Illinois
Fertilizer Conference Proceedings
January 25-27, 1993
| Home | 1993 Index | Search |
| Home | 1993 Index | Search |
David W. Franzen and Ted R. Peck1
Soil sampling density recommendations have been adjusted over the years to reflect the economic situation of growers and the sophistication of fertilizer application equipment. In 1929, a University of Illinois circular recommended 23 surface soil samples and 5 additional subsurface samples as a basis for spot spreading limestone. At that time, all limestone was shoveled out of a railcar into a wagon and from the wagon onto the field by hand. The work required to sample the field this densely was justified by making sure that each scoopful of lime was being applied to an area in need. As fertilizer application and transfer equipment became more automated, the growers and the industry in general saw the spot application of fertilizer as an unneccessary and time consuming activity. The idea of central tendency became the basis of soil testing. Fertilizing on the basis of central tendency required only enough samples to find a field average or perhaps a median value. For years, the Illinois recommendations were eleven samples per forty acre field.
Today, growers are watching their expenses closely. Environmental pressures encourage farmers and fertilizer dealers to fertilize only those areas in need. The equipment manufacturers have over one hundred variable rate fertilizer applicators in the field, with the ability to change fertilizer rates on-the-go. The advances in mapping, digitizing, computers and positional instruments makes variable rate fertilization (VRF) a real option for fertilizer dealers. Many researchers and dealers recognize that VRF makes sense, but few have determined what sampling support is needed to give the detailed information required for the equipment to reach the goal of evening the nutrient levels in a field. Most have reached their sampling recommendation on an economic or even intuitive basis without the intensive sampling required to give their recommendations positive support.
Current VRF sampling is based on either soil type based soil sampling, or sampling
in a 2.5 to 5 acre grid. In soil type based sampling, each soil type is identified
on an SCS map, then field samples are taken in each identified soil type, averaged
for the soil type, and the fertilizer is spread using a blend unique to each
soil type. The other soil test based system has used a regular soil test grid
of between 2.5 and 5 acres in sample area size, which is then used to generate
gridded, kriged and contour maps for applicator use. The purpose of this study
has been to intensively study two forty acre fields, testing which sampling
methods might best represent the real field variability, enabling VRF to be
used to its greatest advantage.
Field 1 is located southwest of Mansfield, Illinois in Piatt county. Most of the field history is known since 1961, when it was first soil sampled in a regular 5-rod (80 foot) grid. The soil type map is shown in Figure 1. In 1963, the south half of the farm received 2 ton/acre of limestone. In 1974, the entire field received 2 ton/acre of limestone. In 1977, the south half received an additional 2 ton/acre of limestone. In 1982, the north half of the field received 3 1/2 ton/acre of limestone. Despite the general liming, grid sampling in the 5-rod grid in 1976, 1982 and 1988 all showed areas in the field which were low in pH and also showed that considerable acreage had been limed for which there was no real need (Figures 2-4).
The current study began in 1989. Soil samples were taken in a measured 5-rod grid during the fall of 1989, 1990, 1991 and 1992. The field was in corn in 1989 and 1991. Soybeans were grown in 1990 and 1992. When the field was in corn, whole plant samples were taken at the 5 leaf stage and leaf samples were taken from the leaves opposite and below the ear at early tassle stage. In 1992, whole soybean plants were taken at the 5 true leaf stage and the third developed leaf from the plant top taken at the early pod stage. Yields were taken in 1991 and 1992 from each plot.
After the 1991 growing season, 4 ton of limestone was made to two areas in the western half of the field. The areas were defined by our soil testing results in the fall of 1991. Flagging outlined the areas for the lime applicator. The lime was applied to each acre of the flagged off area first at 2 ton/acre, then two equal applications of 1 ton/acre to make sure that the spreading was performed by the Agchem Terragator as evenly as possible for the study. The lime was chiseled in within 24 hours by the cooperating grower.
Field 2 is located northwest of Thomasboro, Illinois in Champaign county at the site of an old US Air Force radar installation. This field was not farmed with field crops from 1940 through 1982. When first sampled in 1982, the field was covered with a sparse collection of native grasses. Early aerial photographs suggest that native grass hay may have been removed from the farm during the Air Force period. Still standing at the southeast corner of the site is an old brick building and an antennae. At the very west border of the farm is a cottonwood tree windbreak. Between north and south rows 6 and 7 is another cottonwood windbreak. The farm has been planted in a north and south direction since 1982. Phosphate and potassium fertilizer was applied north and south from 1982 to 1987. The SCS Champaign county soil survey map of the farm is incorrect. The SCS map is reproduced in Figure 5. A more correct soil survey map, based on soil test results and soil type features is shown in Figure 6 .
The Thomasboro field was sampled in a 5-rod grid in 1982,1986,1987 and 1988 prior to this study. When sampling, the distance between each plot was stepped off, not measured. By 1988, it was clear that errors in position were causing poor correlation of sample test results of several plots between years. Therefore, in the soil sampling during the 1989-1992 study years, the distances were measured with a tape as soon as planting was finished and each plot flagged so that the soil and plant sampling would have the same positional value.
Thomasboro was cropped to corn in each year of the study from 1983 through 1992. Whole plant samples were taken at the 51eaf stage, and leaf samples taken opposite and below the ear at early tassle in each year from 1989-1992. Yields were taken in 1989 and 1992. Wet weather in 1990 and a severe drought in 1991 made yield data from those years unusable.
At Thomasboro, where poor correlation in some plots was caused by stepping off rather than measuring distances between plots,we speculated that there was an abrupt border between greatly different soil test values somewhere within these plots, rather than the smooth gradient often assumed. Some of these plots were even more densely sampled to discover the spatial patterns within these border plots. Four different plots were identified, 4-4,4-8,6-12 and 15-4 as border plots. With the center of the identified plot located, samples were taken within a 190 foot square surrounding the plot at 10 foot intervals. Plot number 6-12 was different in that only an 80 feet wide by 110 feet long area was measured. Eighty-eight samples were taken from plot 6-12, while 361 samples were taken from the other three areas.
In 1992, Mansfield and Thomasboro were also soil sampled in a 6X6 grid. The soil test estimates from the maps generated by this grid pattern and a 4X4 possible grid pattern taken from the original data set were compared to the actual test results from the 16X16 grid data sets. The soil survey map at Thomasboro was also superimposed over the original sampling data for P and K. The sample test results within each soil type area were averaged, and the resulting predicted points compared to the original grid.
In each year of the study, soil samples were analyzed for soil pH, Bray P1, and available K using the 1 N ammonium acetate at pH 7 method. Plant samples were microwave digested in a 2:1 ratio of nitric to hydrochloric acids. The digest was then diluted and analyzed for total P,K,S,Ca,Mg,Zn,Fe and Cu using an ICP. Yields obtained from the fields were correlated with soil and plant nutrient levels. Soil test results were correlated with plant nutrient levels.
The spatial variability of the fields and the subplots was studied using two
different geostatistical packages, Surfer and GS +. Surfer is a kriging program
which assumes a linear variogram to develop the kriging estimates. It is a program
which does not care how many samples are taken from a field to develop a map,
is easy to use, and is incorporated into software used to make the maps used
by field applicators today. GS + is a more sophisticated program and has the
ability to determine the best-fit of several possible models for the variogram.
It was used to develop maps made with best-fit models which were then compared
with the Surfer maps which used a forced linear model.
The 1991 pH map for Mansfield, with the area receiving lime outlined by a dark line,is shown in Figure 7. The kriged maps for the Mansfield location in 1992 are shown in Figures 8 through 16. The maps for Thomasboro are displayed in Figures 17 through 25. Included are the 16X16 grid pH, P1 and K maps, and also the associated 6X6 and 4X4 maps for each soil test. When comparing the pH maps from 1992 to those of previous years, the striking point is that following the spot lime treatment of 1991,the 1992 pH map in Figure 8 is the first map which shows correction of the fields' pH problem known since 1961.
To evaluate the appropriateness of different sampling densities in best describing the variability shown with the 16X16 grid, three different methods were used. The first method used was to run a simple regression analysis on the 16X16 predicted points of each sampling density with the actual 16X16 soil test points. The closer to a value of 1.0 the r2 is, the closer each predicted point in a less dense sampling scheme is to the 16X16 grid. Table 1 displays the r2 values from Mansfield and Thomasboro in 1991 and 1992.
To examine the relevance of the correlation to a real world fertilizer recommendation, the soil test levels were combined into ranges which correspond to different lime, phosphate and potassium fertilizer rates. The pH levels were divided into seven ranges, corresponding to different lime rate recommendations: >7.0, 6.5-7.0, 6.2-6.4, 5.9-6.1, 5.6-5.8, 5.3-5.5, and < 5.2. Pl test levels were grouped into the following ranges: > 80, 50-80, 40-49, 30-39, 20-29, and < 20 and K levels were grouped as follows: > 600, 300-600, 270-299, 240-269, 210-239, and < 210. Each of the 4X4 or 6X6 maps so rated were evaluated as follows. If a point on the 4X4 map had a pH of 6.2, and that same point on the original grid showed 6.3, the rating would be 0, meaning that the predicted point was in the same range as the actual point. If the point on the original grid had been 7.8, then the 4X4 estimate would be two ranges less than actual, or -2 levels. The rating scale was also applied to the soil type sampling scheme of Thomasboro. The soil type map was superimposed over the 16X16 grid. Soil test results within each soil type were averaged, and that value rated against the 16X16 sampling grid for each of the 253 data points.
The rating of the maps so evaluated is displayed in Table 2. The differences between the 4X4 and 6X6 grids are not clear in this evaluation. In some of the comparisons of Mansfield P1 and Mansfield K, the 4X4 grid is a little better than the 6X6. However, in the Thomasboro P1 and K, and Mansfield pH, the 6X6 is a little better. The 6X6 grid at Thomasboro is better than the soil type sampling for P and K.
The final evaluation of the sampling patterns is a visual comparison of the 4X4 and 6X6 maps with the 16X16 maps. In a practical sense, being able to pick out soil test differences in a field that reflect the actual variations is the goal of any sampling method. Looking first at Figure 8, the 16X16 map of Mansfield pH, 1992, the map contains two separate high pH areas at the north end, and a large high pH area at the east central area of the field. It also reveals a large area in the central, western and southern parts of the field where the pH's range from 6.0-6.7. The 6X6 map in Figure 9 does a good job not only of identifying these areas, but in defining approximate boundaries. The 4X4 grid map in Figure 10, however, overestimates the high pH area in the northwest area of the field, completely misses the northeast high pH area, minimizes and mislocates the eastern high pH area.
The Mansfield P 1 levels appear in the 16X16 grid in Figure 11. The P 1 map features an area in the northwest with levels in the 30 lb/A range. There is also a similar area in the northeast. Another feature of the northeast is a large area of high phosphate where the old building site used to be many years ago. The rest of the field is above a Pl test of 501bs/acre except for an area in the southwest and the southeast where the P1 test is between 40 and 50 lbs/acre. The 6X6 PI map in Figure 12 contains all of these features, but underestimates the low P area in the southeast. The 4X4 map in Figure 13 identifies the low P area in the southeast and the northwest, but misses the other three important areas of the field.
The Mansfield K 16X16 map in Figure 14 contains only two important features. Most of the samples fall into a narrow range of values between a K test of 250 and 3201bs/acre, except for the old building area in the northeast part of the field. The 6X6 K map in Figure 15 identifies the general K levels of the field, including the high K levels in the building lot. The 4X4 K map in Figure 16 does not identify the high K area at all.
The Thomasboro 1992 pH 16X16 map in Figure 17 shows areas of high pH in the south to central areas and a similar area in the northeast and the southwest. There is also an area of relatively low pH in the northwest and the east. The 6X6 pH map in Figure 18 underestimates the extent of the southern high pH area and overestimates the areas in the central and northeastern areas. However, it succeeds in showing all of these areas even if the actual boundaries are not correct. The 6X6 map also approximates the low pH boundaries in the northwest and east. The 4X4 pH map in Figure 19 approximates the low pH area in the northwest, and identifies the high pH areas in the central and southern areas. However, the high pH area in the south is underestimated and the area in the center is overestimated. The high pH area in the northeast and the low pH area in the east are not identified.
The 16X16 P l Thomasboro 1992 map in Figure 20 shows relatively low P 1 levels in the northwest and also an area ranging from the southern edge, north through the central area of the field. There is also a low P1 area in the east. A high P1 area is in the southeast around the building site, and in the northeast and west sides of the field. The 6X6 P1 map in Figure 21 underestimates the low P area in the northwest, but identifies the other areas of interest in the field. The 4X4 map in Figure 22 overestimates the areas of high P in the southeast, northeast and west. The 4X4 map does not identify the low P areas in the northwest or east.
The 16X16 Thomasboro K 1992 map in Figure 23 contains a high K area in the southeast and the northeast. It also shows a low K area in the south center which continues north to about the center of the field. There is also a low K area in the far east. The 6X6 grid in Figure 24 identifies all of the important features of the K map. The 4X4 map in Figure 25 identifies the high K areas in the northeast and southwest, but does not identify the low K areas of the south or the east. It also shows a large area in the northwest between a K test of 270 - 300 lbs/acre, when most of the samples in the area are well over 300 lbs/acre.
After evaluating all of the sampling grids, the 6X6 grid pattern is superior not only in identifying important features of each field, but also approximates the boundaries of the features better than the 4X4 grids.
The Thomasboro 1992 yield map is displayed in Figure 26. The Mansfield 1991 corn yield map is shown on Figure 27, and the 1992 soybean yield map is displayed in Figure 28. A comparison of yields in the limed portion of the Mansfield field to the yields in the unlimed portion is shown in Table 3. The blacked out area in the Mansfield 1991 yield map is set-aside acres. The blacked out areas in the 1992 Mansfield map are drowned out or severely water damaged areas that were not harvested. Yield comparisons between the two years were made only with the plots that were measured in both years.
It is not possible to determine from this study what difference the liming may have made on the soybean yields at Mansfield in 1992. The yield results are displayed only to show relative yields in areas of the field.
Correlations were made between plant analysis, soil analysis and yields at both Mansfield and Thomasboro. Table 4 displays the relevant comparisons that were found. Correlations between plant S,Fe,Cu,B,A1,Na and Zn were also made, but none were found to be significant. The correlations from Mansfield in 1991 are also included to show the change from the unlimed condition to the limed condition of 1992. In 1992 at Mansfield, pH, P1 and K levels were not significant contributors to yield differences. However, at Thomasboro, pH and the P1 levels were correlated with yield. Based on the soil and plant data available, a phosphate application will probably be made at Thomasboro following harvest in 1992. The plant analysis shows that there is good correlation between soil and plant analysis, with the later leaf sampling a little better generally than the early whole plant sample date. The correlations seem to indicate that the Mn relationships may be more pH related than a possible soil Mn content problem.
Although most of the sampling points from the early sampling years when the distances were stepped off did not vary in soil test level, certain areas varied greatly. Four of these points were intensively sampled in the fall of 1991. The pH, P1 and K maps for one of these areas, plot 44, are shown in Figures 29-31. The pH map shows two abrupt circular areas of high pH. The P1 map shows streaking in a north-south direction and a wide variation between streaks. The K map shows only small differences between points.
Several things can be learned from these subplots. First, the structure of the area surrounding a sample point can be complex. If a field has been farmed and fertilized in one direction for a number of years, it would not be surprising to see some streaking of soil test values in the direction of farming, usually the row direction. It might therefore be advisable in an intensively sampled field, to take each sample not as a composite of five subsamples, as is now advised, with 1 in the center of a box and another four at the corners, but as a transect of three subsamples with one at the center and the other two several paces to the left and right of center, perpendicular to the row direction. Secondly, all soil test variables do not neccessarily vary together. In plot 4-4, the P and pH were variable, but not in the same patterns. The K was very stable across the plots. Other plots which were similarly sampled showed variability in only one or two tests. We also found that soil test values can vary either abruptly in a matter of two to three feet, or gradually over great distances.
In the course of this work, some effort was made to study maps made with the
proper variogram model compared to maps made with a linear variogram model.
To industry, using the proper variogram may not be a great issue, however, to
the statistical community it is a matter for concern. Having examined the maps
made in this study with both the best-fit model and the linear model regardless
of fit, it is evident that the maps made are not exactly the same. However,
the variation in the boundary locations of fertilizer rate changes due to model
differences is much smaller than the variation due to inadequate sample number.
If inadequate samples are taken to represent the field, having the proper model
does not seem to be an issue. In the future, as VRF and geostatistical analysis
become more common, there will be programs available which will give a best-fit
model easily and quickly and can be incorporated into an applicator program.
The type of model resulted in only small boundary changes in our study, and
the linear model still predicted the areas of fertilizer rate changes even though
it was not the best-fit. It was always a good fit.
Two forty acre fields have been sampled and analyzed through soil and plant analysis for 4 years. The Mansfield site has been characterized and the pH corrected for the first time in at least 30 years. The Thomasboro site has been characterized and will soon receive a phosphate treatment. The sampling patterns analyzed suggest that patterns less dense than a 6X6 grid in a 40 acre field may not give enough information to adequately describe a field. Sampling patterns less than 6X6 per 40 may result in over or under application of significant areas of a field as well as missing the important features that a grower is trying to correct when he subscribes to a VRF program. If a field has been inadequately sampled, it is possible that the VRF the map directs will not correct the real field variability. If a grower continues to detect variability in a field after it has been incorrectly fertilized, he may become frustrated and give up on an otherwise rewarding program.
We did not find soil type sampling adequate for developing an informative map
for the two fields. Hidden features such as fertilizer spread patterns, old
field boundaries due to merger of farms, old building and pasture locations
all confound the soil type argument. Soil type maps were never intended to be
quite as accurate as many would like them to be. A soil sampling grid has shown
itself superior to the soil type method at Thomasboro.
Table 3: Mansfield yields relative to field average before and after liming
Table 4: Correlation coefficients (r) of relavant field comparisons
1Graduate Research Assistant and Professor of Agronomy, Dept. of Agronomy, Univ. of IL
Franzen, D.W. and T.R.Peck. 1992. Spatial Variability of Soil pH,Phosphorus and Potassium Levels. IN 1992 Illinois Fertilizer Conference Proceedings, R. G. Hoeft,ed.pp 103-111.
Isaaks, E.H. and R.M. Srivastava.1989. An Introduction to Applied Geostatistics. Oxford University Press.NY.
Linsley, C. M., and F. C. Bauer. 1929. Test Your Soil For Acidity. University of Illinois College of Agriculture and Agricultural Experiment Station Circular 346.
Peck,T.R.,and S.W. Melsted. 1973. Field Sampling for Soil Testing. IN Soil Testing and Plant Analysis. SSSA, Madison,WI. pp67-75.
Peck, T.R. 1991. Variability of Soil pH, Phosphorus and Potassium Levels. pp 107-112. IN 1991 Illinois Fertilizer Conference Proceedings. R.G. Hoeft,ed.
