Soil Science Society of America Journal 67:1180-1185 (2003)
© 2003 Soil Science Society of America
DIVISION S-4—SOIL FERTILITY & PLANT NUTRITION
Comparison of Sampling Designs in the Detection of Spatial Variability of Mississippi Delta Soils
H. J. Buscaglia and
J. J. Varco*
Dep. of Plant and Soil Sci., Box 9555, Mississippi State Univ., Mississippi State, MS 39762
* Corresponding author (jvarco{at}pss.msstate.edu)
 |
ABSTRACT
|
|---|
Precision application of agrichemicals requires an accurate assessment of the spatial structure of soil properties. Spatial structure analysis of soil properties could be influenced by sampling design especially on highly variable alluvial derived soils. The objective of this research was to compare spatial structure analysis between grid-point and grid-cell type sampling for alluvial derived soils, which vary more in one dimension than a second. Soil samples (0- to 0.15-m depth) were taken along two transects in an irrigated cotton (Gossypium hirsutum L.) field located in the Lower Mississippi Valley flood plain and were analyzed for total C and N, extractable Ca, Mg, K, Na, Zn, and P, and pH. Moran's I autocorrelation coefficient was computed at preselected lag distances and correlograms were plotted to examine trends in autocorrelation. A correlation range of near 300 m appeared to be associated with sampling across two soil mapping units, while a shorter range as well as a cyclic spatial structure was likely influenced by alternating soil mapping units. Similarity in autocorrelation trends calculated using grid-point and grid-cell sampling designs suggests spatial pattern detection for soils of the Mississippi Delta can be achieved by either sampling methodology with a sampling resolution of approximately 46 m or less. Grid-point sampling compared with grid cell is more easily facilitated and requires less labor and time, but could be implemented more efficiently by using a grid with reduced sampling intensity in areas with a high probability of long-range autocorrelation.
Abbreviations: SEB, sum of exchangeable bases
|
INTRODUCTION
|
|---|
IDENTIFICATION AND INTERPRETATION of spatial variability in soil properties and its influence on crop productivity presents formidable challenges in addition to environmental and producer induced variability. Variability in soil test properties within fields can now be addressed with variable rate fertilizer technology, but undetected deviations in soil test parameters can result in inaccurate fertilizer recommendation maps (Peck and Melsted, 1973). Accurate detection of field scale variability in soil properties can require sampling densities, which are cost prohibitive (Franzen and Peck, 1995; Wollenhaupt et al., 1994). As a result, grid-point sampling has been recommended for variable rate fertilization (Wollenhaupt et al., 1994).
Variation in physical, chemical, and biological properties especially more in one direction than a second is characteristic of alluvial soils (Campbell, 1979). Soils derived from Mississippi River alluvium vary greatly across relatively short distances, with greater variation generally occurring in a direction perpendicular to the river. Soil properties, which vary in this manner may not be accurately detected if grid-point sampling is used and the period of the grid is similar, but offset to the periodicity of the soil mapping units (Webster and Oliver, 1990). Gaston et al. (2001) working in the Delta region of Mississippi reported spatial variability in soil pH and organic C was primarily short-range or <60 m, their minimum sampling distance. In contrast, clay and sand contents displayed spatial dependence up to around 120 m. Several soil-sampling methodologies have been developed in an effort to accurately describe spatial trends in soil fertility for variable rate fertilization (Cahn et al., 1994; Thompson and Robert, 1995). Soil sampling is routinely based on previously determined management zones such as mapping units (Anderson-Cook et al., 1999; Carr et al., 1991), landscape position (Franzen et al., 1998), or on systematic layouts (Anderson-Cook et al., 1999; Franzen and Peck, 1995; Wibawa et al., 1993).
Traditionally, soil-sampling strategies for field scale fertilization focused on collecting an adequate number of samples to determine the most accurate mean, median, or central tendency of a field. Block, cell, or area sampling (Wollenhaupt et al., 1994) in a grid style manner would have a similar goal, but at a scale relative to the grid size. In contrast, grid-point sampling is most useful when the goal is to produce a more continuous surface of soil test results using interpolation techniques rather than blocks or cells representing the central tendency of an area, but may not adequately describe spatial patterns of highly variable alluvial soils. Also, grid point does not satisfy random requirements as the first selected point can be considered random, but subsequent points are chosen based on preset distances. Random sampling across predefined blocks or cells could help to overcome these shortcomings. The objective of this study was to compare detection of spatial structure in alluvial soil properties between grid-point and grid-cell soil sampling methodologies.
|
MATERIALS AND METHODS
|
|---|
This research was conducted in a center-pivot irrigated cotton field in Bolivar County, Mississippi. The site is located in the Yazoo Basin, which is a physiographic subdivision of the Lower Mississippi Valley floodplain. The most extensive landforms of this region are the ridge and swale (meander scroll) characterized by contrasting textures of sandy ridges and clay-filled swales (Smith, 1996). The following soil units were previously mapped at the site: Commerce (fine-silty, mixed, superactive, nonacid, thermic Fluvaquentic Endoaquepts), Robinsonville (Coarse-loamy, mixed, superactive, nonacid, thermic Typic Udifluvents), Dundee (Fine-silty, mixed, active, thermic Typic Endoaqualfs), and Souva (inactive). Two south-north-oriented parallel transects 1097 m long and 275 m apart were sampled. Soil samples were collected from each transect based on grid-point and grid-cell sampling designs (Fig. 1). Grid-point sampling consisted of compositing four cores taken one per row across four cotton rows (1.016 m spacing) for a total of six representative samples across 24 rows centered on a primary grid point. In addition to obtaining representative point samples, this design allowed for the determination of natural short-range variability as well as possible nonuniformity in historical fertilizer application. This sampling was repeated every 45.7 m along each transect (24 times) with an assumed width of 92 m and an area of 0.42 ha, resulting in a total of 144 samples per transect. Grid-cell sampling was performed by taking a composite sample from three random cores in a 0.22-ha area. To increase the accuracy of the correlograms, the number of paired locations for grid-cell sampling was doubled by taking two sets of composite samples, one from each side of the primary grid point, resulting in a total of 48 sampling areas per transect. For purposes of calculating correlograms, a new centroid for grid-cell sampling was established for each half. A calculated overall average at each primary grid point for each sampling scheme would represent the same area.
View larger version (26K):
[in this window]
[in a new window]
|
Fig. 1. Representation of grid-point and grid-cell sampling designs displaying only two of 24 primary grid points sampled in each transect.
|
|
Soils cores from a depth of 0- to 0.15-m were taken using a 20-mm diam. probe and were air-dried and crushed to pass a 2.0-mm sieve and mixed thoroughly. Soil analysis included total C and N, extractable Ca, Mg, K, Na, Zn, and P, and pH. Subsamples for determination of soil total C and N were ground with a pestle and a mortar to pass a 250-µm (60-mesh) sieve and then oven-dried (105°C). Samples (40–60 mg) were weighed and analyzed using an automated dry combustion analyzer Carlo Erba N/C 1500 (Carlo Erba, Milan, Italy). Extractable cations were determined using the Mississippi Soil Test extraction procedure described by Rasberry and Lancaster (1977). Inductively coupled plasma (ICP 3000, Leco Corp., St. Joseph, MI) was used to determine extracted Ca, Mg, K, Na, and Zn. Sum of exchangeable bases (SEB) was calculated by summing cmolc kg-1 of Ca, Mg, K, and Na. Phosphorus was extracted using the Mississippi Soil Test method, and determined colorimetrically using an ascorbic acid reduction method as described by Murphy and Riley (1962) on a Milton Roy 501 Spectronic Spectrophotometer (Marietta, GA) at a wavelength of 880 nm. The pH of a 1:2 (soil/water) suspension was measured using a Fisher Scientific Model 25 Accumet pH meter (Fisher Scientific, Denver, CO.).
Spatial correlograms, which depict the autocorrelation coefficient as a function of distance were calculated based on measured distances between sample points, or between area centroids when the variable represented a block or an area (Cliff and Ord, 1981). Spatial autocorrelation of sampling variables was measured by Moran's I autocorrelation coefficient (Moran, 1948), defined as:
where MC = autocorrelation for an interval distance class, n = number of localities at an interval class, A = sum of the weights wij, wij = the weight for the relationship between observations i and j, zi = xi - mean(x) is the centered variable obtained from xi, which is the measured sample value at point or centroid i, and zj = xj - mean(x) is the centered variable obtained from xj, which is the measured sample value at point or centroid i + h.
Omnidirectional correlograms were calculated using sampling pairs in both east-west and north-south directions for both sampling schemes. Given the short distance sampled in the east-west direction, it was not possible to assess anisotropy in spatial correlation. Individual autocorrelation coefficients were tested for significance under the randomization hypothesis (Cliff and Ord, 1981). Significance of entire correlograms was evaluated using the Bonferroni technique (Oden, 1984). With the presence of a cyclic spatial structure, comparisons between correlograms were further evaluated by computing average Manhattan distances between pairs across distance classes and then clustering of distances using the Unweighted Pair-Group Method with Arithmetic Average (UPGMA) (Sneath and Sokal, 1973). Statistical analysis was performed using the Statistical Analysis System version 6.12 (SAS Institute, 1996) and S-Plus version 4.5 (MathSoft, 1998) software.
|
RESULTS AND DISCUSSION
|
|---|
Descriptive statistics for grid-point and grid-cell sampling designs for both of the transects are shown in Table 1. Skewness and Kurtosis (Fisher g-statistics) reflect the degree of nonnormality by the amount of departure of their values from zero. Normal probability plots (not shown) for each variable approximated a straight line suggesting a near normal population distribution. Discrepancy between mean and median were most evident for extractable Ca and Mg in Transect I only. Log transformations did not improve the normality of the distributions, therefore analyses were performed with the original untransformed data. No significant differences were found among variable means between sampling schemes of each transect (t test). Variable ranges for grid cell were less than for grid point except for K in the west transect and P and Zn in the east transect. The smaller ranges for grid cell might be because of the homogenizing effect of compositing random samples across a greater area within each sampling region.
View this table:
[in this window]
[in a new window]
|
Table 1. Statistical summary of soil chemical properties for grid-point and grid-cell sampling in Transect I and II.
|
|
Transect I and II correlograms calculated from grid point at lag distances shorter than 24 m (Fig. 2 and 3), showed positive spatial correlation for all chemical elements. Zinc had the lowest MC values. Soil pH was the only variable, which displayed short-range positive spatial correlation and negative MC at lag distances <16 m in both transects. Gaston et al. (2001) concluded variability in pH was primarily short-range in nature, but their minimum sampling distance was 60 m. Both Campbell (1978) and Goovaerts (1998) have reported spatial dependence for soil pH of 10 m or less. A cyclic spatial pattern was not detected in either transect across rows suggesting there was not a gradient caused by fertilizer and amendment application on soil chemical properties within a localized area around a grid point. Cyclic patterns in soil test properties across rows have been shown to exist in other studies and suggest fertilizer-spreading patterns can cause this effect (Mallarino, 1996). This field historically has received broadcast applications of P, K, and lime. Thus, the across row variability appears to be primarily short distance natural variation.
View larger version (38K):
[in this window]
[in a new window]
|
Fig. 2. Correlograms of soil chemical variables for grid cell and composite correlograms for grid point sampling in Transect I.
|
|
View larger version (43K):
[in this window]
[in a new window]
|
Fig. 3. Correlograms of soil chemical variables for grid cell and composite correlograms for grid point sampling in Transect II.
|
|
Transect I correlograms (Fig. 2) for all chemical properties, except Na and pH, were characterized by positive MC for separation distances generally <300 m and negative MC at >300 m. Sodium correlograms showed a cyclic structure, while pH correlograms displayed short range variability at <16 m. Correlograms calculated from grid-point and grid-cell sampling were similar. The shape of the correlograms for total C and N, and extractable P, Ca, Mg, K, and Zn indicated the existence of a spatial gradient. A greater concentration of these elements was observed in the southern portion of the transect than in the northern area. Correlograms for SEB (data not shown) confirmed the existence of a spatial gradient and reflected variations in soil texture. The presence of this trend is likely related to the transversing across soil mapping units, especially the Commerce in the southern region and Robinsonville in the northern region of the transect. The Robinsonville soil is characterized by a coarser texture than the Commerce. Correlograms computed from grid-point and grid-cell sampling for lag distances >45.7 m were highly significant (p < 0.01), except for pH with grid-cell sampling. Because of the strong spatial gradient and great similarity between a majority of measured properties in transect I, no further analysis was performed.
In contrast, the spatial gradient in Transect II for most soil properties was primarily short-range in nature (
46 m) except Na, which extended to near 300 m. The MC for P and Zn decreased sharply for lag distances shorter than 45.7 m, however their values were greater than the MC found at a lag of 45.7 m. Correlograms for C, N, Ca, Mg, K, and SEB (data not shown) in this transect were characterized by alternating positive and negative values. Alternating positive and negative values were in correspondence with areas of relatively high and low elemental concentrations along the transect. The presence of this alternating pattern could be related to the crossing of several different soil map units, such as Robinsonville, Dundee, Commerce, and Souva. Relative to total C, our results agree with those of Gaston et al. (2001) for Transect II, which primarily displayed short-range variability, but not Transect I, which displayed longer-range variability.
Correlograms computed from grid-point sampling for lag distances >45.7 m were highly significant (p < 0.01) for all variables measured except Zn, which was significant (p < 0.05) and pH, which was not significant. Correlograms computed from grid-cell sampling were highly significant (p < 0.01) for Ca, Mg, K, and Na, significant (p < 0.05) for P and not significant (p > 0.05) for pH, C, N, and Zn. The difference in significance between the two sampling schemes in Transect II might be due to the combination of small MC and a lower number of pairs in the grid-cell sampling scheme.
Despite these differences, correlograms for grid-point and grid-cell sampling revealed similar patterns and were further analyzed using UPGMA. Figure 4 shows the dendrogram with the most similar pairs of correlograms based on a dissimilarity matrix of the correlograms' average Manhattan distances. The primary connections grouped in pairs the grid-point and grid-cell schemes for Na, P, pH, K, and Zn. The pairs confirm the similarity between the grid-point and grid-cell correlograms for these elements. Pairs between C and N and between Ca and Mg were formed regardless of the sampling methodology. This grouping might be due to the slight shift between the grid-point and grid-cell correlograms for these variables, and the inherent strong correlation between C and N (r = 0.91) and Ca and Mg (r = 0.96). Cross-correlograms (data not shown) were calculated to determine the extent to which data series exhibit concordant periodic variations. Cross-correlograms between C-N, K-C, K-N, and Ca-Mg were symmetric for positive and negative lags. Cross-correlograms between Ca-N, Ca-C, Ca-K, Mg-N, Mg-C, and Mg-K were not symmetric, confirming a similar cyclic pattern for C, N, and K, but different from the cyclic pattern found for Ca and Mg. Although the alternating pattern could be related to differences in soil properties between soil map units, the relationship between the cycles and the elements suggest that soil-mapping units alone would not adequately describe spatial variability in soil properties. Soil mapping units are not strictly homogenous and variability in soil properties exists within them (Edmonds and Lentner, 1986; Karlen et al., 1990; Wilding et al., 1965).
View larger version (24K):
[in this window]
[in a new window]
|
Fig. 4. Dendrogram of Transect II correlograms based on the dissimilarity matrix of grid point (GP) and grid cell (GC) correlogram's average Manhattan distances.
|
|
|
CONCLUSIONS
|
|---|
The similar shape of correlograms calculated from grid-point and grid-cell sampling suggests spatial pattern detection of soil properties on alluvial soils could be achieved by either sampling design with a grid density of approximately 46 m or less. The presence of long-range variability in Transect I is likely due to the fact that sampling transversed across only two different soil mapping units. The shorter spatial dependence and the presence of cyclic correlograms of Transect II, could be related to the crossing of several soil-mapping units. Differences in elemental cyclic patterns suggest that soil-mapping units could be indicative of differences in soil properties and could be used to modify grid-sampling designs to improve sampling efficiency. However, spatial patterns detected suggest soil mapping units alone are not sufficient to adequately describe spatial variability in soil properties. Grid-point sampling can effectively detect spatial structure of soils of the Mississippi Delta at the level of resolution used in this study. Grid-scale or sampling density requirements appear to vary by field and soil test parameter (Franzen and Peck, 1995; Wollenhaupt et al., 1994). For example, soil pH in this study varied primarily across short distances (<16 m) and would necessitate compositing cores separated by this distance or less. Systematic grid-point sampling is more easily facilitated as it requires less labor than compositing cores across a grid cell and a reduction in sample points (i.e., larger grid size) is possible in areas with low variability in soil properties or long-range autocorrelation.
|
NOTES
|
|---|
Contribution of the Mississippi Agric. and Forestry Exp. Stn., Journal paper no. J10292.
Received for publication December 14, 2001.
|
REFERENCES
|
|---|
- Anderson-Cook, C.M., M.M. Alley, R. Noble, and R. Khosla. 1999. Phosphorus and potassium fertilizer recommendation variability for two mid-Atlantic coastal plain fields. Soil Sci. Soc. Am. J. 63:1740–1747.[Abstract/Free Full Text]
- Cahn, M.D., J.W. Hummel, and B.H. Brouer. 1994. Spatial analysis of soil fertility for site-specific crop management. Soil Sci. Soc. Am. J. 58:1240–1248.[Abstract/Free Full Text]
- Campbell, J.B. 1978. Spatial variation of sand content and pH within single contiguous delineations of two soil mapping units. Soil Sci. Soc. Am. J. 42:460–464.[Abstract/Free Full Text]
- Campbell, J.B. 1979. Spatial variability of soils. Ann. Assoc. Am. Geogr. 69:544–556.
- Carr, P.M., G.T. Carlson, J.S. Jacobsen, G.A. Nielsen, and E.O. Skogley. 1991. Farming soils, not fields: A strategy for increasing fertilizer profitability. J. Prod. Agric. 4:57–61.
- Cliff, A.S., and J.K. Ord. 1981. Spatial processes. Pion, London.
- Edmonds, W.J., and M. Lentner. 1986. Statistical evaluation of the taxonomic composition of three soil map units in Virginia. Soil Sci. Soc. Am. J. 50:997–1001.[Abstract/Free Full Text]
- Franzen, D.W., and T.R. Peck. 1995. Field soil sampling density for variable rate fertilization. J. Prod. Agric. 8:568–574.
- Franzen, D.W., L.J. Cihacek, V.L. Hofman, and L.J. Swenson. 1998. Topography-based sampling compared with grid sampling in the northern great plains. J. Prod. Agric. 11:364–370.
- Gaston, L.A., M.A. Locke, R.M. Zablotowicz, and K.N. Reddy. 2001. Spatial variability of soil properties and weed populations in the Mississippi Delta. Soil Sci. Soc. Am. J. 65:449–459.[Abstract/Free Full Text]
- Goovaerts, P. 1998. Geostatistical tools for characterizing the spatial variability of microbiological and physio-chemical soil properties. Biol. Fertil. Soils 27:315–334.
- Karlen, D.L., E.J. Sadler, and W.J. Busscher. 1990. Crop yield variation associated with Coastal Plain soil map units. Soil Sci. Soc. Am. J. 54:859–865.[Abstract/Free Full Text]
- Mallarino, A.P. 1996. Spatial variability patterns of phosphorus and potassium in no-tilled soils for two sampling scales. Soil Sci. Soc. Am. J. 60:1473–1481.[Abstract/Free Full Text]
- MathSoft. 1998. S-Plus professional version 4.5. Mathsoft, Inc., Seattle, WA.
- Moran, P. 1948. The interpretation on statistical maps. J. R. Stat. Soc. B 10:243–251.
- Murphy, J., and J.P. Riley. 1962. A modified single solution method for determination of phosphate in natural waters. Anal. Chim. Acta 27:31–36.
- Oden. 1984. Assessing the significance of a spatial correlogram. Geogr. Anal. 16:1–16.
- Peck, T.R., and S.W. Melsted. 1973. Field sampling for soil testing. p. 67–75. In L.M. Walsh and J.D. Beaton (ed.) Soil testing and plant analysis. Rev. ed. ASA, CSSA, and SSSA, Madison, WI.
- Rasberry, F.P., and J.D. Lancaster. 1977. A comparative evaluation of the Mississippi soil test method for determining available manganese, magnesium and calcium. Commun. Soil Sci. Plant Anal. 8:327–339.
- SAS Institute. 1996. SAS systems for information delivery for Windows. Release 6.12 SAS Inst., Cary, NC.
- Smith, L.M. 1996. Fluvial geomorphic features of the Lower Mississippi Alluvial Valley. Eng. Geol. 45:139–165.
- Sneath, P.H.A., and R.R. Sokal. 1973. Numerical taxonomy. W.H. Freeman, San Francisco, CA.
- Thompson, W.H., and P.C. Robert. 1995. Evaluation of mapping strategies for variable rate applications. p. 304–323 In P.C. Robert (ed.) Site-specific management for agricultural systems. Proc. 2nd Int. Conf., Minneapolis, MN. 27–30 Mar. 1995. ASA, CSSA, and SSSA, Madison, WI.
- Webster, R., and M.A. Oliver. 1990. Statistical methods in soil and land resource survey. Oxford Univ. Press, New York.
- Wibawa, W.D., D.L. Dludlu, L.J. Swenson, D.G. Hopkings, and W.C. Dahnke. 1993. Variable fertilizer application based on yield goal, soil fertility, and soil map unit. J. Prod. Agric. 6:255–261.
- Wilding, L.P., R.B. Jones, and G.M. Schafer. 1965. Variation of soil morphological properties within Miami, Celina and Crosby mapping units in west-central Ohio. Soil Sci. Soc. Am. Proc. 29:711–714.
- Wollenhaupt, N.C., R.P. Wolkowski, and M.K. Clayton. 1994. Mapping soil test phosphorus and potassium for variable-rate fertilizer application. J. Prod. Agric. 7:441–448.
TOP
ABSTRACT
INTRODUCTION
