Problem Soil Sampling to Optimize Fertilizer Responses Rigas E. Karamanos Western Cooperative Fertilizers Limited BOX 2500, Calgary, AB T2P 2N1 Soil sampling is the first step in the soil testing process. There are two aspects to soil sampling, namely, number of samples taken from a field and depth of soil sampling. The sample that is going to be collected through this process is expected to represent the field from where it is taken. Representation of a field, however, is a subject that has been open to debate for a long time. The common techniques implemented today were developed when fields were either fertilized lightly or not fertilized at all, and nutrient levels were more or less homogeneously distributed within a field. Recognition of odd spots, e.g., saline, knolls, depressions, resulted in a number of variations to a commonly used sampling scheme of randomly taking samples within a field. Rules of thumb, e.g., one sample per ten acres, were introduced, resulting in further deviation from the original intent and violation of the statistical principles on which these sampling schemes were based. Depth of sampling is an integral component of the soil sampling process. Choice of sampling depth often reflects the desire for quickness, simplicity and convenience rather than the need for accuracy and representation or the result of a planning process. A quick scan of the most common sampling scheme in the prairies reveals an apparent confusion: Alberta 0-6 (www.agric.gov.ab.ca), Saskatchewan 0-12 (Karamanos 1996), Manitoba 0-6 and 6-24 (www.gov.mb.ca). Are currently employed soil-sampling techniques appropriate for optimizing fertilizer responses? Literature Review Soil sampling has been previously discussed in this forum (Roberts and Henry 1997). Cline (1944) set the principles of soil sampling over 50 years ago. On the prairies, some of the earlier work in the sixties and seventies set the basis for the sampling techniques commonly used today. The depth of soils sampling was re-visited in Saskatchewan in 1996 (Karamanos 1996). Peck and Soltanpour (1990) have suggested that field sampling is carried out in such a way so that the chemical analyses of the collected samples accurately reflect the field s true nutrient status. This concept is complicated by the fact that commonly only one sample out a field is analyzed for nutrient status. Hence, the principle of field representation has been developed. Representation of the nutrient status has customarily been associated with representation of a more tangible parameter topography; hence, over the years the belief that representative topography or average topography also mean representative fertility or average fertility, respectively, was developed. Recent work on the prairies (Jowkin and Schoenau 1995; McCann et al. 1997; Kryzanowski et al. 1999; Walley et al.
2000; Kryzanowski et al. 2000) has verified the validity of assessing soil fertility based on landscape position. Research on soil sampling for practical field fertilization is admittedly scarce. Studies on spatial variability resulting in precision farming have increased awareness of nutrient variability in the field but have not yet provided us with a practical and economical way of addressing soil sampling. The notion that a sample must be representative of the field or the nutrient status of the field begs the question: what is representative? and representative of what? Sampling approaches Keith (1991) identifies three approaches to sampling (Table 1). Table 1. Primary sampling approaches (Keith 1991). Relative Number of Relative Basis for selecting Approach Samples Bias Sampling Sites Judgemental Smallest Largest Prior history, visual Assessment of Technical judgement Systematic Larger Smaller Consistent grid or pattern Random Largest Smallest Simple random selection An advantage of random sampling is the simplicity of assumptions about the sample population; however, the number of samples required is indeed the largest, consequently resulting in a higher cost. The greatest danger in substituting random sampling with a judgemental sampling (i.e., benchmark sampling) is the difficulty of knowing or proving whether or not various assumptions are acceptable for a particular application (Borgman and Quinby 1988). Since most nutrients show some type of temporal and spatial variability, some form of randomization is desirable. Often, a combination of the three approaches is the most feasible. What does it mean knowing the statistical average of nutrients in a field? Traditional sampling schemes assume that fields are uniform; therefore, they involve the collection of randomly selected soil cores that are mixed together into one sample. The random composite soil samples thus received are subject to judgment in terms of both selection of uniform areas and the path followed to collect the sample (James and Wells 1991). However, in non-uniform fields, this approach can lead to serious deviations. Soil sampling these fields would require large number of samples in order to obtain a proper average. Then, knowledge of the average becomes an exercise in proper statistics but has limited agronomic value. The average of 40 and 60 is 50, but so is of 0 and 100. Treating these two oversimplified situations similarly simply makes no sense. What is the best size of sample? Customarily, fifteen to twenty samples are obtained from a field. Since the majority of fields sampled in western Canada are quarter sections, fifteen to sixteen fields have been interpreted as one sample per 10 acres. This obvious misinterpretation of sampling statistics has led to people obtaining 8 samples out of an 80-acre or 4 out of a 40-acre field, etc. Although there are a variety of methods to estimate the number of samples required, few soil-testing users realize that the sample size is dictated by the desired
accuracy. Mendenall et al. (1994) provide an estimate of choosing the sample size through the bound of error of the estimate (B) and a confidence coefficient (1-α) as follows: z a/2 X (Standard error of the estimator) = B, where z a/2 is the value of z having area α/2 to its right. Therefore, if we the bound was 0.04 (e.g., ±4% of the mean) and the confidence coefficient (1-α) = 0.95 (e.g., 19 of 20 times), α must be equal to 0.05 and α/2 is 0.025. The z-value corresponding to an area equal to 0.025 in the upper tail of the z distribution is z 0.025 = 1.96. Therefore, 1.96σ p = 0.025, or 1.96pq/n = 0.025. To solve for n we must insert a value for p and to ensure that the sample is large enough we should use p = 0.5 which will result in the highest possible solution for n, since p = q = 0.5. Thus, 1.96 (0.5)(0.5)/n = 0.025, or n = (1.96)(0.5)/0.025 = 39.2, therefore, n = (39.2) 2 = 1536 samples. Conversely, if 600 samples are taken, B = 1.96(0.5)(0.5)/600 = 0.04. The same estimate can be simply derived from (n)/n, or 600/600 = 0.04 with 95% confidence. When 20 samples are taken from a field, B = 1.96(0.5)(0.5)/20 = 0.22 with 95% confidence. Appreciation of this also allows us to appreciate that a nutrient recommendation is not written in stone. Further, doubling of samples to 40 does not necessarily double the accuracy of the sample (B = 0.155), but halving the number of samples to 10 has a greater impact (B = 0.31). Certainly taking 4 samples out of a 40- acre field is a fruitless exercise (B = 0.49). Fifteen to twenty samples per field, independent of its size still remains a viable and practical sample size. What is the Correct Depth of Soil Sampling? The guiding principle for choosing the appropriate depth of soil sampling is synonymous to the third step in the soil testing process, i.e., Correlation and Calibration. Thus the depth that provides the best correlation between either nutrient uptake (Correlation) or yield (Calibration) for a particular nutrient should be adopted. Attempting to implement one strategy (or a recipe ) to all soil sampling needs can certainly lead to many errors. Further, following guidelines contained in the literature from areas where assaying mobile nutrients is either meaningless or uncommon can add to the inconsistency of results. In western Canada, all nitrogen databases have been developed based on a 0-24 (0-60 cm) depth. Some of the earlier work by Sopper and Huang (1963) and Sopper et al. (1971) was pivotal to this depth selection. Any database that utilizes less than a 0-24 depth is either based on conversion factors developed for specific areas (Table 2) or is simply wrong. Table 2 provides some common criteria used to convert nitrate-n content from the sampled depth to a 0-24 depth. Karamanos (1996) examined the validity of the factors employed in Saskatchewan and concluded that nitrogen recommendations based on 0-12 depth samples were a reliable alternative to a 0-24 depth, whereas those based on 0-6 depth samples were erratic and tended to overestimate nitrogen requirements at high and underestimate nitrogen requirements at low soil nitrogen levels. The 0-12 sampling depth was introduced in Saskatchewan as alternative to an extremely variable 0-24 sampling depth with the understanding that information at depth was sacrificed for the benefit of sample reproducibility (Table 3).
Table 2. Conversion factors utilized on the prairies to convert nitrate-n from a given depth to a 0-24 depth. Alberta 1 Saskatchewan 2 Soil 0-6 0-12 0-24 Soil Climatic Zone 0-6 0-12 0-24 3.00 1.84 1.20 1.00 Dry Brown 3.00 1.80 1.00 Non-irrigated Brown 3.00 1.75 1.00 Fallow 1.84 1.22 1.00 Irrigated 3.03 1.67 1.00 Dark Brown 2.75 1.70 1.00 Moist Dark Brown 1.85 1.60 1.00 Black 1.80 1.20 1.00 Moist Black 2.10 1.50 1.00 Grey 2.10 1.45 1.00 1 Soil Test Recommendations for Alberta. 1988. Soil Test Technical Advisory Group, Edmonton, AB. 2 Saskatchewan Soil Fertility Sub-council. 1995. Saskatoon, SK. Table 3. Number of samples required for ± 10 lb N/acre from the true mean (Henry, personal communication). Number of samples required Producer 12 inches (30cm) 24 inches (60 cm) 36 inches (90 cm) Pederson 3 47 56 Carlson 12 23 19 Mitchell 41 172 26 Mickelson 5 40 30 Sampling for sulphur is virtually an exercise in futility based on the work of Penny et al. (1996) and the experience that the presence of gypsum crystals in a sample can create false positive results. There is consensus that the sampling depth for non-mobile nutrients (e.g., P, K, Cu, Zn) should be shallower (0-6 ) and sampling depth for deep-rooted crops (e.g., sugar beets) should be deeper (0-48 ). Often users follow some of these guidelines, however, the laboratories are unable to respond by providing recommendations because either their systems do not contain the appropriate information or there is no research to allow development of the appropriate database. Further, specialty crops, e.g., sod, potatoes etc., have sampling requirements that are different from those of mainstream crops. What is the Next Step? Obviously, precision farming provides the theoretical solution to addressing spatial variability. Two main obstacles to its implementation are the cost of carrying out grid sampling and, of course, the interpretation of the results. Current soil testing databases, which have been developed for large area averages are erroneously applied to precision farming, thus their success rate is low. Nevertheless, implementation of any sampling scheme that identifies and rectifies spatial variability must be met with a farmer s willingness as well as capability to address different fertility zones within a field. Certainly, organizing fertility information and implementing a fertility program on a landscape basis appears to be of practical value in western Canada (Walley et al. 2000; Kryzanowski et al. 2000).
References Borgman, L.E. and Quinby, W.F. 1988. Sampling for tests of hypothesis when data are correlated in space and time. In L.H. Keith (ed.) Principles of Environmental Sampling, Amer. Chem. Soc., Washington, DC. Cline, M.G. 1944. Principles of soil sampling. Soil Sci. 58: 275-288. James, J.W. and Wells, K.L. 1991. Soil sample collection and handling: Technique based on source and degree of field variability. p. 25-72 in Soil testing and plant analysis, SSSA, Madison, WI. Jowkin, V. and Schoenau, J.J. 1995. Changes in available nitrogen over a fallow season in an undulating landscape in southwestern Saskatchewan. p. 299-309 in Soils and Crops 1995, University of Saskatchewan, Saskatoon, SK. Karamanos, R.E. 1996. Depth of sampling for soil testing revisited. Proceedings 1996 Soils and Crops Workshop, University of Saskatchewan, Saskatoon, SK. Keith, L.H. 1991. Environmental sampling and analysis-a practical guide. Lewis Publ., Boca Raton, FL. Kryzanowski, L., Goddard, T., Grant, R., Martin, T. and Hunt, A. 1999. Impact of soil landscape variability o crop growth in Alberta. p. 149-153 in Western Canada Agronomy Workshop, Canadian Fertilizer Institute, Ottawa, ON. Kryzanowski, L., Hall, L., Goddard, T., Faechner, T., Hunt, A. and Grant, R. 2000. Landscape influence on soil and agronomic dynamics for precision agriculture in Alberta, Canada. p. 100-105 in Proceedings 2000 Great Plains Soil Fertility Conference, Denver, CO. McCann, B., Pennock, D., Walley, F. and Hnatowich, G. 1997. Soil landscape characteristics as related to precision farming. p. 126-130 in Western Canada Agronomy Workshop, Canadian Fertilizer Institute, Ottawa, ON. Mendenall, W., Beaver, R.J. and Beaver, B.M. 1994. Introduction to probability statistics. 10 th edition, Duxbury Press, Toronto, ON. Peck, T.R. and Soltanpour, P.N. 1991. The principles of soil testing. p. 1-9 in Soil testing and plant analysis, SSSA, Madison, WI. Penny, D.C., McKenzie, R.C., Nolan, S.C., Goddard, T.W. 1996. Use of crop yield and soil-landscape attribute maps for variable rate fertilization. p. 126-140 in Proceedings 1996 Great Plains Conference, Denver, CO. Roberts, T.L. and Henry, J.L. 1997. Soil sampling it pays to do a good job. p. 116-121 in Western Canada Agronomy Workshop, Canadian Fertilizer Institute, Ottawa, ON. Sopper, R.J. and Huang, P.M. 1963. The ffect of nitrate nitrogen in the soil profile on the response of barley to fertilizer nitrogen. Can. J. Soil Sci. 43: 350-358. Sopper, R.J., Racz, G.J. and Fehr, P.I. 1971. Nitrate nitrogen in the soil as a menas of predicting the fertilizer requirements of barley. Can. J. Soil Sci. 51: 45-49. Walley, F., Pennock, D., Solohub, M. and Hnatowich, G. 2000. Agronomic assessment of variable rate fertilization of canola and wheat in Saskatchewan, Canada. p. 52-57 in Proceedings 2000 Great Plains Soil Fertility Conference, Denver, CO.