
larger variance in the apparent distances at high moisture content. The
effect was less noticeable in the 100-mm probe data and not noticeable
in the 125-mm probe data.
[FIGURE 2 OMITTED]
To determine whether the diameter of the wood samples influenced
this variance, the data from the 75-mm probes were broken into three
diameter classes. The wood bolts with the 10 smallest diameters were
classified as "small diameter" (124 to 175 mm), the 10 next
smallest were classified as "medium diameter" (175 to 191 mm),
and the 9 remaining wood bolts were classified as "large
diameter" (191 to 229 mm). Figure 4 demonstrates that diameter was
influential; in the regression for the 75-mm probes, small-diameter
bolts tended to have positive residuals, and large-diameter bolts tended
to have negative residuals. This indicated a need to include diameter as
a covariate in the analysis. Figure 5 shows the residual plot for the
75-mm probes after diameter was included as a covariate. Diameter was
extremely statistically significant in the model (P < 0.0001). This
somewhat alleviated the problem, although some variance problems
remained. Including diameter as a covariate also increased the [R.sup.2]
from 0.7986 to 0.8496. Regression results for all three probe lengths
with diameter included as a covariate can be found in Table 1. Diameter
of the wood bolt was significant in the regressions for both the 75- and
the 125-mm probes.
[FIGURE 3 OMITTED]
[FIGURE 4 OMITTED]
[FIGURE 5 OMITTED]
Given these results, 125-mm probes were the most reliable for
predicting percent moisture content from the apparent distance to the
inflection point of the rising curve ([R.sup.2] with diameter as a
covariate = 0.9426). However, wood sections with a diameter smaller than
125 mm cannot accommodate a 125-mm probe. To determine the effectiveness
of shorter probes in these situations, a linear regression was performed
after removing all wood bolts with a radius larger than 75 mm from the
75-mm probe data set. This ensured that the probes at least reached the
center of the bolt. The results were excellent; in this data set
containing 5 of the original 29 logs, diameter was no longer significant
as a covariate (P = 0.7807 when included), the apparent distance to the
inflection point and the square of the distance were still extremely
significant (P < 0.0001 and =0.0003, respectively), and the [R.sup.2]
of the model without diameter increased to 0.9384. Figure 6 is a plot
&the data included in this analysis with the second-order regression
line, indicating that the regression was appropriate. The residual plot
is given in Figure 7, showing that the variance problem observed when
all diameters were included was no longer apparent.
In addition to the apparent distance to the inflection point of the
rising curve, the apparent distance to the dropoff-point of the signal
showed potential to predict percent moisture content. A linear
regression on the 75-ram probe data proved inappropriate, as the
relationship appeared to be curvilinear; the [R.sup.2] achieved was
0.5481. When a second-order model was applied, it became evident that
outliers were present that had a large influence on the model. Given the
success of prediction with the inflection point, it was determined that
further analysis of the dropoff point was unnecessary.
[FIGURE 6 OMITTED]
To summarize the results, recommendations for TDR studies of wood
moisture content include using 125-mm probes when the diameter of the
wood sample allows and 75-mm probes for samples with a radius of no more
than 75 mm. The most advantageous apparent distance (AD; measured in
meters) to measure along the waveform is the distance from the base of
the probes to the inflection point of the rising curve. The regression
equations for relating this distance to the percent moisture content
(%MC) of the sample follow:
%MC = - 180.65 + 1,344.77AD
- 1,382.39[AD.sup.2] (75-mm probes) (1)
and
%MC = -132.14 + 676.99AD - 407.50[AD.sup.2]
- 0.111 diam. (125-mm probes) (2)
Statistics summarizing the fit properties of these models can be
found in Table 2. The statistics PRESS (predicted error sum of squares),
PRESS RMSE (root mean square error; comparable to RMSE), and [Q.sup.2]
(comparable to [R.sup.2]; see Quan 1988) are related to predictive
abilities of the models. The prediction-related statistics demonstrate
good predictive abilities of the models.
[FIGURE 7 OMITTED]
To further confirm the predictive fit of the model, a cross
validation was performed. The training data set retained 80 percent of
the observations, and the remaining randomly selected 20 percent of
observations became the validation data set. The training data resulted
in the following models:
%MC = - 197.24 + 1,471.94AD
- 1,609.35[AD.sup.2] (75-mm probes) (3)
and
%MC = -131.60 + 659.95AD - 393.09[AD.sup.2]
- 0.092 diam. (125-mm probes) (4)
Model 3 has an [R.sup.2] of 0.9431 and an RMSE of 8.132, and Model
4 has an [R.sup.2] of 0.9410 and an RMSE of 7.324. These statistics are
similar to those of the full model, as shown in Table 2, and should
provide a reliable means of evaluation for the validation data set. The
RMSE of the validation data for Model 3 was 10.387; the RMSE of the
validation data for Model 4 was 7.235. These are similar to the RMSEs of
the training data. A plot indicating the training data, the validation
data, and both complete and training models is shown in Figure 8 for the
75-mm probes. These results provide further evidence of the predictive
validity of Models 1 and 2.
Discussion
This study examined several issues related to the use of TDR as a
means of measuring moisture content in logs. These issues included the
differences in the results obtained by using probes of different
lengths, the part of the waveform reading most correlated to moisture
content, and the nature of the relationship between the TDR measurement
and moisture content.
[FIGURE 8 OMITTED]
Previous research has not studied the impact of different probe
lengths on the moisture content measurements of trees or logs. This
study showed that 125-mm probes performed better for predicting moisture
content than the shorter probe lengths included in the study. This is
possibly due to the longer probes contacting a larger cross section of
the log when inserted. As the estimated moisture content is an average
along the length of the probe, a longer probe provides a more
representative and accurate estimate of the moisture content over the
complete cross section. This also allows the longer probes to perform
more accurately with logs of large diameter.
An interesting finding of this study was that the most desirable
method for predicting moisture content of logs was different than the
method generally used for soil moisture content prediction. The Topp
equation (Topp et al. 1980) is a third-order polynomial equation based
on the apparent length to a particular point of the TDR waveform
reading, which relates that distance to moisture content. Previous TDR
tree moisture content studies considered only the Topp calibration curve
to relate the measurements; this study determined that by using the
inflection point of the rising curve of the TDR waveform, a
statistically better result is obtained with a second-order model.
Recent research based on four hardwood species (two ring porous and two
diffuse porous species) has indicated that a logistic model better fits
the hardwood species data. For each hardwood data set, an upper
asymptote was observed that differed by species and represented a
maximum moisture content. It is possible that our hydration period was
insufficient for our samples to achieve a maximum moisture content,
explaining the success of the linear models.
TDR was shown to be a reliable method for measuring the moisture
content of P. taeda, with a high correlation between the inflection
point of the rising curve of the TDR waveform and the moisture content
of the log bolts. Combined with the durability of the constructed
probes, this makes TDR an ideal method for measuring the moisture
content of wet-stored logs. At present, studies using these probes to
monitor the moisture content of wet-stored pine have been installed at
the wet storage facilities of several southern US paper mills. These
studies will monitor moisture content variation over time, both within
the logs and within the pile, using multiple water regimes. Additional
studies are planned, including studies of hardwood species, that will be
based on a separate calibration study since it is unknown whether the
specific equation or the general form of the calibration curve
calculated here is universal. The results of these studies will provide
information useful for controlling water application; understanding how
water application rates affect log moisture under various conditions
will allow wet storage facilities to maintain log quality and preserve
water by applying it appropriately.
Conclusions
Of the several methods considered for monitoring moisture content
over time in wet-stored logs, TDR was found to be the best suited
because it has the ability to reliably measure high moisture content,
measurement probes can be designed to withstand the conditions of a
wet-stored log pile over time, and several studies have demonstrated its
potential for measuring moisture content in wood. This study showed that
the moisture content of wood can be accurately predicted with TDR with a
very high correlation. Additionally, it indicated that longer probes may
result in more reliable predictions and that the part of the waveform
measurement most useful in predicting wood moisture content differs from
the part traditionally used in soil moisture content measurements, as
does the form of the calibration curve. This study also resulted in the
calculation of a calibration curve specific to P. taeda that will be
employed in future studies of the moisture content of wet-stored logs
over time based on varying water regimes and wood pile dynamics. This
will allow for future decisions regarding appropriate water use while
maintaining moisture content high enough to preserve wood quality while
in storage.
Literature Cited
Constantz, J. and F. Murphy. 1990. Monitoring storage moisture in
trees using time domain reflectometry. J. Hydrol. 119:31-42.
Elowsson, T. and K. Liukko. 1995. How to achieve effective wet
storage of pine logs (Pinus sylvestris) with a minimum amount of water.
Forest Prod. J. 45(11/12):36-42.
Haygreen, J. G. and J. L. Bowyer. 1996. Forest Products and Wood
Science: An Introduction. 3rd ed. Iowa State University Press, Ames. 484
pp.
Nadler, A., E. Raveh, U. Yermiyahu, and S. Green. 2003. Evaluation
of TDR to monitor water content in stem of lemon tees and soil and their
response to stress. Soil Sci. Soc. Am. J. 67:437-448.
Quan, N. T. 1988. The prediction sum of squares as a general
measure for regression diagnostics. J. Bus. Econ. Stat. 6(4):501-504.
Syme, J. H. and J. R. Saucier. 1995. Effects of long-term storage
of southern pine sawlogs under water sprinklers. Forest Prod. J. 45(1):
47-50.
Topp, G. C., J. L. Davies, and A. P. Annan. 1980. Electromagnetic
determination of soil water content: Measurements in coaxial
transmission line. Water Resour. Res. 16:574-582.
Wullschleger, S. D., P. J. Hanson, and D. E. Todd. 1996. Measuring
stem water content in four deciduous hardwoods with a time-domain
reflectometer. Tree Physiol. 16:809-815.
The authors are, respectively, Professor and Codirector, Wood
Quality Consortium, Warnell School of Forestry and Natural Resources,
Univ. of Georgia, Athens (lschimleck@warnell.uga.edu [corresponding
author]); Associate Director, Univ. of Georgia Statistical Consulting
Center, Univ. of Georgia, Athens (krlove@ uga.edu); Research Coordinator
(sanders@warnell.uga.edu), Graduate Student (raybonm@warnell.uga.edu),
and Professor and Codirector, Wood Quality Consortium
(ddaniels@warnell.uga.edu), Warnell School of Forestry and Natural
Resources, Univ. of Georgia, Athens; Division Forester, Molpus
Timberlands Management, LLC, Monroe, Louisiana (jmahon@molpus.com);
Forester, USDA Forest Serv., Southern Research Sta., Athens, Georgia
(eandrews@fs.fed. us); and Senior Research Scientist, National Council
for Air and Stream Improvement, Inc., Southern Regional Center,
Newberry, Florida (eschilling@src-ncasi.org). This paper was received
for publication in May 2011. Article no. 11-00060.
Table 1.--Regression results for three probe lengths with
diameter of wood bolt included as a covariate.
Probe P value of
length (mm) [R.sup.2] Variable variable
75 0.8496 Inflection point <0.0001
Square of inflection point <0.0001
Diameter <0.0001
100 0.9049 Inflection point <0.0001
Square of inflection point 0.0017
Diameter 0.3335
125 0.9426 Inflection point <0.0001
Square of inflection point 0.0003
Diameter 0.0008
Table 2.--Model fit statistics for final recommended Models 1
and 2.1
Model [R.sup.2] [Q.sup.2] RMSE PRESS RMSE PRESS
1 0.9383 0.9319 8.609 8.955 4,410.79
2 0.9426 0.9371 7.235 7.426 3,033.37
(a) RMSE = root mean square error; PRESS =predicted error sum of
squares.