Agronomy Journal Grow Your Career With ASA
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
 QUICK SEARCH:   [advanced]


     


This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Web of Science (4)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Ping, J. L.
Right arrow Articles by Dobermann, A.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Ping, J. L.
Right arrow Articles by Dobermann, A.
Agricola
Right arrow Articles by Ping, J. L.
Right arrow Articles by Dobermann, A.
Related Collections
Right arrow Cotton
Right arrow Crop Models
Right arrow Plant and Environment Interactions
Right arrow Production Agriculture
Right arrow Site-Specific Analysis
Published in Agron. J. 96:1588-1597 (2004).
© American Society of Agronomy
677 S. Segoe Rd., Madison, WI 53711 USA

Cotton

Identification of Relationships between Cotton Yield, Quality, and Soil Properties

J. L. Pinga, C. J. Greenb,*, K. F. Bronsonc, R. E. Zartmanb and A. Dobermanna

a Dep. of Agronomy and Horticulture, Univ. of Nebraska, Lincoln, NE 68583-0915
b Dep. of Plant and Soil Science, Texas Tech Univ., Lubbock, TX 79409-2122
c Texas Agric. Exp. Stn., RR 3, Box 219, Lubbock, TX 79403

* Corresponding author (cary.green{at}ttu.edu)

Received for publication September 8, 2003.

    ABSTRACT
 TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Intercorrelation among soil properties can result in multicollinearity problems regarding relationships between soil properties and crop yield. The objective of this study was to compare statistical methods of examining relationships between cotton (Gossypium hirsutum L.) yield, quality, and soil properties. Soil and plant samples were collected from 1-ha grids on an irrigated production cotton field in Texas from 1998 through 2000. Ordinary least square regression (OLS), partial least square regression (PLS), and principal component regression (PCR) were compared as methods for quantifying relationships between cotton yield or quality and soil properties. The PLS method eliminated multicollinearity problems and resulted in the coefficient estimations with meaningful signs compared with their associations to cotton yield and fiber quality. Furthermore, loadings from linear combinations of variables in PLS allowed identifying soil properties that had the greatest influence on yield. While PCR identified the principal components that maximized the variance of independent variables, it did not improve the modeling of crop–soil relationships. Among the selected soil and landscape properties, sand and clay content, exchangeable Ca2+ and Mg2+, NO3, Olsen-P, pH, relative elevation, and slope were important factors affecting lint yield and fiber quality. Higher lint yields were usually accompanied by higher fiber quality. Magnitudes of influence of different soil properties on yield and quality, however, varied among the 3 yr, suggesting that long-term studies are needed to establish robust relationships for site-specific management.

Abbreviations: CV, coefficient of variation • DCa, depth to caliche • DFCa, depth to free carbonate • DGPS, differential global positioning system • OLS, ordinary least square regression • PCA, principle component analysis • PCR, principle component regression • PLS, partial least square regression


    INTRODUCTION
 TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
PRECISION AGRICULTURE provides an opportunity to increase production efficiency by managing a field based on the variability of soil and other environmental conditions within the field (Pierce and Nowak, 1999). Successful application of this new technology requires an understanding of relationships between crop performance, soil characteristics, and other environmental factors. Numerous studies have indicated that soil fertility and physical properties are usually the most influential factors for cotton yield and fiber quality (Cassman et al., 1990; Morrow and Krieg, 1990; Johnson et al., 1998; Ping and Green, 1999; Bradow and Davidonis, 2000; Elms et al., 2001; Li et al., 2001). Soil properties, however, usually are correlated spatially and among themselves (intercorrelation) because of inherent soil formation processes (Mallarino et al., 1996; Mulla and McBratney, 2000).

Simple correlation and regression analyses are the most common ways to describe the relationship between two or more variables. Correlation coefficients are a measure of the degree of linear association between any two variables when other variables are fixed. The regression model can indicate the extent by which the dependent variable can be predicted by the independent variables; the strength of prediction is expressed as r2, known as variation in dependent variable explained by independent variables or coefficient of determination. The intercorrelation between soil properties does not violate the assumptions of the classical regression modeling, but can cause multicollinearity problems that may result in large variances of parameter estimates, unexplainable signs of explanatory variables, and rejection of regression coefficient estimates that are significant (Morzuch and Ruark, 1991; Neter et al., 1996; Fekedulegn et al., 2002).

The general methods to solve multicollinearity problems between independent variables include removing less important variables; combining variables via principal component analysis (PCA) or factor analysis; and using ridge regression, partial least squares (PLS), or Bayesian methods (Neter et al., 1996). Each method has its own limitations. For example, variable removal can focus on the influential variables, but could lose some information due to reduced number of independent variables (Morzuch and Ruark, 1991). The PLS is designed to construct models that provide effective prediction power (Tobias, 1999). The PCA along with regression analysis has been used in different areas and usually is referred to as principal component regression (PCR) (Martens and Naes, 1989). The PCA can simplify the variance–covariance structure of a set of variables by replacing those with a few linear combinations of the original variables. These linear combinations are uncorrelated with each other while maximizing the variances. As a result, PCA may reveal relationships that previously were not apparent, thereby allowing interpretations not described by classical statistics (Johnson and Wichern, 1999). In a soil classification study, Seeling et al. (1991) reported a better soil class delineation by using PCA. Morzuch and Ruark (1991) demonstrated the merits of PCR over variable deletion for constructing a regression model with the presence of multicollinearity between independent variables in the original data set. Stenberg (1998) indicated that PLS and PCR performed equally well in describing the relationships between soil C, N, and other soil chemical, physical, and biological properties, but PCR produced less interpretable scores and loadings of new principal components than did PLS.

Cotton, the most important crop on the Southern High Plains of Texas, requires more intensive management than do many other crops. Furthermore, cotton fiber quality is also a concern since it can affect the value of cotton and farmers' income. While multicollinearity between independent variables and spatial autocorrelation between observations can exist among soil physical, chemical, and biological properties, the determination of relationships between cotton and environmental factors can be very complex. There is a need to establish accurate relationships between cotton yield, fiber quality, and soil properties for decision-making in site-specific management system (Valco et al., 1998). The primary objective of this study was to compare statistical methods of examining relationships between cotton yield, quality, and soil properties at production level.


    MATERIALS AND METHODS
 TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
This study was conducted on a 49-ha center-pivot–irrigated cotton field (33.7324°N lat; 101.7718°W long) near New Deal, TX, from 1998 through 2000. This field was comprised of three soil types: Acuff sandy clay loam (fine-loamy, mixed, superactive, thermic Aridic Paleustolls), Amarillo fine sandy loam (fine-loamy, mixed, superactive, thermic Aridic Paleustalfs), and Olton sandy clay loam (fine, mixed, superactive, thermic Aridic Paleustolls) (Soil Survey Staff, 1979). Pendimethalin was applied at 1.8 L ha–1 and rows (81 cm) were bedded in early April each year. Cotton varieties Paymaster HS 200 (1998 and 1999) and Paymaster 2200 (2000) were seeded at approximately 156000 seeds ha–1 in mid-May of each year using a John Deere (East Moline, IL) 6100 planter. The field was fertilized with 100 kg N ha–1, 20 kg P ha–1 and 11 kg S ha–1 in 1998; 100 kg N ha–1 and 22 kg P ha–1 in 1999; and 70 kg N ha–1 and 20 kg P ha–1 in 2000.

Rainfall patterns varied during the 3 yr of this study (Fig. 1). The 1998 growing season was characterized by hot and dry weather, while intensive rainfall occurred early in the 1999 and 2000 growing seasons. Cotton in 2000 suffered from drought because of little rainfall from July through September. Additionally, the irrigation system malfunctioned during August 2000. Overall, rainfall in 1998 was much lower than the 90-yr average (Fig. 1). Rainfall in 1999 was higher than the 90-yr average, and 1999 was considered as a relatively wet year. Irrigation amounts of 165 mm water in 1998 and 1999 and 150 mm water in 2000 were applied by the LEPA system. All other agronomic inputs were made by the producer according to local university extension recommendations.



View larger version (18K):
[in this window]
[in a new window]
 
Fig. 1. Accumulative rainfall from April through October from 1998 through 2000 and 90-yr average in New Deal, TX. Notice the abrupt increase in June 2000.

 
Soil and plant samples were taken from 39, 1-ha grids (100 by 100 m), which were equally laid to each quartile of the field with one missing in the southwest quartile because of the occupation of house. Soil samples were collected at depths of 0 to 15, 15 to 30, and 30 to 61 cm by combining three subsamples from an area of about 9 m2 around the grid center. Each sampling position was georeferenced by using an Omnistar (Houston, TX) Model OS7000 DGPS Receiver differential global positioning system (DGPS). Cotton yield and yield component data, including plant number and boll number, were measured from a 4-m2 area from two rows of cotton near the soil sampling locations in each grid. Cotton was hand harvested on 9 Sept. 1998, 10 Oct. 1999, and 9 Oct. 2000. The cotton samples were ginned with a remodeled commercial gin machine, which could clean bur and cotton leaves, and the corresponding lint turnout was used to calculate lint yield.

All the soil samples were analyzed for NO3–N (Mulvaney, 1996); Olsen-P (Olsen and Sommers, 1982); organic matter (OM) (Nelson and Sommers, 1982); cation exchange capacity (CEC) (Thomas, 1982); sand, silt, and clay content (Gee and Bauder, 1986); and pH (Thomas, 1996). Depth to free carbonate and caliche were determined by soil response to 0.1 M HCl and confirmed by B.L. Allen (personal communication, 2001). Relative elevation was determined by a centimeter-accuracy DGPS device and was also used to derive slope using the Spatial Analyst in ArcView (ESRI, 1999). A small portion of lint from each yield sample was used for quality analyses at the International Textile Center, Lubbock, TX, including fiber length, strength, and micronaire (USDA, 1993).

Descriptive statistics and ordinary least square regression (OLS) analysis were computed by using PROC UNIVARIATE and PROC REG, respectively, in SAS (SAS Inst., 1999). Variable selection in regression models was based on subject considerations and statistical criteria such as adjusted r2, which tends to compensate for the optimistic trait in the coefficient of determination (r2) by taking into account the size of the sample, the number of prediction variables, and Mallows Cp values (Neter et al., 1996). Multicollinearity between independent variables was evaluated by examination of the variance inflation factor (VIF), which measured the variance inflation of the estimated regression coefficients compared to when the independent variables are not linearly related. A VIF greater or equal to 10 suggests that a multicollinearity problem exists (Neter et al., 1996). The PLS was performed using PROC PLS while PCR were performed using PROC REG with the scores from PROC PRINCOMP in SAS (SAS Inst., 1999). The loadings and scores of linear combinations of variables that accounted for most of either dependent or independent variable variations in PLS and PCR were selected to further explore the associations between the independent variables and dependent variables.


    RESULTS AND DISCUSSION
 TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Cotton lint yields had larger spatial variations than did fiber quality measurements over the 3 yr (Table 1). Average lint yields were 956, 891, and 662 kg ha–1 in 1998, 1999, and 2000, respectively, with CV values ranging from 18.3 to 24.0%. Similar CV values for cotton lint yields were reported in an irrigated field with 57 grid samples in this region (Elms et al., 2001). Micronaire had relatively larger spatial variation than did fiber length and strength. The average micronaire reading was lowest in 2000 at 3.3; this was also the year in which yields were the lowest. Reduced yields and micronaire in 2000 were attributed to drought from July through September. Lower yields in 1999 might be attributed to excessive soil moisture due to 84 mm of rainfall occurring within 5 d in June and cooler temperatures during that period. Additionally, NO3–N concentrations were <10 mg kg–1 in the northwest field area in 1999 where relative elevation was low and water ponding was observed after the intensive rainfall. This ponding may have resulted in NO3 leaching or denitrification. The northwest region was one of the lowest yield regions with yields ranging from 568 to 1114 kg ha–1 in 1999 but one of the highest yield regions with yields ranging from 1061 to 1473 kg ha–1 in 1998. The high yield in this region in 1998 may have also contributed to lower NO3 due to plant uptake of N.


View this table:
[in this window]
[in a new window]
 
Table 1. Descriptive statistics of cotton lint yield and fiber quality measurement from an irrigated cotton field near New Deal, TX, from 1998 through 2000.

 
A summary of soil physicochemical properties and landscape measurements that potentially influenced cotton yield are listed in Table 2. While the majority of the field appeared flat, elevation and slope changed largely in the southwest region of the field. Consequently, slope had the largest CV value. Slope, elevation, NO3–N, N/P ratio, depth to free carbonate, and depth to caliche exhibited greater spatial variability than did other soil properties (Table 2). A similar trend, noticed by Elms et al. (2001), was that NO3–N, Olsen-P, and Ca had larger spatial variability than K, sand, and clay content. There was a spatial pattern of increasing clay content and decreasing sand content from northwest to southeast. Although the majority of measured soil and landscape properties tended to be slightly skewed, deviations from normal distribution were not significant based on the Shapiro-Wilk's normality test (SAS Inst., 1999).


View this table:
[in this window]
[in a new window]
 
Table 2. Descriptive statistics of selected soil properties and digital elevation model from an irrigated cotton field near New Deal, TX, from 1998 through 2000.{dagger}

 
Correlation Analysis
Lint yields in 1998 and 2000 were significantly correlated to many soil properties. In 1999, however, lint yields were significantly correlated to pH only (Table 3). The differences of correlation coefficients between lint yields and soil properties among the 3 yr may indicate the strong influence of rainfall and soil N changes since these two factors varied during three growing seasons in this study. Morrow and Krieg (1990) concluded water supply and N were critical for cotton production in this area. The number of soil properties significantly correlated to fiber quality in 1998 was greater than in 2000, which in turn was greater than in 1999 (Table 3). Micronaire was negatively correlated to fiber length and strength. Negative associations of micronaire with fiber length and strength were also found by Elms et al. (2001). Micronaire in 1998 and 1999 was correlated positively to NO3, OM, CEC, Ca saturation, and clay content, but negatively to K saturation and sand content. During the 3-yr period, fiber length was correlated with soil pH and Ca saturation. In comparison, fiber strength was less correlated with soil properties during the 3 yr. Micronaire was affected more by environmental factors such as soil physical and chemical properties and landscape characteristics. Johnson et al. (1998) reported that micronaire was correlated more to soil properties than other fiber quality measurements. Their results showed a positive correlation of micronaire to soil P and OM, but a negative correlation to soil pH. Bradow et al. (2000) reported field sites higher in pH or Ca and Mg produced immature fiber with micronaire below 3.5.


View this table:
[in this window]
[in a new window]
 
Table 3. Pearson's correlation coefficients of cotton yield, fiber quality, and soil properties and digital elevation model from an irrigated cotton field near New Deal, TX, from 1998 through 2000.{dagger}

 
Lint yields were negatively correlated to fiber micronaire, but positively correlated to fiber length and strength in both fields in 1998 and 1999. Favorable growing conditions produced both higher yield and fiber quality. The overall correlation coefficients suggested that lower exchangeable Ca2+, lower pH, lower N/P ratio, higher K saturation, greater depth to free carbonate, and caliche layers were favorable for a higher yield and better fiber quality. The Pearson product-moment correlation coefficient, however, is only a measure of the strength of linear association between two variables when other variables are fixed (Neter et al., 1996). Crop growth is affected by many factors and interactions of factors.

Regression Analyses between Cotton Lint Yield and Soil Properties
The Pearson's correlation coefficients, paired plots between independent variables (lint yield and quality measurements) and soil properties, adjusted r2, and Mallows Cp were used to decide which independent variables should be included in the regression model, in which a higher adjusted r2 and lower Mallows Cp indicated a better model selection (Neter et al., 1996). The average values from 0 to 61 cm for NO3–N, pH, exchangeable Ca2+, Mg2+, and K saturation were used for the modeling because NO3–N tended to be moved under plow layer while other variables tended to make more sense for the calcareous soil. Olsen-P, sand, and clay contents at 0 to 15 cm were used for the modeling because soil P was usually measured at this depth to represent its availability for this region, whereas the sand and clay contents in the top 15 cm affected greatly on permeability and release of water from rainfall and irrigation. Availability of limited water in this region was crucial to cotton yield and fiber quality, but measured soil moisture was excluded from modeling because of unexpected rainfall influences on a small portion of samples during sampling. In-season soil water measurements were limited by the fact of a running pivot system. The CEC was excluded from the regression analysis in favor of utilizing exchangeable cations.

Table 4 presents estimators, model p values, r2, and adjusted r2 from three different regression analysis methods. The ordinary least square (OLS) method should have the best-unbiased estimations of linear relationship between the dependent and independent variables if classical assumptions are satisfied (Neter et al., 1996). The OLS results showed that lint yields were modeled better in 1998 and 2000 than in 1999. The model of lint yield in 1999 was not significant (at p = 0.05) (Table 4). Cotton lint yields in 1998 and 2000 tended to be negatively related to pH and exchangeable Ca2+, but positively related to sand and clay, relative elevation, and depth to free carbonate. The signs and magnitudes of exchangeable Mg2+, K saturation, slope, depth to caliche, and N and Olsen-P concentrations differed between 1998 and 2000. The signs and magnitudes in the OLS models may not be reliable, however, because higher VIF values appeared at exchangeable Ca2+ and texture measurements (Table 4). Simple deletion of these variables might reduce VIF, but would be at the risk of information loss because exchangeable Ca2+ and soil texture were important soil properties in this region (Morzuch and Ruark, 1991). Furthermore, some soil properties had already been excluded and remaining variables had different meanings in terms of their contributions to crop growth.


View this table:
[in this window]
[in a new window]
 
Table 4. Regression estimators between lint yield and soil properties and digital elevation model from an irrigated cotton field near New Deal, TX, from 1998 through 2000.{dagger}

 

View this table:
[in this window]
[in a new window]
 
Table 5. Model effect loading of linear combinations and their contributions to model and dependent variable from an irrigated cotton field near New Deal, TX, from 1998 through 2000.{dagger}

 

View this table:
[in this window]
[in a new window]
 
Table 6. Eigenvectors of principal components and their loadings from an irrigated cotton field near New Deal, TX, from 1998 through 2000.{dagger}

 
To reduce multicollinearity between independent variables while keeping meaningful variables, PLS and PCR were performed and resulting models had much lower p values and much higher adjusted r2 values (Table 4). The adjusted r2 values in the PLS models during the 3 yr were increased substantially compared with those obtained with OLS models. With the OLS analysis yield was negatively correlated to N in 1999 and positively correlated in 2000; however, PLS analysis indicated a positive relationship in both years. A similar observation was noted for P. As the overall P level was low in this field, the increase in P concentration should increase cotton yield. Therefore, the negative sign between P and cotton yield obtained from OLS might be wrong because the parameter estimation from this method was probably affected by the multicollinearity (Neter et al., 1996). Furthermore, the magnitudes in the PLS of clay to yield reduced, compared with those in the OLS in 1998 and 2000. The weakly positive association of clay to yield were expected since 1998 and 2000 were drier years. Therefore, PLS method tended to produce the model that showed correct signs and magnitudes in terms of the relationships between cotton yield and soil properties.

In both PLS and PCR, regressions were conducted between dependent variable and linear combinations (LC) extracted from independent variables based on eigenvalues from their original data structure. These two methods, however, had different emphases on the loadings of linear combinations. In PLS, the resultant LCs should predict the dependent variables better (Yu, 2001); whereas in PCR, the LCs mostly tended to represent the variances of independent variables (Johnson and Wichern, 1999). In either case, resultant LCs were orthogonal. The PLS and PCR, however, have been proven to reduce multicollinearity between independent variables greatly in many cases and have advantages over variable deletion (Morzuch and Ruark, 1991; Stenberg, 1998; Tobias, 1999; Ward and Cox, 2000). Furthermore, PLS is also robust against skewness and omission of regressors (Cassel et al., 1999).

The loadings of linear combinations (LC) and their proportions over the total models and variations could help explain what was measured from the extracted LCs or principal components (PCs) and how they influenced dependent variables based on their relatively magnitudes and signs in the PCs. For example, similar magnitudes but opposite signs indicated a contrast between those soil properties while smaller coefficient indicated a small contribution to the LC (Johnson and Wichern, 1999). Table 5 indicates that LC1 in 1998 was basically a measure of contrasts between K saturation and exchangeable Ca2+ and between sand and clay contents according to their signs and relative magnitudes of parameter estimations. Similarly, the LC2 in 1998 was mainly a measure of relative elevation and the LC3 measured the slope. The LC4 and LC5 basically measured pH and Olsen-P, respectively. In comparison, the first LC was very important for the model because it could explain about 42% of the total variation of lint yield and 43% of the total variation of all independent variables alone (Table 5). The first five LC values explained about 55% of the lint yield variation and 79% of variation in all the independent variables. Furthermore, the scores from first linear combination in each observation revealed that higher lint yields appeared at the observations with higher LC1 scores. In other words, higher sand content, lower clay content, lower exchangeable Ca2+, and higher K saturation tended to result in higher yields.

The loadings of LC in 1999 indicated that the first LC was a measure of NO3 and slope. This may indicate NO3 leaching or denitrification because NO3 leaching is related to slope and may have resulted in loss of NO3 in early growing period. This effect, however, was not important in 1998 when there was no heavy rainfall early in the season. The contrast between sand and clay was moved to the second LC in 1999, which indicated that soil texture was less important to lint yield in 1999, a wet year, than it was in 1998, a drier year.

In 2000, the first LC was a measure of exchangeable Ca2+, potassium saturation, depth to caliche, and NO3. The N influence was not included in LC1 for 1998. The importance of N in 2000 compared with 1998 may be attributed to the uniform N applications in 2000 being inadequate to replace the loss of N from leaching that occurred in 1999 in the relatively low and sandy soil regions. The second LC for 2000 mainly measured exchangeable Mg2+, Olsen-P, and clay content. During the drier year of 2000, lint yield was affected more by P. The third LC for 2000 measured relative elevation and slope. Therefore, from the changes of LC orders of independent variables, we could identify the influence of soil properties on lint yield under changing weather patterns during the 3 yr. This may be one advantage of using loading of linear components to determine the weather influence on yields by looking at the orders of soil properties in the linear components during the 3 yr.

The loadings of principal components reflect the measurements of each linear combination that represented the variations mostly. Table 6 lists the loadings of principal components from 1998 to 2000. Similar to PLS, the first PC is very important because it can explain a large portion of the variation of a series of independent variables. The first PC for 1998 was a measure of contrasts between exchangeable Ca2+ and K saturation and between sand and clay contents. The PC2 was a measure of relative elevation and slope while the PC3 was a measure of soil NO3 and Olsen-P. The PC4 was a measure of soil pH and PC5 measured depth to caliche and contrast between NO3 and Olsen-P.

The trend in loading changes for 1999 was similar to that with PLS as NO3 also moved to the first PC. The PC2 for 1999 was a measure of relative elevation and slope, which could closely affect the nitrate leaching after heavy rainfall during the early growing season. In 2000, loadings had similar patterns with those in 1998, except NO3 appeared before Olsen-P, possibly related to NO3 leaching that occurred in 1999. Compared with the loadings from PLS, larger influence was derived from NO3 and Olsen-P in principal component loadings. This is attributed to these two nutrients having much higher variability than other soil properties during this study (Table 1 and 2). While N may have been affected by leaching and denitrification, both N and P were the only two variables analyzed that were affected by different agronomic inputs each year. Large variation can be captured effectively by the first one or two PCs in PCA. As a result, once one or two important independent variables changed, the loadings of other variables changed greatly. In either year, the first five PCs represented >85% of the variation in soil properties (Table 5).

Based on the model p values and adjusted r2 from three regression methods in our data, the PLS tended to produce the best model to represent the relationship between cotton lint yield and soil properties. As indicated in Table 4, the p values with the PLS analysis was statistically significant in all years. The overall adjusted r2 values in PCR, however, were low, suggesting that this method did not improve the modeling. This result further illustrated the conclusion made by Mallarino et al. (1996) that soil factors having high variations were not necessary to explain yield variation successfully.

According to the PLS model, cotton lint yields in 1998 and 2000 were positively related to sand and clay contents, relative elevation, depth to free carbonate, and Olsen-P concentration, but were negatively related to pH, exchangeable Ca2+, and NO3 concentration. This site was featured with high pH values with the average of over 8 and the minimum near 7.8, which would explain the negative association between pH and yield. If pH values were below normal, then a positive association with yield probably would have occurred. Furthermore, high pH values could also reduce the availability of other nutrients like P and Zn. Different signs and magnitudes of the relationships of soil properties and landscape measurements to yields between 1998 and 2000 existed for K saturation, exchangeable Mg2+, slope, and depth to caliche (Table 4). The negative relationship of lint yield to NO3 and positive relation of lint yield to Olsen-P in both 1998 and 2000 matched the results of Reiter et al. (1999), who concluded that the highest cotton yields were achieved with higher P supply from fertigation in high pH and calcareous soils. Oosterhuis et al. (1991) indicated that cotton yield response to stress was associated with soil clay content, which also caused variations in cotton growth due to crusting and emergence problems as well as damping-off disease associated with clay contents. During dry weather patterns, clay soil would supply less water to cotton under limited irrigation and result in lower yield compared with sandy soil. Lint yield in 1999, however, was influenced by slope, relative elevation, and NO3, which all affected N supply.

Although soil pH, texture, exchangeable Ca2+, depth to free carbonate, and caliche could affect lint yield, they are harder to control than N and Olsen-P levels. For example, high pH soil was difficult to correct. Moreover, the influences of soil texture on lint yield could be addressed with water management. The relationships between cotton lint yields and soil properties indicated there was not enough P compared with N based on the observed P levels and the N/P2O5 ratio in this field. Reiter et al. (1999) reported that maintaining proper N/P2O5 ratio by applying more P fertilizer was important for improving cotton yield for the calcareous soil. Therefore, management may need to focus on proper application of N and P and on irrigation based on levels and spatial variability of these nutrients within the field. For example, more N should be applied in those areas where NO3 levels were low in 1999, and more P is required to adjust the N/P ratio to a proper range.

Regression Analyses between Fiber Quality and Soil Properties
Of the three fiber quality characteristics measured, micronaire and length were related more to the soil properties than was strength (Tables 7, 8, and 9). Soil properties had varying influences on fiber quality among the 3 yr. While relationships could be identified in 1998 and 2000, there was only significant association between fiber length and soil properties in 1999, which might also relate to the weather changes during the 3 yr. As with lint yield modeling, the PLS-generated model represented the relationship between micronaire and soil properties better than the OLS and PCR models, as reflected by lower p values and higher adjusted r2 values (Table 7). While principal components could extract soil variables that had large variations, they did not result in stronger relationships with the quality measurements. The results from this study indicated that better fiber quality is associated with higher yields. Therefore, better management may not only increase cotton lint yields, but may also improve fiber quality.


View this table:
[in this window]
[in a new window]
 
Table 7. Regression estimators between micronaire and soil properties and digital elevation model from an irrigated cotton field near New Deal, TX, from 1998 through 2000.{dagger}

 

View this table:
[in this window]
[in a new window]
 
Table 8. Regression estimators between fiber length and soil properties and digital elevation model from an irrigated cotton field near New Deal, TX, from 1998 through 2000.{dagger}

 

View this table:
[in this window]
[in a new window]
 
Table 9. Regression estimators between fiber strength and soil properties and digital elevation model from an irrigated cotton field near New Deal, TX, from 1998 through 2000.{dagger}

 
Micronaire was negatively correlated to exchangeable Mg2+, K saturation, sand content, and slope in both 1998 and 2000. Clay content had different influences on micronaire in 1998 and 2000. This inconsistency may also relate to weather differences. It has been commonly observed that water supply influences fiber quality. Early season drought could result in higher micronaire values (Bradow and Davidonis, 2000). Furthermore, drought and excessive water supply both could reduce micronaire (Ramey, 1986). Another important factor affecting micronaire is the N/P ratio. Reiter et al. (1999) reported that micronaire increased greatly when N/P2O5 ratio changed from 5:1 to 5:3 in a fertigation study.

Significant relationships between fiber length and soil properties existed in all 3 yr (Table 8). The influences of soil properties on fiber length were consistent in 1998 and 2000. As shown in the PLS model, fiber length was positively related to exchangeable Mg2+, sand and clay contents, slope, depth to caliche, and Olsen-P, whereas it was negatively related to pH, exchangeable Ca2+, and depth to caliche. Potassium saturation and NO3 showed different influences on fiber length in 1998 and 2000, which may also relate to differences in NO3 concentrations and weather patterns. Unlike yield and other cotton parameters, a significant relationship existed between fiber length and soil properties in 1999; however, signs and magnitudes of soil property estimators were quite different from those in 1998 and 2000. Although fiber length depends on variety, drought late in the flowering period could reduce fiber length significantly. Grimes and Yamada (1982) concluded that fiber length could be affected once the water deficit was great enough to lower the yield to 700 kg ha–1, which was the case in 2000 (Table 1). Therefore, drought could be one reason for shorter fiber length in 2000 (Table 1).

A significant relationship between fiber strength and soil properties existed only in 1998 (Table 9). Higher VIF values suggested that PLS should be a good choice for describing their relationships. Higher K saturation, sand and clay contents, along with relative elevation, and greater depth to caliche layer tended to increase fiber strength, whereas higher pH, NO3, and slope tended to decrease fiber strength in 1998. It has been previously reported that spatial variations in the levels of P, K, Ca, Mg, or OM were not correlated to fiber strength (Johnson et al., 2002). Bradow and Davidonis (2000) indicated that compared with micronaire and length, fiber strength is related more to genotype. In addition, MacKenzie and Van Schaik (1963) indicated that genotype differences in fiber strength were far more important than was N supply. Our results further illustrated that fiber strength is less affected by soil properties in comparison with other fiber measurements.

Even with the best cases from PLS, only 55% (lint yield in 1998), 62% (micronaire in 1998), 60% (length in 2000), and 50% (strength in 1998) variations were explained by the soil properties tested. The soil properties explained even less the variations of yields and the three quality measurements in the wetter year of 1999. Factors such as soil temperature and solar radiation in early growing season, which were not included in this study, might control cotton yield in years with more rainfall. Furthermore, the magnitudes of soil properties on yield and quality measurements varied between 2 drier years. Johnson et al. (2002) also indicated the difficulty in establishing the relationships between fiber yield and quality with measured soil properties. Corwin et al. (2003) reported that an r2 of 0.61 and an adjusted r2 of 0.55 were obtained between cotton lint yields and electrical conductivity, leaf fraction, gravimetric water content, and bulk density at 0 to 150 cm soil in a central California study, which implied that nearly 40% of cotton yield variation could not be explained by the soil properties investigated. The relatively low r2 or adjusted r2 from our and other studies might also relate to the limitation of empirical regression (Webster, 1997). For precision agriculture, a repeatable model under similar weather pattern is highly desirable. Long-term studies encompassing varying weather patterns may help model generalization and modification to facilitate the requirement for site-specific management.


    CONCLUSIONS
 TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Our study demonstrated that intercorrelations among soil properties were common and affected the empirical modeling of soil–crop relationships. At the early stage of data modeling, some variables at different depths can be combined according to their strong associations. Some variables resulted in high variance of inflation, but they still may have to be included in an empirical model because of their proven biophysical influence on crop growth. The PLS method tended to result in the model having correct signs and magnitudes for describing the relationships between cotton yield or fiber quality and soil properties because PLS was designed to construct models that provided effective prediction power and reduced multicollinearity problems. Although PCR identifies the principal components that maximize the variance of independent variables, it does not necessarily result in a model that can reveal the relationships between independent and dependent variables.

There were similarities in the variables that affected cotton lint yield and fiber quality. Conditions that resulted in higher lint yield also tended to enhance fiber quality into the proper range in terms of their unit price. However, the magnitudes of influence of soil properties on cotton fiber quality were much smaller than those observed for lint yield. Among the soil properties selected, sand and clay contents, exchangeable Ca2+ and Mg2+, NO3, Olsen-P, relative elevation, and slope were important factors affecting lint yield and fiber quality and were further revealed from the loadings in the linear components from the PLS as relatively influential variables. However, fairly large proportions (40% or more) of variations of yields and fiber quality measurements remained unexplained by the soil properties and landscape characteristics. Furthermore, as the coefficients of regression models varied significantly among the 3 yr, it was difficult to establish robust models for decision-making in site-specific management. The relationships between cotton lint yield, fiber quality, and soil properties need to be studied further through use of less empirical techniques.


    ACKNOWLEDGMENTS
 
This research was funded by Cotton Incorporated and the International Cotton Research Center. The Omnistar Company in Houston, TX, donated the GPS device for position georeferencing. The authors are indebted to Mr. Bill Stence for allowing the research to be conducted on his farmland and the A&L Plains Agricultural Labs in Lubbock, TX, for discounts on sample analyses. Special thanks go to Dr. B.L. Allen for delineating soil types. The authors appreciate the three anonymous journal reviewers for their comments and detailed suggestions.


    NOTES
 TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Contribution of the College of Agricultural Sciences and Natural Resources at Texas Tech University Scientific J. Series Paper no. T-4-546.


    REFERENCES
 TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 




This article has been cited by other articles:


Home page
J. Environ. Qual.Home page
J. Lee, J. Six, A. P. King, C. van Kessel, and D. E. Rolston
Tillage and Field Scale Controls on Greenhouse Gas Emissions
J. Environ. Qual., April 3, 2006; 35(3): 714 - 725.
[Abstract] [Full Text] [PDF]


This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Web of Science (4)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Ping, J. L.
Right arrow Articles by Dobermann, A.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Ping, J. L.
Right arrow Articles by Dobermann, A.
Agricola
Right arrow Articles by Ping, J. L.
Right arrow Articles by Dobermann, A.
Related Collections
Right arrow Cotton
Right arrow Crop Models
Right arrow Plant and Environment Interactions
Right arrow Production Agriculture
Right arrow Site-Specific Analysis


HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Crop Science Vadose Zone Journal
Journal of Natural Resources
and Life Sciences Education
Soil Science Society of America Journal
Journal of Plant Registrations Journal of
Environmental Quality
The Plant Genome