| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
School of Natural Resources, The Ohio State University, Ohio Agricultural Research and Development Center, 1680 Madison Ave., Wooster, OH 44691-4096
Corresponding author (calhoun.2{at}osu.edu)
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Abbreviations: Db, soil bulk density • n, number of observations • P, probability • PTF, Pedo-Transfer Function • SD, standard deviation • SE, standard error • SOC, soil organic carbon • SS, sum of squares
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Generally, soil organic carbon (SOC) content, measured in grams per kilogram, and with inorganic particle size expressed as a percent of the 
2 mm fine earth, has been the most important continuous variable in regression models, followed by percent clay and other particle size fractions. The relative importance of SOC and particle size distribution is often reversed when models are developed for B and C horizons, because of the declining impact of SOC with depth (Bernoux et al., 1998).
We believe that soil genetic principles and soil morphology have been underutilized in developing predictive models for Db. Only oblique attention has been paid to the state factors of soil formation (Jenny, 1941, 1980) and their combined impact on Db, a fundamental soil physical property. During the course of progressive soil surveys, morphological data have been routinely collected. It is common knowledge that Db generally increases with depth, but actual depth functions vary with such factors as parent material, vegetation, physiography, and internal drainage. Consequently, field-determined genetic horizon designations might contribute more to a model than will depth alone.
Class-based variables such as site descriptors and morphology are nominal rather than continuous. Lin et al. (1999) used morphology and particle size mass fractions to develop Pedo-Transfer Functions (PTFs) for soil hydraulic properties. Their morphological database was from freshly excavated pits described using modern criteria. The motive for quantifying morphology was the realization that previous soil hydraulic models were primarily particle-size based and were inconsistent predictors because they ignored pore size distribution and continuity. Their approach was to develop a computer-optimized point scale system for each morphological class using hypothetical, structureless, and nearly impenetrable clay as the reference point. The starting points assigned were estimated from literature and empirical knowledge. This approach enabled them to convert variables from nominal to continuous for stepwise multiple regression. A similar approach could be used for Db prediction models, but there is little research on the relationship between state factors, morphology, and Db.
During the last 45 yr in Ohio, we have accumulated characterization and morphological data for more than 2500 pedons. These legacy data are currently being entered into a relational database program (Calhoun et al., 1999). We are now able to interact with 
60% of the morphological data and 100% of the site descriptions and laboratory data. The objective of this paper is to demonstrate that a combination of continuous (laboratory data) and nominal variables (site/state factor and morphological class descriptors) will enable us to develop improved PTFs for Db.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
The western half and northeast quadrants of the State were glaciated during the Wisconsinan. The southern half of the southwest quadrant is Illinoian age glacial till. Both tills are covered by variable thickness of Wisconsinan loess. Outwash deposits are associated with the glacial termini, abandoned glacial lakeshores, and major stream valleys. Locally significant lacustrine deposits are found between the ridge moraines. Lacustrine deposits are also extensive in the northwest quadrant because of ice blockage of northward flowing drainage systems into ancestral Lake Erie. The fine-grained sediments were subsequently exposed by isostatic rebound and ensuing agricultural drainage.
The southeastern and south central parts of the state are unglaciated. The soils there are mostly developed from residuum and colluvium derived from sandstone, shale, and limestone.
Physiographic features include ground moraines, ridge moraines, lake plains, outwash plains, and terraces, floodplains, and the dissected Allegheny plateau. Calcium carbonate content of glacial till decreases from west to east reflecting the bedrock base of limestone in the west and acidic sandstone and shale in the eastern half of the state. Fragipans are common in soils derived from the low-carbonate tills. A more detailed description of Ohio soils, climate, vegetation, and geology can be found in the publication Soil Regions of Ohio (Ohio Department of Natural Resources, Division of Soil and Water Conservation, 1997). A geomorphology map of Ohio can be viewed on the World-Wide Web (Brockman, 1998).
Description of Data Set
Site and Morphology Data
Field descriptions in the Ohio database span the time period from 1950 to 1999. Site and morphological descriptions were made by nearly 200 soil scientists from more than 2500 pedons selected for both soil survey and research purposes. Pedon descriptions were made according to evolving protocols contained in Soil Survey Handbook 18 (Soil Survey Division Staff, 1993) and its predecessors. In 1980, staff of the Ohio Soil Characterization Laboratory standardized terminology for site and morphological description (Smeck et al., 1980). The pedons included in this study that were sampled prior to 1980 have been edited to conform to these standards. The rationale for standardization at the time was to allow consistent coding for computer entry and storage of field data.
The site/morphology population used in this study encompasses a 35-yr span from 1950 to 1984. This population is composed of 937 horizons from 211 pedons. The number of horizons per pedon ranges from 1 to 9. The 211 pedons include 116 soil series, or 24% of the 475 soil series recognized in Ohio. Of the 88 counties in Ohio, 47 are represented in the database. The number of pedons per county ranges from 1 to 16. The data set contains nearly all classes of drainage, physiography, parent material, and vegetation found in Ohio. In certain cases, a class or classes have been combined with a neighboring class to avoid the incorporation of categories with minimal observations in the models.
Physiography classes for depositional landscapes were combined as follows: (i) beach ridges, river or stream terraces, kames, and eskers were included in the class of outwash plains; (ii) slackwater terraces were included with lake plains, and (iii) closed depressions were included in the physiographic class of the surrounding geomorphic surface. Soils developed from residuum and colluvium were placed into the Ruhe and Walker (1968)(p. 551–560) slope categories for erosional landscapes.
Vegetation classes were reduced to three: forest, meadow, and cultivated. Meadow included what was characterized at the time of description and sampling as hay, pasture, weeds, grasses, grasses and shrubs, and park (trees and grasses). This classification grouped pedons that were neither cultivated at the time nor under forest. Cultivated includes what was called "plowed" and "cultivated field".
Parent materials classed as beach deposits, glacio-fluvial, glaciolacustrine, and glacial outwash were combined and classed as "outwash", effectively separating these water-worked parent materials occurring on landscape positions from "alluvium" at lower elevations. Horizons developed from Wisconsinan and Illinoian age glacial till were combined into a single class of "till". Soils developed from limestone, shale, sandstone, and combinations of these (including colluvium from these materials) were grouped into the parent material class of "residuum". Loess was assigned on the basis of parent material information provided in the pedon description. Loess thickness was not always specified. In those cases, horizon descriptions and laboratory data were used as a basis for identification. This reduced the total number of parent material classes to six.
Preliminary analyses revealed that structure grade had less impact on Db than did structure size and type. Soil structure was then reduced to using a combination of size and type with each combination representing 17 classes of structure. There were no descriptions of medium platy structure. Very coarse as a size descriptor was found only with prismatic structure.
Matrix soil color (Munsell) was represented by adding value and chroma, and ignoring hue. The vast majority of hues were either 10YR (90%) or 2.5Y (6%). This created 12 classes of color. These were then reduced to eight when statistical means for Db were calculated because for value + chroma 
10 there was no further upward trend. We determined through simple linear regression (R2 = 0.48) that this set of color classes would be a significant proxy for SOC in the field-based models.
Horizon nomenclature was updated according to definitions in the Natural Resources Conservation Service Field Handbook (Schoeneberger et al., 1998). Vertical subdivisions of master horizons (including subordinate distinctions) were removed (e.g., Btg1 and Btg2 were designated as Btg). This generated 12 horizon classes.
No distinctions were made between fine and coarse sand, nor between fine and coarse loamy sand textures. Silt and sandy clay textures do not occur in this data set, leaving a total of 10 texture classes.
Laboratory Data
Approximately 10% of all pedons described were sampled for bulk density. Between 1950 and the end of 1965, Db was determined from 7.6-cm diam. cores as described in Blake and Hartge (1986). Since 1965, the procedure has used saran-coated natural clods (Brasher et al., 1966). With the advent of the natural clod method, coarse fragment content was determined on all samples. Prior to that, coarse fragments were seldom measured in the laboratory and were mostly estimated in the field. Historically, the clod method was instituted to measure linear extensibility; consequently, all Db values were corrected for coarse fragment content. For most horizons, Db means for natural clods are slightly less than for cores, with the exception of A and Ap horizons, where natural clod means are higher, and for Bg and Btg horizons, where they are equal (Fig. 1)
. This difference, recognizing that these are two separate data sets, may reflect correction for coarse fragments in the clod method and/or compaction associated with the cylinder method. Van-Remortel (1993) compared the two methods on forested coastal plains soils in Virginia. Mean Db values for the clod method were consistently higher than those for the core. The data were not corrected for coarse fragments, and the soils appeared to have relatively high rock fragment contents. Db reported in this paper is restricted to mass/volume obtained following desorption to 33 kPa of either the natural clod or core, and is measured in g cm-3.
|
Of 22 forest sites, only 10 A-horizons were sampled for Db. This resulted from the difficulty of obtaining a representative sample from a root-permeated matrix. This also results in a higher mean Db for the sample population. All 61 horizons classed as residuum were sampled using the coated natural clod method because unglaciated counties were generally the last to be mapped in Ohio.
Statistical Procedures
A query was made of the Ohio Soil Database program to identify all horizons that had both a bulk density measurement and morphological data. These data, along with the laboratory characterization and site data, were then exported to a spreadsheet where they were edited as described above. After editing, the data were exported to a JMP (SAS Institute, 1996) software program (version 3) file for statistical analysis. The resulting table was composed of a mixture of state factor/morphology (nominal) variables and laboratory data (continuous) variables. Three models were generated using all 937 horizons. A stepwise multiple regression platform using only continuous variables generated a Lab Model. Next, a standard least squares platform using only nominal variables generated a Field Model. Finally, a standard least squares platform using a combination of continuous and nominal variables created a Field and Lab Model.
Forward stepwise regression with a probability-to-enter threshold of 0.25 was used for the Lab Model. Log, square root, and squared transformations were assessed, but had no beneficial effect. Some interaction terms were beneficial and were included in the model.
For both remaining models, the procedure used was to first run the standard least squares platform using all nominal variables for the Field model and all variables for the Field and Lab Model. The effect test table was then examined and any variable with a P >F of 0.05 was removed from the model. Following this, the model was reevaluated. Through this iterative process, a final model was accepted which minimized the difference between R2 and adjusted R2, and maximized the Sum of Squares (SS) and F ratio for each effect.
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
|
|
|
|
= 0.05 shows that value + chroma classes 3 and 4 are not significantly different. Classes 7 through 10 are significantly different from 3 through 6 (Fig. 3a)
. The number of observations (n) for each class is a major determining factor for the radius of the Tukey-Kramer means effect circles.
|
The scatter plot for the model shows three distinct features (Fig. 2b). First, there continues to be a grouping of A horizons where predicted Db is greater than measured Db. The Field Model is improved compared with the Lab Model, but it still does not compensate completely. Secondly, the cluster of under-predicted Dbs (primarily Bx and C horizons) between 1.7 to 1.9 found in the lab data model plot are now centered around the whole model regression line. Finally, the predicted Db maxima occur at around 1.75 and are manifested by a vertical line of points perpendicular to the X-axis. This is an artifact of the nominal character of the variables used in the Field Model. Improved dispersion will be noted with the inclusion of continuous variables in the Lab + Field Model to be discussed in the following section.
Coefficients for each class in the overall model generally have expected signs and magnitudes. Because of rounding, the sum of coefficients within each class does not always equal zero. The largest negative coefficient is for forest A horizons, while Ap horizons are 50% less, an eloquent testimony to the impact of agricultural use. Fragipans (Bx horizons) have the largest positive coefficient. The range of coefficient size is related to the magnitude of the class effect. For the Field Model, the ranking of effect from greatest to least, determined on the basis of coefficient range, is horizon, structure, physiography, texture, consistence, value + chroma, parent material, vegetation, and drainage. The position of physiography in the ranking is somewhat questionable because of the low n for classes such as noseslope, backslope (3 horizons, one site), headslope, and shoulder (5 horizons, one site) that seem to be unduly influential in the model. In an alternative model not shown here, the unglaciated landforms were reduced to two classes: summit and slope. This combination decreased the coefficient range from 0.42 to 0.20 and also resulted in a slight decline in R2 (0.68) signaling a small loss of sensitivity in the model.
The Field Model demonstrates the importance of site description and morphological data. This data collection exercise was probably not always accorded equal importance to the laboratory analyses that followed. However, it is encouraging that field data collected by a large number of soil scientists across a 45-yr period, during which description protocols and capability were evolving, still providing a highly significant result.
Field + Lab Model
Addition of SOC and inorganic mass fractions to the model slightly improves R2 (Fig. 2c). SOC, clay and silt account, without transformations, for 49% of the variability and the nominal variables add another 23%. The scatter plot shows improved dispersion around the regression line compared with the Field Model (Fig. 2b).
Value + chroma and drainage in the Field Model are replaced by SOC in the Field + Lab Model (Table 4, Column "FL"). There are strong relationships between soil color and SOC and between SOC and Db (Fig. 3). The relationship between color and Db is not as strong, but is still significant. Soil texture determined in the field is replaced by particle size determined in the laboratory.
The coefficient magnitude and order do not change for structure, moist consistence, and physiography. The coefficient range declines by 50% for horizon, and only slightly for vegetation. The range for parent material is the same, but the signs and magnitudes of some coefficients had changed. For example, alluvium is now negative and loess is positive in the Field + Lab Model. There is no apparent explanation for this shift in coefficient magnitude, other than the impact of silt and clay contents in the model. Substituting depth of sampling for horizon in the Field + Lab Model reduced R2 by 0.025 with P > F of 0.1563. Depth was not significant, which demonstrates that genetic horizons are more important than depth alone in predicting Db.
The importance of genetic horizons as a strong determinant of Db is demonstrated in Fig. 4 by the means of residuals (measured Db minus predicted Db) from the three different Db prediction models. The Tukey-Kramer test is more conservative than the Student's t-test, and appropriate when sample sizes are unequal. The radii of the means comparison circles are a function of the "honestly significant difference". Students t-test generates a more liberally interpreted LSD, which should only be used when sample sizes are equal or, in other words, all horizons had an equal number of Db measurements. The residual means for the Lab Model (Fig. 4a) are negative for A, E, AB, and Bt horizons, indicating that the model overpredicts Db for these horizons. It strongly underpredicts Db for Bx and C horizons. Excluding horizon as a nominal variable, residual means by horizon for the Lab + Field Model are nested within the A horizon circle (Fig. 4b). Within the nested means, the mean of the residuals for Bx and C horizons are significantly different from the mean for the Bt horizons. Including horizon as a nominal variable in the Lab + Field Model now centers the mean of residuals for Db on zero (Fig. 4c).
|
The Lab Model for loess over till (Column L l/t in Table 4) accounts for 63% of the variability in Db. This an absolute 7% improvement in comparison with the Lab Model for all horizons. The entry sequence and variability accounted for was SOC (48%), sand (9%), depth (3%), and SOC/clay (2%). Fine sand and very fine sand accounted for the remaining 1%. Sand content plays a much more important role in determining Db on moraines than when all physiographic components are included. SOC still is the most important determinant along with the interaction of SOC and clay in loess and till soils.
The loess over till Field Model (Column F l/t in Table 4) has an R2 of 0.78, which is an absolute improvement of 15% in comparison with the Lab Model (L l/t). Neither physiography nor vegetation were significant contributors to the model. The largest coefficients are associated with Ap and C horizons. The larger coefficient for Ap horizons compared with A horizons affiliated with forested sites is unexpected. It is possible that fine and medium granular structure, normally found in A horizons under forest, absorbs some of the A horizon effect in the model.
The Field + Lab Model for loess over till (Column FL l/t in Table 4) explains nearly 80% of the variability in Db. This is an improvement of only 2% when compared with the Field Model. The difference between loess and till horizons (parent material) is no longer significant because of the presence of clay free silt in the model. Soil color and drainage class are replaced by SOC. Inclusion of SOC probably was responsible for a decline in magnitude of the coefficients for fine and medium granular structure. Soil texture is replaced by the two particle size interaction variables.
Parent material as a grouping preference may be the most promising categorization because of regional suitability, in contrast to other possibilities such as taxonomy or physiography. By assigning each horizon a parent material class, lithologic discontinuities are recognized and assimilated into the model. This also separates, in this example, the effect of glacial compression on the density of till C horizons versus water-deposited C horizons (Table 5).
|
|
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Received for publication May 1, 2000.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  N. Volland-Tuduri, A. Bruand, M. Brossard, L. C. Balbino, M. I. L. de Oliveira, and E. de Souza Martins Mass Proportion of Microaggregates and Bulk Density in a Brazilian Clayey Oxisol Soil Sci. Soc. Am. J., September 1, 2005; 69(5): 1559 - 1564. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. A. Heuscher, C. C. Brandt, and P. M. Jardine Using Soil Physical and Chemical Properties to Estimate Bulk Density Soil Sci. Soc. Am. J., January 1, 2005; 69(1): 51 - 56. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
Z. Tan, R. Lal, N. E. Smeck, F. G. Calhoun, B. K. Slater, B. Parkinson, and R. M. Gehring Taxonomic and Geographic Distribution of Soil Organic Carbon Pools in Ohio Soil Sci. Soc. Am. J., November 1, 2004; 68(6): 1896 - 1904. [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 | |||
