Predicting corn and soybean productivity for
Illinois soils
J.D. Garcia-Paredes, K.R. Olson *, J.M. Lang
Department of Natural Resources and Environmental Sciences, University of Illinois at Urbana-Champaign, Urbana, IL, 61801, USA
Received 15 January 2000; received in revised form 4 March 2000; accepted 20 March 2000
Abstract
Current corn and soybean productivity data is needed in Illinois for land-use planning, sustainable farm management, and accurate land appraisal. The out-of-date source of soil productivity data is Circular 1156 Soil Productivity in Illinois (Fehrenbacher et al., 1978, Soil productivity in Illinois. UIUC. College of Agriculture. COOP. EXT. SERV. Circular 1156). A new major analysis based on current Illinois farmer crop-yield data is needed to assure the availability of reliable 10-year average corn and soybean yield estimates by soils. The overall objective of this study was to update the corn and soybean yields which serve as a productivity index for Illinois soils since these two crops are grown on approximately 90% of the cropland. An approach based on multiple regression was used to evaluate the relationship between 16 selected soil properties of 34 major soils and established 1970s (1967±1976) corn and soybean yields as published in Circular 1156. Statistical models developed from major soils were tested internally by calculating the 10-year average corn and soybean yields for each of the 34 major soils. The coecients generated from multiple regression were further tested using the soil property values for the additional 165 soils identi®ed in nine counties repre-senting the crop reporting districts and weather districts in Illinois. The 10-year average crop yield trends were determined for 66 counties in the northern region and for 36 counties in the southern region for the 20-year time period between 1976 and 1995. These 20-year yield trend increases were added to the established (Circular 1156) and model predicted 1970s crop yields to estimate 1990s (1986±1995) corn and soybean yields for the average management level for all 199 Illinois soil types in nine selected counties. The 1990s crop yield estimates for the selected counties were weighted by extent of each soil type in the county and compared against 10-year county averages for the 1990s farmer reported Illinois Agricultural Statistics (IAS) corn and soybean yields. Predicted 1990s county crop yields were statistically similar to
0308-521X/00/$ - see front matter#2000 Elsevier Science Ltd. All rights reserved. P I I : S 0 3 0 8 - 5 2 1 X ( 0 0 ) 0 0 0 2 0 - 2
www.elsevier.com/locate/agsy
1990s farmer reported (IAS) county crop yields. The proposed approach to updating corn and soybean yields worked well and should be useful in surrounding states or countries.#2000 Elsevier Science Ltd. All rights reserved.
Keywords:Crop-yield prediction; Soil property models; Illinois soils; Soil types; Yield trends
1. Introduction
Crop yields are the result of environmental factors such as soil, climate, and management inputs. The eect of technology and management on crop yield is determined, in part, by the type of soil. Consequently more speci®c information on the in¯uence of soil properties on crop yields is required. Many scientists have tried to ®nd relationships between soil properties, climate, and crop yields, and grouped soils in order to compare them (Sarkar et al., 1966; Robles et al., 1977; Allgood and Gray, 1978). Most of these agronomic research studies have enhanced the impor-tance of soil depth on crop yields in a direct and indirect way (Shrader et al., 1960; De la Rosa et al., 1981; Reith et al., 1984; Thompson et al., 1991; Craft et al., 1992). Many of the soil properties considered as important for determining crop yields, have been related to moisture holding capacity (Baier and Robinson, 1968; Olson, 1981; Olson and Olson, 1986; Ulmer et al., 1988).
Dierences in crop yield and soil productivity may be represented by productivity indices. Productivity ratings are a good indicator of the suitability of soils for crop production. They are useful in determining optimum soil management and use (Anderson et al., 1938; Fehrenbacher et al., 1970). Accurate and reliable soil pro-ductivity information is desired for crop yield estimates and propro-ductivity indices of each soil type to complement land appraisal and use management. Most soil pro-ductivity data currently adopted were obtained from publication Circular 1016 Productivity of Illinois Soils (Odell and Oschwald, 1970). Much of the data were developed during a period of 1933 to 1950 and updating occurred in the 1960s. Productivity data published in 1978 in Circular 1156 Soil Productivity in Illinois (Fehrenbacher et al., 1978) were updated by numerical adjustment emerging from improved technology.
Crop yields increased signi®cantly in Illinois from 1945 to 1995. These incremental increases of yield were primarily a result of improved technology (Swanson et al., 1977) which included: (1) biological-chemical inputs such as improved varieties, mineral fertilizers, pesticides, and higher plant populations; (2) mechanical resources like machinery; and (3) management. Along with augmented crop yield trend, there were annual ¯uctuations from weather eects.
Crop yield trends were one of the main concerns in the 1970s. Several studies examined crop yield trend movement to determine if it was increasing or leveling o. Many of these studies were focused at or within a state level. Herbst (1975) made a comparison of four 5-year periods (1953±1957 to 1961±1965), (1961±1965 to 1969± 1973), for corn, soybean, wheat, and oat. Greater yield increments were identi®ed in the previous 5-year period (1953±1957 to 1961±1965) compared to the latter period (1961±1965 to 1969±1973).
A study with corn and soybean yields on the Allerton trust farm (Piatt County, IL) for 27 years (1950±1976) demonstrated that yields, unadjusted and adjusted by weather, were accelerated linearly (Swanson et al., 1977). Sonka (1978) studied corn yield trends and variability in Illinois for the period of 1927 to 1977. There were no indications that corn yield had reached a plateau despite yield variability increases since 1970. Chicoine and Scott (1988) evaluated the behavior of corn and soybean yield from 1972 to 1985 in Illinois. They found signi®cant evidence that soybean yield trends may be leveling o. However, the results for corn were inconclusive.
Productivity data published in Circular 1156Soil Productivity in Illinois (Fehren-bacher et al., 1978) were updated through previous number adjustment re¯ecting technology improvements. In 1994 a supplement to Circular 1156 was released to include new soils established between 1978 and 1994 (Olson and Lang, 1994). This supplement used 1970s (1967±1976) management to estimate crop yields. These published yields are very antiquated, more than 20 years old. Changes in crops yield and crop rotation have had an eect. Yield adjustments should be modi®ed by soil type to parallel technology innovation eects on crop production, soil productivity, and subsequent productivity indices.
The overall objective of this study was to update the corn and soybean yields which serve as a productivity index for Illinois soils since these crops are grown on 90% of the cropland. An approach based on multiple regression was used to evalu-ate the relationship between 16 selected soil properties of major soils and established 1970s (1967±1976) corn and soybean yields as published in Circular 1156. The average crop yield trend increases from 1976 to 1995 in farmer reported yields by Illinois Agricultural Statistics (IAS) for the northern and southern regions of Illinois were added to the established (published in Circular 1156) and model predicted 1970s (1967±1976) crop yields to estimate 1990s (1986±1995) corn and soybean yields for the average management level for all 199 Illinois soil types in nine selected counties. The 1990s crop yield estimates for each soil were weighted by extent of that soil type in the county and compared against 10-year county averages for the 1990s farmer reported (IAS) corn and soybean yields.
2. Materials and methods
2.1. Procedures
along with soil properties from 34 major (base and benchmark) soils; (2) to internally check by calculating average corn and soybean yields using the model generated coecients and the soil properties values for each of the 34 major soils; (3) to test coecients generated from multiple regression using the soil property values for an additional 165 soils identi®ed in nine counties representing the crop reporting districts and weather districts in Illinois; (4) to identify any corn and soybean yield outliers (greater than 2 S.D.); (5) determine the reasons for the outliers and propose modi®cation to improve the predictive models; (6) to deter-mine the magnitude of farmer reported (IAS) corn and soybean yield changes from 1976 to 1995 for northern (high productivity) and southern (lower productivity) regions; (7) to use the 20-year crop regional yield increases to predict 1990s crop yields for 199 soils in nine northern and southern Illinois counties; (8) to evalu-ate the model predicted plus 20-year trend increased crop yields and established (Circular 1156) plus 20-year trend increased crop yields for nine selected test counties (Fig. 1) by comparing with the 1990s farmer reported county crop production (IAS).
2.2. Soil types selection
Thirty-four major soil types were chosen for a model development to deter-mine crop yield estimates. These included nine base soils which were selected to represent the best producing soils under basic management which were assigned the highest basic productivity indices (PIs) in Circular 1156 (Soil Productivity in Illinois; Fehrenbacher et al., 1978). Each of these soils have extensive acreage in Illinois. From various soil survey and soil conservation programs, it was determined that a list of 30 benchmark soils represented most of the major soil conditions in the state. There are ®ve major soils on both the base and benchmark lists. This major (base and benchmark) soils list was the basis for developing a crop yield-soil property rating (CYSPR) model (Table 1).
A comprehensive list of physical and chemical properties which aect or appear to aect crop yields in Illinois were identi®ed by multiple regression and included: (1) surface layer thickness (cm); (2) surface layer percent silt; (3) percent organic matter in surface layer; (4) CEC of surface layer; (5) depth (cm) to redoxamorphic (wetness) features drainage class (relates to drainage class); (6) subsoil thickness (cm); (7) plant available water to a depth of 150 cm; (8) rooting depth as a function of soil structure (cm); (9) depth in cm to 2nd parent material (usually thickness of loess); (10) permeability; (11) surface layer pH; (12) subsoil pH; (13) surface layer bulk density; (14) subsoil bulk density; (15) percent Na on the exchange; and (16) percent clay in subsoil.
2.3. Regression analysis of Circular 1156 crop yields
Stepwise multiple regression was implemented to establish the relationship between 10-year crop yield estimates from Circular 1156 and selected soil property values. The soil properties were represented with a numeric value common for each soil property. Only one value was assigned by soil type and property for the appro-priate A horizon, B horizon or soil pro®le.
2.4. Preliminary statistical analysis
A correlation analysis was used to provide information about the nature of the variables used in the multiple regression models, and to identify which variables were more highly correlated. Simple statistical data analyses were evaluated (stem-leaf diagrams, box plot, and normal probability plot) in order to check the usual assumption in regression analysis. The diagrams for most of the predictor variables were acceptable bell-shaped curves. The variable exchangeable sodium was not a bell-shaped curve, since all but one of the soils had values of zero. The Statistical Analysis System (SAS) was applied to analyze the soil and yield data. TheR-square option was utilized with emphasis on maximizingRfor regression.
Table 1
Base (previous most productive) and benchmark (representing major soil conditions) soils in Illinois
Sable siclb Drummer siclb
Ebbert sil
Multiple regression analysis was used to provide estimates of the relationships between crop yield and soil variables. All computations were carried out with respect to the following model:
Yi01X12X2::: iXi; i 1
whereYiis the response or dependent variable, which represents the predicted crop
yields. The explanatory factorsX1,X2,. . .Xiare assumed to be independent.iis the
error due to the fact that the postulated independent variables do not completely account for the variation in Y. The parameter b0, b1 . . . bi are the population
regression coecients for the soil eects.
There are several methods to select an optimum set of independent variables. A criteria for determining how many variables to consider in the model is to use Mal-lows'Cp statistic (Freund and Littel, 1991). The Cp values are calculated with the
following formula:
Cp SSE p=MSE ÿ Nÿ2p 1 2
where MSE, the error mean square for the model; SSE(p), the error sum of squares for the subset model withpindependent variables; N, the total sample size.
If Cp>(p+1) then the model is not completely speci®ed. On the contrary, if
Cp<(p+1) the model is overspeci®ed (too many variables).
The Mallows'Cpcriterion was obtained usingCpvariable selection in SAS. TheCp
variable selection procedures was used to select a subset of predictor variables and to produce an appropriate regression equation model for each crop. Using these criter-ion, those three models with fewer predictors and withCp equal or near to (p+1)
were identi®ed as favorable. The one with the best ®t was selected as the optimum. The soil property models developed from major soils were tested by calculating the average crop yields for all the cropland soils in nine dierent counties representing the nine crop reporting districts (Fig. 1). These counties were selected to represent weather variability in Illinois. The predicted crop yields were compared to the estab-lished (pubestab-lished Circular 1156) yield estimates for each crop, and to the farmer reported crop yields in IAS. Using the adjusted values for acreage and yield, a weighted average by county was calculated and compared with predicted 1970s values and to those crop yields reported in IAS. The coecients generated by multiple regression in the models were used with the soil property values for 165 additional soils found in the nine counties. These were not included in the original 34 major (base and benchmark) soils list. The sign and magnitude of the coecients generated in the multiple regression models did not establish the absolute relationship between speci®c soil properties and crop yields since the variables were not completely independent and since multi-collinearity did not exist between some soil variables.
Circular 1156 Soil Productivity in Illinois (Fehrenbacher et al., 1978). Crop yield data were analyzed for two dierent but overlapping time periods, the ®rst being 1945 to 1995, and the second as 1976 to 1995. Yield trend was measured using the least squares method which employed the following trend equation:
Yabx 3
where Y, predicted value of Y based on the selected year; a, estimated value of YwhereX=0;b,average change inYfor each change in year.
The direct method was used to compute the equation for the linear trend line. In the direct method the point of origin is the initial year (Patchett, 1982). The simple linear trend will give the annual yield increase considering both eects of weather and technology. In this study there was no attempt to evaluate the eect of weather factors on crop yields. The eect of weather could in¯uence yields positively or negatively. However, if the hypothesis for weather trends is regarded as neutral, non-typical weather patterns could be adjusted with dummy variables as suggested by Chicoine and Scott (1988). The dummy variables considered for a model are useful to characterize outliers in critical or atypical years among yield series.
Two dierent forms of trend functions were ®tted to the yield data: (1) linear equation; and (2) quadratic equation. For those data in which larger values of the dependent variable were associated with an increase in variation, a weighted regres-sion was used. The implementation of weighted least squares regresregres-sion was one way to reduce the eect of values with great variability. They are expressed in the fol-lowing equation:
Swi Yiÿ 1X1ÿ . . . mXm2 4
where:wi,positive weights given to the individual observations.
With this equation observation, small weights have less in¯uence on the sum of squares and the estimated parameters. The contrary is true for values with larger weights (Freund and Littel, 1991).
3. Results and discussion
Simple statistics for the response and predictor variables are presented in Table 2. Excluding exchangeable sodium, since all the values except one were equal to zero, the data were widely scattered. The coecients of variation (1sX 100/x) ran from
7 to 100%. The coecients of variation for corn and soybean yields, estimated as an average level of management, were 22 and 21%, respectively.
3.1. Correlation between crop yields and soil property values
correlation coecients between the paired soil factors and crop yields were sum-marized by Garcia-Paredes (1999). In a correlation analysis the ideal situation would have been to have each of the predictors signi®cantly correlated with the dependent variable, and for the predictors to have been uncorrelated with each other. In prac-tice this situation will not likely exist since nearly all variables are correlated to some degree. In general, the variables organic matter (OM), available water (AvW), bulk density of AE horizon (DbAE), and bulk density of B horizon (DbB), thickness of AE horizon (ThAE) cation exchange capacity (CEC) and pH of AE horizons (pHAE) were signi®cantly correlated with corn and soybean yields. The highest correlation between predictor variables was for OM and CEC. Available water, pH and bulk density of AE and B horizons were also correlated with organic matter. Other signi®cant correlations were observed between available water and thickness of AE horizons. High correlations between paired variables, such as organic matter and cation exchange capacity, silt of AE horizons and available water, explain, in part, why cation exchange capacity and available water were not selected in the reduced models.
3.2. Selection of variables and regression analysis
The soil property models derived in this study were based on corn and soybean yield responses to physical and chemical soil characteristics in base and benchmark soils. Most of these soil types occur in only the northern (including central) or southern region. Therefore, the in¯uence of climate had been considered on a regional basis.
Table 2
Simple statistics for response and predictor variables for major soils
Symbol Variable Mean Standard Deviation Range CV(%)
ThAE Thickness of A & E horizons (cm) 38.6 15.1 12.7±81.3 39.0
Silt Silt in A & E horizons (%) 149.4 14.1 77.5±11.6 24.0
OM Organic matter in A & E hor (%) 3.5 1.4 06.0±1.3 39.0
CEC Cation exchange capacity (cmole/kg) 21.8 7.3 39.5±7.5 34.0
DpRx Depth to redox (cm) 59.2 42.9 68.6±152.4 73.0
ThB Thickness of B (cm) 124.9 28.3 0.0±134.6 35.0
AvW Available water to 152 cm 26.2 5.2 14.2±34.3 20.0
RtDp Rooting depth (cm) 124.9 22.8 68.6±152.4 18.0
2ndPm Depth to 2nd parent material (cm) 132.1 32.0 35.6±152.4 24.0
pHAE pH of A & E horizons 6.4 0.5 07.9±5.1 7.0
pHB pH of B horizon 6.5 0.8 08.2±4.6 12.0
DbAE Bulk density of A & E horizons (g/cc) 1.3 0.1 01.6±1.2 8.0
DbB Bulk density of B horizon (g/cc) 1.5 0.1 01.7±1.2 8.0
Na Exchangeable sodium (%) 0.7 4.3 25.0±0.0 586.0
Clay Clay of B horizon (%) 32.1 9.0 50.0±6.5 28.0
Perm Permeability of B horizon (cm) 2.4 2.4 0.08±10.2 100.0
Corn Corn yield (ton/ha) 6.6 1.47 3.32±8.53 22.0
A general examination of the predicted values of the dependent variables (crop yields) versus residuals suggested that some variables might need a curvilinear com-ponent. To identify which variable(s) should have a quadratic term partial regres-sion residual plots were used with the PARTIAL option in the model statement of the SAS program (Freund and Littell, 1991). The results of these analyses illustrated that the variable organic matter had a strong linear trend as well as a suggestion of a slight downward curved pattern, which indicated that this variable might require the addition of the quadratic term. However, the inclusion of the quadratic term in this variable did not signi®cantly improve theR2nor the predictive power of the
equa-tions. Therefore, only linear components were retained and used in the variable selection procedures.
3.3. Regression analysis of Circular 1156 crop yields
Regression models were used to identify the importance of the selected soil prop-erties in predicting common crop yields. The procedure included the following: a preliminary regression analysis was performed adopting the Cp variable selection
procedure to identify the most signi®cant variables in¯uencing crop yields, and to select three of the best regression models based on Mallow'sCpstatistic.Cpselection
procedure printed a total of `n' sets of subsets models evaluated as optimum according to the respective criterion,Cp. The models were then tested by predicting
crop yields based on soil properties for the 34 major (base and benchmark) soils. The most favorable models, identi®ed by theirR-squares, were then selected to pre-dict corn and soybean yields for 167 other soils in the nine counties. The models accounting for the highest variation in each crop yield were selected as the optimum. According to Freund and Littell (1991) an optimum subset model was one which produced the minimum error sum of squares (maximumR2).
The best multiple regression model of the soil property variables and coecients for each crop are listed in Table 3. These equations represented the most signi®cant models developed through multiple regression as identi®ed by theirR2values. Corn
and soybean crops were in¯uenced by dierent soil properties which resulted in distinct model equations. Although, some soil properties were present for each crop yield equation such as organic matter and silt in surface layer. The sign and magni-tude of the coecients generated in the multiple regression models did not establish the absolute relationship between speci®c soil properties and crop yields since the
Table 3
Soil property model for corn and soybean yields under the 1970s (1967±1976) average management level
Corn:
Yc=4.97+0.44(OM)ÿ3.134(DbB)ÿ0.051(Clay)+0.51(pHB)+0.012(RtDP)+0.042(Silt) ÿ0.157(Na)ÿ0.006(2ndPm)
Soybean:
variables were not completely independent and multi-collinearity existed between some soil parameters.
Regression equations were utilized to determine the relative importance of the 16 soil properties in estimating 1970s (1967±1976) yield for corn and soybean. Two equations were created to represent the optimum models for each crop obtained through multiple regression by using maximum R-square values (minimum error sum of squares). Organic matter was the variable accounting for the greatest con-tribution with a partialR2of 0.58 and 0.64 for corn and soybean, respectively. Silt
content of surface layer (A and E horizons) appeared in each equation accounting for a signi®cant portion of the variation, although, this variable was not signi®cantly correlated with crop yields. Bulk density of the B horizon and exchangeable sodium were highly correlated with both crop yields, and they were present in the crop models. Clay content of the B horizon was a non-signi®cant correlated variable with crop yields, however, it was an important parameter in determining variations within both models. Similarly, rooting depth and depth to second parent material were non-correlated variables with crop yields. They did explain variations in the corn yield model.
Estimated crop yields using soil property models produced high R2values when
compared to yield estimates found in Circular 1156. For the major soils, the per-centage of corn and soybean yield variation produced by the models was 90% (Figs. 2 and 3).
3.4. Outliers and studentized residuals
An observation outside of the range of the rest of the observations of a data set was known as an outlier. A quantitative value for detecting outliers was the stu-dentized residual which was obtained by dividing the residual by their standard error (Freund and Wilson, 1998). The residual and studentized residual values were obtained with PROC REG in the SAS System. The crop yields derived from the soil property models for base and benchmark soil were evaluated with a student residual limit of 2.0 to identify outliers (Table 4). The limit value for two standard deviations corresponded very closely to the student residual value of 2 and this value was used to delineate outliers. Three values for corn and two observations for soybean were outside of the speci®ed limit on major soils.
3.5. Predicted crop yield for 165 additional soils in nine counties
The two crop yield models which were developed from the 34 major soils were tested for all additional soils in nine selected counties representing the crop reporting districts. The models were tested initially on 167 other soils found in nine diverse areas of Illinois by using the equations to predict crop yields in a soil property basis. Fig. 3. The 1970s (1967±1976) established (published in Circular 1156Soil Productivity in Illinois) and model predicted soybean yield estimates for 34 major Illinois soils.
Table 4
Number of observations with residuals greater than one and two standard deviations by (S.D.)crop
Crop Major soils 165 other soils Model adjusted 165 other soils
1 S.D.a 2 S.D.a 1 S.D. 2 S.D. 1 S.D. 2 S.D.
Corn 9 (0.46) 3 (0.92) 43 (0.86) 12 (1.72) 19 (0.89) 1 (1.79)
Soybean 10 (0.15) 2 (0.30) 38 (0.29) 8 (0.58) 22 (0.31) 4 (0.62)
The coecients of determination for these soils were lower as compared to the base and benchmark soils (Table 5).
3.6. Outliers and analysis
Two outstanding outliers were identi®ed as organic soils and the soil property models substantially overestimated crop yields. Organic soils were not included in the original 34 base and benchmark soils and these soil property equations should not be used with organic soils. There were signi®cantly more outlier soil types found within the additional 165 soil group from the nine areas (Table 4 and Figs. 4 and 5). However, that was anticipated since organic soils were not used in model develop-ment.
The corn model interpreted 50% of the yield variation on the 165 soils (Table 4). The number of outliers represented 7% of the total observation, which meant that 93% of the values were within the range of 2 standard deviation (S.D.). These values were close to the 95% that would have characterized a normal population
Fig. 4. The 1970s (1967±1976) established (published in Circular 1156Soil Productivity in Illinois) and model prfedicted corn yield estimates for 165 additional Illinois soil types.
Table 5
R-square for selected soil property models using average values
Crop Major soils 165 other soils Model adjusted 165 other soils
Corn 0.90 0.50 0.59
distribution. The soil property equation for soybean resulted in 47 percent of the yield variation (Table 4). From a total of 165 observations, 95% of the observations were within the range of2 S.D.
More outliers were identi®ed with the supplementary soil group from the nine selected counties. Some non-conforming observations were a result of soils with depressional features having drainage problems which were only partially correct-able by surface and subsurface drainage and outlets. Depth to redoxamorphic fea-tures was a variable present in the 16 soil properties evaluated, although, this variable was not chosen in the ®nal equations. The high collinearity between the organic matter selected as the ®rst factor in both yield models and depth to redox-amorphic features resulted in its exclusion.
Based on the pattern of soils among the outliers, many ®t into two distinct groups: poorly drained (including very poorly drained and poorly drained), and soils with rock fragments (coarse fragments greater than 5% in the subsoil). Adjustments (between 5 and 25%) in crop yields were made to re¯ect the impact of these two soil properties.
improved theR2for the 165 soils in nine selected counties to 0.59 for corn, but was
lowered to 0.43 for soybeans (Table 5). The number of outliers (more than 2 S.D.) for corn was reduced from 12 to 1 and for soybeans from 8 to 4 (Table 4).
3.7. Comparison of established 1970s crop yields
There was an attempt to see if the model predicted 10-year crop yield values were consistent with the crop yields obtained by Illinois farmers (IAS) in the 1970s (1967± 1976). The average predicted 1970s corn and soybean yields for all the cropland in the nine counties were estimated and compared to both the farmer reported (IAS) corn and soybean yields and the established 1970s crop yields (published in Circular 1156). These crop yields were estimated from 10-year county averages (1967±1976). Weighted averages were estimated by soil type and county. These county crop yield averages were compared against the predicted 1970s values and the reported county crop yields in IAS. The analysis of variance (anova) showed no evidence to reject the null hypothesis for corn and soybean yields, which meant that yields were similar when compared to the other two data sources. Mean comparisons for 1970s crop yield averages for the nine counties are presented in Table 6.
At a nine county level, the soil property models predicted an average yield within 4±14% of the 1970s farmer reported (IAS) yields for corn and soybean (Table 6). Yield estimates for corn in Circular 1156 were higher than those farmer reported (IAS) corn estimates. However, these dierences were non-signi®cant (Table 6). For soybean, 1970s yields for all the three sources were relatively close to each other and statistically similar (Table 6). In general, in all nine test counties the farmer reported yields in IAS were lower than both the predicted yields by the soil property models and the established crop yields (published in Circular 1156). These crop models were developed to predict the 10-year average yields and not the yields for individual years which are obviously weather depend.
3.8. Crop yield trends in Illinois 1945±1995
From 1945 through 1995 crop yields have increased signi®cantly in Illinois. These increments in crop yields were probably the results of using advanced technology (Swanson et al., 1977) including: (1) biological±chemical inputs such as improved
Table 6
Mean comparisons for nine county of 1970s (1967±1976) crop yields from Circular 1156, model predic-tion, and Illinois Agricultural Statistics Sources
Source Corn Soybean
Ton/ha % Ton/ha %
Circular 1156 6.21 aa 114 2.08 a 105
Model 6.15a 113 2.08 a 104
IL Ag. Stat. 5.46 a 100 2.02 a 100
varieties, mineral fertilizers, pesticides, and higher plant populations; (2) mechanical resources like machinery; and (3) management. To oset the eect of non-typical weather, a rolling 5-year average yield trend was estimated and evaluated with a linear regression model for each crop. Along with the upward trend in crop yields there were annual ¯uctuations because of weather factors. The rolling 5-year avera-ges were used to reduce the variation as a result of weather (rainfall and tempera-ture) in individual years. Figs. 6 and 7 show the crop yield trends in Illinois using rolling 5-year averages from the period 1945±1995. It could be argued that a quad-ratic model or polynomial model should be used to represent the corn yield-trend for this time period (Fig. 6), however, the quadratic coecient was very small and the dierences in r-square were not statistically signi®cant. Therefore, the linear model was used to represent the 1945±1995 crop yield-trend. The average annual yield increment for corn was 115 kg haÿ1, with anr-square of 0.97. For soybean the
linear function was found to better represent the yield trend for this time period (Fig. 7). The average annual yield increment for this time period was 27 kg haÿ1and
the linear model had an r-square of 0.99. Coecients for corn and soybean were smaller in this study than those estimated by Fehrenbacher et al. (1978) in Circular 1156Soil Productivity in Illinois, for the period from 1939±1975.
3.9. Crop yield trends in northern and southern regions in Illinois
An evaluation of regional crop yield trends between 1976 and 1995 was made of farmer reported (IAS) corn and soybean yields (Illinois Agricultural Sta, 1945±1995). A relationship of crop yield versus time (years) was established using regression analysis. The yield response dierences were determined for the 66 counties
in northern region (high productivity) and for the 36 counties in the southern region (low productivity). Garcia-Paredes (1999) reported the test coecients in both regions. None of the coecients were signi®cantly dierent which suggested that the yield increment increase was statistically the same in both northern and southern regions.
The yield trend equations are summarized in Table 7. Regression models were sig-ni®cant for corn only in the southern region. Soybean was sigsig-ni®cant in both regions. In general, both crops showed a higher yield trend in the southern region as compared to the northern region. The yearly increases in corn and soybean yields have declined from the rates provided in Circular 1156 for the time period 1939±1975.
3.10. Comparison of 1990s crop yields
The regional yield trend equations, generated from a 20-year time period, for each crop were used to estimate the yield data for the 1990s (1986±1995) for each soil Fig. 7. Soybean yield trends in Illinois from 1945 to 1995 using a moving 5-year average of 4 previous years plus the numbered year).
Table 7
Yield trend equations for northern and southern Illinois soil regions 1976±1995
Crop Northern Southern
Corn 6.91+0.078(x)a 5.21+0.098(x)
Soybean 2.41+0.026(x) 1.78+0.032(x)
type. Using these equations plus the 1970s (1967±1976) county crop yields, both model predicted and established (published in Circular 1156), a comparison was made with the 10-year farmer reported (IAS) county crop yield averages for the 1990s (Illinois Agricultural Sta, 1945±1995). These models were developed to pre-dict the 10-year average yields and not the yields for individual years which are obviously more weather aected.
From a statistical perspective, the results of this comparison were similar to those for the 1970s county crop yields. In most of the counties farmer reported crop yields were lower than model predicted plus 20-year trend increased yields and established (Circular 1156) plus 20-year trend increased yields, however, these dierences were non-signi®cant (Table 8). As in the 1970s county comparisons, model predicted plus 20-year trend increased crop yields were closer to 1990s farmer reported yields (IAS) than those established (published in Circular 1156) plus 20-year trend increased yields (Table 8).
4. Summary and conclusions
The predictive value of 16 chemical and physical soil properties were tested using 34 major soils which represented most of the major soil conditions in Illinois. Soil prop-erty model equations were developed using the established (Circular 1156) 1970s (1967±1976) yields for corn and soybean for the 34 major soils. These models were validated by testing on 165 additional soils from nine selected counties in Illinois.
The equation for corn explained 50% of the yield variation. The model for soy-bean explained 47% of the yield variation. The model predicted crop yields for nine counties were compared to the established (published in Circular 1156) 1970s (1967± 1976) average yield estimates for each crop, and to the 1970s farmer reported crop yields reported in IAS at a county level. The soil property models predicted an average yield between 4 and 14% of the farmer reported (IAS) yields for corn and soybean.
In this study, crop yield trends behavior was evaluated using yields from 1976± 1995. The relationships between farmer reported (IAS) crop yields and time (years) were established from 1976±1995 using regression analysis. Crop yield trends were estimated for 66 counties in the northern region and for 36 counties in the southern
Table 8
Mean comparisons for the nine county of 1990s (1986±1995) Illinois crop yields from Circular 1156, model prediction, and Illinois Agricultural Statistics sources
Source Corn Soybean
Ton/ha % Ton/ha %
Circular 1156 7.96 aa 108 2.69 a 106
Model 7.90 a 107 2.69 a 106
IL Ag. Statistics 7.40 a 100 2.53 a 100
region for the time period between 1976 and 1995. In general, both crops showed a higher yield trend in southern region as compared to the northern region.
The total projected yield increase for the 20-year period, was added to both the model predicted and established (Circular 1156) 1970s (1967±1976) crop yields for all soils in the nine counties. These 1990s county crop yields were compared against 10-year farmer reported (IAS) averages for the 1990s. Model predicted plus 20-year trend increased crop yields for the nine counties were closer to 1990s farmer repor-ted (IAS) county crop yields than those established (Circular 1156) plus 20-year trend increased crop yields for the nine counties. The soil property models plus 20-year trend increased county crop yields were within 5±7% of the 1990s farmer reported (IAS) county yields for corn and soybean.
Our multiple regression model approach to updating the corn and soybean yields, which serve as a productivity index for Illinois soils, worked well and should be useful in other surrounding states or countries.
Acknowledgements
Published with approval of the Director, Illinois Agric. Exp. Sta. Urbana, IL. Funding for the study was provided by the Illinois Department of Revenue and National Council of Science and Technology (CONACYT) and the Autonomous University of Nayarit in Mexico.
References
Allgood, F.P., Gray, F., 1978. Utilization of soil characteristics in computing productivity ratings of Oklahoma soils. Soil Sci. 125, 359±366.
Anderson, A., Nelson, A.P., Hayes, F.A., Wood, I.D., 1938. A proposed method for classifying and evaluating soils on the basis of productivity and use suitabilities. Univ. of Nebraska. Agric. Exp. Sta. Research Bulletin 98.
Baier, W., Robertson, G.W., 1968. The performance of soil moisture estimates as compared with the direct use of climatological data for estimating crop yields. Agr. Meteorol 5, 17±31.
Craft, E.M., Cruse, R.M., Miller, G.A., 1992. Soil erosion eects on corn yields assessed by potential yield index model. Soil Sci. Soc. Am. J. 56, 878±883.
Chicoine, D.L., Scott, J.T., Jr 1988. Illinois corn and soybean yields: a comparative trend analysis, 1972± 1985. College of Agriculture. Dept. of Agric. Economics. University of Illinois at Urbana-Champaign. AE4631.
De la Rosa, D., Cardona, F., Almorza, J., 1981. Crop yield prediction based on properties of soils in Sevilla, Spain. Geoderma 25, 267±274.
Fehrenbacher, J.B., Pope, R.A., Jansen, I.J., J.D. Alexander, J.D., Oschwald, W.R., 1970. Productivity indexes of soil associations in Illinois. UIUC. Dept. of Agronomy. College of Agriculture. Coop. Ext. Serv. Circular 1041.
Fehrenbacher, J.B., Pope, R.A., Jansen, I.J., Alexander, J.D., Ray, B.W., 1978. Soil productivity in Illi-nois. UIUC. College of Agriculture. Coop. Ext. Serv. Circular 1156.
Freund, R.J., Wilson, W.J., 1998. Regression Analysis: Statistical Modeling of a Response Variable, Academic Press, New York, NY.
Garcia-Paredes, J.D., 1999. Predicting crop yields and productivity indices for Illinois soils. Ph.D. Thesis. University of Illinois Library, Urbana, IL. 122 p.
Herbst, J.H., 1975. Twenty-year trends in crop yields: corn, soybeans, wheat, and oats. Illinois Research No. 3. 17:8-9. UIUC. College of Agriculture.
Illinois Agriculture Sta, 1945±1995. Illinois Agricultural Statistics. Department of Agriculture, State of Illinois, Spring®eld, IL.
Odell, R.T., Oschwald, W.R., 1970. Productivity of Illinois soils. University of Illinois at Urbana-Champaign. College of Agriculture. Cooperative Extension Service. Circular 1016.
Out, S.E., Garcia, P., Pinar, M., 1987. Technological advance, weather, and crop yield behavior. North Central Journal of Agricultural Economics 9, 49±63.
Olson, G.W., 1981. Soils and the Environment: A Guide to Soil Surveys and Their Applications, Chap-man and Hall.
Olson, K.R., Olson, G.W., 1986. Use of multiple regression analysis to estimate average corn yields using selected soils and climate data. Agric. Systems 20, 105±120.
Olson, K.R., Lang, J.M., 1994. Productivity of newly established soils. Supplement to soil productivity in Illinois (Circular 1156). Agron. Mimeo Series No. 1994-11. Univ. of Illinois at Urbana-Champaign. College of Agriculture.
Patchet, S.I., 1982. Statistical methods for managers and administrators, Van Nostrand Reinhold Company, New York, NY.
Reith, J.W.S., Inkson, R.H.E., Caldwell, K.S., Simpson, W.E., Ross, J.A.M., 1984. Eect of soil type and depth on crop production. J. Agric. Sci., Cambridge 103, 377±386.
Robles, C., Gilmore, A.R., Fehrenbacher, J.B., 1977. Site productivity of oaks in relation to soil proper-ties and topography in northern Illinois. Transactions, Ill. State Acad. Sci. 70 (2), 164±172.
Sarkar, P.K., Bidwell, O.W., Marcus, L.F., 1966. Selection of characteristics for numerical classi®cation of soils. Soil Sci. Soc. Am. Proc 30, 269±272.
Shrader, W.D., Shaller, F.D., Pesek, J.T., Slusher, D.F., Riecken, F.F., 1960. Estimated crop yields on Iowa soils. Special Report No 25. Cooperative Extension Service. Iowa State University. Ames, Iowa. Sonka, S.T., 1978. Corn variability since 1927. Illinois Research Fall 1978, pp. 10±11.
Swanson, E.R. 1957. Variability of yields and income from major Illinois crops 1927±1953. Bulletin 610. Agricultural Experiment Station. University of Illinois.
Swanson, E.R., Smith, D.G., Nyankori, J.C., 1977. Have corn and soybean yields reached a plateau? Series E, Agricultural Economics No 77, E±17.
Thompson, A.L., Gantzer, C.J., Anderson, S.H., 1991. Topsoil depth, fertility, water management, and weather in¯uences on yield. Soil Sci. Soc. Am. J. 55, 1085±1091.
