My Account: Log In | Join | Renew
Table of Contents
Select All Chapters
Never before in the history of soil science has the knowledge of soil spatial variability been so germane. It reaches to the heart of the pedology profession and is critical to the success of agronomic practice, agricultural development, land management, and earth science on a global scale. One of the continuing challenges for pedologists and allied earth scientists is to develop integrated system models to scale spatial knowledge of soils from microsamples to pedons, landforms and the pedosphere. Quantification of the magnitude, location and causes of spatial variability is an essential but insufficient ingredient of soil surveys. The final payoff is to communicate this knowledge to user clientele in flexible formats that provide for probability risk assessments and alternative land-use decisions. In this electronics era of exploding information systems, we have an opportunity as never before to (i) impact the direction of pedology, (ii) quantify spatial knowledge, (iii) add quality to soil resource inventories, and (iv) test multiple working hypotheses. Real-time assessment of spatial variability allows differential application and treatment of chemicals, pesticides, irrigation waters, and waste products on small site-specific areas. This chapter elucidates the traditions, relevance, challenges, and opportunities for pedologists to assume leadership roles in augmenting spatial knowledge and its application for wise stewardship of soil resources, the sustainability of the global environment, and the preservation of present and future civilizations.
Historically, use of landscape models has shown that landscapes are predictable; they have a large nonrandom variability component. This nonrandom variability can be used to predict soils on the landscape if the methodology both to describe and to quantify processes that govern landscape development is understood. The bases for our understanding of soils and landscapes are the concepts of Davis and Penck and those of G. Milne, L.C. King and R.V. Ruhe in whose models process plays an important role. The landscape must be defined in three dimensions, the lateral as well as the vertical changes in stratigraphic materials. The hydrologic characteristics of the system, particularly lateral flow, must be determined. A better understanding of stratigraphic control on hydrologic parameters is needed. Neither landscape nor process components are defined adequately by present soil map units. Milne's catena model is an excellent foundation for integration of these concepts.
This paper discusses some of the differences between geologic mapping and soil mapping, and how the resultant maps are interpreted. The role of spatial variability in geologic mapping is addressed only indirectly because in geologic mapping there have been few attempts at quantification of spatial differences. This is largely because geologic maps deal with temporal as well as spatial variability and consider time, age, and origin, as well as composition and geometry. Both soil scientists and geologists use spatial variability to delineate mappable units; however, the classification systems from which these mappable units are defined differ greatly. Mappable soil units are derived from systematic, well-defined, highly structured sets of taxonomic criteria; whereas mappable geologic units are based on a more arbitrary heirarchy of categories that integrate many features without strict values or definitions. Soil taxonomy is a sorting tool used to reduce heterogeneity between soil units. Thus at the series level, soils in any one series are relatively homogeneous because their range of properties is small and well-defined. Soil maps show the distribution of soils on the land surface. Within a map area, soils, which are often less than 2 m thick, show a direct correlation to topography and to active surface processes as well as to parent material.
Reliable soil surveys can be made at reasonable cost because the location of soils are predictable on the landscape. This soil-landscape relationship is the scientific basis that makes it possible to produce a soil mapping model. The soil scientist designs map units based on these models. Saunders County, NE, is used to illustrate the application, design, and redesign of map unit models. Data from pedon descriptions, transects, field notes, and laboratory analyses and knowledge of landscape patterns, geology, and climate are used to develop the models. The model for each map unit is tested during mapping and is adjusted as needed. Well-conceived and tested map units derived from models result in reliable and accurate soil surveys.
The field of statistics is devoted to understanding variability in populations and then using this understanding to compare populations. The objective of this chapter is to give a brief overview of some of the statistical procedures that are used in soil science to quantify differences between soil types and to describe the spatial and temporal variability that is present within mapping units. Procedures that are described include a group of tools called parametric and nonparametric statistics. These all require a prior knowledge of, or an assumption about the probability distribution of the population. These tools also require that the samples be temporally and spatially independent. The second group of tools described is called geostatistics. Geostatistics are based on the theory of regionalized variables that combines our understanding of the continuous nature of geologic properties in space with the random variation that is present in spatially separated samples. Procedures included in geostatistics are extensions of the classical statistical tools with the assumption of sample independence removed. A discussion of the problems and procedures associated with sampling for spatially variable soil properties precedes the description of the various statistical procedures. This chapter provides a direction for further study of the literature by those concerned with this subject.
Map unit composition has traditionally been determined by transecting randomly selected delineations of the map unit in question, recording the soil series at each point on each transect, and calculating the average proportion of each soil that occurs within the map unit. Confidence intervals about the mean for each soil are then determined using either the Student's t-distribution or a binomial method. Data from randomly selected transects were analyzed by both methods and the results were compared. The influence of nonuniform sampling densities, and variations in the number of points sampled in each delineation were also examined. Results indicate that for small sample sizes the binomial method gave narrower confidence intervals; but for large samples the two methods were equivalent. Due to the large number of samples required for geostatistics, its use in evaluating map unit composition is limited.
This chapter reviews the major options available to designers of spatial sampling systems and illustrates some of them with recent case studies carried out in Canada, the Netherlands, and Venezuela. The chapter is concerned only with those attributes of the soil that can be measured quantitatively and does not deal with attributes measured on nominal scales. There is no general optimal sampling design for quantifying map unit composition. The most appropriate sampling design (and hence numbers of samples and cost) depends on the survey aims and the spatial variation of the soil attributes studied. Different sampling approaches must be used depending on whether general or specific information is sought, and whether sampling is to determine the degree of spatial variation and to model it using the variogram before systematic sampling for mapping, or for mapping itself. The variogram must also be estimated by sampling. Sample support size in relation to survey aims and levels of local variation can be critical. Besides sample support size, the spatial resolution (scale) of the units needed to link data gathered from sampling to other spatial data is important. Alternative techniques to more intensive sampling of expensive-to-measure properties are cokriging, regression models and interactive graphics.
One of the most important facts regarding soil maps, soil map units, and associated text/tables is that they are imprecise. It is well to continue characterization of map units and their variability for the benefit of both the makers and the users of soil inventories. Tabulations of statistics (mean percentages, confidence intervals, confidence levels, etc.) should be presented in soil survey reports if available; but such data must be reported and used cautiously. Simple descriptive statistics do not necessarily lend themselves to probability statements and risk analysis. Users need understandable explanations of reported statistics and their proper use. Users also need improved definitions, explanations, and interpretations of similar and contrasting soils. Close cooperation between users and field soil scientists, educational efforts to assist the public in using soil survey information, and interdisciplinary research to improve understanding of the landscape should take priority over radical changes in format or style of soil survey reports.
Spatial variability of soils in the landscape and how this variability is represented by soil maps and portrayed in map unit descriptions are critical for assessing many of today's pressing environmental concerns. Traditional concepts used in mapping, naming, defining, and correlating map units need to be redirected from a focus on taxonomic or use criteria to the use of soil properties as the basis for correlation. Map unit delineations need to represent real segments of the landscape by capturing discrete soil patterns that function as workable stratifications of the landscape and provide detailed soil information for use in environmental assessment. A reliable estimate of the proportionate extent of map unit components is needed. By grouping soils into response classes, minor soil components that are important to understanding landscape dynamics and assessing environmental concerns can be retained in regional-scale databases. Current examples of regional and global soil databases are presented and a global framework approach for nesting mapping databases of different scales is proposed.
Simulation modeling is increasing in importance as a tool for soil survey interpretations. Simultaneous consideration of both agronomic and environmental interpretations will be required as soil management systems are designed that allow optimal, sustainable production while minimizing adverse environmental consequences. Taxonomic data, as currently obtained in soil survey, can be used to estimate basic parameters needed for simulation by defining pedotransfer functions that relate simple soil characteristics to more complicated model parameters. Some important data are not currently being gathered during soil survey, such as properties of the soil surface, measurement of soil bulk densities, and temporal (seasonal) changes in properties. Satisfactory application of models is possible only if such data are obtained in the future.
Characterizing soil spatial variability in soil surveys is critical for maintaining user confidence and soil survey credibility. The purpose of this investigation was to examine the use of USDA-SCS map unit concepts for portraying soil spatial variability and then demonstrate methods for displaying map unit composition in soil survey reports. We examined the composition of four consociation map units in Brazos County, located in the Tertiary Gulf Coastal Plain of Texas. The point-intercept transect method and binomial probability and classical statistical procedures were employed. In all but one case, sufficient observations were obtained to estimate the compositions within ± 15% of the mean at the 80% probability level. Results showed that taxonomic purities of the reference taxa ranged from highs of 49 and 45% for the Crockett (Udertic Paleustalfs) and Spiller (Udic Paleustalfs) series, respectively, to lows of 21 and ltWo for the Robco (Aquic Arenic Paleustalfs) and Rader (Aquic Paleustalfs) series, respectively. The concepts of similar and dissimilar soils, as they related to the major land uses of the area, were used to establish interpretive groupings. Subsequent interpretive purities of the Crockett and Spiller improved to 86 and 81%, while the Robco and Rader improved to 52 and 48%, respectively. The Robco and Rader did not meet the criteria for consociations and were therefore designed as multitaxa map units. Because the named components of these units did not occur in a consistent, coterminous pattern from delineation to delineation, a complex map unit could not be designed. We are therefore proposing the implementation of a new map unit concept to accommodate these conditionalities. In addition, we propose that tables be inserted in future soil surveys that will statistically rate, at given probability levels, the compositional purities of the map units in the survey area
Studies were conducted in three New England states to determine the taxonomic homogeneity of a map unit. In New England, extensive areas of Lyman (loamy, mixed, frigid Lithic Haplorthods) and Tunbridge (coarse-loamy, mixed, frigid Typic Haplor-thods) soils occur in an intricate pattern on over 300 000 ha of upland area. Commonly these soils are mapped as multitaxonomic map units. However, in areas of Lyman and Tunbridge soils, determination by depth to bedrock is complicated due to coarse fragments and the irregular bedrock surface. Four study sites were selected in delineations thought to represent the central concept of the Tunbridge-Lyman map unit. Systematic sampling using ground-penetrating radar (GPR) identified significant portions of the map unit to have characteristics outside the range of identified taxa. Results revealed that soil scientists are consistently underestimating the depth to bedrock. Soil-bedrock models have been reconstructed in order to improve soil map unit design, descriptions, and interpretations.
Estimates of the means of various soil properties within map units may be important to users. It seems appropriate to measutr and report confidence intervals for the means of selected soil properties within map units. Such properties should be important to the expected use and management of eht map unit, or should provide important baseline data for future uses. Data on pertinent soil properties can be collected as part of a systematioc method of randomly selected transects, such as is presently used to determine map unit composition in some soil surveys. Standard statistical procedures can then be used to calculate confidence intervals for the means of these properties. The resulting ranges may or may not be wholly within the bounds of the conceptual series or taxonnomic range. The ranges can be included in the soil survey report in a table, as ranges in the map unit descriptions, and as part of the database for a digitized soil survey, Quantified ranges for the means of soil properties within map unites will significantily improve the quality and usefulness of National Cooperative Soil Survey (NCSS) soil surveys.
The spatial variability of soil organic matter (OM) content as measured by loss-on-ignition, was investigated along several transects and sampling grids in four Massachusetts soil map units. Within four short transects, each located within a single delineation of a soil map unit and sampled at 0.5- or 1.0-m intervals, autocorrelation of the OM content was sufficiently strong to constitute a significant trend that could be removed by curvilinear or nonlinear regression. The regression residuals in these cases evidenced insufficient remaining autocorrelation to merit further analysis using kriging. Organic matter content along a longer transect (1.2 km) located within a single delineation of a map unit sampled at 15-m intervals showed a weaker trend; but the residuals left by regression used to remove trend again showed little autocorrelation. Along a 1.2-km transect crossing the boundaries between three different soil map units, soil OM contents were highly autocorrelated reflecting different OM contents in the soil delineations. These differences formed a pattern too complex to be considered a trend and to be removed by regression. A kriging analysis with jackknifing was able to explain 87% of the variation in OM levels along this transect. Within a two-dimensional grid spanning four soil delineations, OM content was found to follow a positively skewed frequency distribution. The semivariograms of logarithms of soil OM content in different directions within the grid showed evidence of anisotropy, which was removed by dilating the coordinates of the sample locations in one direction. The resulting isotropic semivariogram was used to prepare a map of OM contents within the grid using block kriging. Boundaries on this map resembled map units delineated on the existing 1:15 840 scale soil map.
Land-use planners are requiring site-specific information too detailed for inclusion in published soil surveys. Special-use maps are receiving increasing attention as management alternatives to the soil survey. The special-use maps are often generated with Geographic Information Systems (GIS). Soil survey information frequently is digitized as a part of the data base for special-use maps. Slope and aspect information commonly is generated from Digital Elevation Models (DEM). Field verification of digitized soil-landscape maps is lacking. The opportunity exists to create confusion and ambiguity by layering databases of unknown accuracy and precision. Ground-truth measurements should be obtained to verify precision of computer-generated special-use maps. Soil series, seasonal distribution of soil water, and plant growth and yield are highly correlated to geomorphic surfaces. Digital Elevation Models and GIS have great potential as research tools to enhance our understanding of the process involved in evolution of soil landscapes.