Don Bullock Professor of Biometry and Crop Production Phone: (217) 244-8221 E-mail: dbullock@illinois.edu

Soil fertility is a critical factor in crop production, and experience has taught us well that yield is decreased and profit is lost if soil fertility is not maintained. Soil fertility is measured and monitored by soil sampling, and excellent recommendations for the number of samples to take, the depth of the samples, the sampling patterns, etc. can be found in the Illinois Agronomy Handbook. These recommendations have been tested repeatedly and have been found to be effective and profitable for conventional or whole field management (WFM) systems. Note that I am defining WFM as a system where fertility management decisions are made on a field basis or, at least, on the basis of large portions of fields. In WFM systems, a field is sampled to provide an estimate of the fertility for the whole field.

For example, to determine how much phosphorus (P) fertilizer to apply to a field, the field would be sampled on a 2.5-acre grid, from which an average would be calculated and used to determine the rate for the entire field. If we think about this carefully, we realize that we are actually assuming-or at least behaving as if-the P fertility is uniform across that field, which would make the average a good estimate of the P fertility at every place in the field. In reality, none of us believe that assumption. We recognize that a P fertility map of a real field is not a uniform block, but rather looks more like, in which the darker the color, the higher the soil P test. In the past, we had no ability to easily apply different amounts of fertilizer to parts of fields, so WFM systems was about the best we could do.

With the advent of precision agriculture tools such as variable rate controllers, GPS, and GIS, we now have the ability to apply different amounts of fertilizer to different parts of fields, depending upon the characteristics of the different parts. This type of management is often referred to as site-specific management (SSM). It still remains to be seen whether or not this new SSM technology is profitable, but as we investigate the potential for profitability, one of the first questions that comes up is if we need to sample differently for SSM than WFM, and, in particular, if we have to sample differently from the current recommendation of a 2.5-grid pattern for WFM systems. As we will see, it does matter a great deal.

As we begin this discussion, we need to realize that fields are not uniform. For example, is a P map made from a full section (640 acres), from which we took 1,740 soil samples and found a range of soil P from 5 to 250 lbs/acre. When we group the soil samples into categories based upon the P level and then sum the number of samples in each category, we can produce a graph as shown in. One of the critical things to note about is that it but rather is truncated on the left side and stretched out on the right side due to a number of soil samples with very high P levels. This type of skewed distribution is typical of soil fertility data, and it creates an interesting problem in that there are different ways to calculate an average, and they don't all produce the same number as they would if the distribution were a nice, bell- shaped curve.

Most of us would agree that when we calculate an average, we are attempting to get a single number that is representative or typical of the data. The trick is in deciding what we mean by "typical." If we mean the value that is in the middle (i.e., there are as many samples with a smaller P value as there are samples with a larger P value), then we are calculating what is called the median, which is 34 lb P/acre for this field. If we consider the typical value to be the one that is most common, then we are calculating the mode, which is 30 lb P/acre for this field. If we sum all of the values and divide by the number summed, we are calculating what is called the mean, which is 45 lb P/acre for this field.

Note two things. First, the mean is what we generally think of when we think of the average and is commonly the value reported by soil testing labs. Second, realize that while the median, mode, and mean are ways of calculating the average and identical for a bell-shaped curve, they are not the same for skewed curves, such as we see in soil testing. To be specific, the mean is almost always the largest value, so if we use the mean, we run the risk of thinking that the field, in general, is more fertile than it really is and thus not applying fertilizer when we should. In this example, we would not apply build-up fertilizer based on the mean of 45 lb P/acre, but we would apply build-up fertilizer based on the median value of 34 lb P/acre. The obvious take-home message here is that we should make sure we understand exactly how our testing labs are calculating the average value on a soil testing report. In my opinion, the median is preferable and should be used for both WFM and SSM systems.

Another issue that often arises when talking about soil sampling is the number of samples that need to be taken. Should I take one sample per acre? Is the Illinois Agronomy Handbook recommendation of one sample per 2.5 acres acceptable? What about a sample per 3.3 or even 5 acres? How many samples do we need to get a good estimate of the field's fertility status and thus make a good management decision? How do I make the most money? In answering these questions, there is a very large difference between WFM and SSM systems, and, in general, SSM systems need more samples. Let's take them in turn and see why the differences exist.

When making a decision for WFM systems, we again need to realize that we are looking for a single estimate that best describes the entire field, and that each time we sample, we get different numbers due to natural variability in the field. This variability problem is easily handled for WFM systems if we follow the recommendation of one sample per 2.5 acres. For example, from the above 640-acre field, we get median P values of 34, 34, 33, and 34 lb P acre for the 1-, 2.5-, 3-, and 5-acre grid patterns, respectively. Note that, in this case, the 3- and 5-acre grids give good estimates, but that is because the field is so large that even on a 5-acre basis, we have a large number of samples, and we are only trying to estimate a single number. In general, you should use the 2.5-acre grid pattern to insure the accuracy of the number, although for very large fields that are managed as a homogeneous unit (as in this example), one could make a pretty good argument for sampling something greater than 2.5 acres. Note that the preceding statement is not appropriate for fields of a more typical size, such as 40 or 80 acres. The statement also assumes that you will treat each part of the field the same. That is a very important assumption, and if you treat different parts of the field differently, then the required number of samples becomes a huge issue.

Now, let's look at a case where we want to use SSM and a variable-rate applicator. In SSM systems, we have the stated goal of treating different parts of a field in different ways. We want to put fertilizer where we need it and not put fertilizer in places where it is not needed. That is an admirable goal and intuitively the right thing to do, but to do it successfully, we need good estimates of all parts of the field. In short, we need very good maps, and show maps of a simulated field from which we have pulled samples from every square foot, which would provide near-perfect knowledge or the maps made from 1-acre, 2.5-acre or 5-acre grid patterns. In this simulation, only the lightest areas of the field require fertilizer. Compare the patterns in the maps. It is obvious that while the 1-acre grid pattern looks similar to the actual fertility pattern, the similarity quickly degrades, and by the time we get out to the 2.5-acre grids, they are not very similar to the actual patterns and thus do not provide a good map.

I believe that the inability to map fields well has contributed to the lack of success of variable-rate fertilization. Our field research comparing WFM and SSM technology shows WFM with about a 2 bu/acre yield advantage for corn and a 3/4 to 1 bu/acre yield advantage for soybean. This number is much lower than one would predict from looking at the fields. It is also disheartening that, in our research, the amount of fertilizer often increased with SSM systems because of the high fertility levels in the fields. In cases where the median soil test level exceeds the cut-off level, and thus it is recommended that no fertilizer be applied for a WFM system, there are still areas of the field in need of fertilizer. That is the case above. So when we go to SSM, we are applying fertilizer, while the WFM recommendation would be for no fertilizer. In cases where the median soil test level is less than the cut-off level, SSM systems can result in a recommendation of less fertilizer being applied, because WFM systems apply fertilizer to the entire field, while SSM systems only apply it (or at least attempt to apply it) to areas that need fertilizer.

The take-home message from this is that making the application map is the weak link in SSM systems. It is not that the idea of SSM is incorrect, but rather that the task of making the application map is much more difficult and expensive than many of us had thought. This is not to say that SSM cannot be profitable, but rather that one needs to be careful when presented with a map based on soil samples. The picture may or may not represent reality, and the representation of reality is where all of the profit is found.

Back to Table of Contents