Predictive soil mapping: a review - CiteSeerX

Progress in Physical Geography 27,2 (2003) pp. 171–197

Predictive soil mapping: a review P. Sculla,*, J. Franklina, O.A. Chadwickb and D. McArthura aDepartment

of Geography, San Diego State University, San Diego, CA 92182-4493, USA bDepartment of Geography, University of California, Santa Barbara, Santa Barbara, CA 93106, USA

Abstract: Predictive soil mapping (PSM) can be defined as the development of a numerical or statistical model of the relationship among environmental variables and soil properties, which is then applied to a geographic data base to create a predictive map. PSM is made possible by geocomputational technologies developed over the past few decades. For example, advances in geographic information science, digital terrain modeling, remote sensing, fuzzy logic has created a tremendous potential for improvement in the way that soil maps are produced. The State Factor soil-forming model, which was introduced to the western world by one of the early Presidents of the American Association of Geographers (C.F. Marbut), forms the theoretical basis of PSM. PSM research is being driven by a need to understand the role soil plays in the biophysical and biogeochemical functioning of the planet. Much research has been published on the subject in the last 20 years (mostly outside of geographic journals) and methods have varied widely from statistical approaches (including geostatistics) to more complex methods, such as decision tree analysis, and expert systems. A geographic perspective is needed because of the inherently geographic nature of PSM. Key words: GIS, soil geography, soil survey.

I

Introduction

Soil is a fundamental natural resource; it is the basis of human agriculture. Civilizations rise in regions blessed with rich soil; they fall when humans fail to treat soil with respect. Further, soil plays an essential role in the biophysical and biogeochemical functioning of the planet. On the continents, soil forms a porous boundary where the biosphere, hydrosphere, lithosphere and atmosphere interact. Understanding the *Author for correspondence: Tel., 315-228-7864; fax, 315-228-7726; e-mail: pscull@mail. colgate.edu © Arnold 2003

10.1191/0309133303pp366ra

172

Predictive soil mapping: a review

spatial distribution and management of soil is critical to maintain a productive society and to understand the complex balance of chemical and physical processes that make life possible on Earth. While primarily referred to as pedology and studied in soil science departments, such themes are the focus of many different academic disciplines. For example, soil geography is the study of the location, distribution and pattern of soils on the landscape (Buol et al., 1997). Geography and pedology share a common history in this century. C.F. Marbut, a geographer by training, was responsible for communicating pedological ideas, developed by the Russian school of pedology, within the USA (Cruickshank, 1972). Marbut was a student of W.M. Davis, the first president of the American Association of Geographers (AAG), and later became president himself in 1924. As a result, soil was an active area of research within geography during the early years of the AAG. While soil geography is still occasionally published in geography journals (see Barrett, 1999, for a recent perspective), a great deal of geographic soil research has appeared over the last 20 years outside of geographic journals. From an applied perspective, pedology and soil geography form the basis of soil survey, which remains the primary means by which information on the spatial properties of soil is collected, presented and archived in the USA and throughout the world. The United States Department of Agriculture (USDA), in cooperation with other federal and state agencies, began soil survey work in 1896. The combined state and federal government effort became known as the National Cooperative Soil Survey (Indorante et al., 1996). Its mission was to furnish private landowners, land managers and consultants with soil maps to aid in land-use decision making. More recently, however, soil maps have been used to provide chemical and physical data input within ecological and hydrological process models (Burrough and McDonnell, 1998: 163). The soil survey program was simply not designed to furnish data for such applications. The increasingly sophisticated use of soil data has led to a greater demand for data about soil properties than the conventional soil map can accommodate (Cook et al., 1996). Traditional soil survey concepts are based on qualitative recognition of soil properties in relation to landscape and environmental variables. Although these methods implicitly incorporate the expertise of the soil scientist, they do not make use of geocomputational technologies that are now widely available. Technological advances during the last few decades have created a tremendous potential for improvement in the way that soil maps are produced (McKensie et al., 2000). Remote sensing and photogrammetric techniques provide spatially explicit, digital data representations of the Earth’s surface that can be combined with digitized paper maps in geographic information systems (GIS) to allow efficient characterization and analysis of vast amounts of data. The future of soil survey lies in using GIS to model spatial soil variation from more easily mapped environmental variables. Predictive soil mapping (PSM) begins with the development of a numerical or statistical model of the relationship among environmental variables and soil properties, which is then applied to a geographic data base to create a predictive map (Franklin, 1995). Three main goals of PSM are to: (1) exploit the relationship between environmental variables and soil properties in order to more efficiently collect soil data; (2) produce and present data that better represent soil landscape continuity; and (3) explicitly incorporate expert knowledge in model design. PSM can also potentially advance pedology and soil geography by providing insights into soil forming processes.

P. Scull et al.

173

The State Factor soil-forming model forms the theoretical basis of PSM as well as of traditional soil survey, and has been used in other many other disciplines. Hans Jenny popularized the State Factor theory of soil formation in the USA through his publication of Factors of soil formation (1941). The ideas expressed in this classic text represent a formalization of Dokuchaiev’s ideas of soil formation (Simonson, 1997). Since that time, the theory has provided a paradigm through which soil genesis and distribution can be studied (Schaetzl, 1991). The theory states that soil profile character is a function of climate, organisms, relief, parent material and time, implying that if the spatial distribution of the soil-forming factors is known, soil character may be inferred. This theoretical framework has been used by many authors in pedological research and remains the most popular theory of soil genesis (for a recapitulation of Jenny’s continuing influence on pedology, see McSweeney et al., 1994). Until recently, predictive soil mapping efforts have been constrained by two problems associated with Jenny’s State Factor theory of soil genesis: the mathematical equations derived from theory proved virtually unsolvable (Huggett, 1975) and data describing the soil-forming factors were unavailable, or not widely available. Recent advances in mathematical theory (fuzzy logic) and statistical methods (including machine learning techniques) have helped researchers to better understand the nature of the equations implied by the theory, and to solve them. Relative to the problem of soil-forming factor knowledge, Jenny (1941: 262) wrote, ‘the conversion of such fundamental knowledge (soil–environment relationships) to specific field conditions is impossible unless the areal distribution of the soil formers is known’. Half a century later, remote sensing and photogrammetry along with GIS can now be used to characterize the spatial distribution of soil-forming factors. The foundation has been laid for the emergence of predictive soil modeling as an active area of research. Several articles have been published outside of geographic literature to track progress in the field of PSM over the years. Two earlier articles (McBratney, 1992; Hewitt, 1993) called for an overall change in the philosophy of soil survey and outlined research challenges in the field at that time. In the mid 1990s the term pedometrics was coined (Webster, 1994) to refer to quantitative research in the field of pedology. This term would appear synonymous with PSM as defined above. More recently, reviews have been published on a subset of PSM techniques (McBratney et al., 2000) and general methods of modelling soil variation (Heuvelink and Webster, 2001). Predictive soil mapping has only recently emerged as a research niche, but the pace of research, breadth of knowledge and variety of techniques have expanded rapidly (pedometrics have accounted for 18% of the subject matter in articles published in recent years in Geoderma; Hartemink et al., 2001). A review of recent achievements in the field of PSM is needed because all methods have not been reviewed in past papers. Further, a geographic perspective is needed because of the inherently geographic nature of PSM. First, we review several topics critical to PSM (Section II), including the nature of soil variability (Section II,1), how the soil survey has defined soil mapping (Section II,2), and soil taxonomy (Section II,3). We then briefly describe the components of PSM (Section III): geographic data models of soil variability and soil mapping (Section III,1), digital terrain modelling (Section III,2), fuzzy logic (Section III,3), and remote sensing (Section III,4). Lastly, we discuss the present state of knowledge in PSM (Section IV). Our focus in this review is to evaluate literature in the field of PSM and to reintroduce the subject matter to the geographic audience.

174

II


Background

Prediction of soil properties based on knowledge of the effect of environmental variables on soil formation has always been the basis for all soil mapping. Unfortunately the traditional methods do not yield quantifiable soil–landscape information that robustly describes actual soil variation. In this section, we describe soil variation and explain why traditional methods for defining soil distribution in landscapes are inadequate. It is also our purpose to demonstrate that there is a logical flow from these approaches into the new PSM techniques. All approaches to soil mapping rest on our ability to use knowledge of the process of soil genesis to predict the properties of soils at any point in the landscape. 1

The nature of soil variability

The complex and highly variable nature of soil patterns in landscapes complicates the already labour-intensive process of collecting and presenting soil survey data (Wright and Wilson, 1979). In 1941, Jenny listed three different definitions of soil before he conceded that ‘it is problematic whether any definition of soil could be formulated’ (Jenny, 1941: 2). In the half-century that has passed since Jenny wrote those words there have been numerous attempts to define soil (reviewed by Birkeland, 1999). Most textbook authors describe soil as extraordinarily complex. For example, McKnight (1993: 336) defines soil as ‘an infinitely varying mixture of weathered mineral particles, decaying organic matter, living organisms, gases, and liquid solutions’ (emphasis added). This definition illustrates how complex soil can be. It further demonstrates that the soil landscape is continuous and is not composed of distinct individual soil types. This point has been made repeatedly over the years (Simonson, 1959; Webster and Beckett, 1968; Cambell, 1977; and Moore et al., 1993). A variety of soil genesis models have been proposed in order to account for the high variability of soil and, collectively, they help further illustrate the difficulty of characterizing the soil landscape (for a review see Huggett, 1975; Birkeland, 1999). Three distinctive approaches have been employed: factor models (e.g., Jenny, 1941), where factors affecting soil development are identified; process models (e.g., Simonson, 1959), where soil-forming processes are emphasized; and energy models (e.g., Runge, 1973), where the focus is upon process-driving mechanisms. A host of hybrid models have also been employed. Still, no real consensus exists today as to exactly how to model soil development, partially because of the recent emergence of pedology as an academic discipline (Johnson and Watson-Stegner, 1987). Existing data collection methods do not yield adequate soil information in part because many of the processes that shape the soil landscape are still poorly understood. 2

How the soil survey has defined soil mapping

Traditional soil survey persists as the most popular form of soil mapping and inventory, and in many cases is the only manner in which the highly variable nature of the soil landscape is catalogued. The method consists of three steps (Cook et al., 1996). The first is direct observation of ancillary data (aerial photography, geology, vegetation, etc.) and

P. Scull et al.

175

soil profile characteristics. In the second step, the observations of soil attributes are incorporated into an implicit conceptual model that is used to infer soil variation. The third step involves applying the conceptual model to the survey area to predict soil variation at unobserved sites. Usually, less than 0.001% of the survey area is actually observed (Burrough et al., 1971); a fact reflecting the high cost of field sampling. The conceptual model of soil variation is then transformed into a cartographic model, the choropleth map, by drawing map unit boundaries on aerial photographs. In effect, photographic scale determines the resolution of the soil map. This process has been severely criticized in the scientific literature for two reasons. First, the conceptual model developed by the soil surveyor is primarily implicit, being constructed in a heuristic manner. This results in an excessive dependence upon tacit knowledge and, as such, incomplete information exists relative to the derivation of the ultimate soil survey product (Hudson, 1992). This aspect of soil survey is especially frustrating because it fails to document most of the knowledge that the soil surveyor accumulates during the expensive field mapping process. In essence the soil survey is unfalsifiable and therefore untestable (Hewitt, 1993). The final product of soil survey is a soil map that has unknown assumptions, limitations and accuracy (Burrough et al., 1971; Dijkerman, 1974). PSM techniques are similar in theory to soil survey (they both use knowledge of soil–environment relations to make inferences), but the methods employed often yield quantitative expressions of soil variability with measured levels of accuracy. Clearly one direction of innovation in soil survey is to add objectivity to model development which will allow more explicit scientific communication (Gessler, 1996). The second major criticism of soil survey concerns the role of soil classification. 3

Soil classification

The evolution of PSM techniques has been directly impacted, and continues to be influenced by, the process of soil classification. This is especially true in the USA where the primary focus of the National Cooperative Soil Survey is to develop and map Soil taxonomy (the official classification developed by the U.S. Department of Agriculture; Soil Survey Staff, 1975). The system was influenced by nineteenth century biological taxonomy and the practice of geological survey (Heuvelink and Webster, 2001). The purpose of Soil taxonomy was to provide an objective manner to systematically classify soil and was adopted at a time when soil information had to be abstracted to the level of the modal profile (classified in Soil taxonomy) because it was impossible to catalogue and present the full amount of soil variability (Cambell and Edmonds, 1984). In order to map soil taxa the soil must be perceived as a spatial entity, a ‘pedon’ (a term used in Soil taxonomy to refer to the smallest recognizable unit that can be called ‘a soil’). In practice, this spatial perception of soil results in a map whose classes are homogenous units with unknown variability and sharply defined boundaries (Burrough and McDonnell, 1998). Since the initial development of Soil taxonomy the perception of the soil landscape has changed from a collection of individual soil types to a continually varying mixture of soil components (Dmitriev, 1983). Field observations have shown that in concert with the environmental variables, soil properties vary continuously across landscapes, exhibiting different and complex scales of variation (Simonson, 1959). Therefore, soil distribution is not well represented by choropleth maps

176


(McBratney, 1992; Gessler et al., 1995). While this change has not yet been recognized in Soil taxonomy, PSM research has developed methods of soil inventory that more accurately describe the soil landscape. Such methods are at odds with the traditional approaches because they often involve dismissing the concept of soil as a spatial entity. Rather, the new methods focus on mapping continuously varying soil properties. The implementation of these techniques has been difficult because of the entrenchment of Soil taxonomy. III 1

Components of predictive soil mapping Geographic data models of soil variability and soil mapping

Traditionally, soil maps have been digitized to fulfil the need for soil data within GISbased environmental modelling research. This information from the paper map digitized for the computer is fraught with the same problems of the original choropleth maps – assumed homogeneous units with unknown variability and sharply defined boundaries. GIS allows for a more robust characterization of spatial variability (relative to the cartographic generalizations of the past) by allowing information to be analysed and stored using a variety of data models. As defined in the GIS literature, data modelling is the process of discretizing spatial variation, which entails abstracting, generalizing or approximating geographic reality (defined as empirically verifiable facts about the real world). Unfortunately, data modelling is often confused with issues of data structure and limited by software selection (Kemp, 1992). The process is of crucial importance because it controls the manner in which the data can be processed or analysed (Goodchild, 1994), as well as the view of the data the end user ultimately receives (Goodchild, 1992a; Kemp, 1992). Some data models are more accurate than others at portraying geographical reality (Goodchild, 1992b). In the case of soil, how well do the data, including the data model, represent the highly variable, continuous nature of soil? How can digital computers be used to manage spatial data to best represent the soil landscape? The choice of data models is partially dependent upon how the soil is perceived in geographic space. Objects can be thought of as existing as independent entities in empty space (object view), or as one of an infinite set of tuples (the foundation of geographic information – x,y,z, where z is a measured value and x,y are its location in space) approximated by regions and segments (field view) (Goodchild, 1994). The field view better represents continuous surfaces, such as the soil landscape, but soil data have traditionally been modelled using the feature model of geographic space, the choropleth map. Soil data have probably been managed this way because of the influence of Soil taxonomy, which defines individual soil ‘types,’ and because soil data collection began during a time when few alternatives existed. Using PSM techniques a fundamental change in the soil data model from the choropleth map to the raster grid allows better characterization of actual soil–landscape variability. A raster data model accommodates a ‘field view’ representation of the landscape and is defined as a regular rectangular array of cells with some aggregate value of the field recorded for each cell (Goodchild, 1994). The resolution of the data stored in this format is a function of the grid cell size, which can be made small enough to simulate continuous variation at the landscape scale. The raster has become the most

P. Scull et al.

177

widely used data model for PSM and is also routinely used to manage other environmental information such as elevation (DEMs) and remotely sensed data. Spatial analysis and integration with other types of raster-based environmental data can be easily performed with soil data stored using a raster data structure (Burrough and McDonnell, 1998). 2

Digital terrain modelling

Terrain analysis quantifies the relief component of models characterizing soil formation. Soil development, and its associated profile characteristics, often occurs in response to the way in which water moves through and over the landscape, which is controlled by local relief. Accordingly, terrain analysis will be most useful in environments where topographic shape is strongly related to the processes driving soil formation (McKensie et al., 2000). Digital terrain modelling is a technique for deriving spatially explicit, quantitative measures of the shape character of topography (Weibel and Heller, 1991; Wilson and Gallant, 2000). The spatial distribution of the resulting terrain attributes (characterizing local water flow paths) can also capture the spatial variability of soil attributes. Moore, I.D. et al. (1991) reviewed the analysis of digital elevation data (including DEMs) for hydrological, geomorphological and biological applications. They provided a table that summarized the significance and physical meaning of various terrain attributes to landscape processes. Building on their work, many authors have used terrain attributes derived from digital elevation models (DEMs) as explanatory variables in predictive soil models (Odeh et al., 1991; Moore et al., 1993; Gessler et al., 1995; Skidmore et al., 1996; and others). Methods used to derive terrain attributes have been greatly refined over the last 15 years and future satellites aiding in the development of more accurate DEMs will make terrain analysis an increasingly important component of predictive soils mapping (Moore et al., 1993; Mackay and Band, 1998). Several recent review articles have been specifically devoted to the role of terrain analysis in soil mapping (Ventura and Irvin, 1996; Irvin et al., 1996; McKensie et al., 2000). 3

Remote sensing

Remote sensing data are an important component of PSM because they provide a spatially contiguous, quantitative measure of surface reflectance, which is related to some soil properties (Agbu et al., 1990). Both physical factors (e.g., particle size and surface roughness) and chemical factors (e.g., surface mineralogy, organic matter content and moisture) control soil spectral reflectance (Irons et al., 1989). Surface mineralogy can be derived by wavelength specific charge transfer and crystal field absorptions associated with the presence of iron and iron-oxides (Fe2+ and Fe3+), and vibrational absorptions associated with hydroxyl bonds in clays, adsorbed water and the carbonate ion (Goetz, 1989; Irons et al., 1989). The presence and strength of these absorption features can be used to identify and quantify concentrations of mixed suites of minerals in soil (Johnson et al., 1983; Shipman and Adams, 1987). Organic matter, particle size and moisture content, in contrast, influence soil reflectance primarily through a change in average surface reflectance, and produce only broad spectral

178


expression (Irons et al., 1989). A decrease in particle size tends to increase surface albedo and decrease spectral contrast of absorption features, while an increase in organic matter or soil moisture decreases average reflectance or albedo. Numerous studies have shown the potential benefits of using remote sensing for soil identification and mapping. Comprehensive surveys of soil spectral reflectance include studies by Stoner et al. (1980), Henderson et al. (1992) and Csillag et al. (1993). Remote sensing studies, based on broad-band sensors such as Landsat TM include Agbu et al. (1990), Coleman et al. (1993), Seyler et al. (1998) and Oliveira (2000). Traditionally, remote sensing has been used to classify soil units through photo-interpretation or digital image processing. Combining remotely sensed information with ancillary information such as thematic maps or vegetation cover can yield significant improvements (Wilcox et al., 1994; Cialella et al., 1997; Wanchang et al., 2000). Recent developments in hyperspectral remote sensing offer the potential of significantly improving data input to predictive soil models. Hyperspectral sensors, such as the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) measure a contiguous spectrum in the visible and NIR, and thereby better characterize atmospheric and surface properties (Goetz et al., 1985). The large number of spectral bands permits direct identification of minerals in surface soils. For example, Clark and Swayze (1996), mapped over 30 minerals using AVIRIS at Cuprite, Nevada. Palacios-Orueta and Ustin (1996) showed that enhanced spectral information was suitable for discriminating even subtle spectral changes associated with differences in organic matter and iron content. Other examples of the application of AVIRIS to aid soil mapping include PalaciosOrueta et al. (1998), Okin et al. (1998) and Roberts et al. (1998). Sensors that operate in the microwave portion of the electromagnetic spectrum have also shown promise in soil mapping research. Microwave sensing can be broadly divided into active (e.g., radar) systems and passive systems, and are capable of penetrating the atmosphere under virtually all conditions offering a significant advantage over visible and near-infrared spectroscopy (for a general overview of microwave sensing see Lillesand and Kiefer, 1994: chapter 8). Synthetic aperature radar (SAR) is one example of an active system. SAR has been used to aid soil property mapping, such as soil salinity (Metternicht, 1998) and soil moisture (Engman and Chauhan, 1995; Narayanan and Hirsave, 2001). Active radar systems can also be designed to collect data at varying look angles, providing the opportunity for the acquisition of stereo radar images. Such images can be used to produce high resolution and extremely accurate DEMs (e.g., Fang, 2000). A similar active sensing system is LiDAR (light detection and ranging), which uses pulses of laser light, rather than microwave energy, to illuminate the surface (see Bunkin and Bunkin, 2000 for a review of applications to soil mapping research). While passive microwave systems have seemed to receive less attention in the literature a few examples of soil mapping applications can be found (see Kleshchenko et al., 2000 and Laymon et al., 2001). Regardless of the type of system, remote sensing data, and derived products, are potentially useful explanatory variables in predictive soil mapping models. 4

Fuzzy logic

Fuzzy set theory or fuzzy logic provides an alternative conceptual paradigm within PSM research. The use of this theory has increased greatly in the last few years, making

P. Scull et al.

179

it an important component of PSM. Fuzzy logic is an alternative to Boolean logic that attempts to recognize the concept of partial truth (Brule, 1996). Dr Lotfi Zadeh (1965) introduced the concept and accompanying mathematics in his seminal work ‘Fuzzy sets’. The theory permits partial class membership in contrast to traditional set theory, where set memberships are crisp and binary (i.e., a soil sample is either completely Type A or it is not at all Type A). Central to the fuzzy concept is the idea that objects in nature rarely fit exactly the classification types to which they are assigned (Zadeh, 1965). Rather, they show varying signs of similarity to multiple classes (i.e., an observed soil pedon often resembles more than one of the defined soil series within the area). By using fuzzy membership values (ranging from 0, nonmembership, to 1, total membership) within predictive soil models to express degrees of similarity, generalization problems associated with classification schemes (filtering of information) are minimized, and the complex nature of soil data is allowed to propagate through the modelling process. Similarity values between 0 and 1 are not comparable to proportions and need not add up to 1. Within Boolean logic, probability statements refer to the likelihood of an outcome; the soil sample is either one series or another. With fuzzy logic, a given sample is not definitively a member of the subset of any one particular series. Fuzzy logic is especially useful in soil research because of the continuous and complex nature of the soil landscape. It serves as an important alternative to the subjective rigidity imposed on soils data by Soil taxonomy. Several recent articles provide a thorough review of the use of fuzzy sets in soil science (Burrough, 1989; McBratney and Odeh, 1997; Burrough et al., 1997; De Gruijter et al., 1997). Within PSM research, two different approaches to creating continuous classes using fuzzy logic exist. The first is based on the fuzzy-k-means classifier, which partitions observations in multivariate space into natural classes. This approach is similar to cluster analysis and numerical taxonomy, but the resulting classes are continuous with each observation assigned a fuzzy membership value that characterizes its degree of similarity to each individual class. The concept has been integrated into geostatistical methods and will be discussed in more detail below (see Section IV,1). The second approach is known as the Semantic Import model (SI), and is used in situations when classification schemes are pre-defined and class limits are relatively well understood. The SI model is commonly used in concert with expert knowledge and will be discussed in the expert systems section (see Section IV,4). IV

Recent advances in predictive soil mapping

Within the last decade many authors have sought to model the soil landscape using a variety of methods. Literature in this field could be summarized many different ways, but we concentrate on the literature that directly addresses the goals of predictive soil mapping stated in the introduction (see Table 1). Therefore, we will review research that attempts to exploit the relationship between quantifiable landscape indices and soil character, in order to model the soil landscape in a more continuous and, therefore, realistic manner. The research reviewed here is distinguished from decades of previous research documenting the correlation between landscape position and soil attributes (reviewed by Hall and Olsen, 1991). That body of research is informative, but not useful for

Modelling method

Linear and exponential regression

Punctual and block kriging

Fuzzy mathematical methods

Continuous classification

Factorial kriging

Classification tree

Baysian rule-based methods

Expert systems

Decision tree analysis, neutral networks

A large variety of statistic methods

Linear and logit regression

Factorial kriging

Study

Bell et al., 2000

Burgess and Webster, 1980a,b

Burrough, 1989

Burrough et al., 1997

Castrignano et al., 2000

Cialella et al., 1997

Cook et al., 1996

Dale et al., 1989 (a review)

Ellis, 1996

Gessler, 1996

Gessler et al., 1995

Goovaerts, 1992

Total carbon,

A horizon and solum depth, E horizon presence

Field and laboratory collected physical, chemical and morphological soil properties

Soil erosion class

–

Organic matter

Drainage class

CEC, pH, N, P, K, Na

–

–

Na content, cover loam thickness, stone content

Total soil organic carbon

Dependent variables

None

Curvature, CTI, topo. position

A variety of digital environmental data

Slope, aspect, wetness index, flow length and accumulation, Landsat TM, tree cover

–

Slope, aspect, wetness index

Elev., aspect, NDVI

None explicitly used

–

–

None

Slope, curvature, aspect, hillslope position

Environmental variables

C

E,C

E,C

E,c

–

E,c,X

E,c

C,x

rp

rp

C

E,C,x

Goals attaineda

Table 1 Selected recent literature on predictive soil mapping and mapping (cited in this article), describing the modelling method used, the dependent variables used, and the environmental variables (explanatory) used in the models

180


–

–

–

–

Logistic regression

Kriging, co-kriging, regression kriging

Classification tree

Kriging, splines, trend surface, nearest neighbor

–

Block kriging

–

Fuzzy sets in soil science

Fuzzy-k-means with extragrades

Artificial intelligence/ expert systems

Generalized linear models (logit)

Linear regression

Hartemink et al., 2001

Heuvelink and Webster, 2001

Hewitt, 1993

Indorante et al., 1996

King et al., 1999

Knotters et al., 1995

Lagacherie and Holmes, 1997

Laslett et al., 1987

McBratney, 1992

McBratney et al., 1991

McBratney et al., 2000

McBratney and Odeh, 1997

McBratney and de Gruijter, 1992

McCracken and Cate, 1986

McKensie and Austin, 1993

Moore et al., 1993

A horizon depth, OM and P content, pH

Clay content, CEC, pH, EC, COLE, bulk density and others

–

Fuzzy classes*

–

–

Clay content

–

pH

Mapping unit

Soft layer depth

Presence/absence Noncalc. clay-loam

–

–

–

pH, N, CEC, extractable cations (K, Ca, Mg) –

Slope, wetness and stream power indices, aspect, curvature

Slope, relief, landform, slope position

–

Field collected physical, chemical and morphological soil properties

–

–

None

–

None

Geology, various topo. indices

Hillslope position

Slope, aspect, pot. solar energy

–

–

–

–

E,C

e,c

rp

C

Rp

rp

C

rp

c

E,c

C,x

E,c

rp

rp

rp

rp

P. Scull et al. 181

Regression tree and linear regression

Fuzzy-c-means and kriging

Regression, kriging, cokriging, regression kriging

Bayesian expert system

Kriging, cubic spline

Development of pedometrics

Fuzzy logic expert system (SoLIM)

McKensie and Ryan, 1999

Odeh et al., 1992a,b

Odeh et al., 1994, 1995

Skidmore et al., 1991 1996

Voltz and Webster, 1990

Webster, 1994

Zhu, 1997a,b; Zhu and Band, 1994; Zhu et al., 1997

Soil series, A horizon depth, individual series maps

–

Clay content

Soil landscape unit

Solum depth, depth to bedrock, gravel and clay content

Fuzzy classes

Solum depth, P and N content

Dependent variables

Elev., p.m., aspect, canopy coverage, gradient, curvature

–

None

Veg. type, wetness index, gradient, terrain position

Slope, aspect, curvature

Field-collected physical, chemical and morphological soil properties

Elevation, slope, curvature, CTI, contributing area, downslope means for slope, climate data, Prescott Index, Gamma Radiometry, Landsat TM, and Geology unit

Environmental variables

E,C,X

rp

C

E,c,X

e,C

C

C,E

Goals attaineda

Notes: aLetters refer to the degree to which the goals of PSM defined in the introduction are achieved. Soil–environment relations utilized (letter E), better representation of soil continuity (C), and expert knowledge utilized (X) correspond to goals 1, 2, and 3 respectively from the introduction, and capital letters (E, C, X) indicate the method is relatively more successful than those methods denoted by lower case letters (e, c, x). rp indicates review papers.

Modelling method

Continued

Study

Table 1

182


P. Scull et al.

183

predictive mapping because landscape position is never quantified (rather, position was often qualitatively defined, e.g., toe-slope) and, thus, the documented relationships cannot be generalized using environmental data and digital elevation models to predict soil character at unvisited sites. Table 1 documents modelling methods, model variables, and the extent to which the referenced research satisfies the previously defined goals of PSM (‘Goals attained’ column). Soil-environment relations utilized (letter E), better representation of soil continuity (C), and expert knowledge utilized (X) correspond to goals 1, 2, and 3, respectively, from the introduction, and capital letters (E, C, X) indicate the method is relatively more successful than those methods denoted by lower case letters (e, c, x). Citations with no letters present within the ‘Goals attained’ column do not address the aforementioned goals. For example, the Cialella et al. (1997) received a rating of ‘E,c’, meaning that the methods employed successfully utilized environmental–soil character relations (E) and somewhat successfully presented a better method of representing soil continuity (c). The ratings are provided simply to help organize the literature that was reviewed. Review papers are included within the table, denoted by ‘rp’. Geostatistical methods are not included in the table because they have been comprehensively surveyed elsewhere (Odeh et al., 1994; Burrough et al., 1997; McBratney et al., 2000; Heuvelink and Webster, 2001) and because the objectives and assumptions of geostatistical methods differ slightly from other PSM research. We briefly outline these differences in the following section. 1

Geostatistical methods

Geostatistics are a subset of traditional statistics that deal primarily with spatial data and account for spatial autocorrelation using kriging as the spatial interpolator. The concept is based upon the theory of regionalized variables, which was mainly developed by Matheron (1963) and Krige (1963). Kriging is a form of weighted local averaging that uses a measure of spatial dependence, the variogram, to determine the weights applied to the data when computing the averages. Geostatistical methods have been used in predictive soil mapping research to spatially interpolate soil property values at unmeasured sites from field-collected data. Burgess and Webster (1980a, b) were the first to introduce ordinary kriging to the soil community and since that time an enormous amount of work has been published. For example, ordinary kriging has been used to interpolate many different soil properties, including pollution, trace element deficiencies, salinity and fertility (Heuvelink and Webster, 2001). Ordinary kriging has been criticized for a variety of reasons. For example, Laslett et al. (1987) reported that several authors had criticized geostatistics because kriging is a global rather than local technique, failing to take into account knowledge of soil materials and processes. Other authors have criticized geostatistics because they are excessively data dependent, requiring a large number of closely spaced data points (Zhu, 1997a). As Webster and Oliver (1992) suggest, in excess of a hundred samples may be needed to use geostatistics at the field scale because of high spatial variability of soil in some areas. Geostatistics also assume spatial autocorrelation, which sometimes may be a poor assumption in complex terrain where abrupt changes in soil-forming factors occur (McBratney et al., 2000). Ordinary kriging by itself

184


does not satisfy two of the three goals of PSM presented in the introduction – it does not adequately incorporate expert knowledge and it does not exploit the relationship between environmental variables and soil properties. Ordinary kriging has been modified in a variety of ways to better incorporate ancillary data and known soil–landscape relationships. Block kriging involves determining estimates over meaningful areas rather than at specific points (Burgess and Webster, 1980a; McBratney et al., 1991). Using this method a study area can be stratified into different regions that are reflective of the pedogenetic processes at work. In order to accommodate a trend within a dependent soil variable universal kriging has been used (Webster, 1994). Kriging with external drift is similar to universal kriging, but it uses an ancillary variable to represent the trend (McBratney et al., 2000). Co-kriging takes advantage of correlation that may exist between the variable of interest and other more easily measured variables (Odeh et al., 1995). Regression kriging involves spatially interpolating the residuals from a non-spatial model by kriging, and adding the result to the prediction obtained from that model (Goovaerts, 1997; Castrignano et al., 2000). Factorial kriging is another method to integrate multivariate data into the standard kriging routine to extrapolate soil data (Goovaerts, 1992). Many authors have compared these various methods (Laslett et al., 1987; Voltz and Webster, 1990; Odeh et al., 1994; Knotters et al., 1995). Fuzzy logic has been used with geostatistics by various authors to produce new kinds of fuzzy soil maps with continuous classes (Burrough, 1989; McBratney and DeGruijter, 1992; Odeh et al., 1992a; and reviewed by McBratney and Odeh, 1997). The process entails kriging the matrix of membership values determined by the fuzzy k-means classifier, resulting in a continuous soil surface where individual locations are allowed to belong to more than one class and no rigid boundaries are designated to separate the soil into discrete units or entities. The results of such analysis can be used to assess the pedologic process validity of soil taxonomy by determining whether soils group together into classes that are similar to taxonomic types. Fuzzy classes would presumably reflect the main pedologic features within a given area. In this sense, the classification is quantitative, whereas soil taxonomy is rooted in qualitative discrimination. A comparison between the two could provide insightful and help assist soil taxonomy in making a classification that is more indicative of underlying soil processes. Geostatistics in soil research were originally introduced to quantitatively assess soil variability within soil mapping units (McBratney et al., 1991) in response to criticisms in the early 1970s that soil unit composition was not well quantified (Beckett and Webster, 1971). In this regard, geostatistics have been very useful, having served well the original goals set forth by Burgess and Webster (1980b) when they drew kriging to the attention of soil scientists as a means of spatial prediction. At the field scale soil variation is largely due to the effect that topography has on soil genesis. Geostatistics have been successfully applied in such environments by using terrain attributes as ancillary data within many of the kriging routines described above. Such quantitative within-unit variability of soil properties is very useful in the field of precision agriculture and other situations (e.g., pollutants) where very detailed soil attribute information is needed at the field scale (Heuvelink and Webster, 2001). However, geostatistics have not been applied in a wide variety of environments or at larger scales. In order to be successfully applied in different environments, geostatistics will likely require a different suite of ancillary data. For example, remote sensing data

P. Scull et al.

185

could be used in arid regions where soil toposequences are less well expressed. At larger scales of prediction selection of different sets of ancillary variables is required because different processes define soil character at different scales. The most obvious example is that of climate, which may control soil distribution at large scales (continental), but has little explanatory power at the field level. Regardless of whether adequate ancillary data exist, the amount of data required to use geostatistics for landscape-level prediction would be extremely difficult and costly to collect, given the strict sampling protocol required to characterize spatial dependence. It is also unclear at what landscape scale soils exhibit spatial autocorrelation. Geostatistical approaches do provide a means of creating continuous soil attribute surfaces to better represent soil continuity (Goal 2), and they can be used to exploit the relationship between environmental variables and soil properties in order to more efficiently collect soil data (Goal 1). However, they do not sufficiently utilize expert knowledge (Goal 3), as no attempt has been made in geostatistical approaches to directly integrate expert knowledge. Fundamentally, kriging is a process of interpolation designed to predict attribute values in between locations of measured samples. In this sense, geostatistics represent a middle ground between pure interpolation (e.g., nearest neighbour type classifier) in which only measured points for the variable of interest are used to determine unknown values and other predictive models that primarily use soil–environment correlation to create predictive maps. 2

Statistical methods

Statistical methods can be used to exploit the relationship between quantifiable landscape indices and soil properties to create predictive soil maps. For example, McKensie and Austin (1993) used a regression to account for a large percentage of variation for many soil characteristics (A horizon: clay content, CEC, EC, pH, bulk density, and COLE; B horizon: clay content, CEC, ESP, EC, pH, bulk density, and COLE) using a variety of predictor variables (slope, presence or absence of impeding layer, relief, landform, topographic position). Their results confirm the hypothesis of Moore, I.D. et al. (1991) that soil character is related to quantifiable landscape indices. However, their methods do not provide inference of soil properties at unmeasured sites from mapped environment data because the topographic variables were measured in the field. Linear regression has also been used with terrain variables derived from a 15-m DEM in northeastern Colorado to predict soil attributes (organic matter content, extractable phosphorous, pH and texture) at unvisited sites (Moore et al., 1993). In that particular study 50% of the variance of A-horizon thickness was explained by slope and the wetness index. Gessler et al. (1995) also used regression to model A-horizon thickness from topographic variables in southeastern Australia (plan curvature and wetness index r2 = 0.63, P = 0.001). They modeled solum depth and used logistic regression to model E horizon presence/absence. Elsewhere, logistic regression has been used to model the presence/absence of noncalcareous clay loam horizon in central France using terrain attributes from a 20-m DEM (King et al., 1999). Exponential regression has been used to model soil organic carbon using terrain variables (Bell et al., 2000) in glacial outwash soils in east-central Minnesota. Generalized additive models (GAM) have been used less frequently in PSM research. Gessler (1996) used a GAM

186


model to predict total soil carbon, A horizon depth and solum depth using a variety of environmental predictors (slope, elevation, wetness index, mean annual temperature, precipitation and radiation). This small body of research opened the door to more complex methods by demonstrating the existence of quantifiable relationships. These authors were able to produce soil attribute maps using raster data models whose scale was dependent upon the grid cell resolution of the environmental data. They were successful at exploiting the relationship between quantifiable topographic attributes and soil profile character (Goal 1). The continuous soil attribute surfaces they produce also better represent soil continuity than the choropleth soil maps produced by traditional soil survey (Goal 2). However, the bulk of these methods (excepting GAMs) are limited by their assumed linear relationship between soil and topographic attributes, their assumptions of normally distributed data and their high data requirements. Standard statistical procedures are also not flexible enough to allow robust integration with a variety of potential data sources, such as expert knowledge (Goal 1). Statistical methods do demonstrate in a quantitative manner that terrain analysis can be used to predict soil attributes in relatively small areas with homogeneous parent material. A large proportion of the research using statistical methods was conducted in semi-arid landscapes at small scales (the largest study area of the entire group was ~2000 ha). Obviously, for statistical approaches to be most effective, they need to be more universal. As such, they need to be tested and/or developed at larger scales and in more diverse landscapes. 3

Decision tree analysis (DTA)

The use of decision tree analysis is just beginning to be explored in predictive soil mapping research, although it has been used successfully in the related field of predictive vegetation mapping since the early 1990s (Lees and Ritman, 1991; Moore, D.M. et al., 1991; Franklin, 1998). DTA is a form of divisive classification. The process of tree modelling involves successively partitioning data (called recursive partitioning in the tree modelling literature) into increasingly homogeneous subsets, which, once the partitioning has ceased, are called terminal nodes (Lees and Ritman, 1991). Splits, or rules defining how to partition the data, are selected based on information statistics that define how well the split decreases impurity within the data set (Clark and Pregibon, 1992). Splits are based on threshold values of an explanatory variable, selected by comparing the increase in resulting purity of node membership for all possible thresholds and variables. The process is iterative, growing from the root node (the complete data set) to the terminal nodes in a dendritic fashion (Friedl and Brodley, 1997). Once the tree has been constructed (or grown) it encodes a set of decision rules that describe the data partitioning process. These rules can be used to classify or predict other data sets (Moore, D.M. et al., 1991). Pruning the tree is often necessary to prevent the tree from being overfit to the sample data, and to reduce tree complexity. Pruning entails combining pairs of terminal nodes into single nodes and can be accomplished using cross-validation, which yields an initial indication of how large a tree makes robust predictions (Safavian and Norvig, 1991). Cross-validation involves systematically removing portions of the data set and running the remaining sample through the tree

P. Scull et al.

187

in an iterative manner, eventually yielding estimates of the misclassification rates for each class, each node and the whole model (Breiman et al., 1984). In this manner different sized trees can be compared in terms of parsimony. The term DTA is used to collectively refer to all types of tree-based modelling (the word ‘decision’ is used because it is descriptive, indicating that the analysis eventually leads to a set of decision rules, defining data partitions). The term should be distinguished from classification tree analysis because the latter refers specifically to DTA where the response variable is categorical. The term CART (classification and regression trees – Breiman et al., 1984) is sometimes used but, strictly speaking, refers to specific software. Friedl and Brodley (1997) provide a review of the decision tree algorithms. They divide types of DTA into two classes: (1) homogeneous decision trees, for which a single algorithm is used to estimate each split (e.g., CART); and (2) hybrid decision trees (HDT), for which different splitting methods can be used at different points in the tree (e.g., Quinlan, 1993). They further divide homogeneous decision trees into univariate (UDT), where single features of the input data define splits and multivariate decision trees (MDT), where multiple features of the input data can define splits. According to this naming convention, no distinction is made between methods with different types of response variables, although all methods can be used with both categorical and continuous response variables. In a comparison of these various types of algorithms on a variety of data sets, Friedl and Brodley (1997) found that HDT had the highest classification accuracy. The overall aim of DTA is to design a set of predictive rules (e.g., if geology type A, then soil type B) developed from training data, which can then be applied to a geographic data base to predict the value of a response variable (Michaelsen et al., 1994). Therefore, DTA explicitly uses soil–landscape correlation in model development (Goal 2). The technique appears promising in soil research, but needs to be further tested, as not many of the above types of DTA have been tested in the PSM literature; in fact, only univariate approaches have been employed. For example, Lagacherie and Holmes (1997) successfully used univariate DTA to model a categorical response variable, soil type, within a training set and then, assuming that the training set was representative, extrapolated the model to a much larger region. Their work is interesting because although not a single sample came from the area they eventually mapped they were able to produce a soil map that was more accurate (74% versus 69%) than the existing map produced from traditional methods. Cialella et al. (1997) also used univariate DTA to predict soil drainage class from a variety of terrain attributes and remotely sensed data. They predicted soil drainage class with an average of 78% accuracy – impressive given that the variation accounted for by a typical soil survey ranges from about half the total variance for physical attributes to less than one-tenth for some soil chemical attributes (Gessler et al., 1995). DTA has been compared with other approaches by several authors. In the application of erosion modelling, DTA results were similar to Artificial Neural Networks (Ellis, 1996). Both methods achieved high training accuracy (as measured by the Kappa Statistic), but in terms of prediction accuracy both methods performed poorly. Gessler et al. (1995) compared DTA with generalized linear models (GLM) and generalized additive models (GAM) to predict A-horizon thickness and concluded that GLM was preferable to both DTA and GAM. McKensie and Ryan (1999) compared regression trees and standard linear regression to predict soil properties (total solum depth, soil

188


phophorus and soil carbon) using a large variety of predictors (elevation, slope, curvature, wetness index, climate variables, geology unit, Landsat TM data and Gamma radiometric data) and found that different methods work best for different properties and overall success hinged on the strength of the relationship between soil and environmental variables. Their study of a 50 000 ha forested landscape in southern New South Wales, Australia, is one of the few studies at such a small map scale (large area) or in forested landscape. The most extensive comparison of techniques involving DTA was conducted by McBratney et al. (2000). They compared regression techniques (GLM, GAM, DTA-regression tree), geostatistical techniques (kriging and heterotopic co-kriging), and a hybrid technique (regression kriging). DTA was found to be the poorest performing of the regression techniques because of the unrealistic prediction surface generated by the DTA model. DTA has also been criticized by other authors because of the stepped prediction surface (Gessler, 1996). This phenomenon can be especially apparent in situations where predictor variables have different resolutions. The extent to which DTA yields a better spatial representation of soil continuity is a function of the scale and type of predictor variables used and therefore varies with individual models. DTA is gaining widespread popularity as a means to develop prediction rules that can be rapidly and repeatedly evaluated (Cialella et al., 1997; Franklin et al., 2000). DTA provides the following advantages over standard statistical techniques: (1) it is easier to interpret when explanatory variables are both nominal and continuous; (2) it is invariant to monotone re-expressions (transformations) of predictor variables; (3) it deals more satisfactorily with missing data values and outliers; (4) it is more adept at capturing nonadditive and nonlinear behaviour; (5) it doesn’t make any assumptions about data distribution; and (6) it is easily updateable as more data are collected (Moore, D.M. et al., 1991). The DTA model framework is especially appealing because of its capability to integrate a wide range of data sets as explanatory variables. DTA offers a unique opportunity for interaction between soil experts and soil modellers because the output of the model is a set of rules that can be pedologically interpreted by the soil expert. In this sense, expert knowledge is used in an implicit manner in DTA (somewhat effectively achieving Goal 3). While these rules can often be exceedingly complex, at minimum the expert can decide whether initial splits make sense given their understanding of the landscape. The success of DTA results often hinges on the ability of the modeller to make key decisions during the model building process; there is unfortunately no definitive way to determine the most optimal tree. The aid of the soil expert can potentially elucidate this problem. 4

Expert systems

A variety of expert system approaches to PSM have been developed to utilize expert knowledge. The purpose of such methods is to exploit the information the soil surveyor accumulates while working in the field, by integrating such knowledge into the predictive model (McCracken and Cate, 1986). Unlike the majority of the research reviewed thus far, the dependent variable in many expert systems models is often soil taxa or mapping unit. This apparent disadvantage of expert systems (using classification to characterize soil continuity) does make them easier to integrate into traditional

P. Scull et al.

189

soil survey. In addition, several authors have developed methods to develop continuous soil property maps from the output of expert system models designed to predict soil unit occurrence. Expert systems are composed of data (information on spatial environmental variables, e.g., topography, climate, etc.), a knowledge base (rules and facts related to soil variation supplied from the soil surveyor) and an inference engine (which combines data and the knowledge base to infer logically valid conclusions) (Skidmore et al., 1996). Expert systems differ from conventional models in two ways: (1) they store and manipulate qualitative information (allowing them access to information that cannot normally be used in other modelling frameworks); and (2) they are structured as meta-models (the knowledge is separated from the model) (Davis, 1993). This allows the model to selectively choose which information is relevant at various stages of the modelling process, and it allows for information to be easily updated. Davis (1993) reviews the application of expert systems to environmental modelling research, concluding that the technique is becoming more widely accepted. He further states that the application of expert systems is constrained by an absence of fundamental knowledge for rule generation, a problem that would appear less relevant to soil mapping given the amount of untapped, expert knowledge accumulated by the soil surveyor. The first mention of the use of expert systems in pedology was in a paper presentation at the Northeast Committee Soil Survey Conference, 1984 (Flach, 1985). In her paper, Flach (1985) hinted that recent developments in computer science, especially expert systems and artificial intelligence, could make modelling a practical mapping tool for soil scientists in the near future. A year later, McCracken and Cate (1986) hoped to encourage soil scientists to explore expert systems and its application to soil science through an optimistic article they wrote on the potential use of expert systems. However, little research was actually conducted in the late 1980s and expert systems had not yet begun to fulfil the lofty goals set forth by early practitioners (Dale et al., 1989). In the early 1990s expert systems approaches to predictive soil mapping began to appear in the literature. Skidmore et al. (1991) used a Bayesian expert system to map forest soil into different classes and their results compared favourably with available soil maps and actual field-collected data. Their methods successfully incorporated surveyor knowledge, and remotely sensed and digital terrain attributes, but failed to better characterize continuous soil–landscape variability because their final product was a choropleth map. Skidmore et al. (1996) revisited their earlier research and assessed the mapping accuracy of their results. They found that the soil map produced by the expert system achieved an overall accuracy of 69.8% (sample size = 53), while the map derived from conventional methods had an accuracy of 73.6%. In addition to knowledge provided by a soil scientist, Skidmore et al.’s (1996) expert system used vegetation derived for aerial photography and topographic variables derived from a 10m DEM (wetness index, topographic position and slope). Using an expert system Cook et al. (1996) successfully produced a continuous soil property map for organic matter content using wetness index, aspect and slope as explanatory variables. Their methods were somewhat inefficient because they required a separate expert system for each soil property of interest. However, their research represents the only example of expert systems used to directly predict the spatial distribution of a soil property. All of the other examples of expert systems in the literature

190


deal with soil type or class. The use of expert systems to map soil properties needs to be explored further. The expert systems discussed thus far have all used Boolean logic within their theoretical framework, whereby an observation can belong to one and only one class, and the soil properties of that class are assigned to the observation. With the development of fuzzy logic and the semantic import model (the second fuzzy logic approach to creating continuous classes) the opportunity exists to develop fuzzy logicbased expert systems. Such systems can be used in conjunction with expert knowledge in situations where experts have a good, qualitative idea of how to group data, but have trouble dealing with observations that are not well represented by rigid classification schemes (Burrough and McDonnell, 1998). This method can be particularly useful in situations when taxonomic schemes have been previously developed, as is the case in soil taxonomy. Several examples of this type of approach were published in the mid1990s by A. Zhu and colleagues (Zhu and Band, 1994; Zhu et al., 1996, 1997; Zhu, 1997a,b). Such systems proved useful for mapping soil at unvisited locations using surveyor knowledge and were also capable of producing continuous soil property maps. The use of fuzzy logic within the theoretical framework of the expert system allows the complex nature of soil to propagate through the modelling process, never subjected to classification schemes that filter out potentially useful ‘noise’. The use of fuzzy logic also gives the soil surveyor more latitude during the interview process when the knowledge base is defined for the expert system. Zhu et al. (1997) developed a fuzzy logic-based expert system, called SoLIM, to determine the similarity of each grid cell in a study area to the various taxonomic mapping units delimited by the soil survey. Continuous soil attribute maps were calculated using the similarity values and their relative soil survey determined attribute values (effectively achieving Goal 2). The resulting data (which proved more accurate than soil survey data once field checked) consists of a raster grid, whose resolution is determined by the resolution of the input environmental and digital elevation data. As noted previously, this type of data model is more applicable to environmental modelling than the choropleth map (Burrough and McDonnell, 1998). Expert system approaches to PSM are capable of exploiting soil surveyor knowledge by developing rule-based systems that imitate the surveyor’s conceptual model of soil variability (the primary focus of Goal 3). The method would appear extremely useful for mapping projects (such as those conducted by the NRCS-NCSS), where fieldwork is initially conducted to determine soil–landscape relations. Expert system development could be directly inserted into the traditional soil survey mapping approach as a substitute for the step where the surveyor converts his/her conceptual model into a choropleth map. Rather, that knowledge could be incorporated into the expert system, which could be used to predictively map soil. The resulting raster map would be more scientifically based and explicit than the hand-drawn choropleth maps of the past. Its scale would not be limited to that of the aerial photography, but rather to the scale of the environmental data. Despite these advantages, the expert system approach has some drawbacks. Because the method is dependent upon expert knowledge, it cannot be applied where environment–soil relations are poorly understood. Of course, this criticism can be made of all PSM models, as inductive or deductive knowledge of soil–environment relations is a prerequisite for PSM. However, expert systems are deductive models and, as such,

P. Scull et al.

191

are not driven by any specific field-collected data (although, presumably the soil expert has field experience in the mapping area). Expert systems don’t afford the opportunity to first statistically document landscape–soil relations and then extrapolate the results because expert systems do not directly use sample soil data (‘hard data’) to determine soil–landscape relations. Thus, the utilization of the relationship between environmental variables and soil properties (Goal 1) is only indirectly achieved. Expert systems approaches have been demonstrated to be extremely effective in a small number of case studies. The possibility of satisfying all three goals discussed in this review makes expert systems a predictive soil mapping method that needs to be further tested – both geographically and across different scales of analysis. V

Conclusion

Most of the predictive soil mapping research outlined in this review was conducted at very large map scales (over small areas). In fact the majority of the research was concerned with assessing the spatial variability of soil character within individual fields or across soil toposequences. The primary driving force behind this type of research has been the need to provide accurate soils information for agriculture and ecological models. It is clear that terrain attributes are powerful predictors at the local scale. Geostatistical tools have been successful at using terrain attributes and the spatial dependence of soil properties to interpolate between existing data points within individual fields. Across soil toposequences, statistical approaches provide a useful means of predicting soil character. PSM research has been most successful at the field scale because many of the soil-forming factors are held constant. For example, the net effect of four of Jenny’s five soil-forming factors (climate, organism, parent material and time) was minimal within many of the studies reviewed. While some research has tackled larger areas of study, there exists a deficiency in our ability to predictively map soils at smaller map scales. Since the distribution of soil is scale-dependent, different PSM methods and predictors are likely to work better at different scales. Focus in the future must continue to move toward working over larger spatial extents of study in order to produce landscape-scale soil information. Further, a large proportion of the research was conducted in semi-arid, gently sloping, agricultural landscapes. Humid forestlands, mountainous regions and deserts have received little attention. As such, PSM methods need to continue to be tested and/or developed in a wider variety of landscapes, where spatial soil distributions can be more complex. Different methods will likely be successful to different degrees in different environments. Whereas terrain attributes are dominant predictors of soil character across toposequences in prairie lands, other predictors are likely necessary in areas where soils do not develop into clear toposequences. For example, remote sensing data has been a useful predictor of soil chronosequences in desert landscapes where surface appearance is often related to soil character. Surficial geology is often reflective of soil character in mountain regions where soils are thin and significant bedrock exists at the surface. Focus in the future must continue to determine which methods and predictors work best in which environments. Since the most useful PSM approach will vary across spatial scales and environmental gradients the method used should be driven by the mapping objectives of the

192


project. Whereas the traditional soil survey of the past was expected to meet the needs of a diverse group of end users, PSM methods will vary given the objectives of the survey. As a result, the end users of the soil data need to play a more active role in the survey process. Because precision agriculture and large-scale mapping has been the primary focus of PSM, there exists a deficiency in our ability to predictively map soils for the purpose of general land use planning and management. Even though less detailed soil information is needed for these purposes the development of PSM methods is complicated by the fact that soils exhibit complex spatial variability at small map scales where soil-forming environments vary greatly from one location to the next. A number of alternative methods of characterizing the continuous nature of the soil landscape have been developed. Thus far most of the PSM research has provided soil information in a nonobject form (as opposed to defining soil types as independent entities). Soil data have been generated, organized and presented in the form of either isorithmic maps or fine-scale raster grids. Both of these data models are field-view models of geographic space, which allow the soil to be perceived as a constantly varying surface. Two distinct approaches have been employed: (1) mapping individual soil properties and (2) mapping continuously varying (fuzzy) soil classes. Mapping individual properties is the most common approach and will likely continue to dominate PSM research. The use of fuzzy soil classes in the literature is less common and more difficult to be integrated into standard mapping procedures (such as use of soil taxa) because the concept radically differs from the traditional view of the soil landscape. However, fuzzy soil classes (generated using the fuzzy-k means approach) do have the potential to help further advance Soil taxonomy by identifying taxonomic classes that are more reflective of pedologic processes at work. Expert systems have been greatly underutilized in PSM research, especially considering how effective a small number of case studies have been. Expert systems have the potential to satisfy successfully all three goals discussed throughout the review. They also have the potential to bridge the gap between traditional approaches and PSM methods because field soil scientists do not have to change their conceptual approach to mapping. They can still conduct field reconnaissance to determine soil–landscape relationships. Afterwards though, they can use expert systems to better exploit the knowledge they have garnered. Such an approach could help ease the transition from traditional soil survey to more scientifically explicit methods. Expert systems are also the most fruitful approach to utilizing a wealth of data that has already been collected in a nonexplicit manner, the wealth of expert knowledge (in the form of the conceptual model) that senior soil mappers have accumulated. Focus in the future must also move in the direction of operability. The utility of many PSM approaches has been clearly demonstrated, yet changes in how soils are mapped and perceived has been slow. In the USA various members of the National Cooperative Soil Survey have called for a change in the philosophy of survey (from mapping discrete soil types to mapping continuous varying soil properties; McSweeny et al., 1994), but little change has actually taken place. To realize the potential of PSM in contributing to an overall change in standard soil mapping procedures, practitioners working in the field of PSM need to form working relationships with field soil scientists. PSM methods need also to be presented in a manner that is comprehensible to the soil science trained field mapper. Predictive soil mapping is a relatively recent

P. Scull et al.

193

phenomenon, yet much progress has been made. In the process of mapping, soil knowledge will continue to advance. References Agbu, P.A., Fehrenbacher, D.J. and Jansen, I.J. 1990: Statistical comparison of SPOT spectral maps with field soil maps. Soil Science Society of America Journal 54, 818–18. Barrett, L. 1999: Particulars in context: maintaining a balance in soil geography. Annals of the Association of America Geographers 89, 707–13. Beckett, P.H.T. and Webster, R. 1971: Soil variability: a review. Soils and Fertilizers 34: 1–15. Bell, J.C., Grigal, D.F. and Bates, P.C. 2000: A soilterrain model for estimating spatial patterns of soil organic carbon. In Wilson, J. and Gallant, J., editors, Terrain analysis: principles and applications. New York City NY: John Wiley and Sons, 295–310. Birkeland, P.W. 1999: Soil and geomorphology. Third edition. New York NY: Oxford University Press. Breiman, L., Friedman, J.H., Olshen, R.A. and Stone, C.J. 1984: Classification and regression trees. Belmont CA: Wadsworth. Brule, F.J. 1996: Fuzzy systems – a tutorial. http: //newsgroup: comp.ai; http//www. quadralay.com (last accessed 30 August 2001). Bunkin, F.V. and Bunkin, A.F. 2000: Lidar sounding of water, soil, and plants. Atmospheric and Oceanic Optics 13, 54–72. Buol, S.W., Hole, F.D., McCracken, R.J. and Southard, R.J. 1997: Soil genesis and classification. Ames IA: Iowa State University Press. Burgess, T.M. and Webster, R. 1980a: Optimal interpolation and isarithmic mapping of soil properties; the semi-variogram and punctual kriging. Journal of Soil Science 31, 315–31. –––– 1980b: Optimal interpolation and isarithmic mapping of soil properties: block kringing. Journal of Soil Science 31, 331–41. Burrough, P.A. 1989: Fuzzy mathematical methods for soil survey and land evaluation. Journal of Soil Science 40, 477–92. Burrough, P.A. and McDonnell, R.A. 1998: Principles of geographic information systems. (Revised edition.) Oxford: Clarendon Press. Burrough, P.A., Beckett, P.H.T. and Jarvis, M.G. 1971: The relation between cost and utility in soil survey. Journal of Soil Science 22, 368–81. Burrough, P.A., Van Gaans, P.M.F. and

Hootsman, R. 1997: Continuous classification in soil survey: spatial correlation, confusion, and boundaries. Geoderma 77, 115–35. Cambell, J.B. 1977: Variation of selected properties across a soil boundary. Soil Science Society of America Journal 41, 578–82. Cambell, J.B. and Edmonds, W.J. 1984: The missing geographic dimension to soil taxonomy. Annals of the Association of American Geographers 74, 83–97. Castrignano, A., Giugliarini, L., Risaliti, R. and Martinelli, N. 2000: Study of spatial relationships among some soil physico-chemical properties of a field in central Italy using multivariate geostatistics. Geoderma 97, 39–60. Cialella, A.T., Dubayah, R., Lawrence, W. and Levine, E. 1997: Predicting soil drainage class using remotely sensed and digital elevation data. Journal of Soil Science 62(2), 171–78. Clark, R.N. and Swayze, G.A. 1996: Evolution in imaging spectroscopy analysis and sensor signal-to-noise: an examination of how far we have come. Summaries of the sixth annual JPL airborne Earth science workshop. 4–8 March 1996. AVIRIS Workshop, Vol. 1, 5. Clarke, L.A. and Pregibon, D. 1992: Tree-based models. In Chambers, J. and Hastie, J., editors, Statistical models in S. Pacific Grove: Wadsworth and Brooks, 377–419. Coleman, T.L., Agbu, P.A. and Montgomery, O.L. 1993: Spectral differentiation of surface soils and soil properties – is it possible from space platforms? Soil Science 155, 283–93. Cook, S.E., Corner, R.J., Grealish, G., Gessler, P.E. and Chartres, C.J. 1996: A rule-based system to map soil properties. Soil Science Society of America Journal 60, 1893–900. Cruickshank, J.G. 1972: Soil geography. New York NY: John Wiley & Sons. Csillag, F., Pasztor, L., Biehl, L.L. 1993: Spectral band selection for the characterization of salinity status of soils. Remote Sensing of Environment 43, 231–42. Dale, M.B., McBratney, A.B. and Russell, J.S. 1989: On the role of expert systems and numerical taxonomy in soil classification. Journal of Soil Science 40, 223–34. Davis, J.R. 1993: Expert systems and environ-

194


mental modelling. In Jakeman, A.J., Beck, M.B. and McAleer, M.J., editors, Modelling change in environmental systems. New York NY: John Wiley and Sons Ltd, 3–35. De Gruijter, J.J., Walvoort, D.J.J. and Van Gaans, P.F.M. 1997: Continuous soil maps – a fuzzy set approach to bridge the gap between aggregation levels of process and distribution models. Geoderma 77, 169–95. Dijkerman, J.C. 1974: Pedology as a science: the role of data, models, and theories in the study of natural soil systems. Geoderma 11, 73–93. Dmitriev, E.A. 1983: Continuity of soils and the problem of soild classification. Moscow University Soil Science Bulletin 38, 1–10. Ellis, F. 1996: The application of machine learning techniques to erosion modelling. In: Proceedings, third international conference on integrating GIS and environmental modeling. Santa Fe NM, 16–21 January, 1996. http: //www.ncgia.ucsb.edu/conf/SANTA_FE_CD -ROM/main.html. (last accessed 21 January 2003). National Center for Geographic Information and Analysis, Santa Barbara, CA. Engman, E.T and Chauhan, N. 1995: Status of microwave soil moisture measurements with remote sensing. Remote Sensing of Environment 51, 189–98. Fang, Y. 2000: DEM generation from multi-sensor SAR images. International Archives of Photogrammetry and Remote Sensing 33: 686–93. Flach, K.W. 1985: Modeling and soil survey. Soil Survey Horizons 26, 15–20. Franklin, J. 1995: Predictive vegetation mapping: geographic modeling of biospatial patterns in relation to environmental gradients. Progress in Physical Geography 19, 474–90. –––– 1998: Predicting the distributions of shrub species in California chaparral and coastal sage communities from climate and terrain-derived variables. Journal of Vegetation Science 9, 733–48. Franklin, J., McCullough, P. and Gray, C. 2000: Terrain variables for predictive mapping of vegetation communities in Southern California. In Wilson, J. and Gallant, J., editors, Terrain analysis: principles and applications. New York City NY: John Wiley and Sons, 331–53. Friedl, M.A. and Brodley, C.E. 1997: Decision tree classification of land cover from remotely sensed data. Remote Sensing of Environment 61, 399–409. Gessler, P.E. 1996: Statistical soil–landscape modelling for environmental management. Doctoral Dissertation, The Australian National

University, Canberra, Australia. Gessler, P.E., Moore, I.D., McKensie, N.J. and Ryan, P.J. 1995: Soil-landscape modelling and spatial prediction of soil attributes. International Journal Geographical Information Science 9, 421–32. Goetz, A.F.H. 1989: Spectral remote sensing in geology. In Asrar, G., editor, Theory and applications of optical remote sensing. New York NY: John Wiley and Sons, 491–526. Goetz, A.F.H., Vane, G., Solomon, J.E. and Rock, B.N. 1985: Imaging spectrometry for earth remote sensing. Science 228, 1147–53. Goodchild, M.F. 1992a: Geographical data modeling. Computers and Geosciences 18, 401–408. –––– 1992b: Geographical information science. International Journal Geographical Information Systems 6, 31–45. –––– 1994: Intergrating GIS and remote sensing for vegetation analysis and modeling: methodological issues. Journal of Vegetation Science 5, 615–26. Goovaerts, P. 1992: Factorial kriging analysis: a useful tool for exploring the structure of multivariate spatial soil information. Journal of Soil Science 43, 597–619. –––– 1997: Geostatistics for natural resource evaluation. New York City NY: Oxford University Press. Hall, C.A.S. and Olsen, C.G. 1991: Predicting variability of soil from landscape models. In Spatial variability of soil and landforms. Soil Science Society of America, Special Publication 28, 9–24. Hartemink, A.E., McBratney, A.B. and Cattle, J.A. 2001: Developments and trends in soil science: 100 volumes of Geoderma 1967–2001. Geoderma 100, 217–68. Henderson, T.L., Baumgardner, M.F., Franzmeier, D.P., Stott, D.E. and Coster, D.C. 1992: High dimensional reflectance analysis of soil organic matter. Soil Science Society of America Journal 56, 865–72. Heuvelink, G.B.M. and Webster, R. 2001: Modelling soil variation: past, present, and future. Geoderma 100, 269–301. Hewitt, A.E. 1993: Predictive modelling in soil survey. Soil and Fertilizers 56, 305–14. Horvath, E.H., Post, D.F. and Kelsey, J.B. 1984: The relationships of Landsat digital data to the properties of Arizona rangelands. Soil Science Society of America Journal 48, 1331–34. Hudson, B.D. 1992: The soil survey as paradigm based science. Soil Science Society of America Journal 56, 836–41.

P. Scull et al. Huggett, R.J. 1975: Soil landscape systems: a model of soil genesis. Geoderma 13, 1–22. Indorante, S.J., McLeese, R.L., Hammer, R.D., Thompson, B.W. and Alexander, D.L. 1996: Positioning soil survey for the 21st century. Journal of Soil and Water Conservation Jan–Feb, 21–28. Irons, J.R., Weismiller, R.A. and Petersen, G.W. 1989: Soil reflectance. In Asrar, G. editor, Theory and applications of optical remote sensing. New York NY: John Wiley and Sons, 66–106. Irvin, B.J., Ventura, S.J. and Slater, B.K. 1996: Landform classification for soil–landscape studies. In Proceedings, third international conference on integrating GIS and environmental modeling. Santa Fe NM, 16–21 January, 1996. http: //www.ncgia.ucsb.edu/conf/SANTA _FE_CD-ROM/main.html (last accessed 21 January 2003). National Center for Geographic Information and Analysis, Santa Barbara CA. Jenny, H. 1941: Factors of soil formation. New York NY: McGraw-Hill. Johnson, D.L. and Watson-Stegner, D. 1987: Evolution model of pedogenesis. Soil Science 143, 349–66. Johnson, P.E., Smith, M.O., Taylor-George, S. and Adams, J.B. 1983: A semiempirical method for analysis of the reflectance spectra of binary mineral mixtures. Journal of Geophysical Research 88, 3557–61. Kemp, K.K. 1992: Fields as a framework for integrating GIS and environmental process models. Part one: representing spatial continuity. Transactions in GIS 13, 219–34. King, D., Bourennane, H., Isambert, M. and Macaire, J.J. 1999: Relationship of the presence of a noncalcareous clay-loam horizon to DEM attributes in a gently sloping area. Geoderma 89, 95–111. Kleshchenko, V.N., Komarov, S.A., Mironov, V.L. and Romanov, A.N. 2000: Microwave remote sensing of soil cover. Proceedings – SPIE the International Society for Optical Engineering 4341, 351–57 . Knotters, M., Brus, D.J. and Oude Voshaar, J.H. 1995: A comparison of kriging, co-kriging and kriging combined with regression for spatial interpolation of horizon depth with censored observations. Geoderma 67, 227–46. Krige, D.G. 1963: Two dimensional weighted moving average trend surfaces for oreevaluation. Journal of the South African Institution of Mining and Metallurgy 66, 13–38.

195

Lagacherie, P. and Holmes, S. 1997: Addressing geographical data errors in a classification tree for soil unit prediction. International Journal Geographical Information Science 11, 183–98. Laslett, G.M., McBratney, A.B., Pahl, P.J. and Hutchinson, M.F. 1987: Comparison of several spatial prediction methods for soil pH. Journal of Soil Science 38, 325–41. Laymon, C.A., Crosson, W.L., Jackson, T.J., Manu, A. and Tsegaye, T.D. 2001: Groundbased passive microwave remote sensing observations of soil moisture at s-band and lband with insight into measurement accuracy. IEEE Transactions of Geoscience and Remote Sensing 39, 1844–58. Lees, B.G. and Ritman, A.K. 1991: Decision-tree and rule induction approach to integration of remotely sensed and GIS data in mapping vegetation in disturbed or hilly environments. Environmental Management 15, 823–31. Lillesand, T.M. and Ralph Kiefer, R. 1994: Remote sensing and image processing. New York NY: John Wiley and Sons. Mackay, D.S. and Band, L.E. 1998: Extraction and representation of nested catchment areas from digital elevation models in lake-dominated topography. Water Resources Research 34, 897–904. Matheron, G. 1963: Principals of geostatistics. Economic Geology 58, 1246–66. McBratney, A.B. 1992: On variation, uncertainty and informatics in environmental soil management. Australian Journal of Soil Research 30, 913–35. McBratney, A.B. and De Gruijter, J.J. 1992: A continuum approach to soil classification by modified fuzzy k-means with extragrades. Journal of Soil Science 43, 159–75. McBratney, A.B. and Odeh, I.O.A. 1997: Application of fuzzy sets in soil science: fuzzy logic, fuzzy measurement, and fuzzy decisions. Geoderma 77, 85–113. McBratney, A.B., Hart, G.A. and McGarry, D. 1991: The use of region partitioning to improve the representation of geostatistically mapped soil attributes. Journal of Soil Science 42, 513–32. McBratney, A.B., Odeh, I.O.A., Bishop, T.F.A., Dunbar, M.S. and Shatar, T.M. 2000: An overview of pedometric techniques for use in soil survey. Geoderma 97, 293–327. McCracken, R.J. and Cate, R.B. 1986: Artificial intelligence, cognitive science, and measurement theory applied in soil classification. Soil Society of America Journal 50, 557–61. McKensie, N.J. and Austin, M.P. 1993: A quanti-

196


tative Australian approach to medium and small scale surveys based on soil stratigraphy and environmental correlations. Geoderma 57, 329–55. McKensie, M.J. and Ryan, P.J. 1999: Spatial prediction fo soil properties using environmental correlation. Geoderma 89, 67–94. McKensie, N.J., Gessler, P.E., Ryan, P.J. and O’Connell, D. 2000: The role of terrain analysis in soil mapping. In Wilson, J. and Gallant, J., editors, Terrain analysis: principles and applications. New York City, NY: John Wiley and Sons, 245–65. McKnight, T.L. 1993: Physical geography: a landscape appreciation. Fourth edition. Englewood Cliffs NJ: Prentice Hall. McSweeney, K., Gessler, P.E., Hammer, J., Bell, J. and Peterson, G.W. 1994: Towards a new framework for modeling the soil landscape continuum. In Bryant, R.B. and Arnold, R.W, editors, Factors of soil formation: a fiftieth anniversary retrospective. Soil Science Society of America Special Publication 39. Madison WI: Soil Science Society of America, 127–45. Metternicht, G.I. 1998: Fuzzy classification of JERS-1 SAR data: an evaluation of its performance for soil salinity mapping. Ecological Modelling 111, 61–74. Michaelsen, J., Schimel, D.S., Friedl, M.A., Davis, F.W. and Dubayah, R.C. 1994: Regression tree analysis of satellite and terrain data to guide vegetation sampling and surveys. Journal of Vegetation Science 5, 673–86. Moore, D.M., Lees, B.G. and Davey, S.M. 1991: A new method for predicting vegetation distributions using decision tree analysis in a geographic information system. Environmental Management 15, 59–71. Moore, I.D., Grayson, R.B. and Ladson, A.R. 1991: Digital terrain modelling: a review of hydrological, geomorphological, and biological applications. Hydrological Processes 5, 3–30. Moore, I.D., Gessler, P.E., Nielsen, G.A. and Peterson, G.A. 1993: Soil attribute prediction using terrain analysis. Soil Science Society of America Journal 57, 443–520. Narayanan, R.M. and Hirsave, P.P. 2001: Soil moisture estimation models using SIR-C SAR data: a case study in New Hampshire, USA. Remote Sensing of Environment 75: 385–96. Odeh, I.O.A., Chittleborough, D.J. and McBratney, A.B. 1991: Elucidation of soillandform interrelationships by canonical ordination analysis. Geoderma 49, 1–32. Odeh, I.O.A., McBratney, A.B. and

Chittleborough, D.J. 1992a: Fuzzy-c-means and kriging for mapping soil as a continuous system. Soil Science Society America Journal 56, 1848–54. –––– 1992b: Soil pattern recognition with fuzzy-cmeans: Application to classification and soillandform interrelationships. Soil Science Society America Journal 56, 505–16. –––– 1994: Spatial prediction of soil properties from landform attributes derived from a digital elevation model. Geoderma 63, 197–214. –––– 1995: Further results on prediction of soil properties from terrain attributes: heterotopic cokriging and regression-kriging. Geoderma 67, 215–26. Okin, G.S., Okin, W.J., Roberts, D.A. and Murray, B.C. 1998: Multiple endmember spectral mixture analysis: applications to an arid/semi-arid landscape. Proceedings 7th AVIRIS Earth science workshop. Pasadena CA: JPL, 8. Oliveira, H. 2000: Segmentation and classification of Landsat-TM images to monitor the soil use. International Archives of Photogrammetry and Remote Sensing 33, 1065–72. Palacios-Orueta, A. and Ustin, S.L. 1996: Multivariate classification of soil spectra. Remote Sensing of Environment 57: 108–18. Palacios-Orueta, A., Pinzon, J.E., Ustin, S.L. and Roberts, D.A. 1998: Remote sensing of soils in the Santa Monica Mountains: II. Hierarchical foreground and background analysis. Remote Sensing of Environment 68, 138–51. Quinlan, J.R. 1993: C4.5: Algorithms for machine learning. San Mateo: Morgan Kaufmann. Roberts, D.A., Gardner, M., Church, R., Ustin, S., Scheer, G. and Green, R.O. 1998: Mapping chaparral in the Santa Monica Mountains using multiple endmember spectral mixture models. Remote Sensing of Environment 65, 267–79. Runge, E.C.A. 1973: Soil development sequences and energy models. Soil Science 115, 183–93. Safavian, S.J. and Norvig, P. 1991: A survey of tree classifier methodology. IEEE Transactions Systems, Man, Cybernetics 21, 660–74. Schaetzl, R.J. 1991: Factors affecting the formation of dark, thick epipedons beneath forest vegetation, Michigan, USA. Journal of Soil Science 42, 501–12. Seyler, F., Bernoux, M. and Cerri, C.C. 1998: Landsat TM image texture and moisture variations of the soil surface under the rainforest of Rondonia state, Brazil. International

P. Scull et al. Journal of Remote Sensing 19, 1299–318. Shipman, H. and Adams, J.B. 1987: Detectability of minerals in desert alluvial fans using reflectance spectra. Journal Geophysical Research 92, 10391–492. Simonson, R.W. 1959: Outline of a generalized theory of soil genesis. Soil Science Society of America, Proceedings 23, 152–56. –––– 1997: History of soil science. Early teaching in USA of Dokuchaiev factors of soil formation. Soil Science Society of America Journal 61, 11–16. Skidmore, A.K., Ryan, P.J., Dawes, W., Short, D. and O’loughlin, E. 1991: Use of an expert system to map forest soil from a geographical information system. International Journal Geographical Information System 5, 431–45. Skidmore, A.K., Watford, F., Luckananurug, P. and Ryan, P.J. 1996: An operational GIS expert system for mapping forest soil. Photogrammetric Engineering and Remote Sensing 62, 501–11. Soil Survey Staff 1975: Soil taxonomy: a basic system of soil classification for making and interpreting soil surveys. Agriculture handbook 436. Washington DC: Soil Conservation Service, U.S. Department of Agriculture. Stoner, E.R., Baumgardner, M.F., Weismiller, R.A., Biehl, L.L. and Robinson, B.F. 1980: Extension of laboratory-measured soil spectra to field conditions. Soil Science Society of America Journal 44, 572–74. Ventura, S.J. and Irvin, B.J. 1996: Automated landform classification methods for soil landscape studies. In Proceedings, third international conference on integrating GIS and environmental modeling. Santa Fe NM, 16–21 January, 1996, http: //www.ncgia.ucsb.edu/conf/ SANTA_FE_CD-ROM/main.html (last accessed 21 January 2003). National Center for Geographic Information and Analysis, Santa Barbara, CA. Voltz, M. and Webster, R. 1990: A comparison of kriging, cubic splines and classification for predicting soil porperties from sample information. Journal of Soil Science 41, 473–90. Wanchang, Z., Yamaguchi, Y. and Ogawa, K. 2000: Evaluation of the effect of pre-processing

197

of the remotely sensed data on the actual evapotranspiration, surface soil moisture mapping by an approach using Landsat, DEM meteorological data. Geocarto International 15, 57–68. Webster, R. 1994: The development of pedometrics. Geoderma 62, 1–15. Webster, R. and Beckett, P.H.T. 1968: Quality an usefulness of soil maps. Nature 219, 680–82. Webster, R. and Oliver, M.A. 1992: Sample adequately to estimate variograms of soil properties. Journal of Soil Science 43, 177–92. Weibel, R. and Heller, M. 1991: Digital terrain modeling. In Goodchild, M.F. and Rhind, D., editors, Geographic information systems, principles and applications. New York NY: Taylor and Francis, 269–97. Wilcox, C.H., Frazier, B.E. and Ball, S.T. 1994: Relationship between soil organic carbon and Landsat TM data in eastern Washington. Photogrammetric Engineering and Remote Sensing 60, 777–81. Wilson, J. and Gallant, J., editors, 2000: Terrain analysis: principles and applications. New York NY: John Wiley and Sons. Wright, R.L. and Wilson, S.R. 1979: On the analysis of soil variability, with an example from Spain. Geoderma 22, 297–313. Zadeh, L.A. 1965: Fuzzy sets. Information and Control 8, 338–53. Zhu, A. 1997a: Measuring uncertainty in class assignment for natural resource maps under fuzzy logic. Photogrammetric Engineering and Remote Sensing 63, 1195–202. –––– 1997b: A similarity model for representing soil spatial information. Geoderma 77, 217–42. Zhu, A. and Band, L. 1994: A knowledge-based approach to data integration for soil mapping. Canadian Journal of Remote Sensing 20, 408–18. Zhu, A., Band, L., Dutton B. and Nimlos, T.J. 1996: Automated soil inference under fuzzy logic. Ecological Modelling 90, 123–45. Zhu, A., Band, L., Vertessy, R. and Dutton, B. 1997: Derivation of soil properties using a soil land inference model, SoLIM. Soil Science Society America Journal 61, 523–33.

Predictive soil mapping: a review - CiteSeerX

Predictive soil mapping: a review - CiteSeerX

Suggest Documents

Predictive soil mapping with limited sample data - SoLIM

Predictive Soil Mapping in Arizona's Basin and Range Province - NRCS

Soil-phosphorus extraction methodologies: A review - CiteSeerX

Soil-phosphorus extraction methodologies: A review - CiteSeerX

A Review of Predictive Modelling from a Natural Resource ... - CiteSeerX

Data governance in predictive toxicology: A review - CiteSeerX

Digital Soil Mapping - European Soil Data Centre

Digital Soil Mapping - European Soil Data Centre

Predictive soil mapping as a means to aggregate and improve existing

Forecasting the Future of Predictive Crime Mapping - CiteSeerX

CONCEPTUAL MAPPING TO FACILITATE REVIEW OF ... - CiteSeerX

CONCEPTUAL MAPPING TO FACILITATE REVIEW OF ... - CiteSeerX

JDI Briefs: Predictive mapping (predictive policing) - Semantic Scholar

JDI Briefs: Predictive mapping (predictive policing) - Semantic Scholar

A systematic mapping review of Randomized Controlled ... - CiteSeerX

Methods for prediction of soil dielectric properties: a review - CiteSeerX

Economic Model Predictive Control - A Review

Digital Soil Mapping - Semantic Scholar

Distributed Model Predictive Control: A Tutorial Review

Distributed Model Predictive Control: A Tutorial Review

soil ...www.researchgate.net › publication › fulltext › Mapping-

Remote Predictive Mapping: An Approach for the Geological Mapping

Mapping of soil surface crusts using airborne ... - CiteSeerX

Mapping surface roughness and soil moisture using multi ... - CiteSeerX