Completion of Soil Fertility Paradigms Evaluated through Collaboration On-Farm and On-Station

Amendments and Input Costs Tables 1-3 in the Appendix show the types of inputs used for each treatment at all locations during the three years of study. The types and rates of amendments varied from site to site and year to year, since recommendations were based on annual soil tests and the interpretation philosophy of each treatment. The tables also show the costs, minus delivery, for each treatment. Total fertilizer costs shown are those for lime, calcium, potassium, and zinc, but not for manure or nitrogen, which may have been applied to the whole field. Some phosphorus fertilizer of either organically approved or synthetic origin was applied, but we tried to keep P application rates the same across the two treatments, and therefore these applications are not included in the overall treatment costs. Input cost was the dominant difference between the two treatments. As Figure 1 (Appendix) shows, the average difference in input cost between the treatments averaged $10.42 per acre; with the CR approach requiring $9.61 more in fertilizer and $0.81 more in lime. These figures are based on local prices for limestone, although in parts of Iowa calcitic limestone is not locally available. Producers in those areas can expect to pay approximately $0.15 per ton per mile additional to import this lime. Results for Individual Variables In the following discussion of individual variables, the criterion for statistical significance is set at $B&A (B=0.11, or 89% confidence. Traditional confidence levels of 95% or 99% may have passed over relationships that were submerged in the $B!H (Bbackground noise $B!I (B associated with these on-farm trials. On the other hand an even more liberal standard would have resulted in numerous statistically $B!H (Breal $B!I (B differences that would only make interpretation more difficult, increasing the risk of declaring significant differences when none really existed (type 2 error). Moreover, even at $B&A (B=0.11, many relationships were significant for only one year, calling their reality into question. Soil Mineral Analysis Tables 4 and 5 in the appendix show selected soil mineral analyses for 2001 and over the whole period of the study, respectively. Because treatment effects may not have evidenced themselves immediately, data from the last year of the project are shown separately. Eighteen parameters were analyzed, and the tables show a scatter pattern of treatment responses. Cells are highlighted in yellow if the corresponding nutrient was part of the treatment regimen on that farm at some time during the study. If the alpha level for a treatment difference was less than 0.11, the probability of a greater F-statistic appears in the tables in green, and the means of the two treatments are shown in red and blue, red being used for the larger of the two numbers and blue for the smaller. The same protocol also applies to the tables that follow. When the data are examined across all sites, nine of the 11 farms showed a significant treatment difference for zinc, a frequently used nutrient in the CR treatments. Six farms showed treatment differences in sulfur, a component of both gypsum (calcium sulfate) and the zinc and potassium sulfates that were applied (Table 5, Appendix). Other parameters related to treatments are calcium, magnesium, potassium, and acidity. A number of farms showed significant treatment differences for three nutrients, but the differences occurred in some cases where nutrient was not applied to either treatment. A number of other differences appeared for nutrients other than those applied; of these, four farms showed manganese differences, two farms showed iron differences, two farms showed treatment differences in copper, and one farm showed a phosphorus difference (Table 5, Appendix). Weed Biomass Weed biomass was collected because it has been asserted that soil calcium/magnesium ratios affect certain weeds, perhaps through soil structure effects. The Appendix provides broadleaf and grassy weed biomass, with Table 6 showing 2001 results for each farm and Table 7 showing results combined over years. Data generally appear only if the difference between treatments was significant at close to the 11 percent level of significance. In 2001, three of eight farms exhibited what might be considered significant differences in grassy weed biomass, with two being greater for the SLAN treatment and one for the CR plots. This last farm (# 11) also showed a significant difference in grassy weed biomass when the data were analyzed across the years. Farm #8 also showed a significant difference over years in grass biomass, but in the other direction, i.e. SLAN>CR . Farm #1, which only participated in year one, had significantly greater broadleaf weed biomass in the SLAN treatment, as did Farm #7. Overall, though, grassy weed and broadleaf weed biomass were similar for the CR and SLAN treatments. Grain Quality Analysis Grain samples were analyzed for 15 parameters related to feed quality (Tables 8 and 9, Appendix) as proxy for grain quality. In general, grain quality was fairly uniform across the treatments, but there were farm-to-farm variations. Taken over all three years, grain from Farm #2, for example, showed significantly higher crude fiber, acid detergent fiber, and crude protein and lower crude fat in the CR treatment. Farm #9, in contrast, showed greater neutral detergent fiber, crude fat, and net energy for lactation in the CR treatment than the SLAN. There was no consistent pattern to these differences, as illustrated in the summary statistics in Tables 8 and 9. Moreover, few of the farms provided significant treatment differences in the parameters related to applied amendments. Crop Leaf Tissue Analysis Tables 10 and 11 in the Appendix provide overall and by-farm values for 12 leaf nutrients for 2001 and for the whole study, respectively. It should be noted that the overall data by farm are averages of two or three different crops in the rotation. In 2001, three of eight farms did show elevated leaf potassium in the CR treatments; in two of the three this is consistent with fertilizer rates applied, and in the third (#6) a high rate of lime in the CR may have increased potassium availability. Leaf potassium was a significant overall treatment effect in 2001 (Pr>F=0.0907), but not for the study as a whole. In three of nine farms overall and in three of eight in 2001, leaf magnesium was significantly lower in CR treatment crops; however leaf Mg was higher on one farm in 2001. Two of eight farms in 2001 and two of nine in the study overall showed elevated leaf iron in the CR treatment; however, in the same number of cases SLAN yielded higher leaf iron levels. The apparent statistical significance is in part a function of the uniformity of soil test iron readings; the average levels of leaf tissue iron in the two treatments were only 1.2 percent apart. Leaf tissue nitrogen was significantly greater in the CR treatment in 2001 (Pr>F=0.0741) but not for the study as a whole. Even though zinc was a frequently applied nutrient in the CR treatment, zinc levels were significantly higher in the CR treatment on only one farm in the overall analysis and on no farms in 2001. Particulate Organic Matter Although organic matter was not directly involved in the treatments, several parameters related to the organic fraction of the soil were measured. Particulate organic matter (POM) is a component of organic matter that is related to soil aggregation (tilth) and the potential for mineralization of nutrients into plant-available forms. Tables 12 and 13 in the Appendix show, respectively, POM-related parameters for 2001 and the study overall. There were not many treatment differences that could be considered statistically significant. Water pH was different on two farms. Soil aggregate mass, an indication of aggregate stability and thus soil structure, was greater in SLAN than CR on Farm #3 in 2001 and in the overall analysis on Farm #9, but the overall statistics for aggregate mass show no treatment effect. This is not surprising since on any given farm, crop rotations and yields were the same for the two treatments. Grain Yield Table 14 provides grain yields by farm for each year of the study. There were three site-years in which the CR treatment had higher yields (Pr>F≤0.11), and one in which the reverse was true. Farm #6 had greater yields in the CR two years out of three. This was the farm in which soil potassium and phosphorus levels were higher in the CR treatment, although levels were in the ISU $B!H (Boptimum $B!I (B range throughout the field. Also, the pH measured in 2001 was approximately 6.1 in SLAN and 6.4 in the CR experimental units, a small but statistically significant difference. Change in Soil Test Parameters Figures 2-7 in the Appendix present changes in several soil test parameters over the course of the study. The figures are based on averages across farms and thus show only general trends. Each graph separates the five farms that participated all three years ( $B!H (BCR 3 $B!I (B and $B!H (BSLAN 3 $B!I (B) and the three farms that joined the study in 2000 ( $B!H (BCR 2 $B!I (B and $B!H (BSLAN 2 $B!I (B). For each group, initial soil samples were taken prior to the first crop, either the previous fall or in the early spring. A single, bulked sample was taken for the field overall, and this sample established the initial readings for both treatments. Soil pH remained stable on the farms that participated in the study the entire three years (Fig. 2, Appendix). In the three farm fields that joined for the second year of the experiment, soil pH declined in the final year of the study to approximately 6. This is a level that may or may not affect soybean yields, depending on the pH of the subsoil. Differences in pH were a function of farm more than of treatment. Soil test zinc was clearly affected by treatment (Fig. 3, Appendix). From initial levels of 2.0 ppm (three-year fields) and 1.3 ppm (two-year fields), the CR levels climbed to 2.5 ppm after three years and to 2.2 ppm in the two-year fields, while Zn levels remained flat in the SLAN. This reflects the frequent application of Zinc sulfate in the CR treatment of most farms. No zinc was applied in the SLAN field strips. Soil test potassium in parts-per-million remained fairly constant on the five farms that participated in the study all three years (Fig 4, Appendix). Both treatments in the three farms that joined the study in 2000 exhibited somewhat lower soil test potassium prior to that cropping year, but levels increased subsequently. Most samples tested in the ISU $B!H (Boptimum $B!I (B range or greater throughout the study. At the final sampling, in 2001, CR field strips overall showed significantly greater soil test potassium than SLAN strips (Pr>F=0.0138; Table 4, Appendix). The CR soil fertility paradigm focuses on the relative base saturation of K, Ca, Mg, and H. Like absolute potassium levels, percent potassium on the exchange sites was stable for the three-year farms and increased from a low level on the two-year farms (Fig. 5, Appendix). In 2001, percent K was significantly greater in CR strips than SLAN strips (P>F=0.0065). During the experiment, average K saturation levels changed from 2.8 percent to 3.2 percent in CR strips of the three-year farms, while corresponding strips on the two-year farms went from 1.7 to 2.6 percent. At those rates of change, 8-17 years would be required to increase cation exchange potassium to the 5 percent saturation sometimes cited as the ideal according to the cation ratio approach to fertility (Bear et al., 1945). Percent calcium on the soil cation exchange was stable in the 64-66% range on the three-year farms, with a gradual increase in the CR field strips. No treatment difference was evident on the two-year farms, but percent calcium declined from 69 to about 60 percent over the period. Since the two treatments were similar in this regard, the suggestion is that pH and field conditions at the final sampling may have contributed to the low numbers. The ideal proportion for calcium has been set, variously, at 65% (Bear et al., 1945) and 65-85% (Graham, 1959). From initial levels around 25.5% (three-year fields) and 24% (two-year fields), magnesium ion saturation declined to 22-23 percent at the end of the study in both treatments. The ideal proportion for magnesium was set at 10% by Bear et al. (1945). By that standard, and at that rate of change, an additional 11 or 12 years of CR treatment would be required to meet the 10% target. However, magnesium saturation decreased steadily in both treatments during this study, probably for the same reasons that reduced calcium levels. Averaged over all farms, end-of-project magnesium levels were nearly identical in the two treatments. A longer period of observation would have been required to distinguish a treatment effect from the overall effects of liming and sampling conditions. In 2001, the two-year farms declined from the previous year approximately one percent in saturation of magnesium and eight percent in saturation of calcium. These declines were accompanied by an increase in hydrogen on the cation exchange of 8-9 percent (Fig. 8, Appendix), consistent with a reduction in soil pH in both treatments. Multivariate Analysis The pattern of treatment responses was inconsistent except for those parameters directly related to inputs applied (e.g. soil test Zn). In an effort to discern response patterns, the data were submitted to principal components analysis. The principal components approach seeks the best combination of parameters to account for observed variability. The weightings on the parameters define a vector in multidimensional space. Subsequently a second combination of the parameters is sought to define a vector orthogonal (perpendicular) to the first and best accounting for the remaining variability. A series of these eigenvectors may be generated, each accounting for a diminishing portion of the observed variability. Ideally, the first few eigenvectors will identify most of the important relationships among the data elements. A feature of principal components analysis is its intolerance of missing data. If a farm-year was missing observations, it was necessary to either withdraw those variables from the entire multivariate analysis or to eliminate that farm-year from the analysis. For this reason, the parameters were divided into two groups. The $B!H (Bblue $B!I (B parameters were present in the greatest number of farm-years. The $B!H (Bgreen $B!I (B parameters comprised a more comprehensive set of the variables measured (including $B!H (Bblue $B!I (B variables), but the data from fewer of the farm-years included all these parameters. Tables 15 and 16 in the Appendix show the first eight principal components for the blue variables and the green variables, respectively. Tables 1 and 2 below summarize the most heavily loaded parameters for the first eight eigenvectors in the blue and the green sets of variables, respectively. Among the $B!H (Bblue $B!I (B variables, the most strongly weighted variable in principal components 1 through 8 is, respectively: soil test calcium; soil pH; grain crude protein; particulate organic matter carbon; soil test sulfur; POM carbon; magnesium saturation of the soil cation exchange; and grain net energy for lactation. In Table 2 of the $B!H (Bgreen $B!I (B parameters, the most strongly weighted indicator in principal components 1 through 8 is, respectively: soil test calcium; grain crude protein; leaf iron content; soil aggregate mass; calcium saturation of the exchange; magnesium saturation of the exchange; grain dry matter content; and soil test sulfur. [Tables here in hard copy] Tables 15 and 16 in the Appendix also show the Eigenvalue for each principal component and the proportion of total variability explained by the Eigenvector. In addition, the table provides results of paired-comparison t-tests for treatment differences in each of the eight principal components. For the $B!H (Bblue $B!I (B group of parameters (Table 15), significant treatment differences in the Eigenvalue occurred for principal component 5 (with soil test sulfur being the dominant variable) and for principal component 7 (with percent magnesium saturation on the soil cation exchange being dominant). These two variables are related to the treatments. Sulfur was applied to both treatments on some farms as potassium sulfate, but it was applied at greater rates as calcium sulfate (gypsum) as part of the CR treatment on several of the farms (Tables 1-3, Appendix). Magnesium application, even as a component of lime, was avoided in the CR treatments of all farms as part of the readjustment of soil cation ratios. Similarly with the green variables shown in Table 16, principal component 8 provided a significant difference in Eigenvalues between the two treatments. The most heavily loaded parameter was once again soil test sulfur. It is true that the principal components yielding significant t-tests represent only a minor amount of the total observed variability. However, they are at least logically related to the two treatments, and probably more so than the other principal components. The first principal component of both the $B!H (Bblue $B!I (B and the $B!H (Bgreen $B!I (B parameters is most heavily weighted on soil test calcium. That could be considered a treatment effect except that soil test magnesium is also positively weighted in both these principal components; in principal component 7 of the blue variables, calcium and magnesium are oppositely weighted, consistent with a treatment effect. CR in this study attempted to reduce magnesium by increasing calcium saturation. Other Eigenvectors in both groupings seem to relate to general characteristics of soil testing, leaf analysis, grain quality, and organic matter. Consistent with this, the plots of principal component 1 versus principal component 2 (Figures 9 and 10, Appendix) show that values cluster according to farm much more than – and by year as much as – treatment. Discussion Individual Parameters The consistent treatment effects observed in this study were those directly related to the soil amendments used. Thus calcium, magnesium, sulfur, potassium and zinc were the elements associated with overall treatment differences in soil tests and/or leaf tissue tests. There were farms that did not show significant treatment differences for these nutrients, and there were farms that showed significant treatment differences for nutrients that were not included in any application. Grain analysis identified no significant treatment effects overall, although individual parameters on individual farms were significantly associated with treatment. Similarly, while there were instances of significant treatment differences in grass or broadleaf weed biomass, overall weed levels were similar both in the study overall and in 2001, the final year of the project. Neither did the soil organic matter analyses show significant treatment differences for the study as a whole or for 2001. Soil water pH (1:1 soil:water), analyzed for the study as a whole, was the only variable showing significant treatment effects. The CR approach to fertility seeks to bring cation nutrients into conformity with an ideal set of ratios. For the Midwestern soils that were part of this study, the CR treatment was designed to increase potassium, calcium, and zinc and to decrease magnesium on the soil cation exchange. Zinc was the nutrient most responsive to applications, although soil test results were not expressed as percent zinc on the cation exchange. Potassium increased for both treatments, both in absolute terms and as a percent of cation saturation, and CR was significantly higher in potassium saturation than SLAN by the end of the study. Based on the observed rate of change, 8-17 years would be required to increase cation exchange potassium to the 5 percent saturation sometimes cited as the ideal according to the cation ratio approach to fertility (Bear et al., 1945). A more lenient target of 2-5 percent was suggested by Graham (1959). By that standard, the farms that remained in the study for three years were already adequate in potassium when they began the project, and the three farms that joined the project in 2001 had adequate potassium by the end of the project. The ideal proportion for magnesium has been stated as 10% (Bear et al., 1945) or 10-20% (Graham, 1959). By the first standard, at least an additional 11-12 years of treatment would be required to meet the target in the CR treatment. Actually, since magnesium levels fell similarly in both treatments for other reasons, there is little evidence of a treatment effect on magnesium saturation during the period of the study. By the standard of Graham, magnesium saturation was already approaching acceptable levels at the end of the project; however, as mentioned, most of that change was pH related. Yields Considering statistical significance to be $B&A (B<0.11 (89% confidence), there were three instances of greater yields in the CR treatment and one of greater yield in the SLAN treatment. In none of the years, nor for the study as a whole, were there significant treatment effects on crop yield. Overall, there were small, nonsignificant, yield differences for each of the crops. These will be considered in the summary of economics. Assuming for the sake of discussion that these small yield differences were actually treatment effects and not merely random, the question as to the cause arises. One way of regarding these results is as a vindication of the cation ratio approach to soil fertility. An alternative explanation is based on the fact that the CR often applied fertilizer in cases where the SLAN approach recommended none. The SLAN recommendations were based on Iowa State University calibrations. ISU soil test categories, for example, represent $B!H (Ba decreasing probability of an economic yield response to applied nutrients. $B!I (B (ISU Extension bulletin PM-1688). $B!H (BEconomic yield $B!I (B recognizes that there may be a crop response to amendments beyond economically optimum levels – just not a profitable response. In terms of crop yields, this study does not vindicate one treatment or the other, since each paradigm accounts for the yield observations. Moreover, we do not have sufficient information to distinguish the factors that made CR yields greater on Farm #6 two of three years. Treatment strips remained the same on each farm for the duration of the experiment, so the persistent effect on yields could either be due to treatment or perhaps variations in the field. (Variability due to replications, however, was low.) Zinc, a frequent CR amendment, was never applied on Farm #6. The first yield difference appeared in the first cropping year, by which time the chief difference in treatments had been 800 lbs per acre of dolomitic (magnesium-containing) limestone versus 2,000 lbs of calcitic limestone. By the end of the 1999 cropping season, however, the field strips differed in magnesium and calcium percent saturation by only a percent or two. As Figures 9 and 10 (Appendix) show, Farm #6 scored lower than other farms in the first principal component, but this principal component bears no obvious relationship to the treatments of this study. Patterns Observed Multivariate analysis was employed to detect patterns in the data. Most of the patterns found in soil, crops, and weeds relate to farm differences, not to differences between the two treatments. These farm-to-farm differences likely resulted from both physiographic characteristics such as soils and from the cumulative effects of farm management. Some treatment-related patterns in the data did emerge, but these related to nutrients applied rather than to secondary characteristics such as other nutrients, soil organic matter, grain quality, or weeds. Treatment-related associations in the data were overshadowed in magnitude by farm-related differences.