Small-Scale Spatial Variability of Soil N and P
and Implications with Regard to Sample Core Number
D.W. Franzen and D. Berglund

North Dakota State University
Centrol Inc., Twin Valley, MN


Previous research has shown that soil N and P is variable in North Dakota fields (Franzen and Swenson, 1996; Franzen et al., 1997A; Franzen et al. 1997B). Field fertility patterns may be represented by dense grid sampling (220 foot grids or denser), by topography based sampling, or a combination of the two methods (Franzen et al. 1997B). Regardless of the method, each grid sample or topography zone sample is represented by one or more soil cores. If small-scale spatial variability is not present, one soil core would be representative, however, the higher the variability, the greater the number of cores required to represent the area.

In representing fields with a composite sample, Swenson et al. (1984) determined that twenty sample cores taken from representative areas within a field would be adequate. This type of sampling minimized the variability of sampling by recommending that sample cores not be taken in unusual areas, such as saline areas, eroded places, and areas of extreme soil texture or organic matter.

Given that the sample areas will be at least as homogeneous as a representative area of a field, the number of sample cores needed to represent an management zone or grid is probably between 1 and 20. Franzen and Swenson suggested from a limited data set taken from two fields in 1994 that perhaps 3-5 soil cores were sufficient to represent a sampling area. However, a data set taken by Centrol Inc. of Twin Valley, MN called this observation into question.

The source of small-scale close spatial variability may be either inherent through pedogenic and normal soil transformation processes or imposed through human activity. Inherent variability may be caused by micro-landscapes, while imposed variability may be caused by band applied fertilizer history, improper application of high rates of broadcast fertilizer, non-uniform manure application, or livestock activity. This study was conducted to investigate the spatial variability of soil N and P at distances less than those normally studied for variable-rate fertilizer application in order to determine the sample core number appropriate to represent grids or sampling zones, and to determine what soil sample results may really mean to producers.

Methods and Materials

In Illinois, a forty-acre field was sampled in an 80 foot grid each year from 1982 to 1992 while in continuous corn. High rates of fertilizer P and K were applied in the first four years of the study. No fertilizer was applied from 1987 through 1992. In 1986, 1987 and 1988, sample locations were located by stepping off each 80 foot grid north and south, and using row widths as a guide. Several grids varied greatly in either P, K or soil pH from year to year. In 1989 and subsequent years, the grids were located using a tape measure, and the number of grids that were greatly different decreased to about four. These four sample locations were sampled separately and intensely in 1992 to reveal the variability that was causing differences in values between years as a result of apparent small differences in sample location. These areas were sampled in an 8 foot grid. Most areas sampled were 160 feet square. Two cores, 0-6 inches in depth were taken within a few inches of each other to represent each sample location.

In North Dakota, 60 foot by 60 foot square areas within site-specific sample study fields were sampled in a 10 foot grid. Four study fields were represented. A total of 22 sites were analyzed for NO3-N and 18 sites were analyzed for P. Sample core depth was 0-24 inches for NO3-N and 0-6 inches for P.

North Dakota sample core number was evaluated using two methods. The first method was to compare the percentage of values represented by 1, 3, 5 or 8 sample core means taken in established patterns (Figure 1) compared to the number of sample cores in the whole grid that represent the whole grid means plus or minus 5%, 10%, 20%, 30% or 50% of the mean. For example, if 6 out of 32 mean values of a pattern of 3 sample cores lie between the grid mean 20% , than the pattern of 3 cores represents the grid mean 20% , 16.7% of the time.

The second method compares the percentage of values of 1 through 30 sample core means compared to the whole grid mean 20% only. Instead of a set pattern described in the first method, a random combination of values is used.

Results and Discussion

The Illinois work demonstrates the effect of imposed variability on close spatial variability. Figure 2 shows that in one area high levels of P are present every 50 feet, which approximates the width of fertilizer applicators used to apply the high levels of buildup P and K in the field. In Figure 3, soil pH in the same area is highly variable in spite of no limestone being applied during the study period. There is every indication that the small-scale variability of soil pH is inherent in this location in the field.

Imposed variability brings challenges to sampling. Some studies have focused on sampling to compensate for variability due to band applications of fertilizer. Most of these studies recommend either very high levels of sample cores, or avoiding the band applications altogether. One study suggests using a chain saw sampling device perpendicular to the band direction. The avoidance is confounded by the lingering effects of the band in subsequent years. However, problems such as overlapped edges and under-fertilized centers of application patterns could be minimized by careful application. Spreaders should be calibrated properly, and high rates should not be applied in single application unless carefully calibrated. High rates of broadcast fertilizer using a spinner spreader may be better applied by two applications of a lower rate, applying the fertilizer by splitting the centers. Manure applications should also be applied as uniformly as possible. Application of manure using turned compost helps to minimize the clumps found in more fresh manures. Rates of P applied in a band application should be low to prevent streaking.

In the North Dakota study, no streaking due to imposed variability was found at any site. In Figure 4, a sampling of four sites shows that high and low regions were clustered, but not streaked through the sites. Ranges of variability are shown in Table 1. The range of variability was very low at some locations and relatively high in others. The range of values was generally higher at sites with higher mean values.

Figure 5 shows graphs of a Valley City and a Colfax site, with 1, 3, 5, and 8 patterned means compared to a 5, 10, 20, 30 and 50% range around the mean of each site. At the Valley City site, both the 5 and 8 sample core means were within 20% (7 lb/acre NO3-N) over 80% of the time, while they were within 30% (11 lb/acre NO3-N) in 100% of comparisons. At a Colfax site with mean of 67 lb NO3-N, the patterned means of 5 and 8 sample cores were within 20% in 56.3% each of comparisons, and 30% in about 70% of comparisons.

At the Valley City site, based on the 36 sampling values, the mean of the values is 36.4. If 5 to 8 cores are taken and the mean is predicted, what does the value really mean? The original site consisted of 36 sampling points. If 20% of the mean is acceptable, then all sample values between about 29 lb/acre and 43 lb/acre are represented by the mean. There are 13 sample values (36.1% of the area) below this level and 10 values (27.8% of the area) above this range. Four out of 10 values above the range were within 1 lb/acre of the range, 5 out of 10 values were within 5 lbs/acre of the range and one value was 33 lb/acre above the range. In the sample values below the range, 5 samples were within 2 lb/acre of the range, four values were within 4 and 10 lb/acre lower, and four values were within 14 and 18 lb/acre. So even if the composite 5- 8 sample core is taken and represents the mean by an acceptable level, only about 38% of the area is actually represented by the mean. A more reasonable range of acceptability at this site is 30%. The area represented by 5- 8 cores is 100% of the comparisons between 26 lb/acre and 48 lb/acre. At this level, only three out of 36 points have a higher value than this range, and eight have values lower than this, so about 70% of the area is represented by the mean 30% at this site using 5-8 cores.

At the Colfax site, if the acceptable range is 30% (47-87 lb/acre) only 14 of 36 points are within the range (38.9%). At this site, which is located on the border of the highest and lowest values in the field, the mean value only represents a small segment of values. Some values are as much as 100 lb/acre greater than the mean and 60 lb/acre less. A significant area of this site will be under or over fertilized despite core number to represent the mean.

At other sites, however, the mean value represents the area very well. At Mandan in 1996, both sites had 100% of the sample points within 20% of the mean. Many of the sites with lower means contained most sample points within a narrow range of values.

In the comparison of random sample means consisting of 1-10 cores, the results are shown in Table 2. When the values of the areas were < 30 lb/acre, the 5-8 cores were within 20% of the mean over 80 % of the time. When the values were >30 lb/acre, 5-8 cores were within 20% only 66.9% to 71.9% of the time.

In sampling, the sites that were located in the middle of a sampling zone were more uniform than those in "sampling transition zones" (STZ's). Sites at Valley City, Mandan and Gardner tended to be zones in the middle of what were relatively homogenous areas of nutrients. The Colfax sites were areas bordering the transition from the highest N and P levels to the lowest N and P levels at the location (Figure 6). The problems associated with locating sampling locations suggests that when sampling by management zone, gathering sample cores from the interior of the zone may be more meaningful than gathering them near the boundaries. Also, if sampling by grid, it may be more likely that samples are gathered from STZ's because the sampling is not directed away from these areas as it is in more directed zone sampling.


Data from many sites were analyzed for P and NO3-N from a 10 foot grid in North Dakota and for P and soil pH in an 8 foot grid in Illinois. Variability due to both imposed and inherent sources were found. When sampling sites for site-specific application of fertilizers, more cores is generally better. One core is not acceptable, whereas 5 and especially 8 cores represents the mean most frequently. In sites with a low soil N mean (<30 lb/acre), the mean represents a large area of the sample location. However, when variability is large (>30 lb/acre) the area represented by the mean of the location is relatively smaller. A sample analysis should be viewed as not a concrete value, but a range of possible values which is small at low sample levels and large at high levels. To minimize the uncertainty in soil sampling, samples taken in a directed strategy should be gathered from the interior of the zone, and not at the edges.


Franzen, D.W. and Swenson. 1996. Soil sampling for precision farming. p 129-134. In:1995 Sugarbeet Research and Extension Reports. Vol. 26.

Franzen, D.W., V.L. Hofman, and L.J. Cihacek. 1997A. Evidence for a relationship between soil nitrate-N levels and topography and comparison of topography and grid soil sampling for soil nutrients. p. 118-120. In:1996 Sugarbeet Research and Extension Reports. Vol. 27.

Franzen, D.W., A.D. Halvorson, V. Hofman. 1997B. Variability of soil nitrate, phosphate, chloride and sulfate-S under different landscapes-a report of work in progress. NDSU Ext. Report 35.

Swenson, L.J., W.C. Dahnke, and D.D. Patterson. 1984. Sampling for soil testing. Dept. of Soil Sci. Res. Report No. 8.

1997 Sugarbeet Research and Extension Reports. Volume 28, pages 143-153