| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a Huxley College of Environmental Studies, Western Washington Univ., Bellingham, WA 98225-9181
b USDA Forestry Sciences Laboratory, Corvallis, OR 97331
c Dep. of Forest Resources, Oregon State Univ., Corvallis, OR 97331
Corresponding author (homann{at}cc.wwu.edu)
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Abbreviations: CV, coefficient of variation • MDD, minimum detectable difference between treatments • MS, mean square
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Concentrations of C and N (g kg-1) are commonly evaluated in soil assessments (Grigal et al., 1991; Knoepp and Swank, 1997; Trettin et al., 1999). Incorporation of information about rock volume and either soil mass or bulk density allows examining masses of C and N (kg m-2) to specified depths (Homann et al., 1995; Trettin et al., 1999). Soil C and N masses can also be expressed as amounts associated with a given quantity of inorganic material, which provides a constant basis for comparison in the absence of erosion (Jenkinson, 1971; Bormann et al., 1995; Homann and Grigal, 1996). This latter approach is particularly applicable to account for changes in bulk density and soil volume associated with changes in organic matter.
Evaluating effects of management practices on forest soil C, N, and other properties with a rigorous, conclusive, and statistically valid approach is difficult because of sampling and measurement challenges, the inherent spatial variability in soil properties, slow changes in soil properties, and large land areas required for stand manipulations. These difficulties have been successfully overcome by examining previously established, replicated vegetation treatments, for which no pretreatment soil data exist (Binkley and Valentine, 1991; Son and Gower, 1992), assessing soil changes over time in plots or watersheds (Mroz et al., 1985; Johnson et al., 1995; Knoepp and Swank, 1997; Trettin et al., 1999), and combining both pre- and post-treatment measurements with replicated treatments in the tropics, where changes in soil properties can be rapid (Fisher, 1995; Binkley and Resh, 1999).
Pretreatment sampling of replicated treatments is useful for four reasons. First, in combination with post-treatment measurements, it allows the magnitude and direction of change to be quantified (Fisher, 1995; Binkley and Resh, 1999). Second, pretreatment measurements can be the basis for estimating MDD or minimum detectable changes for a given sampling intensity (Huntington et al., 1988). Conversely, statistical properties and simulation can indicate the number of samples required to detect a specified change (Johnson et al., 1990). Third, stored pretreatment samples may prove valuable when new analytical techniques become available or when currently unanticipated analyses are deemed to be important. Finally, assessing differences between post- and pretreatment measurements may enhance the sensitivity of detecting treatment differences compared with evaluating post-treatment values only. We address this final concept in this study.
The objective of this study was to determine if pretreatment soil C and N measurements will help us detect differences among forest treatments. The treatments are overstory species and woody debris manipulations in a newly established, large-scale, long-term experiment in a Pacific Northwest Douglas-fir forest. To ascertain the statistical sensitivity of this experiment, we combined statistical simulation and pretreatment data to determine (i) how MDDs vary among different approaches of quantifying soil C and N and (ii) how MDDs change when pretreatment data are combined with future post-treatment measurements vs. using post-treatment values only. Because of their association in organic matter, we hypothesized soil C and N would have similar MDDs relative to their current values, and that those relative MDDs would decrease by similar amounts when pretreatment data are considered in the analysis.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
200 cm, of which only 15 cm falls between June and September. Climate at the site is typically cooler and snow is more common than on the coast, although on-site weather records are not available (Little et al., 1995). The overstory vegetation is dominated by 70- to 100-yr-old Douglas-fir, which established naturally after a fire in 1881 (Little et al., 1995).
|
Early successional vegetation is being created by clearcutting, followed by planting knobcone pine (Pinus attenuata Lemm.) and Douglas-fir, and leaving sprouting hardwoods. Mid successional vegetation is being created by clearcutting, followed by planting only Douglas-fir and controlling competing tree species. Late successional vegetation is being created by thinning the current forest and underplanting with sugar pine (Pinus lambertiana Dougl.). Small quantities of woody debris will be implemented by removing stems greater than 10-cm diam. and 3-m length following initial and subsequent harvesting or thinning. Large quantities of woody debris will be produced by leaving fallen stems equivalent to 15% of the preharvest standing volume. Over multiple harvests, this will yield a different quantity and quality of woody debris on each vegetation type. Therefore, we considered the treatments to be independent, rather than crossed between vegetation and woody debris, and used a completely randomized block analysis (Zar, 1999), with the plot serving as the experimental unit. Treatment is a fixed factor with six degrees of freedom, block is random with two degrees of freedom, and the residual has 12 degrees of freedom. The F ratio of mean squares (MS) to test the treatment effect is MStreatment/MSresidual.
Soil Sampling and Analysis
Each treatment plot consists of a central 2-ha measurement area surrounded by a 4-ha buffer. On each plot, pretreatment soil sampling was conducted at 15 or 16 grid points on a 25 by 25 m grid in the measurement area. At sampling points located on trees, tree roots, logs, or large rocks, up to two additional points within 1 m of the initial sampling point were attempted. Overall, samples were successfully collected at an average of 13 points per plot.
Pretreatment soil sampling was conducted during July through October 1992. At each sampling point, slope angle was measured. Samples were taken perpendicular to the slope. The O horizon, excluding wood >2.5-cm diam., was collected from within a 30-cm-diam. ring. Then a steel, rectangular corer with a 10 by 15 cm cross-sectional area was driven into the mineral soil to a depth of 35 cm, excavated, and removed. When compression of the core occurred, the distance from soil surface to core surface was recorded. From the open-faced side of the corer, three layers of mineral soil were extracted: A horizon, bottom of A horizon to 15-cm depth, and 15- to 30-cm depth. Whole samples, including rocks, were dried at 60°C.
The O horizon samples were hand sorted into rocks with diameters >4 mm, wood fragments with diameters >6 mm, and the residual material. These components, including rocks, were dried to constant mass at 65°C and weighed. Subsamples of the residual material were ground to <0.8 mm (20 mesh), and analyzed for loss-on-ignition and ash concentration by heating at 550°C for >12 h, and for Kjeldahl N by the method of Bremner and Mulvaney (1982). The CVs for replicate analyses were 1.0% for loss-on-ignition and 5.1% for Kjeldahl N. Eight ground (<0.25 mm, 60 mesh) subsamples of residual materials were analyzed for total C and N with a Carlo-Erba NA 1500 Series 2 analyzer (Carlo Erba Strumentazione, Rodano, Italy). Comparison with corresponding loss-on-ignition and Kjeldahl N values yielded a mean ratio of total C to loss-on-ignition equal to 0.53 (CV = 4.3%) and mean ratio of total N to Kjeldahl N equal to 1.19 (CV = 9.2%). These ratios were used to convert loss-on-ignition to total C and Kjeldahl N to total N for all other residual-material samples.
Mineral soil samples were sieved with a 4-mm sieve to yield <4-mm soil. A 4-mm sieve, rather than the standard 2-mm sieve, was used because of our desire to quantify total soil C and N but our inability to efficiently break up 2- to 4-mm organic-containing aggregates (Cromack et al., 1999). The >4-mm material was hand sorted into >4-mm rocks and >4-mm wood, which included both roots and woody debris. Soil and rocks were weighed, and the masses were converted to a 105°C equivalent based on drying of subsamples. Wood was dried to constant mass at 65°C and weighed. Subsamples of <4-mm soil were ground to <0.25 mm (60 mesh) and analyzed for total C and N with a Carlo-Erba NA 1500 Series 2 analyzer. The CVs for replicate analyses were 5.2% for C and 5.6% for N.
Data Analysis
Soil properties were calculated for the O horizon, 0- to 15-cm mineral soil and 15- to 30-cm mineral soil. Masses of soil constituents per unit area were determined by dividing masses in a sample by the cross-sectional area of the sample; values were slope-corrected to facilitate comparison with vegetation measurements made at the site (Little et al., 1995). The A horizon was not considered separately because of the subjective assessment of the poorly defined boundary between the A and B horizons. Compression of the mineral soil occurred in 8% of the cores and averaged 1.9 cm for those cores. This compression was assumed to have occurred evenly over the 30-cm sampling depth.
Masses of O-horizon residual material and <4-mm mineral soil were multiplied by C and N concentrations to yield C and N masses for each measured layer. Whole or partial mineral soil layers were summed to yield rock, <4-mm soil, wood, C, and N masses for the 0- to 15- and 15- to 30-cm mineral soil depths. Concentrations of C and N in these aggregated, slope-corrected layers were calculated as masses of C and N divided by mass of <4-mm soil.
Masses of C and N associated with the upper 87 kg m-2 of <4-mm inorganic material were calculated to provide a comparison based on inorganic mass rather than layer depth. This method accounts for temporal changes in layer thickness, soil volume, or bulk density that could yield misleading results in the depth-based approach (Jenkinson, 1971; Bormann et al., 1995; Homann and Grigal, 1996). Masses of inorganic material and associated C and N were summed through O horizon and whole or partial mineral soil layers until the 87 kg m-2 inorganic mass was reached. This is the smallest amount of inorganic material found at a single sampling point. Inorganic mass was calculated by subtracting organic matter mass from total mass. Organic matter mass was determined by dividing the total C by 0.53, which is the ratio of total C to loss-on-ignition developed for O horizon residual material. Similarly, we also calculated masses of C and N associated with the upper 205 kg m-2 of total inorganic mass, which included rocks as well as the inorganic material in the <4-mm fraction.
Bulk density of the <4-mm soil was calculated by dividing <4-mm mass by <4-mm volume. The <4-mm volume was calculated as total sample volume minus rock and wood volumes. These latter volumes were determined from rock and wood masses, rock particle density of 2.4 Mg m-3 based on water-displacement measurements of nine rock samples, and a wood density of 0.4 Mg m-3, which is the value used for sound wood in western coniferous forests (Harmon et al., 1986). Using an alternate wood density of 0.2 Mg m-3, which is more typical of decomposed wood (Harmon et al., 1986), yields bulk density values that are 1% greater than those reported here.
Statistical Analysis and Simulation
We characterized variability with CVs among all sampling points, among points within plots, pooled across plots, and among plots within blocks pooled across blocks. A plot value was determined by averaging the points within the plot.
We calculated the MDD between the two most extreme population means with the following formula, which is reformatted from Zar (1999)
 | (1) |
is 1.8, based on power = 0.8, 
= 0.05, and the 12 degrees of freedom associated with the MSresidual (Zar, 1999). We determined the MDD for two scenarios: (i) when only post-treatment measurements would be available and (ii) when differences between post- and pretreatment measurements would be available. For the first scenario, we assumed post-treatment measurements would have the same inherent variability among replicate experimental units as the pretreatment measurements and, thus, would have the same MSresidual. This assumption can be evaluated only with future post-treatment measurements.
For the second scenario, we simulated differences between post- and pretreatment measurements using within-plot pretreatment variability and the t distribution. The t distribution is defined by
 | (2) |
1 and 2 are means of random samples from the same normal population, s2p is the pooled variance of the observations within the samples, and p1 and p2 are the numbers of observations in each sample (Zar, 1999). Rearranging the equation and assigning 1 = post-treatment and 2 = pretreatment yields the difference between post- and pretreatment values for a single plot
 | (3) |
For s2p we used the variance of the pretreatment sampling point values within a plot, pooled across plots within a block, and assumed to be the same for pre- and post-treatment. ppre is the number of pretreatment sampling points within a plot, and ppost is the number of post-treatment sampling points within a plot, which was assumed to be 13. The value for t was randomly selected from the t distribution with degrees of freedom associated with s2p and entered into Eq. [3] to yield one of an infinite number of possible differences between post- and pretreatment samples for a single plot. From simulated differences for each of the 21 plots, we calculated the MSresidual for the randomized block design (Zar, 1999). The t distribution and the resulting MSresidual are applicable when there is no change from pre- to post-treatment. The MSresidual is also applicable if a change occurs and that change is the same magnitude for all plots of the same treatment, which we assume in this analysis. Such a consistent treatment effect would increase the MStreatment, but it would not alter the MSresidual upon which the MDD is based (Eq. [1]).
We repeated the process of simulating differences and calculating MSresidual a total of 250 times for each variable to yield an average MSresidual for each variable within 10% of its expected value with 95% confidence and used this average to determine the MDD with Eq. [1]. Variables that had skewed distributions of point values within plots were log-transformed prior to analysis to make the distributions more normal. Relative MDD was calculated as
 | (4) |
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
5%) of C and N concentrations was relatively small compared with the variation among sampling points (19–31%). Masses of C and N in the O horizon were twice as variable as in the mineral soil, because the O horizon depth varied, whereas the mineral soil values were for specified depths. Wood was two to nine times as variable as other properties in both O-horizon and mineral soil. There was considerable variability in O-horizon rock mass, probably resulting from both sampling procedures and natural mixing with the mineral soil. For variables in general, the within-plot CV was on the order of two-thirds of the CV among all sampling points (Table 2), indicating most of the variability was caused by within-plot variation. Knowledge of soil variability can guide decisions about the intensity of sampling (Grigal et al., 1991). Comparing within-plot CVs for the 2-ha Siskiyou measurement areas with those of similar-sized plots in other forests indicated contrasting trends. The CVs of O-horizon organic matter and N concentrations within 0.4-ha plots in Lake States forests (Grigal et al., 1991) were similar to those at the Siskiyou site, but CVs of the mineral soil were substantially greater than at Siskiyou site. In contrast, CVs for C concentration in O horizon and mineral soil in a Maine Spodosol (David et al., 1990) were similar to the Siskiyou site, but CVs for N concentration were greater than at Siskiyou site.
|
Minimum Detectable Differences
Evaluating O horizons is common (Covington, 1981; Snyder and Harter, 1987; Hendrickson et al., 1989; Mattson and Swank, 1989), but accurately assessing changes in O-horizons and differences between treatments may prove difficult (Federer, 1982). We do not consider the O horizon to be a good basis for examining treatment effects in this study, because of pretreatment characteristics. O-horizon mass and ash concentration were strongly related (Fig. 1) , indicating O horizons on different plots contain different amounts of mineral soil contamination. The large variability in rock mass (Table 2) substantiates this. Variable soil mixing by large mammals, soil animals, and insects could contribute to this high variability. Alternatively, separation between the O horizon and mineral soil might not have been consistent during sampling, which is a common problem (Federer, 1982; Alban and Perala, 1990).
|
Relative MDDs were 26 to 57% for the various approaches of quantifying soil C, if only post-treatment measurements were made (Table 3). They were substantially smaller for C masses than for C concentration (Table 3), indicating the added effort to measure rock volume and either soil mass or bulk density increases the sensitivity to determine treatment differences. The C mass of the O-horizon plus 0- to 30-cm mineral soil had the smallest relative MDD. The C mass based on fixed depths had smaller relative MDDs than C mass associated with specified amounts of inorganic material.
|
Pretreatment measurements reduce the MDD for only some soil properties because of the magnitude of pretreatment variability. In general, the larger the CVs among plots within blocks, the greater the decrease in MDD that can be obtained by accounting for this pretreatment variability. For example, in the O horizon plus 0- to 15-cm mineral soil, the pooled CV among plots within blocks was 12.2% for pretreatment N mass (Fig. 2) . Accounting for this variability by examining the difference between pre- and post-treatment measurements decreased the MDD from 47 to 24% (Table 3). In contrast, accounting for the smaller variability of C mass, with a pooled among-plot CV of 8.4%, did not change the MDD (Table 3).
|
8%, accounting for the real variability among plots begins to more than compensate for the compounded measurement error and the MDD begins to decrease.
|
Treatment differences in soil C and N properties will have to be substantial to be detected at the Siskiyou site. Extreme treatments need to differ by 20 to 50% to be detected (Table 3). These differences are within the range of possible influences of tree species (Challinor, 1968; Alban, 1982) and forest management practices (Johnson, 1992; Bormann et al., 1994).
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Analysis of multiple pretreatment sampling points combined with statistical simulation allowed us to ascertain the value of pretreatment measurements in detecting treatment differences at the Siskiyou Integrated Research Site. The minimum detectable difference between treatments will be decreased substantially for both N concentrations and masses by assessing the difference between post- and pretreatment measurements, rather than by assessing only post-treatment values. However, pretreatment data do not always help detect treatment differences. Pretreatment measurements will not substantially decrease minimum detectable differences for soil C properties, because there is less relative variability among plots for soil C than for soil N. Soil C and N masses, which incorporate rock volume and soil mass, can yield smaller MDDs than C and N concentrations. Evaluation of other experimental sites is required to determine if these conclusions can be used to design sampling strategies in other Pacific Northwest forests, or if site-specific characteristics will dictate when pretreatment measurements will or will not enhance our ability to detect differences among treatments.
|
ACKNOWLEDGMENTS |
|---|
Received for publication January 3, 2000.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  C. H. Shaw, J. R. Boyle, and A. Y. Omule Estimating Forest Soil Carbon and Nitrogen Stocks with Double Sampling for Stratification Soil Sci. Soc. Am. J., September 30, 2008; 72(6): 1611 - 1620. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. B. Harrison, A. B. Adams, C. Licata, B. Flaming, G. L. Wagoner, P. Carpenter, and E. D. Vance Quantifying Deep-Soil and Coarse-Soil Fractions: Avoiding Sampling Bias Soil Sci. Soc. Am. J., September 1, 2003; 67(5): 1602 - 1606. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. T. Conant, G. R. Smith, and K. Paustian Spatial Variability of Soil Carbon in Forested and Cultivated Sites: Implications for Change Detection J. Environ. Qual., January 1, 2003; 32(1): 278 - 286. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Rothe, K. Cromack Jr., S. C. Resh, E. Makineci, and Y. Son Soil Carbon and Nitrogen Changes Under Douglas-fir With and Without Red Alder Soil Sci. Soc. Am. J., November 1, 2002; 66(6): 1988 - 1995. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| The SCI Journals | Agronomy Journal | Crop Science | |||
| Journal of Natural Resources and Life Sciences Education |
Vadose Zone Journal | ||||
| Journal of Plant Registrations | Journal of Environmental Quality |
The Plant Genome | |||
