| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a Texas A&M Univ., current address: Dept. of Soil Science, North Carolina State Univ., Raleigh, NC 27695
b Texas A&M University–USDA-ARS, 3810 4th Street, Lubbock, TX 79415
c Texas A&M University, 1509 Aggie Drive, Beaumont, TX 77713
d Texas A&M Univ., Rt. 3, Box 219, Lubbock, TX 79403
e Economics, Texas Tech University, Lubbock, TX 79409
* Corresponding author (r-lascano{at}tamu.edu)
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
-0.64). In 1998, lint yield, SWC, clay, sand, and SE were cross correlated within 60 to 80 m. Multivariate state-space analysis showed that lint yield at position i was weighted on lint yield, SWC, clay, sand, and SE at previous position i - 1. It is concluded that 75% ET and N rate at 90 kg ha-1 would be the basis to consider variable water and N rates related to field conditions, and the state-space model quantified spatial interdependence between irrigation, fertilization, and field heterogeneity.
Abbreviations: ET, evapotranspiration • ICB, incomplete block design • LEPA, Low Energy Precision Application • RE, relative elevation • SD, standard deviation • SE, site elevation • SWC, soil water content
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
While soil water was associated with cotton lint yield (Bordovsky et al., 1992; Moore, 1998; Lascano and Li, 2001), few studies have attempted to quantify the effects of N runoff on cotton yield and N uptake because of tillage (Soileau et al., 1994), and slope and SE (Li et al., 2001a, 2001b). Soil texture and topographic features affected spatial distribution patterns of soil water (Stone et al., 1985; Halvorson and Doll, 1991; Lascano and Hatfield, 1992; Brubaker et al., 1994; Li et al., 2001a), organic C, particle-size (Kabrick et al., 1997; Merrill et al., 1998), N and P distribution and use (Brubaker et al., 1994; Soileau et al., 1994), and grain yields of wheat (Triticum L.), corn (zea maize L.) and soybean (Glycine max L. Merr.) (Halvorson and Doll, 1991; Timlin et al., 1998; Kravchenko and Bullock, 2000; Stevenson et al., 2001). Li et al. (2001a) reported that cotton lint yield, measured on the semiarid High Plains of Texas, increased by 34% on low positions, where soil in the rooting zone contained 9% more water compared with upslope areas. However, that study focused on spatial soil water, lint yield and landscape relations, and N fertilizer management was not considered.
Since spatial soil heterogeneity poses problems on field experiments, Van Es et al. (1989) suggested to use an incomplete block (ICB) of size two design, based on the principles of regionalized variable theory, to minimize the influence of field heterogeneity on the outcome of experiments in large fields. Van Es et al. (1989) also showed that the use of incomplete blocks with constrained randomization estimates treatment effects by means of equal average-distance comparison, reduces the average error mean square by 44% compared with using complete blocks. Given that many soil properties do not vary randomly, multiple linear regression models (Brubaker et al., 1994), topographic parameters such as surface elevation and curvature (Timlin et al., 1998), and a topographic factor determined from neighboring site slopes (Halvorson and Doll, 1991; Li et al., 2001b) have been used to explain soil and crop variability. However, these methods were limited for site-specific water and N management purposes.
Crop response varies within a field because the reciprocal relationships between crop growth and underlying soil processes are variable in both space and time (Cassel et al., 2000; Wendroth et al., 2001). Spatial or temporal domain methods based on state-space models have been used to estimate parameters, test hypotheses, and describe spatial correlations among soil water, nutrient and crop variables (Morkoc et al., 1985; Shumway et al., 1988; Wendroth et al., 1992, 2001; Nielsen et al., 1994; Cassel et al., 2000; Li et al., 2001a; Stevenson et al., 2001). Spatial correlation between soil temperature and moisture (Morkoc et al., 1985), alfalfa (Medicago sativa L.) and ryegrass (Lolium L.) yield, effective soil N and N fixation (Wendroth et al., 1992), wheat-grain yield, soil-base saturation, and water-storage capacity (Cassel et al., 2000), and cotton lint yield, soil water, NO3-N, and SE varying along transects (Li et al., 2001a) were described using multivariate autoregressive state-space analysis. However, none of these state-space models included landscape position or soil texture variables.
Recent developments on new approaches for cotton site-specific water and N management indicate the need for application of water and N based on site-specific soil and field conditions. There is also a need for information on variability in soil properties, yield, N uptake, and fertilizer N recovery with different irrigation and N rates. Although variability in soil water, lint yield, and N uptake across the landscape (Li et al., 2001a), effects of irrigation and N rates on lint yield (Bronson et al., 2001), soil water ET patterns (Lascano and Li, 2001), and in-season cotton N uptake and plant growth have been analyzed (Li et al., 2001c), the effects of different N rates, interaction between irrigation and N treatment, and influence of soil texture and topographic features on water and N use requires further study. The interdependence between cotton growth, soil properties, irrigation, fertilization, and field heterogeneity must be both understood and quantified to develop a variable water and N application framework.
We hypothesized that SE and soil texture are factors controlling water and N effects on cotton growth, N uptake, and thus lint yield. Our objectives were to (i) determine the effects and interaction of irrigation and N treatments on cotton N uptake and lint yield using a mixed model analysis, (ii) identify the major soil and site parameters that control water and N effects on lint yield, and N uptake using simple, cross and canonical correlation, (iii) use a state-space approach to analyze the major factors and underlying processes that affect lint yield, and (iv) compare the state-space forecasted lint yield with that measured the following year.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
The study began in May 1998 on an area 32 m wide by 700 m long within the center pivot (45 ha). Treatments consisted of irrigation at 50 and 75% of estimated cotton ET, and N fertilizer at rates of 0, 90, and 135 kg ha-1. The applied ET was estimated as 50 or 75% multiplied by (cotton ET - rain), where cotton ET was calculated by multiplying the reference ET by a cotton coefficient (Lascano, 2000). The N treatments were in the range of N rates applied for cotton (Soileau et al., 1994; Moore, 1998; Varco et al., 1999). Treatments were arranged in an ICB design with blocks of size two, as designed by Van Es et al. (1989), to equalize the average comparison distance of N treatments for reducing any influences of field heterogeneity on outcome of field experiments (Van Es et al., 1989; Van Es and Van Es, 1993). For each ET level, there were four replicates for the control and N treatment rate of 135 kg ha-1, and five replicates for the N rate of 90 kg ha-1. Plot size was sixteen 1-m rows wide and 50 m long. Cotton (‘Paymaster Roundup Ready 2326’) was seeded at a rate of 16.8 kg ha-1 on 8 May 1998 and 10 May 1999. Fertilizer N (urea, 32-0-0) was fractionally chiseled into the soil at a rate of 45 kg ha-1 at emergence, bloom, and first square stage, respectively, based on local recommendations given by the Texas Agricultural Extension Service. Other requirements of fertilizers and weed and insect control were done according to regional recommendations. Irrigation (5.1–15.2 mm 3 d-1) was applied using the LEPA system (Lyle and Bordovsky, 1981). In-season rain was below (91 mm) the 30-yr average (320 mm) in 1998, and was unusually wet (130 mm) in June, but dry thereafter, in 1999. Total irrigation water was 242 and 323 mm in 1998, and 190 and 286 mm in 1999 for the 50 and 75% ET, respectively.
Soil Water Monitoring, and Plant and Soil Analysis
Soil water distribution in space and time was monitored along transects in the south-north direction across the field. There were four neutron access tubes (5-cm diam. and 2 m long each), installed on row 7 and 14 of each plot, for a total of 52 soil-water monitoring points for each ET level. The average distance was 12 m between tubes along irrigation circles. Monthly SWC was measured in 0.3-m increments to a 1.8-m depth at each tube using a neutron probe (Model 503 Hydroprobe, CPN Corporation, Martinez, CA) calibrated for this field (Li et al., 2001a). Surface (0–0.06 m) SWC was measured using a ThetaProbe (Delta-T Devices Ltd., Cambridge, England) in 1999, and site altitude measurement and calibration are also given by Li et al. (2001c).
Cotton rooting depth was measured at plant maturity (late August) in soil profiles excavated at three locations (south, center, and north) for each ET level. Since cotton roots were only observed within the 0- to 0.9-m depth, we evaluated soil tests in two zones: (i) the rooting zone (0–0.9 m), and (ii) below the root zone (0.9–1.8 m). Soil was sampled on the row, within 0.5 m from each neutron access tube, to a depth of 1.8 m in 0.3-m increments at emergence, bloom, and harvest. Soil samples were air-dried and sieved to 2 mm. Soil texture was measured using the hydrometer method (Gee and Bauder, 1986). The 0.1 M KCl extractable NO3-N was measured using a Technicon Auto-Analyzer II C (Technicon Instruments Corp., Tarrytown, NY) (Li et al., 2001a).
Total N uptake by the crop was determined on 10-d intervals beginning at an early vegetative stage (2 wk after emergence) until harvest. At the vegetative stage, 32 plants per plot were taken from four 5 by 25 m sampling regions around each neutron access tube, and 16 plants per plot were taken from the same areas during bloom and boll growth stages. Root length and fresh and dry weights of roots, leaves, stems, bolls, and seeds were measured. Plant samples were dried at 70°C to a constant weight and ground to pass through a 0.5-mm sieve. Plant N was determined using a LECO FP-528 Analyzer (Leco Corp., St. Joseph, MI) (Li et al., 2001a). Cotton lint was hand harvested on 5 October in 1998 and on 12 October in 1999. Lint yield was determined at each plot within four 1 by 4 m areas, 10 m apart on rows 10 and 11, which had not been sampled.
Total N uptake was calculated by multiplying N content by total plant dry weight. The amount of fertilizer N used by the cotton was calculated as the difference between total N uptake in fertilized plots and the control (zero fertilizer N) plots. Fertilizer N recovery was defined as the ratio of the amount of fertilizer N used to N applied. We assumed that (i) cotton use of residual soil N and N from other sources (ryegrass mineralization, atmospheric deposition, and irrigation water) was the same in control and N treatments, (ii) there was no enhancement of mineralization caused by adding N fertilizer, and (iii) there was no significant immobilization of applied fertilizer N. Thompson et al. (2000) made similar assumptions for estimating N balance.
Mixed Model, Cross Correlation, and State-Space Analysis
Mixed model analysis (Little et al., 1996) was used to determine effects of irrigation and N fertilization on response variables. Regression relationships and linear correlation among the soil and cotton crop variables was determined using PROC GLM and PROC CORR procedures (SAS Institute, 1990). The spatial correlation between two variables x and y (lint yield, SWC, sand, clay, and SE) was determined with the cross-covariance function (Cxy[h]) and the cross-correlation function (
xy[h]), as described by Cassel et al. (2000) as follows:
 | [1] |
 | [2] |
x and y are the standard deviations of x and y, respectively. The ARIMA procedure (SAS Institute, 1993) was used to estimate the cross correlation function xy(h) between lint yield, soil water, clay, sand and SE.
Cotton lint yield, SWC, texture, and SE were identified in state-space analyses as quadrivariate first-order autoregressive processes. We described how the state vector xi of cotton yield at a location i was related to the system state vector xi-1 of all measured variables at location i - 1 as described by Cassel et al. (2000) as follows:
 | [3] |
is a p x p matrix of state coefficients, and 
i is the model error vector. The state-space model (Eq. [3]) assumes errors to be zero-mean, uncorrelated, and included in the model error vector 
i with common r x r covariance matrix Q. The error vector i is assumed to be independent. The elements of the state vector xi are determined via a sequence of canonical correlation analyses of the sample autocovariance matrix (state-space coefficient matrix) by approximate maximum likelihood (SAS Institute, 1993). The Proc StateSpace procedure was used to determine the mean vector µ, p x p covariance matrix 
, r x r covariance matrix Q, and state coefficients with the p = 4 dimensional vector process, which led to a state-space model with the estimated measurement matrix (i) taken as a 4 x 4 transition matrix as follows:
 | [4] |
[4]
The state-space forecasts were calculated using the Kalman filtering technique (SAS Institute, 1993), and the forecasted lint yield obtained with the lint yield measured in 1998, was then compared with the lint yield measured the following year (1999). This was an attempt to show how and where future lint yields could be forecasted under local field conditions.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
|
|
In 1998, temporal SWC profiles decreased from the vegetative (June) to boll growth stage (August) and increased again towards harvest (October) (Fig. 2) . The top 0- to 0.3-m soil was dry throughout the growing season. The SWC in the rooting zone (0–0.9 m) was on average 42% (or 85 mm m-1) of the total soil water measured within 0- to 1.8-m depth. As expected, analysis of variance indicated that temporal SWC between measurement date (P < 0.0001), irrigation level (P < 0.0014), and soil depth (P < 0.0006), and interaction between measurement date and soil depth was significantly different (P < 0.0001).
|
|
There were larger SD in clay and sand contents (0–0.3 m) in the 50% ET areas than the 75% areas (Table 1). Unlike the SWC patterns, soils on the northern upslope areas contained more clay (194–275 g kg-1) compared with the footslope areas (140–235 g kg-1) between 300 and 400 m in the center field as measured in July 1998 (Fig. 3e). It appeared that sand content varied with position because of deposition in low lying areas, in the order: center areas > southern area > northern upslope (Fig. 3e,f). As a result, soils (0–0.3 m) contained more sand (785–816 g kg-1) in low position areas (south-center field), compared with the northern upslope areas (740–785 g kg-1). Soil NO3-N in the root zone (0–0.9 m) declined in 1999 compared with 1998 (Table 1). Soil NO3-N (0–0.9 m) patterns showed trends similar to clay content distribution on the northern upslope areas (Fig. 3e), where the soil NO3-N (measured on 10 July 1998) was similarly high, but with different rates of N fertilizers applied (Fig. 3e). For example, in this area the control plot (zero N fertilizer) contained a NO3-N level (148 kg ha-1 in 0–0.9 m) as much as its neighbor plots, where soils were fertilized at 90 or 135 kg ha-1 (Fig. 3e,f).
Mixed Effects of Treatments
The main effects of irrigation, determined with the mixed model procedure, were significant on cotton lint yield, followed by N uptake, and SWC during the 2 yr (Table 2). There was no interaction between irrigation and N treatment on response variables. The fixed effect of N was significant only on lint yield in 1999 (Table 2), where the lint yields averaged 824, 911, and 847 kg ha-1 for the N treatments at rates of 0, 90, and 135 kg ha-1, an increase of 10.6% (vs. control) and 7.6% (vs. 135 kg N ha-1) for the N treatment at 90 kg ha-1.
Random effects of the mixed model variance components (block, and block x treatments) were not significant (Table 2). However, Z-statistics showed that the residual of the mixed model was significant on all measured parameters (Table 2). The Z-statistics for the model variance components was based on asymptotic normality (Little et al., 1996). The mixed model residual effects consisted of the covariance of other independent variables such as SE, SWC, texture, or other nutrients rather than N.
Linear Regression and Correlation Between Variables
Cotton lint yield was linearly correlated (Fig. 4a,b)
with SWC in the rooting zone (0–0.9 m), which explained as much as 56% of the lint yield variation for the 50% ET irrigation level (Fig. 4a). A large portion (80%) of the lint yield greater than mean was produced with a SWC >0.10 m3 m-3. Cotton lint yield was negatively correlated with clay (Fig. 4c), and positively correlated with sand (Fig. 4d). There was smaller correlation coefficient between lint yield and soil NO3-N (r = 0.21). Cotton lint yield was quadratically related to soil NO3-N with yields decreasing at NO3-N content >140 kg ha-1 in the rooting zone (0–0.9 m). The N uptake was less influenced (11–16% of the variations) by the SWC.
|
Cross Correlation Between Lint Yield, Soil Texture, and Site Elevation
Site elevation and soil texture were identified as the important factors contributing to variability in SWC, cotton lint yield, and N uptake. There was a cyclic, positive or negative, feedback relationship between lint yield, SWC, and elevation as measured in 1998. As illustrated in the scatter diagram (Fig. 5)
, the cross-correlation functions (xy), calculated with Eq. [1] and [2], of cotton lint yield, SWC, clay, sand, and SE ranged between -0.72 and 0.64. Considering the 95% confidence level as described by Cassel et al. (2000), the lint yield was positively cross correlated with SWC (Fig. 5a) but negatively cross correlated with SE (Fig. 5b) across a lag distance of 60 m (±30 m); cross correlation between lint yield and clay was negative across a distance of 80 m on the 50% ET transect (Fig. 5b). On the 75% ET transect, lint yield was negatively cross correlated with elevation (Fig. 5e) and positively with sand content (Fig. 5f) across a lag distance of 80 m (±40 m). Clay was also positively cross correlated with elevation across a lag distance of 80 m on the 50% ET transect (Fig. 5d), where clay content increased with elevation. As the lag distance increased, these cross-correlation functions became positive at a lag distance of ±180 m (Fig. 5). However, there was no cross correlation between N uptake and SWC (results not shown). As a result of larger variation of SE on the 50% ET, the cross correlation distance between SWC, lint yield and elevation was 20 m shorter on the 50% ET than on the 75% ET transect (Fig. 5e,f).
|
) of the state-space equation corresponding to each irrigation level along the 50% ET transect (50%ET) and 75% ET transect (75%ET) are given below:
 | [5] |
 | [6] |
The above estimated 4 x 4 transition matrix showed that the weight relations between these state-space variables become alternatively negative or positive as the position i changed. The estimated transition matrix leads to a multivariate state-space cotton lint yield model shown in Fig. 6
. Via the state coefficient , both multivariate state-space equations (Eq. [7] and [8], Fig. 6) demonstrated that cotton lint yield yi, was positively weighted on soil water and negatively weighted on elevation at previous position i - 1 on both transects (Eq. [7] to [8]). However, lint yield, yi, was negatively weighted on the clay content Ci-1 along the 50% ET transect (Eq. [7]), but positively weighted on the sand content Si-1 along the 75% ET transect (Eq. [8]). Differently, 50% ET lint yield at position i is much more influenced by the previous SE (Eq. [7]) compared with the 75% ET lint yield at position i (Eq. [8]).
|
are shown in Table 3 for the two state-space cotton lint yield equations given in Fig. 6. The standard error for the yield, SWC and elevation are comparable. The t-statistics showed that all parameter estimates were significant (t > |2|) for both models (Table 3).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Because of sand deposition in lower areas, summit areas had more clay. Therefore correlation between sand and elevation was negative. Since the impact of SE on SWC was greater than soil texture, soils in lower position areas contained more water (Fig. 1) through run-on water from upslope areas (Li et al., 2001b). As a result, higher cotton lint yield and N uptake were measured on lower positions (Fig. 3). Because of possible water runoff, described using a topographic factor calculated using slopes and distance from neighboring site points (Li et al., 2001b), lower cotton lint yield and N uptake were measured on northern upslope and summit areas (Fig. 3). Other studies have also shown the impacts of elevation and slope length on crop yield. For example, soil water was not evenly distributed and spring wheat yield and water use were linked to elevation and slope length (Halvorson and Doll, 1991), and surface elevation and curvature, and soil organic matter, P, and K contents strongly contributed to the spatial and temporal variability of maize yield on a hillslope (Timlin et al., 1998). Topography influences hydrologic and agronomic processes complicating water and N application decisions.
Soil NO3-N and Nitrogen Treatments
In the study site, N fertilizer has been applied to the soil at a rate of 190 kg ha-1 yr-1 for the last 5 yr, explaining the lack of cotton response to N treatments. Large amounts of soil NO3-N (including residual and fertilizer N) were measured in July 1998 (Table 1). The preseason soil residual NO3-N in the effective root zone (0–0.9 m) remained moderately high before seeding in May 1999 (Table 1). However, assuming N from other sources (rye residue mineralization, atmospheric deposition, and irrigation water) was the same for control and N treatments, total N inputs at 90 kg ha-1 for 1999 year were reasonable, based on N uptake in 1998 (Table 1) and preplant NO3-N assessment (Table 1). Larger amount of residual NO3-N caused the lack of N treatment effect on lint yield and N uptake in 1998. Also, rain variability could have influenced cotton responses to the treatments within years. Rain at the vegetative growth stage in 1999 favored plant growth. Therefore there was an increase of 14% on lint yield with 50% ET in 1999 compared with 1998, and the irrigation level became more important on lint yield in 1998 than in 1999 as compared by the P values (Table 2).
As expected, significant effects of N treatments and interaction between irrigation and fertilizer N on lint yield occurred in 1999, a wet year, when residual soil NO3-N was at moderate levels (Table 1). In other areas, cotton lint yield increased significantly with N rates between 90 and 135 kg ha-1 in a deep sandy soil in Georgia (Harris, 1998). Higher amounts of N (112–134 kg ha-1) were needed for cotton production on a high pH alluvial soil of Louisiana (Moore, 1998). Profitable N rates ranged between 60 and 105 kg ha-1 for no-tillage cotton grown in a loamy soil in Mississippi (Varco et al., 1999), and between 110 and 165 kg ha-1 for cotton production in a clay loam in California (Plant et al., 2000). These conclusions were based on the relationship between lint yield and N rates. Our results showed that N rate of 90 kg ha-1 could be a basis rate for farmers. While one in two years there was significantly higher lint yield with the N rate at 90 kg ha-1 as mentioned above, the mean fertilizer N recovery was consistently higher with this N rate compared with the rate of 135 kg ha-1 (25.9 versus 14.3% in 1998, and 25.8 versus 19.0% in 1999). Higher fertilizer N recovery means less residual fertilizer N left in the soil. Further, low input of N is agronomically, economically, and environmentally beneficial in this semiarid area (Segarra et al., 1989; Morrow and Krieg, 1990; Bronson et al., 2001), which has also been shown in other areas (McConnell et al., 1996; Varco et al., 1999; Thompson et al., 2000).
State-Space Description of Cotton Lint Yield Variability
The measured lint yield (1998) decreased toward the upslope area on the 50% ET transect, (Fig. 6a), so that lint yield at position i was negatively weighted on yield at the previous position i - 1 (Eq. [7]), and therefore the negative weight of SE was significant (Table 3). Conversely, on the 75% ET transect, lint yield at position i was positively weighted on the yield at the previous position i - 1 (Eq. [8]), since the measured lint yield tended to increase on lower areas (Fig. 6b). As a result, the influence of SE at position i - 1 on the lint yield at position i was more significant on the 50% ET than the 75% ET (Table 3), where site positions are on average 0.5 m lower than on the 50% ET (Table 1).
The positive weight on SWC and negative weight on elevation (Eq. [7] to [8]) indicated that lint yield increased at position i were weighted based on an increase of SWC and decrease of elevation. Clay (50% ET) and sand (75% ET) at the previous position i - 1 also had significant weight on lint yield at position i (Table 3). The linear regression analysis showed the dispersion of lint yield versus SWC, clay, and sand (Fig. 4). However, all measured lint yield values were within the 95% confidence limits of lint yield forecasts given by the state-space model (Fig. 6). Future cotton yields in such an undulating field, given by the autoregressive state-space analysis, are weighted on soil water, texture, and topography variability. Higher lint yields would be expected on lower positions (Fig. 6), where water and nutrient run-on is expected (Li et al., 2001b). Conversely, lower lint yields are predicted on upslope areas (Fig. 6), where water and nutrient runoff should be high (Li et al., 2001b).
The forecasted yield values Yi, given by the state-space models with data measured in 1998, were related to cotton lint yield measured in 1999. The forecasted values Yi ranged from 528 to 942 and 696 to 1144 kg ha-1, and the measured lint yield values varied between 507 to 1097 and 578 to 1326 kg ha-1 on the 50 and 75% ET transects, respectively. The correlation coefficients between these forecasted and measured lint yield values were 0.71 and 0.74 for the 50 and 75% ET, respectively. This result is important as it further shows the validity and applicability of state-space models to predict crop yield from 1 yr to the next. In this case, the soil NO3-N amounts and weather conditions were different in 1999 compared with 1998. Because cotton lint yield had greater correlations with soil water, sand, clay, and SE, and a much smaller coefficient with soil NO3-N, our state-space lint yield model, including lint yield, soil water, sand, clay, and SE as model parameters, gave a reliable lint yield prediction (r > 0.70) from 1 yr to the next.
The autoregressive state-space model appears to be a utile forecast tool to predict future yields related to soil properties and field conditions. Many studies have used linear and multiple regression models to describe and estimate soil, water and crop variables. For example, Halvorson and Doll (1991) and Brubaker et al. (1994) used such regression models to describe and predict crop yields and soil properties by landscape position. However, the state-space model is a result of a multivariate autoregressive analysis. A difference in estimate of crop yield or soil properties between the state-space model and linear model analysis is that the locations of forecasted values are given by the state-space model. In our state-space cotton lint yield models (Fig. 6), the spatial association of lint yield, soil water, clay, sand, and SE are described, and the future lint yields are quantified and located. Our results suggest that variability of field measurements could be combined with state-space models to describe the association of crop yields with soil properties and site topography in space, and to understand the interdependency between irrigation and fertilization in heterogeneous soils.
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
Received for publication March 28, 2001.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  P. Lu, Y. Su, Z. Niu, and J. Wu Geostatistical Analysis and Risk Assessment on Soil Total Nitrogen and Total Soil Phosphorus in the Dongting Lake Plain Area, China J. Environ. Qual., May 25, 2007; 36(4): 935 - 942. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. F. Bronson, J. D. Booker, J. P. Bordovsky, J. W. Keeling, T. A. Wheeler, R. K. Boman, M. N. Parajulee, E. Segarra, and R. L. Nichols Site-Specific Irrigation and Nitrogen Management for Cotton Production in the Southern High Plains Agron. J., January 5, 2006; 98(1): 212 - 219. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Iqbal, J. J. Read, A. J. Thomasson, and J. N. Jenkins Relationships between Soil-Landscape and Dryland Cotton Lint Yield Soil Sci. Soc. Am. J., May 6, 2005; 69(3): 872 - 882. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. B. Zeleke and B. C. Si Scaling Properties of Topographic Indices and Crop Yield: Multifractal and Joint Multifractal Approaches Agron. J., July 1, 2004; 96(4): 1082 - 1090. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. N. Kravchenko, K. D. Thelen, D. G. Bullock, and N. R. Miller Relationship among Crop Grain Yield, Topography, and Soil Electrical Conductivity Studied with Cross-Correlograms Agron. J., September 1, 2003; 95(5): 1132 - 1139. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. F. Bronson, J. W. Keeling, J. D. Booker, T. T. Chua, T. A. Wheeler, R. K. Boman, and R. J. Lascano Influence of Landscape Position, Soil Series, and Phosphorus Fertilizer on Cotton Lint Yield Agron. J., July 1, 2003; 95(4): 949 - 957. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||
