| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Dep. of Soil Science, University of Saskatchewan, 51 Campus Drive, Saskatoon, SK, S7N 0E1 Canada
* Corresponding Author (pennock{at}sask.usask.ca)
 |
ABSTRACT |
|---|
 TOPNOTES ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSION AND CONCLUSIONSREFERENCES |
|---|
Abbreviations: 0xN, N treatment referring to zero times the recommended soil test rates • 0.5xN, N treatment referring to 0.5 times the recommended soil test rates • 1xN, N treatment referring to one times the recommended soil test rates • 1.5xN, N treatment referring to 1.5 times the recommended soil test rates • and 2xN, N treatment referring to two times the recommended soil test rates • AIC, Akaike information criterion • DEM, digital elevation model • EONR, economically optimum N-fertilizer rate • LEC, landform element complexes • PWP, permanent wilting point
|
INTRODUCTION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSION AND CONCLUSIONSREFERENCES |
|---|
Any agronomic and economic evaluation of variable rate inputs must be placed within the larger context of all factors that affect yield. Bullock et al. (1998) state that there are three broad categories of factors affecting crop yields: site characteristics, uncontrolled stochastic factors, and inputs to production. They define site characteristics as factors that are not controlled by the farmer and which remain constant over the course of the growing season; factors such as soil texture, slope, or soil organic matter. Uncontrolled stochastic factors are not controlled by the farmer but are also not constant over time—for example, weather conditions during the growing season are a major determinant of yields in the northern Great Plains. Inputs to production are the only category that is completely controlled by the farmer.
There is, however, a final category of factors not explicitly discussed by Bullock et al. (1998) which are of great importance to variable rate fertilization—those factors that change over the growing season but the value of which is known (to varying degrees of precision) at the time of seeding. These factors include plant available nutrient levels and soil moisture. The value of these factors at the time of seeding can be assessed by soil sampling and can be viewed as a site characteristic, albeit of a more dynamic nature than soil texture or slope.
The initial stage in the evaluation (or application) of variable rate fertilization is the division of the field into segments (termed sections by Bullock et al., 1998) that are small enough to have uniform (or at least a limited range of) site conditions. Each of these segments then receives the appropriate level of inputs to meet the farmer's production goals. The delineation of these segments has been a major focus of precision farming research to date. Three broad approaches can be discerned: (i) segments defined by yield mapping, (ii) by soil sampling of dynamic site characteristics such as soil N, and (iii) by static site characteristics such as topography.
Blackmer and White (1998) argue that simply mapping yield variability within a field even for a number of years provides very little information that can be used to improve N management. They state that yields are influenced by many factors beyond N availability, and hence, additional information on these other factors is required if a reliable evaluation of N requirement is to be made.
The remaining two methods are closely related. Delineation of segments based on soil sampling of dynamic soil properties in the northern Great Plains has been based on sampling for the major yield-limiting nutrients such as soil N or P (Franzen et al., 1998; Clay et al., 2000). Blackmer and White (1998) argue that segments defined by soil N levels have limited potential for variable rate fertilization because the current N recommendations based on these field samples fail to consider N mineralized over the growing season from organic sources and N lost by leaching, gaseous loss between seasons, or early in the growing season. As well, the high variability of measured N and P levels in the field limit the confidence with which reliable prescriptions can be made. Beckie et al. (1997) concluded that variable rate N applications based on spring N levels were not effective because of the high variability of soil N levels experienced at their research site in Central Saskatchewan. On more practical grounds, the large number of hectares per farm in the northern Great Plains greatly limits the number of soil samples that can be used to develop the field segments (Franzen et al., 1998).
The final approach to delineate segments uses static site characteristics as a surrogate for the more difficult-to-measure dynamic site characteristics such as available N. McCann et al. (1996) and Blackmer and White (1998) use soil organic matter (SOM) levels (as assessed using black and white aerial photography) to delineate field zones. Attempts to relate crop yields to soil taxonomic units have only been marginally successful, as the major taxonomic criteria in soil classification are generally unrelated to crop growth (Wibawa et al., 1993). Clay et al. (2000) found that a variable rate P fertilization program based on soil type and field history was the most economically successful of the examined approaches, but it only increased profitability by only $3.74 ha-1, as compared with a uniform P application.
There is extensive literature from the northern Great Plains that suggests the possibility of using topography to delineate practical, agronomically meaningful field segments. Distinct levels of the plant-growth controlling soil properties such as soil moisture and soil N are associated with different topographical units throughout the region (Hanna et al., 1982; Wibawa et al., 1993; Corre et al., 1996; Stevenson and van Kessel, 1996; Beckie et al., 1997; Franzen et al., 1998). In a comparison of four approaches to delineating management units for variable rate fertilization, Beckie et al. (1997) concluded that topographically defined units were most effective at their research site.
It is important to note, however, that the relationship between dynamic soil properties and segments defined on the basis of site characteristics such as topography will not be constant through time because of uncontrolled stochastic factors such as weather. For example, soil moisture differences between slope positions are most strongly expressed in spring (following snowmelt recharge) and after major rainfalls, and typically are lost towards the end of the growing season after crop utilization of soil moisture (Zebarth and de Jong, 1989; Androsoff et al., 1995). The temporal dependence of the relationship between terrain attributes and soil moisture has also been noted by researchers in other regions (Western et al., 1999). Hence, the variability of yields due to the uncontrolled stochastic factors may overwhelm the predictive values of the topography-yield relationship. Thus, the objectives in this study were to assess the yield response of canola in three slope positions to five N fertilizer rates. This agronomic evaluation then provided a basis for an economic assessment of the potential returns from the adoption of variable rate nitrogen fertilization.
|
MATERIALS AND METHODS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSION AND CONCLUSIONSREFERENCES |
|---|
|
In early spring 1996, one composite sample (from 10 to 15 subsamples) to a 30-cm depth was taken from linear LEC at the sites and sent to Enviro-Test Laboratories (124 Veterinary Road, Saskatoon, SK, S7N 5E8 Canada) for a standard soil N test via a 2 M KCl extract (Keeney and Nelson, 1982) and N recommendation. The linear LEC represented 
35% of the site, and are the recommended slope positions for soil sampling under provincial guidelines. The fertilizer recommendation based on the soil N test involves the calculation of a target yield typically based on the 30-yr average precipitation for the region the sampled field occurs in. The fertilizer recommendation is the nutrient level required to achieve this yield goal (adjusted for the measured N content of the soil) using a production model ( Enviro-Test Laboratories, personal communication, 2001). The soil test recommendation was used to determine the rates of fertilizer inputs, including N, required at the site for all 3 yr of the study.
In 1996, the soil test recommended N rate was 84 kg ha-1 and the P rate was 25 kg P2O5 ha-1. Nitrogen treatments were 0xN, 0.5xN, 1xN, 1.5xN, and 2xN rate, all with P and S at the recommended rate. The same procedures were used in 1997 and 1998. Soil test recommended rates in 1997 and 1998 were 90 kg N ha-1 and 31 kg P2O5 ha-1.
Six replicated blocks of treatments were laid out on the surface in each year, where each block of treatments spanned the three landform complexes (Fig. 2) . Each treatment strip was a single pass by the modified air seeder (described below). The treatments were randomized within each block, but were not rerandomized by landform position within each block (Fig. 2). In both 1997 and 1998 the six blocks were moved to land immediately adjacent to the area used in the previous year to eliminate carry-over from the treatments (Fig. 1).
|
In each of the treatment strips, a 10-m long, 1.52-m wide harvest sample was taken from each of the three landform complexes using a small plot combine. In addition, a 1.0-m2 area was hand harvested for determination of total aboveground biomass production.
Seeding dates were 4 June 1996, 12 May 1997, and 15 May 1998. Seeding was delayed in 1996 because of wet soil conditions. Harvest dates were 2 Oct. 1996, 23 Aug. 1997, and 24 Aug. 1998.
Soil Sampling and Laboratory Analysis
Two 30-cm long, 7.2-cm diam. soil cores were taken to a 60-cm depth the week before seeding from each LEC in the 0XN, 1XN, and 2XN treatment strips in each block of treatments. Spring soil samples were analyzed for nitrate N, and gravimetric moisture was determined in both the 0- to 30-cm and 30- to 60-cm depth. Gravimetric moisture values were converted to a volumetric basis using the bulk density of the sample. Field moist soil samples (20 g) were shaken for 1 h in 200 mL of 2 M KCl and the solution filtered through Whatman No. 40 filter paper (Whatman Ltd., Maidstone, Kent, UK) manufacturer and location) (Keeney and Nelson, 1982). Nitrate in the soil extract was measured colorimetrically using an AutoAnalyzer (Technicon AutoAnalyzer II, Tarrytown, NY). Permanent wilting point (PWP) was determined by applying 1500 kPa of pressure to samples in a pressure plate apparatus (Klute, 1986). The PWP values were corrected to a volumetric basis using the bulk density of the sample and were then subtracted from the volumetric soil moisture of the sample to give available moisture at the time of sampling to 60-cm depth.
Statistical and Economic Analysis
Because the N treatments were based on a gradient treatment design (i.e., a set of treatments that are an increasing dosage of a quantitative factor) the mean separation technique was regarded as inappropriate (Swallow, 1984), and an equation was fit to represent the relationship between canola yield and N rates for each slope position in each year. Cerrato and Blackmer (1990) state that too often researchers have confined themselves to fitting only a quadratic model to the yield data, whereas their examination of optimum curve fitting for Iowa corn yields indicated that other models, notably the quadratic plus plateau model, gave a superior fit. We fit five models to our yield data in each year: linear, quadratic, linear plus plateau, quadratic plus plateau, and square root following the model specifications given by Cerrato and Blackmer (1990).
The SAS NLIN procedures (SAS Institute, 1990) were used to determine the model parameters for all but the linear model, for which linear regression was used. The weighted linear squares method for the linear model and weighted nonlinear least squares methods for the remaining models were adopted to obtain model parameters that minimize the difference between the predicted yield and the average yield of the six replicates as a function of fertilizer applied. The reciprocal of the variance of estimated yield was used as the weight in the nonlinear and linear least squares methods. Once the model parameters have been obtained, the validity of the fitted response models was tested by analyzing the statistical properties of the normalized residual errors, rs:
 | [1] |
Under the hypothesis that the model is correct, elements of rs are zero independent normal deviates. The following criteria was used to test whether rs are zero normal deviates at the 99% confidence level,
 | [2] |
It is also necessary to test whether rsi taken as a whole appear to conform to the hypothesized rs distribution. Since rsi are normal independent deviates with zero mean and unit variance, the sum of squared rsi (SSE) should have a 
2 (chi) distribution with a degree of freedom N-p. Thus at 99% confidence level:
 | [3] |
These validity tests are useful in guarding against lack of fit. Models passing these tests can be considered as candidates for the best model. The sum of the squared error can be used for further model discrimination if the number of parameters is the same; however, for the selected models, the linear model has two parameters and the other four models have three parameters and discrimination among models requires a different criterion.
As a rule of thumb, the best model is the model that is physically meaningful, best fits the data, and has a minimum number of parameters (Bard, 1974). For a physically meaningful model, one of the widely used criteria for model discrimination is the Akaike information criterion (AIC) (Schweppe, 1973), which reflects both the lack of fit and complexity of the model (number of parameters). Akaike information critierion is derived from the negative log likelihood of the fitted model and can be estimated from the residual sum of squares of deviations from the fitted model:
 | [4] |
The economically optimum N-fertilizer rate (EONR) was derived from the yield response curves. The analysis assumes that producer wants to maximize expected profits with no consideration of the variance of profits (or risk) resulting from the adopted practice (Bullock and Bullock, 1994). The economic optimum is determined by considering the price ratio between the unit cost of inputs, namely N, and the unit value of yield (Bullock et al., 1998). The price ratio is defined as the cost of the input divided by the value of the crop. A constant price ratio (based on 1997 values) was used for all 3 yr based on a cost of N fertilizer of $0.424 kg-1 and canola at $0.231 kg-1 for a price ratio of 1.8344. The predicted economic optimum rates of N were calculated by equating the first derivative of the response equations to the price ratio (Cerrato and Blackmer, 1990). All dollar values were initially calculated in Canadian dollars and converted to U.S. Dollars. The exchange rate used was the average rate in 1997 (1.00 U.S. dollar = 0.7225 Canadian dollar).
The net return at the EONR for each LEC in each year was compared with the net return at a constant application of the soil test recommended rate in that year. The constant 1xN application represents the current fertilization practice in the region. This comparison is, however, based on perfect hindsight—the net returns are calculated based on knowledge generated after the crop is harvested and as such it represents the maximum possible return that could be achieved. These calculations also do not take into account any extra costs resulting from the adoption of variable rate equipment by the producer. A complete assessment would require that the purchase and maintenance cost of the variable rate fertilizer (VRF) equipment be included in the analysis.
|
RESULTS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSION AND CONCLUSIONSREFERENCES |
|---|
Mean plant available soil water at the time of seeding decreased in each successive year of the study relative to 1996 levels (Table 1). As well, the differences in mean plant available soil water between the concave segments and the convex segments decreased in each successive year of the study (Table 1). In the drier conditions of 1998, available water was very low in both the convex and linear complexes. The lower levels of preseeding available water in each successive year were largely because of the postharvest precipitation and snowfall received in each year. The normal (1969–1990) precipitation received in the October through April period was 77.4 mm. In the fall and winter of 1995 and 1996, 122.9 mm were received at the Saskatoon station (20 km south of the research site); in the same period in 1996 and 1997, and 1997 and 1998, 97.9 mm and 47.3 mm, respectively, were received (data from Saskatchewan Agriculture and Food, Regina, SK, 2000).
|
|
|
|
|
|
|
Clearly a critical aspect of the implementation of a VRF N program is the ability to select a particular rate based on the information available to the producer at the time of seeding. If the effect of uncontrolled stochastic factors overwhelms the predictability because of static or dynamic site characteristics then the design of a VRF program is greatly complicated.
Plant available water and soil nitrate in spring (the week before seeding) were regressed against the N rate required to achieve the economically optimum yield. The yield at economic optimum was related to spring available water (SAW [in cm of water to 60 cm of soil]):
 | [5] |
The relationship between soil nitrate measured in spring (to 60-cm depth) and EONR was not significant.
|
DISCUSSION AND CONCLUSIONS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSION AND CONCLUSIONSREFERENCES |
|---|
The differences between the LEC narrowed over the 3 yr of the study, but the relative ranking of the three LEC remained the same. The narrowing of the differences between the LEC is due to the state-dependence of the spatial pattern of soil moisture in the landscape (Western et al., 1999). In years of average or above-average precipitation (1996, 1997 in this study), the soil moisture in each landform position is controlled by surface and lateral redistribution of water, and a clear difference between landform position occurs. In years of below-average precipitation (1998), recharge of the soil moisture that was depleted in the previous crop year does not occur, and the differences between positions narrow, or in years where precipitation fails completely, vanish (Western et al., 1999).
Landform variability of spring soil nitrate was less obvious due in part to the relatively large variability in point-to-point estimates of spring soil nitrate (Table 2). Generally the concave landform complexes had the greatest spring N levels, but considerable variability occurred. The lack of a consistent relationship between landform complex and available soil N observed in our study is also consistent with the literature in Saskatchewan and elsewhere. Androsoff et al. (1995), van Kessel et al. (1993), Corre et al. (1996), and Beckie et al. (1997) also were unable to detect a clear, temporally stable association between soil nitrate and topographic position. The lack of a pattern for soil nitrate may be because of the highly dynamic nature of N mineralization (Blackmer and White, 1998).
Clear and consistent differences in canola yields were associated with the landform complexes used to stratify the research site (Table 4). Averaged across treatments and years, the yield of canola was 0.5 t ha-1 in the convex LEC, 1.0 t ha-1 in the linear LEC, and 1.4 t ha-1 in the concave LEC. Others have reported a similar influence of topography on crop yields, with greater yields in concave positions observed for corn (Kravchenko and Bullock, 2000; Simmons et al., 1989; Stone et al., 1985), soybean (Glycine max [L.] Merr.) (Kravchenko and Bullock, 2000) and wheat (McConkey et al., 1997; Sinai et al. 1981; Beckie et al., 1997). These observations are consistent with early research on the topographic-yield relationship by de Jong and Rennie (1969) and Spratt and McIver (1972), both of whom showed that soil moisture differences were the major control on wheat yields in different slope positions.
We also observed considerable year-to-year differences, with the highest average yields achieved in 1996 and lowest in 1998 (Table 4). This yield pattern is consistent with the decrease in spring soil moisture over the 3 yr of the study but do not reflect mean spring nitrate levels, which were lowest in 1996 and highest in 1997.
The differences between landform complexes were also evident in the response of canola to the N fertilizer treatments (Table 5). It is notable that the 0XN yields accounted for over 50% of the maximum yields achieved through fertilization in all but one case, yet the yields were not strongly correlated to spring N levels. This indicates that although the soil was able to provide considerable available N to the growing plant, this N supplying ability was not captured by the N fraction measured as plant available nitrate N in spring (i.e., the cold KCl extract).
In all 3 yr, the maximum yield because of fertilization achieved on the convex LEC was well below the check (i.e., 0XN) yield on the concave LEC. Spratt and McIver (1972) had previously found that increased fertilizer inputs on the convex slope positions could not compensate for the locally arid conditions associated with these positions. This finding runs counter to the common perception that the lower productivity of the soils in the convex landform can be remedied by the addition of higher N fertilizer inputs.
Slopes of the linear yield relationships for concave LEC in 1996 and 1997 were well in excess of the price ratio, indicating that economic yield response occurred up to the maximum N rates used in the 2 yr (i.e., 168 kg ha-1 in 1996 and 190 kg ha-1 in 1999). Research from several regions cited by Grant and Bailey (1993) suggest that canola will respond to N additions in excess of 200 kg ha-1 when other soil factors are not limiting. Jackson (2000) also found that peak seed yields occurred at N rates of 180 to 220 kg N ha-1.
The dominant control (of those we measured) for the year-to-year differences was clearly spring soil moisture. The differences between LEC were greatest in years with a high range in soil moisture conditions such as 1996 and were least in relatively dry years such as 1998. The influence of weather conditions on the expression of the topography-yield relationship has been noted by other researchers in this region. For example, Halvorson and Doll (1991) reported that topography had a limited impact on wheat yields in drought years, presumably because of limited redistribution of rainfall in the landscape. In contrast, Simmons et al. (1989) reported that landform complex had the greatest influence on corn yield during a year with severe early season drought. They reasoned that the early season drought exacerbated the differences in soil moisture between different landform complexes. The influence of growing season precipitation also differs between years; Kravchenko and Bullock (2000) reported that wheat yields in highly concave areas characterized by large flow accumulation were negatively correlated with early season precipitation in relatively wet years whereas positive correlations were reported during dry summers.
Despite the influence of annual weather on yields, the strong relationship between spring soil moisture and the N rate required to achieve the economically optimum yield indicates that it may be possible to derive an LEC-specific N recommendation for canola if the relationship presented here can be shown to be consistent for other sites and at other price ratios. The recommendation of higher N levels in the concave LEC in above-average moisture years also presupposes that other agronomic factors do not overwhelm the yield-fertilizer relationship. For example, Beckie et al. (1997) worked in a zone of higher precipitation in Saskatchewan and they found that a combination of disease and weed pressure caused reduced canola yields in the concave landform complex at their site.
In conclusion, the landform stratification used in this study was useful for the development of landform-specific N recommendations. The landform complexes are dominantly surrogates for the spring soil moisture conditions, which are the major control of canola yields at the site. The results further suggest that an intensive sampling for available N conditions in spring would not be a useful diagnostic tool for predicting optimum N application rates. As an alternative, a spring soil moisture sampling program appears to be the most viable predictive measure for determining optimum N application rates.
|
ACKNOWLEDGMENTS |
|---|
|
NOTES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSION AND CONCLUSIONSREFERENCES |
|---|
Received for publication June 5, 2000.
|
REFERENCES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSION AND CONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  B. C. Si and R. E. Farrell Scale-Dependent Relationship between Wheat Yield and Topographic Indices: A Wavelet Approach Soil Sci. Soc. Am. J., March 1, 2004; 68(2): 577 - 587. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 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 | |||
