SSSAJ Journal of Natural Resources and Life Sciences Education
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via ISI Web of Science (1)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Ishida, T.
Right arrow Articles by Ando, H.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Ishida, T.
Right arrow Articles by Ando, H.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Ishida, T.
Right arrow Articles by Ando, H.
Related Collections
Right arrow Pedology
Right arrow Soil Fertility and Productivity
Right arrow Spatial Distribution
Soil Science Society of America Journal 67:190-197 (2003)
© 2003 Soil Science Society of America

DIVISION S-5—PEDOLOGY

Drainage Network Analysis for Regional Partitions of Alluvial Paddy-Field Soils

T. Ishida*,a, S. Itagakia, Y. Sasakib and H. Andob

a Department of Agricultural Engineering, Faculty of Agriculture, Kagawa University, Ikenobe 2393, Miki, Kagawa, 761-0795 Japan
b Department of Agronomy, Faculty of Agriculture, Yamagata University, Tsuraoka, Yamagata, 997-8555 Japan

* Corresponding author (ishida{at}ag.kagawa-u.ac.jp)


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Terrain data on floodplains might be a useful source of ancillary information about soil properties at a regional scale. However, terrain data are not believed to be suitable for obtaining information on paddy-field soils on floodplains, because the contrast in the topography of floodplains is only slight. This study was conducted in the Tsuruoka area, a part of the Shonai Plain in the northern part of Japan, to evaluate whether a division of floodplain into landform elements can provide useful information on delineation of paddy-field soils. As source data for the division, a digital elevation model (DEM) was constructed based on a closely spaced differential global positioning system (DGPS) survey and geostatistical interpolation. By using the DEM, drainage network analysis was performed to delineate the landform units. To evaluate correspondence of soil classification with the land division, soil chemical data from 154 soil samples were classified by using a set of numerical procedures. The numerical procedures included principal component analysis and unsupervised classification techniques. In the land partition map created, each landform unit could satisfactorily represent a separate soil group as characterized by soil chemical properties. A statistical analysis revealed that the regional partition map was better than the conventional soil map inasmuch as the partition map described more satisfactorily the spatial distribution of soil chemical properties. However, this conclusion depends largely on the spatial resolution and precision of the DEM. The useful regional partition map could not be produced by different DEM that has lower resolution and accuracy.

Abbreviations: AIC, Akaike's Information Criteria • BSP, base saturation percentage • CEC, cation-exchange capacity • EXCH, exchangeable • DEM, digital elevation model • DGPS, differential global positioning system • GPS, global positioning system • PAC, Phosphate Adsorption coefficient • TN, total N


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
CONVENTIONAL SOIL MAPS based on low-intensity surveys neither delineate all of the inherent local variability nor represent specific variations in soil attributes. To overcome these defects in conventional soil maps, in previous papers Ishida and Ando (1994a)( 1994b) applied a set of statistical and numerical classification procedures to soil chemical data collected in paddy fields of the Aidu Basin in the northern part of Japan and then produced a region-partitioning map on the basis of the numerical procedures. Inclusion of elevation data as an attribute made the region-partitioning map useful—the soil groups classified with the data set that included the elevation data were able to satisfactorily represent regional differences in rice (Oryza sativa L. )yield; this was not possible with the soil groups based on the data set without elevations or with the soil groups of a conventional soil map. The landform of the Aidu Basin is classified into three major topographic features: two middle terraces and a floodplain, indicated by a striking contrast in elevation data. This is why the elevation data were useful for the soil classification and, therefore, terrain attributes derived from DEM were unnecessary. However, in regions where soil-landform models are more complex, the terrain attributes might be important. In the present study, the study area involves only a floodplain; therefore, the contrast in the topography of the terrain is only slight. Shoji et al. (1973a) supposed that the main source of information on soil variability in floodplains might be landform variations that may reflect underlying changes in parent materials and differences in age of topographical development. However, the supposition is not quantitatively validated. In addition, the landform variations are prescribed by watersheds and overland flow paths. Algorithms traditionally included in deriving most terrain attributes use neighborhood operations to calculate attributes, such as slope, aspect, and points of inflection, based on the values in adjacent points (Troeh, 1964; Acton, 1965; Walker et al., 1968; Moore et al., 1993). While watersheds and overland flow paths are closely related to slope, aspect, and inflection information, they also present nonneighborhood problems, such as determining direction of flow in the interior of a large flat area, in which terrain attributes cannot be determined by the values in adjacent points (Jenson and Domingue, 1988). These nonneighborhood problems are critical in studying relations between floodplain soils and landforms. The nonneighborhood problems involve the identification and treatment of local depressions and relief increments of flat areas. Drainage network analysis has been used in hydrological studies, such as hydrological surface runoff models, but little attention has been given to its use in regionally partitioning soil groups.

Terrain data as ancillary information needs to be obtained cheaply. In Japan, there is a nationwide DEM that can be easily utilized (Miyazaki and Tsukahara, 1987). However, it is doubtful whether its spatial density and precision are sufficiently high to extract topographic structures useful to analyze soil-landform processes within floodplains.

The objectives of this paper are to determine whether automatic drainage network analysis, which takes into account methods to overcome nonneighborhood problems, can be used to regionally divide paddy-field soils within a floodplain, and to examine the differences between the topographic structures extracted from the DEM obtained by the global positioning system (GPS)—which is accurate, but expensive—and those from the nationwide DEM.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Site Description
Our study was conducted in the Tsuruoka area, located in the southern part of the Shonai Plain, in the western part of Yamagata Prefecture in northern Japan. The Tsuruoka area is regarded as a region representative of Japanese paddy fields (Sasaki, 2000). Almost all of the arable land is utilized as paddy fields. Most parts of the Tsuruoka area are alluvial plains of fluvial origin. The southern edge of the Tsuruoka area is bound by the peaks of the Asahi Mountains, the southeastern edge is bound by the long stretch of the Dewa Hills, the northern edge by Mt. Chokai, and the western edge by the Sea of Japan. The Akagawa, Ohyama, and Ohto Rivers flow north through the study area (Fig. 1) . The Akagawa and the Ohyama Rivers flow parallel, originating from the peaks of the Asahi Mountains and emptying into the Sea of Japan. The Ohto River flows from the west of the Ohyama River and joins the Ohyama River in the middle of the study area. The geological features of the upstream regions of the Akagawa and Ohto Rivers can be classified, respectively, into granitic rocks and volcanic tuffs, which were formed during the Miocene Epoch of the Tertiary Period. The geological features of the upstream regions of the Ohyama River are mixed volcanic tuffs and granitic rocks. Soil groups in the study area generally correspond to Entisols (Soil Survey Staff, 1998).



View larger version (36K):
[in this window]
[in a new window]
  		Fig. 1. Location map of the study area showing contour lines and the locations of soil sampling points.

 
The usual size of a section of paddy field in the study area is 0.3 ha; this size has been used in all paddy land consolidation projects in the study area. In Japan, a holding of paddy field is called a field lot. The field lot is usually 100 m long by 30 m wide.

Determination of Soil Chemical Properties
Soil samples were collected from the Ap horizons of 154 sites (Fig. 1). Soil sampling was designed so that soils were normally sampled at the intersections of a grid of 600 by 600 m spacings. We measured only chemical soil characteristics, because the potential fertility of paddy fields depends largely on these chemical properties (Kyuma and Kawaguchi, 1973). After the soil had been air-dried and sieved, soil fractions <2 mm were used for the laboratory determination of soil chemical properties. Soil reaction measurements were made by means of an electrometric method using a glass-indicating electrode (Date, 1986). Total nitrogen (TN) was measured with the semi micro-Kjeldahl method (Bremner and Mulvaney, 1982). Phosphate adsorption coefficient (PAC) was evaluated as follows: 672 mg of P2O5 in 50 mL of ammonium phosphate dibasic solution was added to 25 g of soil and the content of the phosphate in the filtered solution was determined by colorimetric analysis (Nanjyo, 1986). Cation-exchange capacity (CEC) was measured as follows (Kamata, 1986): First, soils were saturated with 1 M NH4OAc and the adsorbed NH+4 was extracted. The contents of exchanged cations (EXCH cations; Ca, Mg, K, and Na), which were extracted with 1 M NH4OAc, were estimated by inductively coupled plasma optical emission spectrometry (Muramoto et al., 1987a, b). In addition, we computed base saturation percentage (BSP), because this soil property is considered to be important for representing paddy-field soil fertility.

Methods of Clustering Soil Sampling Points
Each chemical attribute in the soil sampling dataset was standardized to unit variance and zero mean. To avoid redundancy of information, we applied principal component analysis to the soil sampling data set. We identified the principal components, whose eigenvalue was >1, as useful. Since a value of 1 indicates that the variance of the component is greater than that of the original dataset and therefore the component might be significant (Rao, 1964), we selected 1 as the threshold. The component scores, corresponding to the components identified as useful, of each soil sample were calculated and used as input data for the following clustering method.

Hierarchical and nonhierarchical clustering methods were successively applied to numerically classify the sampling points using Ward's method (Ward, 1963) and the ISODATA method (Ball and Hall, 1965). One of the potential advantages of the ISODATA method is the use of heuristic devices for adjusting the number of clusters to the apparent natural structure of the data set.

The component scores were applied to Ward's method to determine the number of groups whose classes were identified as the upper part of the dendrogram and to estimate their centroids. Ten groups were found to be suitable, based on a subjective examination of the dendrogram. Second, the group centroids obtained by employing Ward's method were set as the initial seed points used as cluster nuclei for the ISODATA method, and the ISODATA method was then applied. The steps of the ISODATA method were repeated until a convergence was achieved. This convergence was evaluated using the Akaike's Information Criteria (AIC) index (Akaike, 1974). The AIC index is estimated by

[1]
where n is the number of sampling points and nc is the number of groups. R is defined as follows (Ohsumi, 1976). Since the lower AIC value indicates goodness of classification, the number of groups selected corresponds to that for which the AIC value is the least.

[2]
where np is the number of principal component scores and ngi is the number of soil samples allocated to the ith group. The variable zijk is the value of the kth component score for the jth soil sample allocated to the ith group. The variable zik is the mean value of the kth component score over all the soil samples allocated to the ith group.

GPS Survey and DEM Production
Elevation data were collected at 5489 different paddy-field lots over the 4300-ha site using Trimble mobile GPS receivers (Trimble, Sunnyvale, CA) mounted on a range pole and a stationary receiver mounted on a tripod; these receivers were operated in the kinematic DGPS mode to define terrain variability (Fig. 2) . The kinematic DGPS mode uses the GPS carrier signal and information on relative position of reference station to correct various inaccuracies in the GPS system. The reference station was at the Shonai Branch of the Yamagata Agricultural Experiment Station, located near the study area. The survey system used is able to provide location information including elevation for each paddy field to an accuracy of a few centimeters. These measured elevations were interpolated into a regular grid using an ordinary kriging procedure with a linear detrend to create a 10-m DEM. The linear detrend is the deletion of a linear trend from elevation. The linear trend is a linear function of elevation. That is, it has the form z = a0 + a1x + a2y, where the a's are coefficients determined by least square method, z is elevation and (x,y) is the location.



View larger version (49K):
[in this window]
[in a new window]
  		Fig. 2. Location map of the study area showing contour lines and the locations of the GPS survey points. The gray "lines" are strings of DGPS points.

 
Taking into account topographic discontinuities because of the presence of rivers, the study area was divided into three watersheds: Watershed A covering the east side of the Ohyama River, Watershed B covering the west side of the Ohyama River below the convergence of the Ohyama and Ohto Rivers, and Watershed C lying between the Ohyama and Ohto Rivers. In the kriging procedure, a variogram of each watershed was estimated by using the elevation data measured within each watershed. A model of variogram and its key parameters were selected by fitting the model to data using a weighted least square method, as described by McBratney and Webster (1986).

As referential data to the DEM, we also used a nationwide DEM produced by the Geographical Survey Institute (Miyazaki and Tsukahara, 1987). The digital elevation data in the national DEM are on a 50 by 50 m grid scale. The elevation data are in 1-m increments with a typical root mean square error of 5 m.

Drainage Network Analysis
We followed Martz and Garbrecht (1992), whose approach is based on overland flow simulation across the landscape, to numerically define drainage networks and watershed boundaries from the DEMs. The goal of the drainage network analysis in this paper is to produce a set of drainage divides such that the divides completely partition the watershed into subwatershed polygons and to evaluate whether the subwatershed polygons are useful for delineating soil groups. A drainage network to which each grid cell is connected with a treelike structure must be composed to determine the subwatershed polygons corresponding to each grid cell on the DEM.

The basic points of the drainage network analysis are based on the following views. From the rule that the flow direction for a grid cell is the direction of the steepest downward slope to an adjacent cell (Fairchild and Leymarie, 1991), a data set of flow vectors was created, and the path of steepest descent between adjacent cells was followed until the edge of the DEM was reached. If the steepest downward slope was encountered at more than one adjacent cell, flow direction was assigned to be toward the cell taking the largest local relief into account. Therefore, by using the data set of flow vectors and streamlines, drainage networks were represented by a treelike structure such that a downstream junction or a loop in a stream could not exist. The flow vectors data set was converted to a flow accumulation data set, where each cell was assigned a value equal to the number of cells that flow to it. The subwatershed was defined as the entire upstream drainage area starting from the watershed outlet cell located at the edge of the DEM. The method of Martz and de Jong (1988) was used for the algorithm determining the subwatershed area. Treatments of depressions and flat areas were included in the drainage network analysis. The analytical treatment of depressions was essentially the one used by Martz and de Jong (1988). The treatment was to fill depressions by increasing the values of cells in each depression to the value of the cell with the lowest value on the depression boundary. The analytical treatment of flat areas is similar to the flow direction modification procedure used by Jenson and Dominigue (1988). The treatment was to impose relief on flat areas by adding a small incremental elevation value to each cell in the flat area to allow an unambiguous definition of flow lines across these areas.


   RESULTS AND DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
CONCLUSIONS
 REFERENCES
 
Digital Elevation Model Production and Drainage Network Analysis
Experimental variograms for elevation data, preprocessed with the linear trend, were described in each watershed (Fig. 3) . The variograms were represented by spherical or exponential model and their key parameters are given in Table 1.



View larger version (20K):
[in this window]
[in a new window]
  		Fig. 3. Experimental variograms of elevation data with detrend. The lines represent values calculated by fitting spherical or exponential models: •, Watershed A; {blacktriangleup},Watershed B; {blacksquare}, Watershed C.

 

View this table:
[in this window]
[in a new window]
  		Table 1. Key parameters of theoretical models fitted to experimental variograms.

 
In the subwatershed map (Fig. 4 , the solid lines represent the pattern of ridges or the boundaries of the subwatersheds. The ridges correspond to cells having a flow accumulation value of zero (cells into which no other cells flow). The dashed lines indicate the fully connected drainage network. The networks were delineated by the pattern formed by highlighting cells with accumulation values higher than an arbitrary threshold value. Therefore, although the drainage network depends on the threshold value, the pattern of ridges is independent. Since the drainage network is not essential to this study, the threshold values were arbitrarily determined. As shown in Fig. 4, Watershed A was divided into four subwatersheds, Watershed B into three subwatersheds, and Watershed C could not be divided.



View larger version (68K):
[in this window]
[in a new window]
  		Fig. 4. Map of subwatersheds delineated by automatic drainage network analysis.

 
To display the usefulness of the kinematic DGPS method on topographical divisions in alluvial lowland areas, the drainage network for subwatershed A-1 created by using the kinematic DGPS data was compared with that created using the nationwide DEM, as described above (see Fig. 5) . In this figure, the dashed and solid lines represent the drainage networks produced by the former and the latter DEMs, respectively. It is clear at first sight that the "scratch" patterns of the drainage network created from the nationwide DEM produced too many subwatersheds and the numerous subwatersheds are practically meaningless to the regional partition of paddy fields. Terrain maps of the particular topographical features within the rectangle in Fig. 5 (Fig. 6) highlight the differences between the drainage network patterns created by the two DEMs. The x and y coordinates are the distances from the upper left corner of the rectangle. For the nationwide DEM (Fig. 6b), flat areas extend over the rectangle as a result of the low relief of the floodplain and the low vertical resolution of the DEM. Flat areas are not unusual in Japanese alluvial lowlands. Therefore, a DEM more precise and dense than the nationwide one is a prerequisite for evaluating soil-landscape processes from a DEM.



View larger version (58K):
[in this window]
[in a new window]
  		Fig. 5. Comparison between drainage networks in Subwatershed A-1 produced by kinematic DGPS survey and nationwide DEM. Solid lines show drainage networks obtained by the nationwide DEM; dashed lines show drainage networks obtained by kinematic DGPS survey. The rectangle is the area depicted in Fig. 6.

 


View larger version (34K):
[in this window]
[in a new window]
  		Fig. 6. Terrain maps produced by (a) kinematic DGPS survey and (b) nationwide DEM.

 
Clustering for Soil Sampling Points
As indicated in Table 2, we used the two components, whose eigenvalue is >1, to estimate the corresponding component score of each soil sample that is equal to summation of each attribute's value multiplied by the corresponding eigenvector. The two components accounted for 71% of total variance in the sample. Large contributions to the first component were derived from PAC, CEC, and EXCH Ca and Mg. The second component was mainly related to pH, EXCH Na and K, and BSP.


View this table:
[in this window]
[in a new window]
  		Table 2. Components 1 and 2 eigenvalues and eigenvectors for the soil sampling data set.

 
First, to evaluate the results from the clustering analyses for the soil sampling points, we will briefly describe the chemical characteristics of Japanese paddy-field soils. In general, the chemical properties are related to the parent materials or to the toposequence of the soil (e.g., Shoji et al., 1973b; Matsuzaka, 1969). A high content of organic matter is related to parent materials of volcanic ash or peat. Alluvial soils show a low PAC. The value of CEC is closely related to the clay-mineral composition and soil texture of alluvial lowland soils, which depend on both the upstream geological system and topography (Shoji et al., 1973b). The parent material or toposequence of soil, however, is not a dominant factor in the amount of EXCH cations and BSP, because heavy applications of inorganic soil amendments and chemical fertilizers, such as KCl, result in higher amounts of EXCH cations and increased BSP values. The mean values for the attributes of soil groups and the spatial distribution of the soil sampling points are presented in Table 3 and Figure 7 respectively. Soil groups are divided into two types according to the values of CEC and PAC. Groups e and f are characterized by high CEC and high PAC, while the others have low CEC and low PAC. Soils belonging to Groups e and f are located along the Ohto River or the upstream section of the Ohyama River; the other soils occur along the Akagawa River or the downstream section of the Ohyama River. Parent materials from the upper reaches of the Ohyama and Ohto Rivers differ from those from the upper reaches of the Akagawa River. The upstream regions of the Akagawa River are characterized by granitic rocks; on the other hand, the alluvial deposits of the Ohyama and Ohto Rivers are derived from volcanic tuffs. Clay minerals in the alluvial deposits originating from the volcanic tuffs are mainly 2:1 types and those derived from granitic rocks are predominantly 1:1 types (Sasaki, 2000). This difference in clay-mineral compositions can account for the difference in the values of CEC and PAC among the soil groups. The PAC value of Group e is close to the upper limit (6.555 g kg-1 soil) for the PAC of Japanese alluvial paddy soils (Matsuzaka, 1969). The higher PAC value shows that the origin of soil is volcanic ash. Thus, compared with Group f, Group e is more influenced by volcanic ash, not through a direct ejection from volcanoes, but through resedimentation.


View this table:
[in this window]
[in a new window]
  		Table 3. Mean values of soil groups belonging to each cluster.

 


View larger version (40K):
[in this window]
[in a new window]
  		Fig. 7. Map of the study area showing soil sampling points classified according to soil chemical information.

 
Among Groups a, b, c, and d, TN contents are higher in Groups a and c than in Groups b and d. Since there are only three soil samples belonging to Group c and the locations of these soil samples are dispersed, it might be said that we can conclude nothing about Group c. Soils belonging to Group a are mainly located in the area nearest the Akagawa River, while those belonging to Groups b and d are sited along the Ohyama River. In addition, the CEC value of Group a is higher than that of Groups b or d. Based on the fact that the variation of the C/N ratio is not great over the study area (Sasaki, 2000), it might be reasonable to conclude that this difference in CEC is affected by the TN content, because TN content is fairly indicative of organic matter content. These facts indicate that there is a difference in parent material between Group a and the other groups. Consequently, it might be concluded that the parent material of soil Group a is mainly derived from the Akagawa River, and that of Groups b and d is more affected by the Ohyama River than the Akagawa River.

Correspondence Between Landform Division and Soil Classification
The Economic Planning Agency produced nationwide soil maps with a scale of 1:50 000, to obtain basic information on general land utilization or development, irrespective of land cover types such as forest and agricultural land (Economic Planning Agency, 1970). Afterwards, the soil maps were converted to digital soil map with spatial resolution of about 1 km (Geographical Survey Institute, 1992), so that we could easily utilize soil map information. The conventional soil map (Fig. 8) was compiled based on the digital soil map. The detailed soil survey to make the soil map was performed at intervals of 50 ha together with ancillary soil physicochemical measurements of soil samples in the surface plow layer (0–15 cm) every about 10 ha (Economic Planning Agency, 1970). There are six soil series in the conventional soil map: Nishiyama soil series (strong gley soils, fine textured type); Kotohama soil series (strong gley soils, medium-, and coarse-textured types); Kaneda soil series (gray lowland soils, fine-textured and grayish-brown types); Morohashi soil series (the same types as the Kaneda soil series); Asozu series (gley soils, gravelly type); Nagatomi soil series (lowmoor peat soil).



View larger version (41K):
[in this window]
[in a new window]
  		Fig. 8. Conventional soil classification map.

 
By visual comparison of Fig. 4 and 8, the Watershed A appears to be divided into soil series extending north and south in a similar way to the land division. The Nishiyama soil series distributes over the Watershed A. The areal distribution of the Nishiyama soil series does not correspond with the soil groups based on soil chemical properties, judging from the fact that CEC and PAC are different between the northern part and southern edge of Watershed A as mentioned above. The delineated line of soil series on the Watershed B appears to be similar in shape to the land division. Different soil series from Nishiyama series are located at the eastern and southern edges of the Watershed B, as well as the land division. Although the Watershed C could not be divided into subwatersheds delineated, the Watershed C is composed of two soil series. The conventional soil map may be more useful than the topographical division map, in the sense that soil chemical properties at the northern edge of Watershed C is different from those at the southern parts.

The soil group corresponds, to some extent, with the subwatershed (Table 4). For example, it might be regarded that the correspondence of the subwatershed C-1 with soil Group e is tolerably good. The boxed cell in this table indicates that the correspondence is >60%. The number of soil samples belonging to each soil group found in each soil series was also estimated in a similar way to Table 4 (Table 5). The number of the boxed cell in Table 5 is only three, indicating that the topographical division map represents soil fertility more satisfactorily than the conventional soil map.


View this table:
[in this window]
[in a new window]
  		Table 4. Correspondence between topographical divisions and soil classification groups. Italics indicate that the correspondence is >60%.

 

View this table:
[in this window]
[in a new window]
  		Table 5. Correspondence between conventional soil series and soil classification groups. Italics indicate that the correspondence is >60%.

 
To obtain a more quantitative index for the suitability of the landform division map, we applied the AIC index to the classification results according to soil chemical information, the topographical division map, and the conventional soil map. The AIC is a theoretical index for better classification. The smaller the value of AIC, the better the results of classification. For the soil sampling data, the AIC values of the conventional soil map and topographical division map were 118.7 and 89.2, respectively. The AIC values were computed by using a data set that was normalized but to which principal component analysis had not been applied because comparing the AIC values driven from total variance are unbiased. The AIC result reveals that the topographical division map is more useful than the conventional soil map for representing soil fertility levels. The similar AIC value based on the ISODATA classification of the soil sample data was 76.9, indicating that the topographical division map could satisfactorily reproduce the distribution of soil chemical characteristics.


   CONCLUSIONS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
REFERENCES
 
The landform unit according to the drainage network analysis was related to separate soil group identified from numerical classification of soil chemical properties, which largely determine the fertility capability of paddy fields. The landform partition map also corresponded, to a certain extent, with the conventional soil map based on soil survey for general land utilization and development. However, the statistical comparison between the landform division map and the conventional soil map revealed that the landform division map was more suitable inasmuch as the delineation of the landform units represents quite well the spatial distribution of the soil group. The inability to estimate soil productivity has led to the criticism of various classification systems (Isbell, 1990). In further study, the correspondence of landform units with the spatial characteristics of rice yield must be examined to verify the hypothesis that the regional partition may enable to resolve the problem encountered in soil classification.

This study demonstrated the feasibility of using the drainage network analysis in regionally partitioning soil groups. The feasibility might depend on the DEM compiled by precise DGPS survey and geostatistics. Since landform changes on floodplains are generally very slight, the drainage network analysis requires collection of accurate and closely spaced topographical data in the similar way to this study, judging from the fact that more inaccurate and more sparse spaced data on the nationwide DEM created the meaningless "scratch" patterns of the drainage network for the regional partition of paddy-field soils. However, a DGPS survey might not be a cheap method of obtaining the DEM and, therefore, the terrain data would not be practically useful as an ancillary source of information. Other modern methods (e.g., high resolution remote sensing) could allow a DEM to be rapidly generated with very fine spatial resolutions and to very high accuracy standards.

Received for publication April 18, 2002.


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 





This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via ISI Web of Science (1)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Ishida, T.
Right arrow Articles by Ando, H.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Ishida, T.
Right arrow Articles by Ando, H.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Ishida, T.
Right arrow Articles by Ando, H.
Related Collections
Right arrow Pedology
Right arrow Soil Fertility and Productivity
Right arrow Spatial Distribution

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Crop Science
Journal of Natural Resources
and Life Sciences Education		 Vadose Zone Journal
Journal of Plant Registrations Journal of
Environmental Quality		 The Plant Genome