| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a Center for Development Research (ZEF), Univ. of Bonn, Walter-Flex-Str. 3, 53113, Bonn, Germany
b Dep. of Geography, Univ. of Durham, Science Laboratories, South Road, Durham, DH1 3LE, UK
* Corresponding author (spark{at}uni-bonn.de)
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Abbreviations: DEM, digital elevation models • ECEC, effective cation-exchange capacity • LOI, loss-on-ignition • PC1, Principal Component 1 • PC2, Principal Component 2 • PC3, Principal Component 3 • PC4, Principal Component 4 • PC5, Principal Component 5 • PCA, principal component analysis • SS, sum of the squares • TEB, total exchangeable bases • TOd, extractable total oxides
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Although great progress has been made in the methodology of soil-landscape analysis with the recent development of spatial statistics and geographical information systems (see McBratney et al., 2000), interpretation of results continues to rely heavily on establishing statistical associations between terrain attributes and soil properties. For example, Cook et al. (1996) developed a rule-based soil mapping system and demonstrated a method to predict the spatial distribution of organic matter in topsoil. The four rules specified in their model are mostly topography-related attributes: wetness index, aspect, localized slope angle, and the association of field drainage. Zhu et al. (1997) used four terrain attributes (elevation, aspect, slope gradient, and profile curvature) and two additional environmental variables (geology and surface vegetation) to predict the distribution of soil types in their soil-land inference model (SoLim). A similar emphasis on topography occurs in many other soil-landscape models (e.g., Moore et al., 1993; McSweeney et al., 1994; McBratney et al., 2000).
This emphasis on using geomorphological variables to predict spatial variations in soil properties can be linked to both theoretical and practical considerations. Theoretically, landforms may be the best indicator of soil attributes in places where the variation of other environmental factors is relatively small (Moore et al., 1993). Soils on slopes often vary in response to the way in which water and soil materials move through and over the land surface, and this movement is in turn controlled by the geometry of the land surface (Huggett, 1975). Many investigations have identified statistical relations between landform variables (e.g., slope angle, distance from the divide, slope aspect, plan, and profile curvature) and soil attributes. Ruhe and Walker (1968) and Vreeken (1973) are some early examples where statistical relationships were sought (mostly through regression analysis) between soil properties and one or several topographical parameters (see Gerrard, 1992). This approach has rapidly gained popularity with the recent development of techniques for the direct calculation of terrain parameters from digital elevation models (DEM) and spatial statistics (Pennock et al., 1987; Moore et al., 1993a; Odeh et al., 1991, 1994; McKenzie and Austin, 1993; Gessler et al., 1995; McBratney et al., 2000). In terms of practical considerations, a topographic map is still the most easily available information in many parts of world. In developing countries, where relatively expensive hydrogeological and soil surveys have not yet been made, such soil-landscape analysis is considered as a key technique in natural resource and biological surveys (Gessler et al., 1995).
Despite recent developments in analytical methodologies, some theoretical questions of systematic correlation between soil properties and landform geometry have not been fully investigated. First of all, basic process-response relationships of soils have not been fully understood in many previous studies. For example, it may be that topographic control on soil properties operates differently in different landscapes, and that not all soil properties depend on topographic parameters to the same extent (Gerrard, 1992). Furthermore, each soil attribute has a unique spatial distribution on a slope, as a result of differences in geochemical and geomorphological mobility and different responses to various pedological and ecological processes (Huggett, 1975). Secondly, it is well known that soils are anisotropic in both a vertical and lateral context, and in most places vertical soil variation is more pronounced than lateral (Wilding et al., 1994). However, most previous research has only dealt with lateral variation at one or two depths, because of either the difficulty of sampling or added complexities in statistical analysis. Thirdly, it is difficult to prove that current soil properties are in steady state with the current form of the hillslope, something which most statistical approaches assume (Gerrard, 1992). Moreover, since the soil-forming factors at a given point may change through time, this makes it more difficult to relate soil properties to landform geometry (Vreeken, 1973; Park et al., 2001).
The aim of this study was to identify process-response relationships between pedogeomorphological processes operating on a hillslope and the distribution of soil properties using ordination techniques, and so assess the possibility of predicting the spatial distribution of soil properties using terrain attributes. Despite the complexities of the soil system, it can still be assumed that the spatial distribution of soil attributes at the landscape scale is controlled by certain environmental factors. If enough attributes are analyzed using appropriate multivariate statistical techniques, the relevant environmental factors may be revealed with some confidence. This is the assumption underlying the use of ordination techniques in ecological and environmental gradient analysis (Jongman et al., 1995) and factor analysis in behavioral science (Cattell, 1978). In a similar way, ordination techniques can provide a means of characterizing the spatial distribution of soil properties to infer the soil-forming processes involved, and to establish predictive relationships between soil properties and environmental factors.
Many different ordination procedures have been used in ecological studies (Jongman et al., 1995), and some previous pedological investigations have applied these techniques to infer pedogenic processes from soil properties. Anderson and Furley (1975) used PCA to deduce the slope processes responsible for the distribution of surface soil properties in southern England. Sondheim et al. (1981) applied PCA to identify dominant pedogenic processes from eleven chemical and physical soil properties in British Columbia. Odeh et al. (1991) compared various ordination techniques to elucidate the relationship between soil properties and terrain attributes in south Australia. They concluded that PCA and Redundancy Analysis (RDA) are better than other unimodal models such as correspondence analysis and canonical correspondence analysis, because interrelations among soil variables or between soil and landform attributes are more linear than unimodal.
The specific aims of this study are (i) to infer pedological and geomorphological processes through an ordination analysis of measured soil properties on a hillslope; (ii) to characterize soil properties according to their differential response to the pedogeomorpholological processes identified; (iii) to assess the influence of surface geometry on the spatial distribution of soil properties; and (iv) to investigate the possibility of predicting soil properties using terrain attributes. Given the diverse nature of both soil and geomorphological processes, site-specific representation of soil-landscape relationships may well be identified. The main emphasis in this study, however, is to clarify some neglected aspects of many previous soil-landform relationship studies and to provide further insight for future soil-landscape modeling.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
convex slope steep straight slope (weak) concave foot slope (see Fig. 1)
. A hollow extends upslope from the base of the slope, almost reaching the flat interfluve. The straight slopes have an average maximum angle of about 23° throughout the valley; this reduces to 20° at the base of the hollow, but approaches 30° on the lower spur. The three-dimensional arrangement of the hollow and spurs governs subsurface drainage (Anderson and Burt, 1978), and in turn strongly influences catenary soil development (Park et al., 1996) and long-term slope denudation (Park and Burt, 2000). Average annual rainfall, based on 20-yr meteorological records at a nearby location, is 1030 mm and mean air temperature is 9.2°C.
|
There is a well-developed podzolic catena on the slope (Park et al., 1996; see also Fig. 1). The soil classification for this research was based on the Soil Survey of England and Wales (Avery, 1990), but U.S. soil taxonomy equivalents at the subgroup or great group level were also identified and are presented in Fig. 1 (Soil Survey Staff, 1998). According to the latter classification, most soil profiles surveyed were identified as loamy-skeletal, siliceous, mesic, but a wide range of taxonomic types were found. Stagnoluvic gley soils have developed on the flat interfluve, with stagnogleyic podzolic brown soils on the upper convex shoulder slopes. On the steep part of the slope, podzols upslope connect to orthic brown soils downslope via podzolic brown soils. Stagno-gley podzols have developed along the convergence line of the upslope hollow and upslope flanks, and merge into a type of podzolic soil where sesquioxides and fine soil materials accumulate in subsurface and surface horizons. This podzolic soil merges downslope into the seepage soils of the lower hollow.
After the examination of various field and laboratory evidence (temporal soil moisture variation over the slope, soil morphologies, and the alloy of secondary Fe oxides), Park and Burt (1999) concluded that such a podzolic catena is the result of subsurface lateral flow (throughflow) and the existence of paleopedological subsurface horizons such as indurated horizons and subsurface stone layers. Throughflow is mainly generated by anisotropic soils related either to pedogenic horizonation or changes in parent material. When throughflow occurs, its spatial distribution and flow rate are greatly influenced by slope configuration (Anderson and Burt, 1978; McDaniel et al., 1992; Park and Burt, 1999). On the study slope, the throughflow intensity is highest for soils developed on the convex slope and steepest part of the hollow because of the existence of compacted and indurated horizons at shallow depth (usually at 30–60 cm). The pedogenic process for the indurated horizon is not clear, but overall morphologies resemble fragipans: high bulk density, brittle soil consistency without cementation; an abrupt boundary with the overlying horizon; entire thick coatings of fine material on upper parts of coarse fragments; and fine platy structure within massive structure (Park et al., 1996). These horizons are commonly found on the steeper upper slopes, where the stagnoluvic gley soils, stagnogley podzols, and nonhydromorphic podzols appear, but disappear in the orthic brown soil zone. As a consequence of the loss of the indurated horizonation, the dominant hydrological process also changes from shallow subsurface throughflow to vertical percolation.
Field Sampling and Laboratory Analysis
Fifty-four soil pits were excavated on a 25-m square sampling grid (January–May 1995). A further 10 soil profiles were investigated in the middle of the grid along the flank slopes where the soil changes rapidly. Pits ranged in depth from 40 to 110 cm depending on the amount of stones in the subsoil. Approximately 2 kg of soil was collected at each 10-cm sampling interval. Even-depth sampling was used for two reasons: first, the main focus of this research is the vertical comparison of soil property variation among different soil profiles and soil types; and second, the lateral continuity of soil horizonation, even within the surveyed profiles, is poor because of coarse soil texture and high slope angle. A total of 502 samples were gently ground to pass through a 2-mm sieve.
Table 1 lists the 32 soil properties measured and the laboratory techniques used. The soundness of the assumption underlying the ordination technique—that the variance characteristics of soil attributes reflect pedological and geomorphological processes at work on hillslopes—mainly depends on our ability to choose and analyze enough environmentally significant soil properties. The main objective of this research is to identify the dominant hydrological, geomorphological, and pedological processes on the study slope; therefore, we included those soil properties most frequently used to identify such processes. Consequently many other soil properties, especially those related to agricultural productivity, were not considered. Although this does not ensure that these data sets are sufficient to decipher environmental processes, careful application of ordination procedures does offer the chance to understand relationships between soil attributes and pedogeomorphological processes.
|
The wetness index and stream power index were calculated using the following equations (Moore et al., 1993):
 |
Ordination and Correlation with Terrain Attributes
To identify underlying mechanisms for complex environmental and ecological variables, two successive statistical procedures were used following Jongman et al. (1995): first, an ordination to summarize and arrange the data; and second, correlation and multiple regression analysis to interpret the results in the light of what is known about the environment at the sites. Among many other ordination techniques, PCA was used because of its mathematical simplicity and its proven robustness in soil-landform relationship interpretation (Odeh et al., 1991). A detailed description of PCA is beyond the scope of this research: Webster and Oliver (1990), Odeh et al. (1991), and Jongman et al. (1995) provide full accounts.
Principal components analysis generally consists of two steps: the first extracts principal components and the second rotates extracted components to help clarify their interpretation. The first principal component accounts for the largest amount of variance in the sample; successive components explain progressively smaller portions of the total sample variance. The number of components extracted was decided using the rules described in Cattell (1978): eigenvalues greater than one and the interpretation of the scree plot. Before applying PCA, soil variables that were not normally distributed were transformed using an appropriate transformation function (Table 2). Every variable was then normalized to have a zero mean and unit variance to remove the effect of different measurement scales.
|
To interpret the physical meaning of each component and to characterize how terrain attributes are correlated with individual soil properties, Pearson's correlation was first used to assess the association between the component scores of each variable and terrain attributes. Then stepwise multiple regression analyses relating each component to terrain attributes were also performed. The inclusion criterion in the stepwise regression was an F-ratio probability greater than 0.05; a probability <0.01 was used for the exclusion criterion. To estimate the relative influence of individual terrain parameters, the regression coefficient was standardized (ß) and the change in R2 was estimated at each step. The change of R2 is only a relative estimate of the contribution of an individual terrain attribute to the total explained variance because of the possible interaction of independent variables.
The soil attribute with the highest factor loading on each component was further analyzed by spatial interpolation and by stepwise multiple regression. Five soil depths (0–10, 20–30, 40–50, 60–70, and 80–90 cm) were selected to compare vertical and lateral distribution patterns, and the same multiple regression procedures described above were used for each soil depth. The number of sampling points vary from 48 to 64, depending on depth. Spatial interpolation was performed using kriging in the WINSURFER package (Golden Software Inc., Golden, CO). Neither nugget effect nor zonal anisotropy was considered in the interpolation procedure as there were an insufficient number of samples for accurate variogram modeling.
Partitioning Soil Variance into Lateral and Vertical Components
Soil properties in natural landscapes vary according to sampling position and sampling depth. The variation of soil properties over the landscape occurs due to different environmental factors, but vertical horizonation and translocation processes are two unique characteristics of soils. To facilitate further understanding on the pedogeomorphological processes inferred from the ordination, a two-way ANOVA was applied to partition the total variance of each soil attribute into lateral and vertical components.
The total observed variation in soil properties is subdivided into three components: the sums of squares (SS) due to lateral variation across the slope, the SS due to vertical variation with sampling depths, and the remainder:
 |
The first two components can be restated as the lateral and vertical components in the total variance of soil properties. The effect of these two factors was assessed by unbalanced two-way analysis of variance without replication (Milliken and Johnson, 1984; Zar, 1996). While ANOVA procedures are primarily used for experimental hypothesis testing, they can also provide a way of assessing sources of variation (Zar, 1996). Since the purpose of this analysis is the assessment of variability of soil properties at the hillslope scale, it is more appropriate to consider the model as a random-effect model (Model II, ANOVA). Because only one sample was measured (unreplicated), the possible interaction between two factors and the possible measurement error cannot be estimated in the above function. Therefore, F ratios for each factor are determined by mean square of factor to mean square of remainder (Zar, 1996).
Two-way ANOVA procedures were performed in SPSS MANOVA following the unbalanced random model (Model II) (Milliken and Johnson, 1984). The key assumptions of the analysis of variance, normality, and the homogeneity of variance, were tested by the calculation of Cochran's C statistics and the examination of residual plots (Norusis and SPSS Inc., 1994). Some soil properties violate the assumption, but no further statistical consideration has been introduced, since it is generally considered that the analysis of variance is robust with respect to the violation of assumption unless the data deviate severely from the underlying assumptions (Zar, 1996). The goodness of fit of the model was assessed by a regression analysis.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
The soil variables were sorted and regrouped after rotation based on the factor loadings of each component (Table 2). Table 3 is a correlation coefficient matrix of each component with terrain attributes. In most cases, the correlations were low (<0.5), but given the complex distribution of soil properties, higher correlations between single terrain attribute and soil properties are perhaps unlikely. The relatively high r (>0.3) with depth (DEPTH), implies that vertical processes of pedogenesis may play a more important role than lateral processes in total soil variation.
|
|
Most F ratios, both for vertical and lateral components, have a sufficiently high significance level (p < 0.01) to reject the null hypothesis that there is no difference in soil attributes between soil profile location and different sampling depths. The ratio (a/b) between two F ratios for soil profiles (a) and soil depth (b) was calculated to estimate the relative importance of vertical and lateral components to the total variance of each soil property. The dominance of the vertical component (low a/b) in Group 1, interpreted below as a podzolization effect, is mirrored by Group 2 (nutrient dynamic effect), and Group 5 (soil acidification effect). Group 3 (‘solute leaching effect’) generally shows greater lateral variation than vertical variation, while Group 4 (erosion-weathering effect) shows similar variance in both the vertical and lateral components.
The average goodness of fit (R2) of the two-way ANOVA model is 0.65, with a range from 0.37 to 0.87. Groups 2 and 3 show higher R2 than the others. This accords with the high R2 in the multiple regression model (Table 4), which might indicate that soil attributes belonging to these two groups have a clear spatial structure that can be more easily described using a linear statistical model. It is worth noting that soil attributes belonging to these two groups commonly show high F ratios for the lateral component, even though the F ratio for the vertical component is highly variable.
Interpretation of Components
Podzolization Effect (Component 1)
Principal Component 1 (PC1) may be described as a podzolization effect. This is because this component is strongly correlated with the oxide-related soil properties, excluding those related to secondary Mn (Mno, Mnd, Fed/Mnd, and Feo/Fed). A detailed investigation of the lateral and vertical distribution of Fe and Al oxides at the study slope revealed that the distribution of these attributes is determined by the intensity of podzolization, which in turn is governed by the intensity of throughflow in topsoils (Park and Burt, 1999). While an elevated amount of oxalate extractable Si (Sio) has been observed in some podzolic horizons (Bs) and possible geochemical processes involved in translocation of Si during podzolization have been hypothesized (Farmer et al., 1980), Sio has low communality (0.6), which indicates that Sio seems to have a rather different spatial distribution from other oxides on the study slope.
In the interpolated maps of CBD extractable total oxides (TOd), there is a clear increase of total oxides from the upper convex shoulder slope towards the foot of the slope in the surface soil layers, except in the lower hollow where surface water seepage occurs (Fig. 3A) . This spatial pattern is closely related to the intensity of throughflow over the slope, which is highest on the upper slope because of the presence of indurated horizons and partially cemented podzolic B horizons, but becomes weaker towards the base of the slope because of the lack of impermeable horizons at depth (Fig. 1). Along the hollow where convergent throughflow occurs, the lower TOd content extends further downslope. A similar spatial distribution of TOd is also clear in the map of the 20- to 30-cm depth (Fig. 3 B), but below this there is no clear pattern (Fig. 3C–E). Relatively low oxide content persists in the middle of the hollow where the most intensive subsurface hydromorphic processes occur at the 40- to 50-cm depth (Fig. 3C). There are patches of high oxide content in the podzolized upslope soils and at the base of the hillslope in subsurface soil layers.
|
Nutrient-Dynamic Effect (Component 2)
Principal Component 2 (PC2) has the highest factor loading with exchangeable bases (Ca2+, Mg2+, K+, and Na+) and related soil properties [base saturation (BS), divalent cation saturation (CMS), and total exchangeable bases (TEB)]. Aluminum saturation (AS) has a high negative factor loading (-0.87) with this component. It is a well-known geochemical principle that divalent base cations (Ca2+ and Mg2+) and Al compete for cation-exchange surfaces in acid soils (Robarge and Johnson, 1992). The average soil pH at the study site is 3.94 and Al occupies about 80% of total exchangeable surfaces. The competition between divalent base cations and Al is also clearly seen in a high negative correlation between base saturation and Al saturation (r = -0.98, p < 0.01).
The PC2 can be described as a nutrient-dynamic effect. On a nutrient-poor slope, the spatial distribution of major nutrient cations, such as Ca, Mg, and K, is found to be tightly controlled by the soil-vegetation system despite the steep hillslope and catenary variations of soils and vegetation (Trudgill, 1988). Because of this, our results show a relatively homogeneous content of exchangeable bases in the topsoils all over the slope except in the lower hollow (Fig. 4A) , and a rapid decrease in concentration with depth. A noticeable increase of TEB in the lower hollow may be caused by the accumulation of base cations leached from upslope by throughflow. In subsurface soils, however, the cations are no longer subject to uptake by vegetation, but instead subject to downslope redistribution according to hydrological flowpaths. A deficit of TEB upslope and an accumulation downslope with depth, especially within the saturated wedge in the lower hollow, can be observed in Fig. 2B through 2E . In a detailed examination of hydrochemical processes at the study slope, Burt and Park (1999) showed that Ca2+, Mg2+, and Na+ carried by throughflow are stored in the saturated wedge at the base of the slope and gradually released into the stream at times of high flow.
|
|
The soil properties belonging to this group show clear variations in both the vertical and lateral components (Table 2). The F ratio in the two-way ANOVA for depth frequently exceeds 60. The F ratio for the soil sampling position is also the highest for any soil attribute except some belonging to Group 3. The ANOVA model explains more than 80% of total variance (Table 2), and 61% of total variance is predicted by terrain attributes (Table 4). In the regression model for TEB in the topsoil layer (Fig. 4), plan curvature and upslope area make significant contributions to the explained variance, reflecting the increase of TEB along the lower hollow. Elevation, omitted in the regression model of the topsoil layer, becomes dominant in the other regression models explaining the general spatial distribution of TEB, low upslope and high downslope.
Manganese Leaching Effect (Component 3)
Most Mn-related properties, including exchangeable Mn, have a high factor loading with Principal Component 3 (PC3). The main spatial characteristic of secondary Mn is a clear differentiation downslope in acid environments (McDaniel et al., 1992), a pattern which is clearly reflected here. The comparison between F ratios in Table 2 confirms the dominance of lateral variation of Mn-related soil properties. The marked accumulation of secondary Mn at lower slope positions in acid soil environments occurs as follows: secondary Mn, once released by chemical weathering, moves downslope in reduced form and accumulates in the A horizon at lower (especially convergent) slope positions in the form of relatively stable Mn2+–organic matter complexes under higher reduction potential (EH) conditions (McDaniel et al., 1992).
Clear lateral differences of Mnd are easily seen in the interpolated maps (Fig. 5) . Unlike many other soil properties, the downslope difference remains clear at depth, even though the effect becomes weaker. The vertical distribution of Mn oxide content is unique: soils upslope show an increase with depth, but soils downslope show a steady decrease with depth. Given the high intensity of throughflow processes on the study slope, this vertical pattern is indicative of lateral removal from upslope and deposition in surface layers downslope. There is no significant influence of podzolization processes on the vertical translocation of Mn oxides (see also Jersak et al., 1995).
|
Even though the communality and factor loadings are relatively low, soil pH was also included in this group (Table 2). Both the regression analysis (not shown here) and the ANOVA model for soil pH show a similar spatial distribution to secondary Mn: clear differentiation downslope without any significant influence of vertical pedogenesis. In regression models, elevation and slope angle are also the main independent variables explaining soil pH with similarly high R2 values (0.66 and 0.61 at 0- to 10- and 20- to 30-cm depth, respectively).
Erosion-Weathering Effect (Component 4)
Principal Component 4 (PC4) has a high factor loading with soil texture properties. The Feo/Fed ratio is also included but has lower communality. There is a distinct difference in the distribution of soil texture between shallow (<40 cm) and deep soils on the study slope. In the upper 40 cm, there is a clear differentiation of soil texture from the interfluve to the base of the hillslope: finer soil on the interfluve changes to coarse, sandy soil at the convex shoulder slope, and then becomes finer downslope (Fig. 6A,B)
. High silt and clay content are also observed at the base of the hollow. Active subsurface washing processes through relatively permeable A horizons may selectively remove fine soil particles, such as silt and clay, and result in a sandy soil texture in the upslope area.
|
The PC4 is identified as an erosion-weathering effect, given the dominant slope processes. In general, the vertical and lateral differences of soil properties are comparatively small in the ANOVA model (Table 2). Considering the characteristics of the parent material (arenaceous sandstones and siltstones), the compositional changes of soil texture in the soil samples may be small, but the dominance of the lateral component over the vertical for the case of sand and silt contents seems to indicate selective transportation of fine particles along the hillslope as seen in Fig. 6A and B.
The regression models for individual soil layers accord with the spatial distribution described above (Fig. 6): a clear difference in topsoil associated with elevation, and a high silt content in the hollow associated with plan curvature; nevertheless, R2 is low (<0.38) in most soil layers. The regression model for the component, in which plan curvature and depth were included (Table 4), also has a low R2 (0.26).
Acidification Effect (Component 5)
As might be expected with a more ‘minor’ fifth factor, this component is more difficult to interpret than the previous four components. Effective cation-exchange capacity (ECEC), total exchangeable cations, exchangeable Al, and LOI are strongly correlated with Principal Component 5 (PC5). The spatial distribution of these soil properties is believed to be connected to soil acidification processes in highly acidic soils. The average soil pH is 3.94, and Al occupies more than 80% of total soil exchangeable sites. Consequently, this component is interpreted as an acidification effect.
In the ANOVA analysis, a relatively strong vertical component was recognized, but the model fits (R2) are lower compared with other soil properties (Table 2). Their spatial distribution patterns are also poorly modeled by terrain attributes (Table 4 and Fig. 7) .
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Our attempt to establish regression functions to predict soil properties using terrain attributes has had varying degrees of success, not only in the amount of variance explained by terrain attributes (0 < R2 < 0.77), but also in the combination of terrain attributes brought into the regression models. In general, the fit of regression models is highest in topsoil (0.32 < R2 < 0.77) and decreases with depth for most soil properties examined. This implies that soil properties in shallow surface soils are more predictable from terrain information than those in subsurface soils, possibly because pedogeomorphological processes are more clearly controlled by soil horizonation and slope form at shallow depth. The dominant contemporary slope process on the study slope is throughflow, which is generated by vertical anisotropy of soil hydraulic characteristics and steep slope gradient, and its intensity shows good correspondence with the three-dimensional arrangement of slope geometry and soil development (Anderson and Burt, 1978; Park and Burt, 1999). Active throughflow processes result in the spatial differentiation of soil properties; these have a high correlation with terrain attributes. On the other hand, the limited influence of shallow throughflow on the steep slopes (excluding the hollow) and the heterogeneity of the parent material reduce the systematic catenary variation of soil properties at depth, which in turn makes them difficult to predict solely from terrain information.
One difficulty involved in modeling soils is that individual soil properties, even in topsoil, have very different combinations of terrain attributes in their regression models. For example, elevation and upslope area jointly explain 77% of the variance of total oxide content in surface soils, but do not explain other soil properties so successfully. For many soil properties at different depths, elevation is the predominant predictor regardless of sampling depth. The contribution of other primary and secondary terrain attributes are variable depending on the soil attribute chosen. A large variation in the kinds of terrain attributes in regression models and their relative contribution to R2 is the most frequently observed result of statistical modeling of soil-landform relationships (see Pennock et al., 1987; Moore et al., 1993; McKenzie and Austin, 1993; Odeh et al., 1994; McBrateny et al., 2000). This problem has partly been attributed to imperfections in the recording and analysis of terrain attributes (Moore et al., 1993; Gerrard, 1992). This research, however, has clearly shown that the fundamental problem is the fact that some soil properties show a remarkably varied spatial distribution according to their differential involvement in pedological and geomorphological processes. This implies that a given set of terrain attributes that show a good statistical relationship with one soil attribute may have less relevance for other attributes.
In a more theoretical context, an assumption in most attempts to seek a functional correlation between soil properties and landform geometry is the downslope movement of soil materials in response to slope or the downslope hydraulic gradient (Huggett, 1975; Moore et al., 1993). In this analysis, it is apparent that few soil properties follow a simple downslope transport process as governed by slope form alone; only Mn comes close to this. Most other soil properties deviate significantly from the assumption, which greatly reduces the possibility of predicting soil properties using terrain attributes.
Both pedological and geomorphological processes exhibit complex variations in time and space. The results of this research may be criticized as a site-specific presentation of soil-landform relationships on a single hillslope. It may also be true that the specific choice of soil properties in this research had a strong influence on the statistical grouping of soil properties and the identification of process domains. However, this research has the one clear implication, that any future attempts to predict and model soil properties using environmental variables must put a greater emphasis on the selection of the soil properties modeled and the detailed controls on the selected soil properties.
|
CONCLUSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  M. D. Tomer, C. A. Cambardella, D. E. James, and T. B. Moorman Surface-Soil Properties and Water Contents across Two Watersheds with Contrasting Tillage Histories Soil Sci. Soc. Am. J., February 27, 2006; 70(2): 620 - 630. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
F. M. Ziadat Analyzing Digital Terrain Attributes to Predict Soil Attributes for a Relatively Large Area Soil Sci. Soc. Am. J., August 25, 2005; 69(5): 1590 - 1599. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. D. Tomer and D. E. James Do Soil Surveys and Terrain Analyses Identify Similar Priority Sites for Conservation? Soil Sci. Soc. Am. J., November 1, 2004; 68(6): 1905 - 1915. [Abstract] [Full Text] [PDF]
|
|
| |||||
