Illinois
Fertilizer Conference Proceedings
January 24-26, 2005
| Home | 2005 Index | Search |
| Home | 2005 Index | Search |
T. R. Ellsworth, R. L. Mulvaney, T. J. Smith, S. A. Khan, and C. W. Boast
1
A major impediment to N management has been the lack of a soil N test that can reliably predict crop yield response to fertilizer application. For instance, research has repeatedly shown that soil nitrate tests are rarely correlated with crop yield response. It has long been postulated that certain organic forms of soil N are readily converted to plant-available forms and account for the lack of response to fertilizer application. However, traditional methods to fractionate soil organic N and relate these fractions to crop yield response have not been successful. Recent research at the University of Illinois led to new diffusion techniques that permit quantitative determination of different forms of hydrolyzable soil N (Mulvaney and Khan, 2001). These techniques showed a higher concentration of amino sugar N for soils that are non-responsive to N fertilization than for responsive soils (Mulvaney et al., 2001). Subsequent work led to a simple, rapid soil test for estimating an alkali-labile fraction of soil N (i.e., amino sugar N), which is commonly referred to as the Illinois Soil N Test or ISNT (Khan et al., 2001), with further evaluations through 105 on–farm N–response experiments funded by FREC (Mulvaney et al., 2004). This latter work clearly demonstrated the utility of the new soil test for identifying sites were corn did not respond to N fertilization, and suggests a more ambitious application for responsive sites, based on significant correlations with optimum N rate and check–plot yield. Moreover, there are indications that spatial variability in test values may be related to field variability in crop N response, which suggests that the ISNT could be utilized successfully as a basis for site–specific N management.
Since the ISNT is designed to measure a portion of soil organic N that is capable of relatively rapid mineralization, one could anticipate that test values would vary with space and time due to microbial activity in response to organic and inorganic substrates, and variation in soil moisture and temperature. This obviates the need to identify appropriate sampling strategies to account for both temporal (i.e., spring or fall, etc.) and spatial variabilities (i.e., identifying a suitable grid scale for composite sampling as well as the number, sample depth, and spatial arrangement of soil cores within a composite sample, for practical application of the ISNT). The focus of the present research effort was in large part a consequence of an increased understanding arising from earlier investigations, as well as an increased awareness of the limitations of our current knowledge base. Several of these findings and limitations are reviewed here.
Weekly sampling in our first year of study within three field sites in east central Illinois (Boast et al., 2003, see especially Figures 4 and 5) suggested a trend of decreasing ISNT values beginning in March through May with an apparent subsequent increase in late Fall. However, there were considerable fluctuations in plot-scale ISNT values during this initial study that led to tentative conclusions regarding temporal sampling. In addition, a subsequent study by Hoeft et al. (2004) suggested conflicting trends regarding temporal variations in ISNT values. As part of the initial work reported by Boast et al. (2003), dense spatial samplings at sub-meter scales were performed within each of three field sites. From these data, we concluded that much of the tentative nature of the initial temporal studies was a consequence of inadequate soil compositing methods, implying a need for a more robust temporal analysis.
Another aspect related to ISNT soil sampling was discovered by Mulvaney et al. (2004), who observed that in some instances subsoil ISNT values (i.e., 12–24 inches in depth) played an important role in determining corn crop response to N fertilization. Further, it appeared that this effect was related to soil management (i.e., tillage, manure history, etc.). In addition, with respect to estimating the mean ISNT value for small field plots, the first-year study indicated the importance of sampling method (i.e., the number and spatial arrangement of soil cores included in a composite sample), indicating the need to understand how sample design influences the accuracy of spatial maps of ISNT values for grid-based sampling strategies.
The objectives of the proposed work are to quantify variations in the ISNT under a variety of agricultural management environments. This will allow us to accomplish the overall goal of our research: to identify a strategy for using the new N test in production agriculture. The sub-objectives proposed are to characterize:
To achieve objective 1, in addition to the continued monitoring of the temporal trends in ISNT values in the three fields studied in year 1 (designated "North", a corn-soybean rotation without a recent manure history, "Manure", a corn-soybean rotation with annual manure application, and "South", a continuous corn rotation, see Boast et al., 2003), two new management systems were included in the present study. These were designated "No-Till", a no–till corn–soybean rotation, and "South S–C", which was an expansion of the study within the South field to include a transition from continuous corn to a corn–soybean rotation. The soil at the South field is a Sabina silt loam, a light-colored, low organic matter Alfisol, while the other three sites represent Mollisols. The Manure site is located on a Drummer silty clay loam, the North site on an Ipava silt loam, and the No-Till site on a Catlin silt loam. During year two, within each of the five study sites, three 100-m2 (25 feet by 43 feet) unfertilized plots were established, including two that were cropped to corn and one that was fallow. The second year of temporal sampling commenced on 25 October 2002 and continued through 17 November 2003. Soil samples were collected at bi– weekly intervals for each of three depths (0 to 3 inches, 3 to 7 inches, and 7 to 12 inches) and analyzed using the ISNT. The sample for each depth in a given temporal plot consisted of a composite of eight subsamples obtained as follows. Each plot was first divided into four 11–feet–by–20–feet subplots on a 2 × 2 pattern, and then a 7 × 8 grid of locations was marked with twine to define each of the four subplots. On a given sampling day, two of the 56 (7 × 8) locations within each subplot were chosen in a stratified random sampling, without replacement. For each depth, a soil sample was taken from these two locations in each of the four subplots within a given temporal plot, and the eight samples were composited. This approach provided a relatively uniform coverage of the corresponding temporal plot area on each sampling date and doubled the number of soil cores included in a composite sample relative to the first year of the study. To achieve objectives 2 and 3, during April 2004 soil samples were collected, and ISNT values measured, within three ~40–ac fields in east central Illinois. These fields were in corn-soybean rotations and were cropped to corn in 2004. Both individual “point” core and composite core soil sampling was employed, with point sampling within sub-regions of the field at a grid scale of 30 × 60 ft2, and a 5-core composite sample at a grid scale of 60 × 200 ft2, with sampling depths of 0-12 and 12-24 in. Following soil sampling, a randomized block design was employed to conduct corn N response trials within each field. Fields contained 12 to 36 blocks, with block sizes of 240 × 400 ft2 and 200 × 240 ft2, each of which contained 8 and 6 plots (60 × 200 ft2), respectively. One of 8 (or 6) N rates was randomly assigned without replacement to each plot. Yield was obtained as the combine yield monitor average value for the central 20– × 150-ft2 area within each plot. The sampling method provided an estimate of the corresponding ISNT value for each plot and depth.
The revised compositing method greatly enhanced our ability to quantify temporal trends in ISNT values, as evidenced in Figures 1 and 2. Figure 1 shows the biweekly average ISNT value for the two cropped temporal plots in the North Field for each sampling depth. The temporal pattern evident in this figure was remarkably similar to that for all 5 field sites and sampling depths, indicating an annual low in ISNT values occurring roughly in mid-May, with annual peaks during late summer and early spring. The depth and cropped temporal plot average biweekly ISNT value for all 5 sites is shown in Figure 2. The trend evident here is relatively consistent with that observed during the first year of study (Boast et al., 2003), indicating that seasonal fluctuations are significant and care should be taken to obtain representative samples for ISNT analysis.
Note that fluctuations in ISNT values tend to increase with field average ISNT, being greatest for the North field, for which variations between minimum and maximum observed average ISNT values are approximately 100 ppm. These results are consistent with previous conclusions stated by Mulvaney et al. (2003), that soil sampling for the ISNT is best done in the fall after harvest, but if necessary can be postponed until early spring. Fall sampling has the advantage that ISNT values tend to be somewhat lower, thereby reducing the risk of underfertilizing a responsive site.
A first focus in objective 2 was to characterize subsoil ISNT values (12–24 in) relative to 0–12 in depth values and to explore the feasibility of estimating the former via indirect methods. The impetus for this effort was the observed relationship between the 12–24 in depth ISNT values and corn crop responsiveness to N fertilization (Mulvaney et al., 2004) as well as the relative increased difficulty in sampling below depths of 12 in. One option we examined for estimating subsoil ISNT values was co–kriging, relying on a spatially more dense sampling of the 0–12 in depth. The practical utility of this approach relies strongly on the correlation of ISNT values between depths of 0–12 and 12–24 in. We found this correlation to be relatively strong for samples in fields without a recent manure history (correlation of Ρ > 0.60 employing the stated compositing strategy and Ρ > 0.4 for point samples) while the relationship was much weaker for samples collected within a manured field ( Ρ ~ 0.3 for composite samples and 0.2 for point samples). Scatterdiagrams illustrating these relationships are shown Figure 3.
Scattergrams for other non-manured field sites were similar to that shown here, with a maximum correlation for composite samples of approximately 0.79, suggesting that co–kriging is a viable option for reducing subsoil sampling in these situations.
The spatial variability of ISNT values was characterized with variograms, which provides an average measure of the difference in ISNT values as the spatial separation between the measurements increases. The point samples (i.e., 1 in diam. soil cores) for fields studied indicate that roughly 40% of the field-scale variability in ISNT values occurs within relatively short spatial separations (< several feet). This implies that compositing offers a great opportunity over point sampling for estimating the within field variations in "plant-scale" ISNT values. (We hypothesize that for management purposes, the minimal scale of interest for characterizing ISNT variability corresponds to the average value encountered by a growing corn plant, which is likely on the order of several feet, i.e, the "plant-scale". However, sub-plant scale variability in ISNT values may prove important if such variability results in a differing crop response to N fertilization, a topic that may warrant future investigation.)
We examined the influence of compositing with respect to estimating the spatial distribution of ISNT values. This required estimating a "point" scale variogram using both the composite and point measurements. Figure 4 provides an example of this estimation for a non–manured field in a corn–soybean rotation.
This point variogram provides a basis for evaluating the influence of compositing on the average error incurred in mapping ISNT values. For example, consider the three compositing schemes shown in Figure 5, where L is the separation between individual soil cores.
The hexagonal composite design was shown to be optimal in terms of minimizing the estimation error, however practical implementation of such a design may not be feasible. For the field corresponding to the variogram shown in Figure 4, the average estimation errors for alternative compositing schemes are shown in Figure 6
This approach of estimating the average error requires knowledge of the model variogram. Despite this limitation, several general conclusions for sampling for any arbitrary soil property can be made. First, for a given sampling density (i.e., grid scale), the work demonstrates the importance of compositing for reducing estimation uncertainty, as well as the necessity of properly accounting for the support effect in variogram estimation. Second, typical soil testing practices employ a composite sample for a relatively small value of L (~ 10 ft). The research reported here demonstrates the comparatively poor performance of such an approach. Rather, the optimal configuration for collecting "n" core composite samples should be that which maximizes the spatial uniformity between the individual core sample locations. The figures depicting the spatial distributions of errors illustrates the need to either minimize these variations directly via optimizing the sampling design, or taking this uncertainty into account when making management recommendations. Finally, for all three grid scales (0.5, 1.0, and 2.5 ac) we examined and for the variogram shown in Figure 4, the 7 core-composite, hexagonal sampling scheme consistently outperformed the 9 core– composite optimal sample arrangement. However, practical implementation of the hexagonal approach may be more considerably more challenging.
Our analysis of the several field sites with respect to objective 3 is in a very preliminary stage. We have not yet begun to analyze one of these sites, of the two we have partially analyzed, we present the more detailed analysis here. Although ISNT was important in explaining yield for both of the studies we have examined, the case we present here indicates a stronger spatial relationship between ISNT corn yield response to nitrogen fertilization. This study site is located within the C. U. Williams field owned by the University of Illinois at Urbana-Champaign and located in McLean County, Illinois. The dominant soils within the field are Typic Argiudolls (Varna series), Typic Haplaquolls (Drummer), and Aquic Argiudolls (Elburn, Flannagan, and Shenoa). Our study site was within an approximately 30 ac area in the northern region of this field. The experimental design consisted of twelve blocks (240 × 400 ft2), with 8 plots (60 × 200 ft2) in each block. Within each block, one of 8 N rates was randomly assigned without replacement to each plot, with rates of 0, 40, 80, 120, 160, 160, 200, and 240 lbs/ac. A five core composite sample was collected along the center transect within each plot with a spacing of 30 ft between adjacent soil cores at depths of 0–12 in and 12–24 in and analyzed for the ISNT. The yield data were collected as described previously.
We emphasize that this is a preliminary analysis of the results, which ignores the spatial correlation in error residuals. A simple linear regression between the 96 yield and N rate observations was performed, which provided an adjusted R2 of 0.50, or 50% of the variation in yield accounted for after the fact by N rate. Note that this is a calibration as opposed to a prediction and represents N response during 2004 only. Adding the 0–12 in depth ISNT values to the regression significantly improved the results, with an adjusted R2 of 0.68 (i.e., ~68% of the variability explained after the fact by N rate and 0–12 in ISNT). Employing N rate and the 12–24 in depth ISNT values explained 66% of the variability. Since the ISNT values between depths are correlated, we subsequently used an appropriately weighted linear sum of the two in the regression procedure. This "weighted" ISNT value and N rate accounted for 70% of the variations in yield.
An obvious limitation to the preceding approach is that it assumes that yield is linearly related to N rate, which of course is not generally observed. As a first step to overcome this limitation, a linear plateau N response model was employed. We assumed that two of the three coefficients of this model (i.e., Check Plot yield and Maximum Yield) are simple functions of the weighted ISNT value. Check Plot yield was assumed to be linearly related to ISNT, and a three–parameter Mitscherlich equation was employed for Maximum Yield as a function of ISNT. As a first approximation, the slope of the linear plateau model was assumed to be independent of ISNT. This led to a 6 parameter model for describing the 96 combinations of yield, N rate, and weighted ISNT values. The resulting calibrated model explained 83% of the variability in yield, the calibration results are shown in Figure 7.
The revised sample design for the second year of temporal monitoring significantly improved our ability to delineate seasonal variations in ISNT values, suggesting that fall is the most optimal sampling period. Results support Mulvaney et al. (2004) in highlighting the importance of subsoil ISNT values relative to corn yield response to N fertilization, and indicate the possibility of reducing subsoil sampling requirements within non-manured fields via co–kriging. Additionally, work demonstrated the benefit that results from defining an appropriate soil compositing scheme and the relative poor performance of current ~10 ft scale compositing methods.
With respect to objective 3, our results are tentative in that only one field site has been analyzed in the fashion demonstrated in Figure 7, and we have not yet investigated alternative formulations of the N response function, nor other response models, and there are limitations to the approach such as neglect of correlation in error residuals and possible bias in the calibration. Further, it is not yet clear whether the Williams Field site is a typical field situation. However, these findings are extremely gratifying and remarkable, and suggest a great potential for soil based, site-specific N management.
Figure 1. Temporal variations in ISNT (ppm) in the North field at three sampling depth intervals.
Figure 2. Cropped plot, 0-12 in depth-averaged ISNT values for all 5 field sites.
Figure 4. Observed point and composite variograms and the corresponding point and composite model.
Figure 5. An illustration of alternative compositing schemes.
Figure 6. The kriging estimation error corresponding Figure to 4 and a sampling grid of 0.5 ac.
Figure 7. Linear plateau model calibrated and observed corn yield within Williams Field.
1 T.R. Ellsworth is an Associate Professor, R.L. Mulvaney is a Professor, T.J. Smith is a Visiting Research Specialist, S. A. Khan is a Research Specialist, and C.W. Boast is a Professor Emeritus, Department of Natural Resources and Environmental Sciences, University of Illinois, Urbana, IL.
Boast, C. W., T. R. Ellsworth, T. J. Smith, R. L. Mulvaney, S. A. Khan, E. M. El-Naggar, and R. G. Hoeft. 2003. Spatial and temporal variability in the Illinois N test. In: Illinois Fertilizer Conference 2003 Proceedings (R. G. Hoeft, ed.). pp. 15-19.Khan, S. A., R. L. Mulvaney, and R. G. Hoeft. 2001. A simple soil test for detecting sites that are nonresponsive to nitrogen fertilization. Soil Science Society of America Journal, 65:1751–1760.
Mulvaney, R. L., and S. A. Khan. 2001. Diffusion methods to determine different forms of nitrogen in soil hydrolysates. Soil Science Society of America Journal, 65:1284–1292.
Mulvaney, R. L., S. A. Khan, R. G. Hoeft, and H. M. Brown. 2001. A soil organic nitrogen fraction that reduces the need for nitrogen fertilization. Soil Science Society of America Journal, 65:1164–1172.
Mulvaney, R. L., S. A. Khan, R. G. Hoeft, J. J. Warren, and L. C. Gonzini. 2003. Field and laboratory evaluations of the Illinois N test. In: Illinois Fertilizer Conference 2003 Proceedings (R. G. Hoeft, ed.). pp. 3–8.
Hoeft, R. G., R. L. Mulvaney, S. Khan, E. D. Nafziger, J. J. Warren, L. C. Gonzini, T. K. Lehman, and A. Gulso. 2004. Illinois soil N test: Temporal and spatial variation and prediction of N response. In: Illinois Fertilizer Conference 2004 Proceedings (R. G. Hoeft, ed.). pp. 39–46.
Mulvaney, R. L., S. A. Khan, J. J. Warren, L. C. Gonzini, T. J. Smith, and R. G. Hoeft. 2004. Potential of the Illinois soil nitrogen test to improve nitrogen fertilizer management for corn production. In: Illinois Fertilizer Conference 2004 Proceedings (R. G. Hoeft, ed.). pp. 29–37.
