| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a Agriculture and Agri-Food Canada, Lethbridge Research Centre, 5403 1st Ave. S. Lethbridge, AB, Canada T1J 4B1
b Alberta Agriculture and Food, 5401 1st Ave. S. Lethbridge, AB, Canada T1J 4V6
* Corresponding author (ZvomuyaF{at}agr.gc.ca).
 |
ABSTRACT |
|---|
 TOPNOTES ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Abbreviations: Cmin, mineralizable carbon • EC, electrical conductivity • LF-C, light fraction carbon • LF-N, light fraction nitrogen • Nmin, mineralizable nitrogen • PLS, partial least squares • PRESS, predictive residual sum of squares • SOM, soil organic matter • TIC, total inorganic carbon • TN, total nitrogen • TOC, total organic carbon • VIP, variable importance in the projection
|
INTRODUCTION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Soil quality criteria and corresponding standards vary with soil function, which complicates the use of a common definition of soil quality. Thus, soil quality is best defined in relation to the function ascribed to the soil and interpreted using soil quality indicators based on quantitative measures selected according to that function, e.g., sustaining biological productivity, maintaining environmental quality, and promoting plant and animal health (Doran and Parkin, 1994; Nortcliff, 2002; SSSA, 1997). A function particularly relevant to agricultural ecosystems is biological productivity, as measured by crop yield. The capacity of soil to sustain productivity is a function of intrinsic soil properties (soil quality) and extrinsic factors (landscape quality factors, e.g., precipitation, temperature, topography, and hydrology) (Janzen et al., 1992b). Monitoring productivity and long-term sustainability of agricultural ecosystems relies on selecting a suite of intrinsic soil properties or soil quality indicators (Larson and Pierce, 1994) that are measurable surrogates of physical, chemical, and biological soil attributes that determine how well a soil performs (Burger and Kelting, 1999). These indicators can be classified as either inherent or dynamic (Wienhold et al., 2004). Inherent soil quality indicators are those attributes related to a soil's natural composition and properties as influenced by the factors and processes of soil formation, while dynamic indicators relate to soil properties and processes that change on a human time scale as a result of land use and management decisions.
While the relationship between productivity and extrinsic factors is well established and considerable research effort has been directed toward determining the influence of management practices (e.g., tillage, crop rotation, organic amendments) on soil quality indices (Cambardella et al., 2004), less is known about the pattern of the relationship between soil quality and productivity. Herrick (2000) emphasized the need, inter alia, to demonstrate causal relationships between soil quality and ecosystem functions such as biomass production.
Soil quality–productivity relationships have traditionally been quantified by sampling a number of points in a landscape and correlating yield with selected soil properties (Rezaei et al., 2006; Wright et al., 1990). While this approach has generated useful information, establishing causality has been complicated by the confounding effects of extrinsic factors such as topography, climate, and hydrology. For example, higher soil moisture in depressional areas may enhance productivity, which, in the long term, results in higher organic matter accumulation compared with upland areas (Olson et al., 1996). In such a scenario, regression analysis may indicate a strong correlation between yield and soil organic matter (SOM) content, but it is not possible to definitively establish whether the correlation indicates an effect of SOM on yield or merely reflects the accumulation of SOM at sites that favor higher yield.
In an alternative approach (Olson et al., 1996), the relationship between soil quality and productivity can be quantified under conditions where extrinsic factors such as precipitation, temperature, topography, and hydrology are uniform. Variability in soil quality is artificially introduced by deposition of diverse soils on a common subsoil at the same location, thus eliminating the confounding effects of the extrinsic factors. Yield is then defined as a function of intrinsic soil properties. This approach focuses on surface soil, which, under conditions like those in southern Alberta, is most influential in dictating productivity and most subject to agronomic manipulation. Additionally, disturbed land reclamation schemes, such as those on abandoned oil and natural gas wellsites, focus on topsoil replacement as an essential measure to restore the soil's productivity (Larney et al., 2005; Zvomuya et al., 2006). The practical implications of this work have previously been highlighted (Olson et al., 1996).
The overall objective of this study was to describe the relationship between measurable properties and crop productivity. Specific objectives addressed in this study were to: (i) assess and compare temporal changes in productivity among diverse soils subjected to the same landscape quality factors and management regime; (ii) identify soil indicators linked to productivity of dryland spring wheat; and (iii) describe the relationship between these factors and wheat yields under the semiarid conditions of southern Alberta.
|
MATERIALS AND METHODS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Site Description
The 14-yr study was established in 1990 at the Agriculture and Agri-Food Canada Research Centre at Lethbridge, AB (49°42' N, 112°50' W), in a field that had level topography and uniform original soil characteristics. The soil at the site was an Orthic Dark Brown Chernozem (Typic Boroll) with a 20-cm Ap horizon, a thin B horizon (
10 cm), and a calcareous C horizon. The site had last been cropped in 1988.
Soil Collection
Thirty-six soils differing widely in chemical, physical, and biochemical characteristics (Table 1
) were selected from sites with different properties and management histories (Olson et al., 1996). All soils were Chernozems (Borolls) developed under grassland. The soils were collected mostly from the Ap or Ah horizon using a Bobcat skid-steer loader (Melroe Co., Fargo, ND) and transported to the experimental site by truck (typically one truckload per plot).
|
2 cm deep) with a disk harrow to roughen the interface between the subsoil and the subsequently transplanted soil. Following plot demarcation, soil samples were taken (0–15-, 15–45-, and 45–75-cm depths) from each plot for bulk density (mean 1.41, 1.52, and 1.51 Mg m–3, respectively, for the three layers) and particle size analysis (soil texture for the 0–75-cm depth was silty clay loam). Temporary wooden frames (6 by 5 m) were placed around each plot to avoid interplot soil mixing during deposition. Three replicates of the 36 transplant soils were then placed onto the 108 plots. The mean depth of the deposited soil layer, measured in 1992 following 2 yr of settling, was 19 cm (range 15–24 cm).
The treatment layout was a split plot, with whole plots arranged in a randomized complete block design with three replications. Soils (36) were the whole plots (6 by 5 m) and fertilizer N rate was the subplot (0 kg N ha–1, 6 by 3 m; 60 kg N ha–1, 6 by 2 m).
Agronomic Practices
The site was seeded (67 kg seed ha–1) to spring wheat (‘Leader’) on 24 Apr. 1991 following broadcast application (by hand) of N fertilizer (56–62 kg N ha–1 as NH4NO3) on designated subplots. All plots received P, applied with the seed, at rates sufficient to prevent deficiency (13–20 kg P ha–1 as triple superphosphate). Reseeding was performed using a single-row seeder in rows where germination was poor. The site was similarly fertilized and seeded in subsequent years. The wheat cultivar was ‘Biggar’ in 1992, ‘Lancer’ in 1993, and ‘Katepwa’ thereafter. Preseeding, in-crop, and post-harvest weed control followed local practices with recommended herbicides. Strict no-till practices were adopted to minimize mixing of soil among plots. Soil erosion control measures were applied during the first winter as described by Olson et al. (1996). Seedling establishment in all plots was poor in 1992 and 1997 due to early-season drought; the crops were, therefore, terminated with glyphosate [N-(phosphonomethyl)glycine] and the site chemically fallowed for the remainder of the growing season. Summer fallowing (leaving land unseeded for an entire growing season) is common in this semiarid region because it allows replenishment of soil water. Although the practice is now waning with the advent of no-till, in parts of western Canada the land is still left idle once every 2 to 3 yr.
The crop was harvested by hand each year from a representative section of each subplot at maturity. The harvested crop biomass samples were dried at 70°C and weighed for dry total aboveground biomass yield determination. The remaining crop in all plots was swathed, baled, and removed. Regular straw removal, while not widely recommended, is still a common practice in some western Canadian areas because the straw is used for livestock bedding (e.g., Campbell et al., 1991).
Soil Sampling
At the time of soil deposition, a large composite soil sample was taken from each of the 108 main plots and archived after air drying. Additional soil samples were collected in 1994 for determination of aggregate stability. In summer 1997, four soil cores (diameter 3.35 or 3.6 cm) were collected from each subplot and composited for measurement of changes in selected soil properties. The depth of sampling (10 cm) was less than the original depth of deposition (mean 19 cm) but represented the layer of soil subject to most rapid change. All samples were air dried and ground (<2 mm) before analysis. Subsamples were finely ground (<0.15 mm).
Laboratory Analyses
Total C and TN concentrations were determined by dry combustion with a CNS analyzer (Model 1500, Carlo Erba Instruments, Milan, Italy). Ammonium- and NO3–N were extracted with 2 mol L–1 KCl and determined colorimetrically using an AutoAnalyzer II SC Colorimeter (Technicon Industrial Systems, Tarrytown, NY). Bicarbonate-extractable (Olsen) P (Olsen et al., 1954) was measured colorimetrically (Murphy and Riley, 1962) with a Pye Unicam PU 8650 spectrophotometer (Philips Scientific, Cambridge, UK) at 860 nm. Total inorganic C (TIC) was measured by the method of Amundson et al. (1988). Total organic C was calculated as the difference between total C (TC) and TIC. Light fraction SOM was isolated from air-dried soil (<2 mm) using a density fractionation method adapted from (Strickland and Sollins,1987; Janzen et al.,1992a). Total C and TN in the LF SOM were determined by automated dry combustion as described above. Mineralizable C was estimated from the cumulative release of CO2 during a 10-wk moist incubation of 75 g of oven-dry soil. The CO2 was trapped in dilute NaOH and analyzed by infrared gas analysis after acidification. Traps were replaced and analyzed after 1, 2, 4, 6, and 10 wk. Mineralizable N was determined by measuring the increase in KCl-extractable N during 10 wk of incubation in the same sample used to determine Cmin. Additional chemical measurements were soil pH in 0.01 mol L–1 CaCl2 (1:2 w/v), electrical conductivity (EC) in saturation paste extracts (Rhoades, 1982a), and cation exchange capacity (CEC) by the 1 mol L–1 NH4OAc (pH 7.0) method (Rhoades, 1982b). All analyses were performed on coarse-ground samples (<2 mm), except for TC, TN, and TIC (<0.15 mm).
Soil physical measurements were particle size analysis by the pipette method (Gee and Bauder, 1986), water retention at –30 and –1500 kPa (Klute, 1986), bulk density (Culley, 1993), and aggregate stability by the wet-sieving method of Kemper and Rosenau (1986).
Statistical Analyses
Descriptive Statistics
Descriptive statistics for summarizing the soil quality parameters were estimated using the MEANS procedure in SAS (SAS Institute, 2004). Geometric mean and geometric CV were computed for all properties except pH.
Analysis of Variance
Treatment effects were evaluated on multiply imputed data using the GLM procedure because the size of the experiment (36 soils x 2 N treatments x 14 yr x 3 replications) made it impractical to estimate covariance parameters using the preferred PROC MIXED for repeated measures (Littell et al., 1996; SAS Institute, 2004). Multiple imputations for missing data (1.3% of total) were performed with the Markov chain Monte Carlo simulation in SAS using the MI procedure (SAS Institute, 2004). The procedure generated five data sets that were analyzed separately using PROC GLM with the multivariate (Wilk's 
statistic) approach to evaluate within-subject main and interaction effects. The Tukey–Kramer pairwise multiple comparison procedure was used with the LSMEANS statement to maintain an experimentwise significance level of 0.05 for all pairwise comparisons. Estimates and standard errors from the PROC GLM ANOVA of multiply imputed data sets were combined using PROC MIANALYZE (SAS Institute, 2004) to generate valid statistical inferences that properly reflected uncertainty due to the missing values. When the year x soil x fertilizer interaction was significant, the data were analyzed separately for each year using PROC MIXED. Effects were considered significant at P 0.05 unless otherwise stated.
Test for Homogeneity of Coefficients of Variation
The test for homogeneity of CVs (Zar, 1999) was performed for total aboveground yield across all years (with yield data for each year analyzed separately) to determine whether the soil productivity was becoming more similar (i.e., converging). Coefficients of variation used to estimate the test statistic, which approximates the 
2 distribution with k – 1 degrees of freedom, were generated using the MEANS procedure in SAS. The z-test (Zar, 1999) was used to compare any two CVs.
Partial Least Squares Analysis
Quantitative relationships between biomass production and soil quality indicators were explored using the method of partial least squares projection to latent structures (PLS). The PLS analysis was performed using PROC PLS in SAS (SAS Institute, 2004; Tobias, 1995). Initially, all soil quality indicators measured in the study (Table 1) were included as predictor variables in the PLS model. Predictors with the greatest influence in explaining variability in yield, or variable importance in the projection (VIP), were then selected based on the criterion of VIP > 0.8 (Wold, 1995). This was followed by sequential exclusion of predictor variables with the least impact on the model, based on loading weights and scores, until the highest q2 (cross-validated r2) was obtained. The number of PLS factors (latent variables) was selected using a cross-validation method in which the original data set was divided into two groups: a training or calibration set and a test or validation set. The number of extracted factors with the minimum predictive residual sum of squares (PRESS) statistic was chosen as the optimum. Using the CVTEST option of PROC PLS, the optimum or minimizing number of factors was compared with the PRESS for fewer factors to test whether there was a significant difference. In the absence of a significant difference, the model with fewer factors was chosen. The predictive strength (r2) of the model was assessed by linear regression of measured values in the response variables vs. the predicted values obtained in the cross-validation procedure.
Preliminary PLS analysis indicated similar results when the 1990 soil measurements were assessed against yields from each of the years 1991 through 1996 or when the assessments were made vs. the cumulative yield for 1991 to 1996. The same was true for the 1998 to 2004 yield data vs. parameters measured in 1997. No significant relationships were found, however, when the PLS model was applied to 1998 to 2004 yields vs. 1990 measurements, while a weaker relationship resulted when the cumulative yield for 1991 through 2004 was modeled using the 1990 measurements. Therefore, soil indicators measured in samples taken in 1990 were used to model the 1991 to 1996 cumulative yield, while the 1998 to 2004 cumulative yield was modeled using measurements from the 1997 soil samples plus some less dynamic parameters assessed in the 1990 samples.
Regression Models
Linear-plateau and quadratic-plateau functions were used to describe biomass yield response to each soil quality indicator identified by PLS analysis as an important contributor to variability in productivity. Least-squares optimization was performed with PROC NLIN in SAS (SAS Institute, 2004) to obtain these functions. To test whether the overall model for the combined data for yield without N fertilizer and yield with N fertilizer described yield response better than modeling each N treatment separately, we computed the F ratio from the sum of squares (SS) and corresponding degrees of freedom (df) for the combined (SScomb, dfcomb) and separate (SSseparate, dfseparate) models:
 | [1] |
The corresponding P value was determined using the FDIST worksheet function in Microsoft Excel. When PROC NLIN failed to converge, a simple linear function was tried. Yield was modeled as unresponsive to the soil properties when P > 0.05.
|
RESULTS AND DISCUSSION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
The variability was still evident for selected (mostly biochemical) soil properties in samples taken in 1997. Of these, TIC had the highest geometric CV (102%), while bulk density had the lowest (12%), indicating relatively uniform settling of the deposited soils.
Productivity
Repeated measures ANOVA for total aboveground wheat biomass yield indicated highly significant (P < 0.001) main (soil, fertilizer, and year) and interaction (soil x fertilizer, soil x year, fertilizer x year, and soil x fertilizer x year) effects. For both N treatments, the year-to-year variation in yield (Fig. 2
) closely reflected growing-season rainfall totals (Fig. 1). Yield increased linearly with increasing growing-season rainfall (r2 = 0.95) when data from 1995 and 2002 (poor seedling emergence) were omitted. The positive correlation between wheat yield and growing-season precipitation under rainfed conditions has previously been reported (Camara et al., 2003). Further analysis of our data by year revealed significant soil x fertilizer interactions (P 0.02) in all years except 1991, 1998, and 2001 (Table 2
). The number of soils responding to N application differed, however, among the years that had significant soil x fertilizer interactions (Fig. 2). For example, N application increased yield in 34 of the 36 soils in 2002 (Fig. 2s and 2t), but only in five soils in 2000 (Fig. 2o and 2p). The limited response to fertilizer N in 2000 was expected because of low rainfall during the growing season (61 mm in 2000 compared with 338 mm in 2002). Only Soils 20, 26, and 27 showed higher yields with N application in all years with a significant soil x fertilizer interaction. These soils had among the lowest inherent fertility, as evidenced by low yields without N application (Fig. 2). Soil 20 was from a site that had been artificially eroded (20 cm of topsoil removed) by leveling in 1957. Soils 26 and 27 were B and C horizon subsoils, respectively, from native grassland. Soils 8 and 9, which ranked among those with the highest inherent fertility, responded significantly (higher yields) to N application only in 2002 and 1996, respectively.
|
|
|
Identifying Significant Soil Quality Indicators for Biomass Yield
The 24 soil properties assessed in this study (Table 1 plus available water capacity and ratios C/N, C/P, LF-C/LF-N, and N/P) include many of the indicators recommended in minimum data sets proposed earlier for soil quality assessment(Doran and Parkin, 1994; Larson and Pierce, 1994). Thirteen soil properties included in a four-latent-variable PLS model (root mean PRESS = 0.75) contributed significantly (VIP 
0.8) to the variability in cumulative yield for 1991 through 1996 among soils receiving no fertilizer N (Table 3
). Total inorganic C, Cmin, and the LF-C/LF-N ratio were the most important variables, with VIP > 1. The model accounted for 74% of the variability in cumulative yield and 87% of the variability in the independent variables. The model had good predictive power, as evidenced by the significant relationship (P < 0.001) between observed and predicted yield values, obtained by cross validation (predicted yield [Mg ha–1] = 10.9 + 0.66 x measured yield, r2 = 0.69). For N-fertilized subplots, a four-factor PLS model consisting of 19 soil quality indicators (Table 3) was the best estimator for cumulative yield during the same period (1991–1996), explaining 64% of the variability in yield and 83% of the variability in the predictors. The model also had good predictive power (predicted yield [Mg ha–1] = 16.2 + 0.56 x measured yield, r2 = 0.51, P < 0.001). In both models, TN and TOC as well as LF-C and LF-N concentrations had equal contributions, and one of the two in each pair can be used as a proxy for the other in the model.
|
The lower explained variance for the zero-N cumulative yield for 1998 to 2004 (55%) compared with that for 1991 to 1996 (74%) could be due to the absence of some of the more important biochemical indicators such as LF-C and LF-N in the models fitted to the 1998 to 2004 data. The 1998 to 2004 cumulative yield from N-fertilized plots was not significantly related to any combination of the available indicators, implying that supplemental N may have partly offset limitations imposed by some facets of soil quality (e.g., N-supplying capacity of organic matter).
Total inorganic C was negatively correlated with cumulative yield (see below) and had the highest VIP of all indicators included in the final models (Table 3). Higher TIC concentrations (and corresponding lower yields) were associated with artificially eroded and subsurface donor soils, thus highlighting the negative impact that erosion can have on yields in soils underlain by calcareous material. Similar results have been reported by Papiernik et al. (2005), who recorded a 50% reduction in wheat yields in highly eroded soils where calcareous subsoil material had been exposed.
Although soil pH, EC, and textural classes (sand, silt, and clay contents) are understandably included in many minimum data sets suggested in the literature (e.g., Arshad and Martin, 2002; Brejda et al., 2000; Rezaei et al., 2006), our PLS analysis identified some significant effects only for the 1991 to 1996 cumulative yield from N-fertilized soils (Table 3). Similarly, aggregate stability did not have a significant effect on yield (VIP < 0.8) in our study, contrary to findings in those studies. This observation, in part, reflects the narrow range represented for these parameters; for example, most of the soils in our study were of neutral pH, low EC, and medium texture. The absence of a significant effect, therefore, does not indicate that these parameters are not important, only that they did not have a large influence in our study.
Most of the indicators identified as significant in explaining biomass yield variability in our study echo those reported as key attributes in minimum data sets established to predict soil productivity in other studies (e.g., Govaerts et al., 2006). Although our data did not include direct measurements of soil biota, such as microbial biomass C and N, biochemical parameters included in our models (i.e., SOM, Cmin, Nmin, LF-C, and LF-N) represent an important subset of possible biological attributes of soil quality and provide significant insight into soil biological processes (Gregorich et al., 1997). Other parameters, such as infiltration, were also not included because, as noted in other studies (e.g., Brejda et al., 2000), the time and labor costs required to measure such indicators were prohibitive, given the size of the study. Nonetheless, for this study we selected those indicators that are widely used in soils of this region and that have been shown in extensive preceding work to be responsive to management practices (Biederbeck et al., 1994; Janzen et al., 1992a).
Our PLS modeling results may be particularly important in identifying suitable soils for topsoil replacement in degraded-land reclamation schemes in agroecosystem settings similar to those in this study. Topsoil replacement is widely used in reclamation of lands disturbed by mining activity (e.g., Bowen et al., 2005) and oil and natural gas wellsite construction (Larney et al., 2005; Zvomuya et al., 2006). The multivariate procedure offers the opportunity to identify key indicators related to crop productivity and allows a reasonable estimate of biological productivity associated with a given candidate donor soil based on the key indicators.
Critical Soil Property Levels for Maximum Yield
For all indicators, cumulative biomass yield response was described better when fertilizer treatments were analyzed separately rather than together. Significant positive correlations were detected between cumulative yield without N fertilizer for 1991 through 1996 and 9 of the 24 baseline soil properties tested (Table 4
). Only TIC was significantly negatively (and linearly) correlated with biomass yield, with the yield decreasing by 0.67 Mg ha–1 for each 1 g kg–1 increase in soil TIC (r2 = 0.23, P < 0.001). The negative effect of TIC on wheat yields has previously been reported (e.g., Papiernik et al., 2005). Response functions were linear plateau (e.g., Fig. 4
) for TN, TOC, Cmin, KCl-extractable N, LF-C, LF-N, and HCO3–extractable P, and quadratic plateau for Nmin (Table 4).
|
|
The relationships between yield and levels of individual soil properties were weaker in N-fertilized than in unfertilized soils, as evidenced by lower coefficients of determination for the former (Table 4). Similarly, the first regression coefficient, b, was lower with N fertilizer for the five indicators common to the two fertilizer treatments that shared the same model type. These observations imply that N fertilizer can partially compensate for deficiencies in soil quality indicators, particularly those related to N supply. With added N, for example, yield showed no significant response to initial KCl-extractable N because fertilizer optimized this parameter in all soils. Similarly, N fertilizer decreased the responsiveness of yield to labile fractions of SOM (LF-C and LF-N) and their activities (Cmin and Nmin), all of which are related to soil N supply. Previous research under similar climatic and soil conditions in southwestern Saskatchewan showed that these parameters were enhanced by N fertilizer in a continuous wheat cropping system (Campbell et al., 1997). Similarly, Körschens et al. (1998) reported no positive yield response to TOC when mineral fertilization was optimal. Loveland and Webb (2003) noted that breakdown of SOM is necessary to maintain yields in unfertilized soils, and suggested that, at subthreshold TOC concentrations, nutrient mineralization may be insufficient to sustain satisfactory yields. Thus, the ability to supply N is an important aspect of soil quality.
None of the soil physical properties measured were significantly correlated with aboveground biomass yield, perhaps reflecting the generally lower variability in the physical indicators among the soils compared with chemical properties (Table 1). The relative insensitivity of crop, pasture, and forestry productivity to some relatively constant soil physical properties such as particle size distribution has previously been reported (Schipper and Sparling, 2000). Contrasting results, however, were reported in a semiarid rangeland study, which indicated that plant variables were more sensitive to soil physical properties than to chemical properties (Rezaei et al., 2006).
The critical level for Olsen P (47.2 mg kg–1) obtained in our study is consistent with previous observations in the region. For example, McKenzie et al. (1995) reported that wheat yield showed little or no response to soil test P levels greater than an equivalent of 50 mg Olsen P kg–1 in a wide range of Alberta soils. Comparable critical concentrations have been reported for unfertilized wheat yields in Australia (Holford et al., 1985). Similarly, for TOC, various earlier reports suggested 20 g kg–1 as a threshold below which a potentially serious decline in soil quality will occur (Loveland and Webb, 2003).
Although the above indicators individually explained a significant proportion of the variability in aboveground yield (Table 4), yield response to the level of any one indicator depends on its degree of limitation relative to the availabilities of others (Kho, 2000). Arshad and Martin (2002) emphasized that a critical limit of a soil indicator can be ameliorated or exacerbated by limits of other soil properties and the interactions among soil quality indicators. Körschens et al. (1998) proposed TOC thresholds, for crop production, that increased with increasing clay content, from 1% TOC at 40 g kg–1 clay to 3.5% TOC at 380 g kg–1 clay. In our study, response functions for each indicator were determined at varying levels of all other indicators. Thus, possible interactions among indicators may have influenced the slopes, critical levels, and plateaus associated with a given indicator. Of the important ratios tested in our study, however, such as C/N, C/P, LF-C/LF-N, and N/P, only LF-C/LF-N made a significant contribution (VIP > 0.8) to the final model. Nevertheless, in a study such as this, soil quality indicators are confounded with each other and cannot be changed independently (Kho, 2000). Thus, empirically derived relationships between productivity and individual soil quality indicators, such as those in the present study, are best seen as correlations rather than as causal relationships.
The lack of significant biomass yield response models for the rest of the indicators could be due to the baseline levels of the indicators (e.g., CEC and pH) being within the range required to maximize yield under the conditions of the study. Janzen et al. (1992b) noted that the yield response to a quality factor depends on the initial magnitude of that factor. Additionally, between-soil variability for some of these parameters may have been too low to elicit a significant response or breakpoint in the response.
Because of interactions between soil properties and environmental, landscape, and management variables, the soil indicators identified as most important for explaining yield variability in our study may have different significance under another set of conditions. Although some indicators such as TOC are common to a wide range of soils and geographic regions, there may be no universal optimum set of soil quality indicators for all agroecosystems (Brejda et al. 2000). Similarly, critical indicator levels for maximizing productivity may vary under different management, landscape, and climatic settings. Knoepp et al. (2000) noted that, while identification and quantification of soil quality indicators for soils with a single primary function, such as agricultural or plantation soils, is broadly feasible, quantifying these indices in complex, multifunction ecosystems is much more difficult.
Our approach, like others, is not without limitations (Olson et al., 1996). Results based on the approach are probably best seen as supplementary to those derived from alternative approaches to the study of soil quality.
|
CONCLUSIONS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Partial least squares analysis identified a suite of soil properties, including TOC, TN, TIC, labile fractions of SOM (LF-C, LF-N, Cmin, and Nmin), and extractable nutrients (N and P) that could serve as definitive quality criteria, specific to the function of plant biomass production, for soils in the semiarid regions of western Canada. Using segmented-model regression, we estimated the critical level of each indicator above which yield response was unlikely. With some exceptions, applying N fertilizer tended to reduce the slope of the response to the various soil quality indictors, but often did not significantly affect the critical level. For most of the indicators, fertilizer application diminished the goodness of fit (r2) of relationships between yield and the indicators.
Our findings provide evidence that selected indicators can provide a definitive, quantitative assessment of soil quality. The findings also lend credence to the value of our soil-transplant approach in quantifying relationships between soil function and indicators for specific areas while circumventing the limitations traditionally imposed by divergent landscape quality factors.
|
ACKNOWLEDGMENTS |
|---|
|
NOTES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Permission for printing and for reprinting the material contained herein has been obtained by the publisher.
Received for publication May 16, 2007.
|
REFERENCES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 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 | |||
