| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a USDA-ARS, 119 Keim Hall, Lincoln, NE 68583-0934
b Univ. of Nebraska, 103 Miller Hall, Lincoln, NE 68583
c Colorado State Univ., C130 Plant Sci., Ft. Collins, CO 80523
d Usda-Ars, Aerc-Csu, Ft. Collins, Co 80523-1325
* Corresponding author (cjohnso2{at}bigred.unl.edu)
Received for publication February 14, 2002.
 |
ABSTRACT |
|---|
 TOPNOTES ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Abbreviations: EC, electrical conductivity • ECa, apparent electrical conductivity • FICS, Farm-Scale Intensive Cropping Study • MS, mean square • OM, organic matter • SDAMP, Sustainable Dryland Agroecosystem Management Project
|
INTRODUCTION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Some types of research, or research goals, require the precision achievable only through controlled plot-scale experimentation; other research goals are well suited to on-farm investigations. Experiments best conducted on farm include those (i) requiring specific soil types or environmental conditions not found on an experiment station, (ii) involving the study of farm management, (iii) analyzing integrated systems such as crop and livestock production, (iv) evaluating performance of a management system under real farm conditions, (v) examining occurrences requiring large land areas (e.g., runoff, erosion, and pest infestation), (vi) studying the long-term effects of specific management practices, and (vii) evaluating farmer innovations (Lockeretz, 1987). Many of these examples require field-scale analyses. New technologies used in site-specific management and other sustainable management practices, including global positioning systems, geographic information systems, and field-scale sensors, are also best evaluated at the field scale (Vanden Heuvel, 1996).
Field-scale research addresses issues of operational scale and soil variability to produce outcomes different from those of experiment station–based research (Table 1). It promotes broad-based investigations that address not only technical, but also economic and social factors; increases farmer involvement, interest, acceptance, and adoption of successful outcomes; and facilitates a systems perspective, wherein multiple components are evaluated. It has been suggested that farmer-vested field-scale research can reverse research direction and emphasis (Sumberg and Okali, 1988). Instead of functioning merely as a means to validate experiment station findings, these experiments allow us to begin with the system. Research questions that originate from system outcomes can then be investigated using the more sensitive experiment station plot-scale approach.
|
Precedent exists for unreplicated experiments, particularly within the specific research disciplines of engineering, plant breeding, and landscape ecology. Examples of nonreplicated experiments take several forms. Multiple locations of identical treatments (Moreau et al., 1999; Johnson et al., 1992) are commonly used as replicates. Time-series experimental designs compare changes in treatment units with those of a reference unit over time (Hawkins, 1986; Stewart-Oaten et al., 1986). Before-and-after comparisons are used in environmental impact studies (Wiens and Parker, 1995). Multiple independent experimental results are sometimes combined to simulate replication (Hannah, 1999). Other researchers have used preliminary or separate tests to derive an estimate of experimental error that is applied in subsequent experiments, making experimental error derived from replication unnecessary (Sahagün-Castellanos and Frey, 1994; Box et al., 1978, p. 374–418). Beyers (1998) suggested the use of causal inference supported by simple descriptive statistics, including tables, graphs, estimates of means and standard errors, regression, and multivariate analyses to evaluate experimental results.
Although a plethora of analyses and design approaches exist for nonreplicated experiments, clear and widely accepted solutions to this problem are lacking (Stewart-Oaten et al., 1992). This is because the omission of replication in experimental design can have serious repercussions. Conclusions stemming from nonreplicated experiments may be transferable to only a small population of experimental units, sometimes to only the original experimental area. Assumed hypotheses may not be those actually tested, the degree of precision may be overestimated, perceived treatment differences may merely reflect variation among experimental units rather than treatments, and the effects of treatments and experimental units may be confounded. Statistical designs not incorporating replication must address these issues.
Increasingly, agronomic investigators are exploring the use of computer and satellite technologies applied as field-scale tools, including georeferenced crop yield monitors, remotely sensed data, and ECa sensors. This technology is appropriate to a broad-based and large-scale approach to agricultural experimentation that focuses on spatial patterns across a field. As a result, much current agronomic research is directed toward understanding temporal and spatial interrelationships among physical, chemical, and biological soil properties and their combined contributions to crop productivity at the field scale.
Soil clay type and percentage, moisture (in conjunction with pore size, tortuosity, and water-filled pore space as they vary with depth), salinity of the soil solution, and temperature can affect ECa measurement (Rhoades et al., 1989; McNeill, 1980). One or more of these factors will dominate ECa in specific soils. Significant correlations have been documented between ECa and soil properties affecting its measurement, including soil moisture (Khakural et al., 1998; Sheets and Hendrickx, 1995), salinity (Lesch et al., 1992; Rhoades and Corwin, 1981), and depth to claypan (Sudduth et al., 1995).
Previous experiments at a semiarid experimental site, the Farm-Scale Intensive Cropping Study (FICS) (Johnson et al., 2001), revealed that soil properties (0- to 7.5- and/or 0- to 30-cm depths) associated with erosion [percentage clay, bulk density, pH, and laboratory-measured electrical conductivity (EC)] were positively correlated with ECa (approximately 0–30 cm depth of measurement) while soil properties indicative of crop productivity [soil moisture, total and particulate organic matter (OM), total C and N, extractable P, microbial biomass C and N, and potentially mineralizable N] were negatively correlated with ECa (Table 2). Some of these soil properties are directly measured by ECa while others are correlated with them. Variable levels of individual soil parameters can be associated with a single ECa value because of the buffering effect of corresponding variations in opposing soil parameters affecting ECa. For this reason, at the FICS site, ECa appears to be most useful as a tool for integrating the multiple soil physical, chemical, and biological properties that underlie production potential.
|
0.06). The effectiveness of ECa for delineating yield potential at the FICS was corroborated by strong relationships (r = -0.97 to -0.99) between mean ECa (0–30 cm depth of measurement) within ECa class and mean yields within ECa class for winter wheat in both typical- and high-yielding years (Johnson et al., 2003). Similar correlations with ECa were found for both winter wheat and corn yields when ECa was measured at deeper depths (0–90 cm). In published reports, the relationship between ECa and yield is often significant within crop treatments and fields but inconsistent across years (Jaynes et al., 1993; Sudduth et al., 1995; Kitchen et al., 1999). These studies have been conducted in humid, high-precipitation regions where yields are limited by both insufficient and excessive precipitation. In these regions, variable precipitation inputs can alter or even reverse the relationship between the soil properties driving ECa and crop yields. In semiarid cropping systems such as the FICS, where yields typically reflect only the degree of drought stress, ECa may more consistently predict yield.
It is important to note that while the magnitude of measured ECa fluctuates over time, spatial patterns or zones of ECa remain constant (Lund et al., 1999; Sudduth et al., 2001). The ability to map patterns of productivity across a landscape makes possible novel research opportunities. Zones of within-field variability based on ranges of ECa may be applicable to the statistical evaluation of treatment effects in field-scale experiments. It may be possible to estimate experimental error based on within-field observations instead of replication through random sampling across a field or by using new technologies, such as ECa classification. In soils where ECa–classified zones explain a significant amount of the variability in production potential within a field, zone-based sampling schemes may reduce error variance compared with samples taken at random. It is also possible that the variances of field measurements taken within ECa zones may provide an estimate of small-plot experimental error. This assumes a traditional plot-scale randomized complete block design established within the experimental site where the ECa zones function as blocks.
The dilemma presented by the lack of feasible replication and blocking in field-scale research is the focus of this paper. We examined the relationships among field-scale within-field variability, field-scale replication, and plot-scale blocking and the implications of these relationships for statistically evaluating field-scale experiments. Our primary objective was to determine whether field-scale experimental error can be estimated using within-field variability in soil condition; if this is feasible, traditional replication may be unnecessary. A secondary objective was to evaluate ECa classification as a basis for estimating plot-scale experimental error. This may permit the screening of treatments and treatment effects at the field scale for further investigation in plot-scale work.
|
MATERIALS AND METHODS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
The FICS site is managed as eight approximately 31-ha fields to include two replicates of each phase of the 4-yr rotation each year (Fig. 1) . It is gently sloping (0–5%) and comprised of a mixture of soils, including Platner (fine, smectitic, mesic Aridic Paleustolls), Weld (fine, smectitic, mesic Aridic Argiustolls), and Rago loam (fine, smectitic, mesic Pachic Argiustolls). Regional climate is cool and semiarid with a mean annual temperature of 10°C and mean annual precipitation of 420 mm. Precipitation is highly variable, with 75% falling between April and September, and highest amounts in May, June, and July.
|
A stratified soil-sampling strategy was developed wherein strata were allocated into four classes based on ranges in ECa. Using ERDAS Imagine (ERDAS, Atlanta, GA), ECa maps from each of the eight fields in the study site were individually interpolated by inverse-distance weighting. Next, ECa data in each interpolated map was spatially clustered using unsupervised classification (ERDAS, 1997) to form 12 classes of ECa (10-m2 grid-cell resolution), which were then recoded (combined) into four classes. Recoding was done by adjusting within-class ECa ranges to mimic the dominant visible spatial patterns observed in the original 12-class gray-scale ECa maps (Fig. 2) . This classification procedure aggregates ECa data points into naturally occurring clusters to minimize within-class variance.
|
Soil Sampling and Analysis
The experimental site was sampled in two phases based on crop status. Wheat and fallow fields were sampled in mid-August 1999 following wheat harvest while millet and corn fields were sampled in mid-November 1999 following corn harvest. At each of the 96 sampling points, seven 4-cm-diam. soil cores were taken at 0- to 7.5- and 7.5- to 30-cm depths, composited by depth and mixed well. Surface soils (0–7.5 cm depth) were sieved to pass a 2-mm screen. At this point, a portion of the soil was stored at 4°C while the remainder was air-dried. Due to their higher water content, deeper soils (7.5–30 cm) were sieved to 4 mm. Once again, a portion of the soil was stored at 4°C. The remainder was air-dried and ground through a soil grinder (M.G. Johnston Industries, Lakeville, MN)1 to pass a 2-mm sieve. This type of grinder crushes soils to leave residues intact for particulate OM analyses.
Soil was assessed using physical, chemical, and biological parameters as proposed by Doran and Parkin (1996). Soil physical parameters included bulk density (Blake and Hartge, 1986), texture (Kettler et al., 2001), and gravimetric water content. Chemical measurements consisted of total and particulate OM (0.053- to 0.5- and 0.5- to 2-mm size fractions) by loss on ignition (Cambardella et al., 2000); pH and laboratory-measured EC, using a 1:1 water/soil mixture; 2 M KCl-extracted NO3–N and NH4–N, measured on a LACHAT FIA autoanalyzer (Zellweger Analytics, LACHAT Instrument Div., Milwaukee, WI); total C and N, analyzed with a Carlo Erba NA 100 (CE Elantech, Lakewood, NJ); and extractable P, by the Bray-1 method (Bray and Kurtz, 1945). Microbial biomass C and N, by microwave irradiation (Islam et al., 1998), and anaerobically incubated potentially mineralizable N (Waring and Bremmer, 1964; Keeney, 1982) analyses were conducted to assess soil biological function. All testing was performed on air-dried soil with the exception of microbial biomass C and N, pH, laboratory-measured EC, and anaerobic potentially mineralizable N, which were assayed using fresh soil within 2 wk of collection.
Statistical Analyses
The data in this study were analyzed as a complete block with two blocks (or replicates) and four rotational phases that functioned as treatments within each block. When the study was initiated in 1999, treatments were assigned to each of the eight fields to maintain continuity between historical and newly imposed treatments (Fig. 1). Electrical conductivity class was used as an additional blocking variable (Table 3). All data were analyzed on a volumetric basis with the exception of KCl-extracted NO3–N and NH4–N (µg g-1 soil) and water content (g g-1 soil). Although soil samples were collected and analyzed using 0- to 7.5- and 7.5- to 30-cm soil depths, statistical comparisons were made on 0- to 7.5- and 0- to 30-cm depth increments. Data from the two analyzed depths were combined and weighted to calculate 0- to 30-cm values that best corresponded to the depth of ECa measurement (0–30 cm).
|
If nonadditivity was significant for an individual soil property, a second step was taken in the analysis. Nonadditivity effects in the rep x crop term were removed, and the residual MS was applied for an improved estimate of experimental error (Tukey, 1949; Lentner and Bishop, 1993). We then tested to determine if the residual MS was significantly larger than the MS (within field). The MS (within field) was considered to be a reasonable estimate of experimental error for a given soil parameter if this F test was not significant.
Small-plot experiments are typically conducted as randomized complete block designs where plots are grouped into blocks based on soil properties. Assuming ECa classifications are reasonable surrogates for blocks, a small-plot randomized complete block design might be set out as in Fig. 3C . In this case, experimental error for the small-plot experiment could be estimated by the variability among sampling points in each ECa classification as obtained by the MS error (64 df) in Table 3.
|
Soil and residue data collected from the SDAMP site were used for comparison with those of the FICS. Soil data included total C and N concentration, analyzed on a Leco analyzer (Leco Corp., St. Joseph, MI)1; pH, using a 1:2 water/soil mixture; P, analyzed by the NaHCO3 method (Olsen and Dean, 1965); and bulk density. With the exception of bulk density, all soil data from this site were analyzed as a complete block design with two blocks (or replicates), 10 rotational phases (treatments) within each block, three slope gradient classes, and 8 to 12 yr (determined by available data) as a time variable. Residue comparisons were made in the same manner, except that only data from the wheat–fallow treatment were used. This treatment most closely resembled FICS residue measurements taken the first year following conversion from wheat–fallow management. Bulk density measurements in the SDAMP were collected from randomly selected points within the three slope aspects at a different time than other soil tests. For each soil parameter, equality of MS errors from the FICS and SDAMP were tested at the P = 0.05 level of significance. All statistical analyses were conducted using SAS (SAS Inst., Cary, NC).
|
RESULTS AND DISCUSSION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
|
In situations where the elimination of replication is proven appropriate, an additional treatment(s) could be added to the experiment. For example, the FICS site (Fig. 1) could be split from north to south into two treatment areas. Different tillage regimes could then be assigned to the east and west sides of the study site and compared over time for their impact on soil parameters and yield. Data could be analyzed as a complete block, with ECa classes functioning as replicates. Given this scenario, it is important to note that tillage treatments would be confounded with fields, so conclusions regarding treatment differences would be based on the assumption that each field is representative of the population of fields of interest.
Estimating Plot-Scale Experimental Error from Field-Scale Experiments
Within-field ECa classification of soil condition presents interesting possibilities for agronomic field designs. Soil condition has been defined as the combination of soil characteristics that establish the level of soil function as a medium for crop production and a contributor to air and water quality (Johnson et al., 2001). For the FICS site, classification based on ECa delineates distinct zones of soil condition that are related to yield variability within a uniformly managed field (Johnson et al., 2001, 2003). Therefore, ECa classification can be used as a basis for blocking to control experimental-error variance where classes function as experimental blocks. Blocking by ECa class is appropriate because classes are related to outcome (yield) differences expected in the absence of treatments (the rationale for blocking).
The disparity in scale between a typical plot-scale experiment and the section of farmland (250 ha) comprising the FICS site is illustrated in Fig. 3. The image on the left (A) is an aerial photograph of the site with an example of a plot-scale experiment shown as a black square near the center of the southeast field. An ECa map of this same field, classified into four conductivity ranges, is shown on the lower right (B). The selected plot-scale site encompasses three of the four conductivity classes, likely providing an excellent basis for blocking. Although this is shown as a traditional layout (C), because the blocks (ECa classes) are homogeneous, there is no reason that they must be adjacent to one another. Plots could be scattered throughout the field, randomly applied to all four ECa classes. It is now a small step to conceptualize the entire experimental site as an enlarged rendition of the pictured plot-scale experiment where variability within ECa class represents the experimental error of the plot-scale experiment.
The presence of the plot-scale SDAMP, within close proximity and comprised of the same soil types found in the FICS, provided an opportunity for testing these relationships between plot-scale experimental error and the variance of field-scale ECa classified within-field variability. Comparisons between soil and residue analyses made from the two sites are shown in Table 6. For surface soil (0–7.5 cm depth), the MS within ECa class errors for surface soil did not differ from MS error for the plot-scale experiment for any measured parameters except bulk density (P 
0.05). At the 0- to 30-cm depth, only MS for pH was the same for plot-scale error and within ECa class field-scale analyses. However, while plot- and field-scale MS's for C and N and extractable P were significantly different, they showed only three- to fourfold differences. This degree of heterogeneity is not excessive because it has little effect on ANOVA (Scheffe, 1959).
|
Although preplant residue MS did not differ (P 0.05) between the SDAMP and FICS sites, the magnitude of residue levels was greater for SDAMP than FICS (Table 6). Data from SDAMP were collected from wheat–fallow treatments only (the best basis for comparison) over a period of 12 yr. Because this treatment was managed using no-till, these data reflected residue accumulation not found during the first year of the FICS experiment. Postharvest residue measurements had threefold greater variance at the plot scale than at the farm scale. This may also reflect differences in experiment age at the time of residue collection. Residue levels at the SDAMP site resulted from multiple-year accumulations that have been exposed to varying rates of decomposition and wind and water erosion, factors increasing variability in surface residue cover and biomass production. It is difficult to make clear-cut comparisons of measurements from two different experimental sites. Yet, even though MS differences were likely falsely elevated due to different sampling times relative to the age of each experiment, MS compared well among the two studies (two levels of scale).
|
CONCLUSIONS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Classified maps based on ECa can sometimes be used to delineate soil spatial heterogeneity at large experimental scales. At the FICS site, ECa–classified within-field variability was similar to MS error at the nearby SDAMP plot-scale experiment, indicating that field-scale MS (within field) variability is synonymous with soil heterogeneity partitioned by blocking at the plot-scale. Within-field ECa classes do not constitute traditional blocks, defined as sets of homogeneous experimental units. In our experimental design, each field to which a treatment (crop) was applied represents one experimental unit. Instead of comprising a set of experimental units unto themselves, the four ECa classes assigned to each treatment fall within one experimental unit, i.e. within-field blocking.
The relationship between experimental error derived from ECa–classified zones and plot-scale blocking supports an alternative in experimental design. Hargrove and Pickering (1992) suggested using nonreplicated large-scale experiments to develop hypotheses; these hypotheses could then be applied to smaller-scale experiments from which large-scale processes could be inferred. Within-field variability, delineated by ECa classification, may facilitate such an approach. We propose using ECa–classified MS (within field) in field-scale experiments to estimate plot-scale experimental error. Treatment differences and their standard errors could then be used in a systems approach to roughly evaluate treatments and identify research questions requiring further study in plot-scale experiments.
Research conducted on farm at the field scale can improve the accountability of agronomic research to the producers it serves. Incorporating unclassified and ECa–classified within-field variability in the design of field-scale experiments may advance on-farm research by offering alternative methods for statistical analyses.
|
ACKNOWLEDGMENTS |
|---|
|
NOTES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
1 Mention of a trademark, proprietary product, or vendor does not constitute an guarantee of or warranty of the product by USDA nor imply its approval to the exclusion of other products that may be suitable. 
|
REFERENCES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  M. S. Grigera, R. A. Drijber, K. M. Eskridge, and B. J. Wienhold Soil Microbial Biomass Relationships with Organic Matter Fractions in a Nebraska Corn Field Mapped using Apparent Electrical Conductivity Soil Sci. Soc. Am. J., August 3, 2006; 70(5): 1480 - 1488. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
W. K. Jung, N. R. Kitchen, K. A. Sudduth, and S. H. Anderson Spatial Characteristics of Claypan Soil Properties in an Agricultural Field Soil Sci. Soc. Am. J., June 21, 2006; 70(4): 1387 - 1397. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| The SCI Journals | Crop Science | Vadose Zone Journal | |||
| Journal of Natural Resources and Life Sciences Education |
Soil Science Society of America Journal | ||||
| Journal of Plant Registrations | Journal of Environmental Quality |
The Plant Genome | |||
symbols represent selected soil-sampling sites. Ranges of apparent electrical conductivity for each of the four classes from low to high are 0 to 0.17, 0.17 to 0.23, 0.23 to 0.29, and 0.29 to 0.56 dS m-1.