Agronomy Journal Journal of Natural Resources and Life Sciences Education
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 ISI 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 ISI Web of Science (2)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Phillips, S. B.
Right arrow Articles by Mullins, G. L.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Phillips, S. B.
Right arrow Articles by Mullins, G. L.
Agricola
Right arrow Articles by Phillips, S. B.
Right arrow Articles by Mullins, G. L.
Related Collections
Right arrow Nitrogen
Right arrow Wheat
Published in Agron. J. 96:591-600 (2004).
© American Society of Agronomy
677 S. Segoe Rd., Madison, WI 53711 USA

NITROGEN MANAGEMENT

Estimating Winter Wheat Tiller Density Using Spectral Reflectance Sensors for Early-Spring, Variable-Rate Nitrogen Applications

S. B. Phillips*,a, D. A. Keaheya, J. G. Warrena and G. L. Mullinsb

a Eastern Shore Agric. Res. and Ext. Cent., 33446 Research Drive, Virginia Polytechnic Inst. and State Univ., Painter, VA 23420
b Dep. of Crop and Soil Environ. Sci., 424 Smyth Hall, Virginia Polytechnic Inst. and State Univ., Blacksburg, VA 24061

* Corresponding author (stevep{at}vt.edu).

Received for publication January 23, 2003.

   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Adequate tiller density is critical for attaining optimum grain yield in winter wheat (Triticum aestivum L.). To ensure maximum tiller development, several states in the Mid-Atlantic recommend split-applying N in the spring based on tiller density at Zadoks Growth Stage 25. However, this strategy requires that several labor-intensive measurements be made in each field. Recent work has suggested that remote sensing might eliminate this problem. The objectives of this study were to estimate winter wheat tiller density using an on-the-go, spectral reflectance sensor and to determine the effect on grain yield of tiller density–based, variable-rate N applications at a 1-m2 resolution. Twenty-two site-years of data were collected from diverse locations across Virginia from 2000 to 2002. The normalized difference vegetation index (NDVI) was a reliable predictor of tiller density across environments (0.67 ≤ r2 ≤ 0.99), with 18 of 22 sites having slopes and intercepts that were not different from one another. Nitrogen fertilizer rates and grain yields resulting from using sensor-based estimates of tiller density were not different from those when using the standard practice for the Mid-Atlantic region at four out of six locations. At two of the six sites, sensor-based N recommendations were 11 kg N ha–1 lower than standard recommendations with no effect on grain yield, resulting in higher N use efficiencies at these locations. These results show that on-the-go, optical sensor technology can be used to accurately estimate winter wheat tiller density for determining and applying appropriate N fertilization rates at a 1-m2 resolution with minimal ground truthing required (one physical tiller count for each major soil type in a field).

Abbreviations: ESAREC, Eastern Shore Agricultural Research and Extension Center • GS, growth stage • LAI, leaf area index • NDVI, normalized difference vegetation index • NIR, near infrared • NUE, nitrogen use efficiency • R, red • TAREC, Tidewater Agricultural Research and Extension Center • VI, vegetative index


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
TILLERING IN WINTER WHEAT is the formation of lateral shoots out of axillary buds in the plant. Maintaining adequate tiller density is a critical component of wheat management as grain yield is influenced significantly by the number of grain-bearing, headed tillers (Klepper et al., 1982). A timely planted winter wheat crop will germinate, emerge, and tiller before the dormancy period that usually begins in December. Production of fall tillers is critical as these will begin to grow first in the spring, generally producing heads with more kernels; however, further tillering will occur in late winter and early spring, before stem elongation (Alley et al., 1996). Tiller formation in winter wheat is often affected by the amount of stress experienced during early plant development (Krenzer and Nipp, 1991). Additional factors that can impact tiller density include soil characteristics, crop residues, planter problems, and soil drainage patterns (Flowers et al., 2001). Scharf and Alley (1993) indicated that if tiller density is low when wheat begins to grow in the spring (<1000 tillers m–2), fertilizer N should be split-applied between Zadoks (Zadoks et al., 1974) Growth Stages (GS) 25 and 30 to stimulate formation of additional tillers and achieve optimum yield. However, at higher tiller densities (>1000 tillers m–2), optimum yield can be attained using a single N application at GS 30. In similar studies in North Carolina, Weisz et al. (2001) identified a critical tiller density in no-till wheat of 550 tillers m–2, below which an N application at GS 25 is recommended for optimum yields.

Although applying N based on tiller density at GS 25 has proven beneficial, this system is not widely utilized by growers due to the amount of time and difficulty involved in counting tillers and the variability of tiller density across a field (Flowers et al., 2001). To encourage making necessary N applications at GS 25, an accurate and efficient method to estimate tiller density must be available to growers. Results from several studies indicate that remote sensing may be a possible solution.

Plant reflectance is determined by leaf surface properties, internal structure, and the concentration and distribution of biochemical components; therefore, remote analysis of reflected light can be used to assess plant biomass and the physiological status of a plant (Peñuelas and Filella, 1998). Red (R; 671 ± 6 nm) and near infrared (NIR; 780 ± 6 nm) are often among the wavelengths of interest for indirect measurements of plant characteristics as R-reflected energy is absorbed by chlorophyll and NIR reflectance is sensitive to water content and leaf cell structure (Wood et al., 1999). A vegetative index (VI), which is the mathematical combination of spectral reflectances from two or more wavelengths, usually has greater sensitivity to plant vegetation than individual wavelengths (Wanjura and Hatfield, 1987). In the study by Wanjura and Hatfield (1987), a ratio vegetation index (NIR/R) was most sensitive to high levels of crop biomass, but during early vegetative growth (e.g., GS 25 in winter wheat), a normalized difference vegetation index [NDVI; (NIR – R)/(NIR + R)] provided more accurate estimates. This finding was supported by Serrano et al. (2000), who reported that NDVI was more suitable than a simple ratio for assessing wheat growth at the initial stages. Lukina et al. (2000) examined the effect of row spacing on spectral irradiance in winter wheat and reported high correlation (0.81 ≤ r ≤ 0.98) between NDVI and vegetative coverage. Wood et al. (1999) regressed GS-25 tiller density in winter against airborne digital photography–derived NDVI values collected on two dates from four fields and reported coefficients of determination ranging from 0.57 to 0.95. Flowers et al. (2001) also examined the relationship between GS-25 tiller density and several VI and digital counts at various wavelengths obtained from color or color infrared aerial photographs. While NIR digital counts were consistently correlated with tiller density (0.67 ≤ r ≤ 0.87), this relationship varied from field to field and from year to year. To account for this variability, the researchers computed relative NIR digital counts using areas of high and low tiller densities from each field, which resulted in high correlation (r = 0.88) across environments. Flowers et al. (2001) indicated that factors such as wheat variety, soil type and color, and atmospheric effects could account for some of the differences among environments. They also stated that frequent and intensive measurements needed to be made across the field to account for tiller density variability. Recently, Flowers et al. (2003) validated their technique across a wide range of environments including six soil types, six wheat varieties, and two tillage systems and found that NIR remote sensing accounted for 76% of the variation between predicted and measured GS-25 tiller densities. However, they also stated that factors such as bidirectional reflectance, the lack of radiometric correction, digitization processes, camera settings, film exposure, and film processing can influence measured reflectance and can complicate the use of color infrared aerial photographs to estimate GS-25 tiller density (Flowers et al., 2003).

Optical sensor–based technology may eliminate complicating factors such as atmospheric conditions, camera and film inconsistency, and spatial variability when using spectral reflectance to estimate tiller density. Raun et al. (2002) demonstrated that crop reflectance measurements using optical sensors could be used to efficiently determine variable fertilizer rates for topdress N applications to winter wheat based on in-season estimates of grain yield potential. The procedures developed were applied at spatial scales of 1 m2, which is the finest resolution where differences in soil test parameters and plant production characteristics are found (Raun et al., 1998; Solie et al., 1999). Commercial-scale, on-the-go, variable-rate applicators that employ the techniques of Raun et al. (2002) have been developed and are currently available to growers and fertilizer dealers (www.ntechindustries.com; verified 20 Jan. 2004). It is hypothesized that this same technology can be used to make on-the-go estimates of tiller density for each 1 m2 in the field.

The objectives of this study were to estimate winter wheat tiller density at GS 25 using an on-the-go spectral reflectance sensor and to determine the effect on grain yield of tiller density–based, variable-rate N applications at a 1-m2 resolution.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Site Descriptions
Twenty-two site-years of data were collected from 20 experimental locations in Virginia from 2000 to 2002. The sites represented a range of coastal plain, piedmont, and valley soil types; cropping systems (no-till vs. conventional, continuous wheat vs. rotation); row spacings; and wheat varieties (Table 1). Wheat seeding rate was fairly constant ({approx}400 seeds m–2) among locations and years.


View this table:
[in this window]
[in a new window]
  		Table 1. Brief site descriptions of locations in Virginia used for correlating winter wheat tiller density at Zadoks Growth Stage 25 with spectral reflectance measurements, 2000 through 2002.

 
Growth Stage–25 Tiller Density Measurements
In 2000, GS-25 tiller density data were collected in established wheat fields at the Eastern Shore Agricultural Research and Extension Center (ESAREC) near Painter, VA; the Tidewater Agricultural Research and Extension Center (TAREC) near Suffolk, VA; and in farmer fields near Port Royal, New Kent, and Warsaw, VA. At the Painter and Suffolk sites, 45 and 20 measurements, respectively, were made every 1 m2 along transects in the fields (Painter-TS 2000 and Suffolk-TS 2000) by counting all tillers with three or more leaves. At Port Royal, New Kent, Warsaw, and an additional field at the ESAREC (Painter 2000), tiller density was determined for 5 to 10 (depending on field size) randomly selected 1-m2 areas within the fields. In 2001, 40 tiller density measurements were made every 1 m2 along a single transect at the ESAREC (Painter-TS 2001). Random tiller density measurements were collected as described above in established fields at the ESAREC and the TAREC (Painter 2001 and Suffolk 2001, respectively) and in farmer fields near Chatham, Culpeper, and Stuarts Draft, VA. In 2002, random tiller density measurements were collected from three fields at the ESAREC (Painter 2002, Painter-HR 2002, and Painter-VT 2002), two fields at the TAREC (Suffolk 2002 and Suffolk-DF 2002), and farmer fields near Charles City and New Kent, VA.

A replicated study was also conducted at the ESAREC for the 2000–2001 cropping season. Three seeding rates (194, 387, and 581 seeds m–2) planted in 17-cm rows and two planting dates (25 October and 15 November) were included in a randomized complete block design, replicated four times. Plot size was 2.1 by 7.6 m. In the spring of 2001, tiller density measurements were collected from two randomly selected 1-m2 areas in each plot (except the late-planted, low-seeding-rate treatment) on 13 February and again on 7 March (Painter-TD and Painter-WR, respectively). The plots that were planted on 15 November at a rate of 194 seeds m–2 had stands that were too thin in February for accurate tiller density estimates; thus, no data were collected from these plots.

Spectral Measurements
For each 1 m2 where GS-25 tiller density was measured, spectral reflectance data were also collected. Reflectance data were collected under natural lighting using a hand-held, photodiode-based optical sensor equipped with R (671 ± 6 nm) and NIR (780 ± 6 nm) interference filters oriented to measure both incident (solar irradiance) and reflected (plant surface irradiance) light. A detailed description of the instrument used in this study can be found in Raun et al. (2001). Spectral measurements were collected by moving the sensor across each targeted 1-m2 area in the same direction the rows were planted. An additional set of readings (Painter-AR 2001) were collected from the same areas as Painter-WR 2001 to evaluate any difference in measured reflectance values due to the sensor traveling perpendicular to planted rows (AR) compared with parallel to the row orientation (WR). Bare soil or natural background reflectance measurements were also collected at each location. The reflectance values measured were used to calculate a modified NDVI, which compensated for light interference due to cloud cover, shadows, sun angle, etc., as follows:

[1]
where NIRref and Redref are the magnitude of reflected light and NIRinc and Redinc are the magnitude of incident light (Raun et al., 2001). A single regression equation relating tiller density and the modified NDVI was derived across environments, as were independent models for each location using PROC NLIN (SAS Inst., 1990). Slopes and intercepts for the independent regressions were compared to determine if a single relationship could be used across environments to predict tiller density using NDVI measurements collected at a 1-m2 resolution (PROC REG; SAS Inst., 1990).

Variable-Rate Nitrogen Applications
Additional field studies were established in 2001 at the ESAREC, Chatham, and Stuarts Draft and in 2002 at the ESAREC, the TAREC, and Charles City. The objective of these studies was to evaluate the wheat grain yield response to variable-rate N applications made at a 1-m2 resolution according to sensor-based estimates of GS-25 tiller density. The trials included five treatments arranged in a randomized complete block design with four replications. Plot size was 5 by 5 m except at Charles City and Chatham where plots were 3 by 5 m. All fields received 28 to 34 kg N ha–1 preplant and had not received any additional N fertilizer at the time tiller density measurements were collected. At GS 25, N fertilizer was applied to plots as urea ammonium nitrate solution (30–0–0) using pressurized backpack sprayers.

One fertilizer treatment was based on average tiller density measurements collected within the experimental area. According to the Virginia fertilizer recommendations for winter wheat, fields with tiller densities below approximately 700 tillers m–2 should receive 67 kg N ha–1, and fields with densities greater than approximately 1000 tillers m–2 do not need to receive any N at GS 25 (Alley et al., 1996). Nitrogen rate recommendations between these critical levels were made using the following linear relationship adapted from Scharf and Alley (1993):

[2]
A second fertilizer treatment was composed of variable N rates applied at a 1-m2 resolution according to sensor-based estimates of tiller density. A two-point calibration technique, similar to the procedure suggested by Flowers et al. (2001) for aerial photography–derived estimates of tiller density, was used to predict tiller densities used in the variable N rate treatment at each location. The intercept was established by determining NDVI for bare soil (0 tillers m–2) at each location or the residual background in a no-till system. The second point was NDVI plotted against a single tiller density measurement collected from an area of the field representing a high tiller density. These two points were used to generate a linear regression equation specific for each site. Predicted tiller densities for each 1 m2 within the plot were entered into Eq. [2] to determine recommended N rates on a 1-m2 basis. The resolution of the variable-rate treatment integrated the possibility of 15 or 25 different N rates plot–1 (depending on plot size). Also included was a single rate that was equivalent to the average of the variable rates and two check plots that received no topdress N at GS 25. At GS 30, all plots received an additional 50 to 67 kg N ha–1 as 30–0–0. Pesticides were administered throughout the season as needed at each location. At maturity, grain was mechanically harvested from each plot for yield determination. Differences among treatments were determined using PROC GLM and mean separation procedures (SAS Inst., 1990).


   RESULTS AND DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
CONCLUSIONS
 REFERENCES
 
Tiller Density Variability
At Painter in 2000, various seeding rates planted on different dates resulted in significantly different (p < 0.05) tiller densities in the spring of 2001 (Table 2). By the mid-February sensing date, the plots planted on 15 November had not fully reached GS 25, and tiller densities following the high and medium seeding rates for the earlier planted wheat were not different. However, by 7 March, all plots had reached GS 25, and all seeding rates for both planting dates had resulted in significantly different tiller densities (p < 0.05; Table 2). These data also indicate that considerable spring tillering can occur in late February and early March in Virginia. Tiller density data presented in Table 3 demonstrate the inherent variability in tiller density that can exist in fields even at a constant seeding rate. Tiller densities at these sites averaged from <500 to >950 tillers m–2, with standard deviations ranging from 43 to 258 tillers m–2 (Table 3). This degree of variability is consistent with that found by Flowers et al. (2001), who reported a standard deviation of 191 tillers m–2 for an on-farm site in North Carolina. The spatial variability of tiller density within randomly selected linear transects at three sites is illustrated in Fig. 1 . These data indicate that tiller density can fluctuate greatly across short distances in the field. It is also apparent that N rate recommendations would vary considerably if applied at a resolution conforming to the variability in the field.


View this table:
[in this window]
[in a new window]
  		Table 2. Variability in winter wheat tiller density following different planting dates and seeding rates at Painter, VA, 2000.

 

View this table:
[in this window]
[in a new window]
  		Table 3. Variability in tiller density at Zadoks Growth Stage 25 in established wheat fields in Virginia, 2000 through 2002.

 


View larger version (38K):
[in this window]
[in a new window]
  		Fig. 1. Inherent variability in winter wheat tiller density and subsequent N rate recommendation across linear transects at three locations in Virginia.

 
Using Eq. [2], the N rate recommendations for the three sites in Fig. 1 based on average tiller density would be 20 to 30 kg N ha–1 (Table 4). However, if recommendations were based on individual measurements of tiller density in each field, N rates would range from 0 to 67 kg N ha–1 (Table 4). Assuming that the variability measured in a transect was representative of the entire field, applying N based on field averages of tiller density would have resulted in only 10 to 51% of the field receiving the optimum N rate (±10 kg N ha–1; Table 4). The inherent spatial variability of tiller density in Virginia fields and the subsequent effect on N fertilization accuracy indicate that sensor-based, variable-rate N applications may be useful in Mid-Atlantic wheat production systems.


View this table:
[in this window]
[in a new window]
  		Table 4. Minimum, maximum, and average N rate recommendations based on Zadoks Growth Stage 25 tiller density for each 1 m2 along linear transects in three established wheat fields in Virginia, and fertilizer application accuracy at a 1-m2 resolution based on the average N rate recommendations.

 
Correlation of Normalized Difference Vegetation Index and Tiller Density
Sensor-based NDVI measurements were not affected by direction of travel relative to row orientation (Fig. 2) . This consistency in the optic field [slope and intercept not significantly different from 1.0 and 0.0, respectively (p < 0.05)] was an important observation as ground coverage at GS 25 is often less than 50%, which could increase the potential for soil or vegetative background to influence spectral reflectance (Lukina et al., 2000). Across site-years, NDVI was well correlated with tiller density (Fig. 3) . The curvilinear response observed in Fig. 3 is similar to those reported by Asrar et al. (1984) and Serrano et al. (2000), who plotted NDVI against leaf area index (LAI). This similarity in curve structure is not surprising as LAI and tiller densities are both representative of biomass in a given area. Serrano et al. (2000) noted that this saturation created a problem when using NDVI for estimating LAI in well-developed canopies. While the same plateau exists when using NDVI to estimate tiller density, the joint occurs beyond the critical tiller density (1000 tillers m–2) where a N recommendation would be made (Fig. 3). Thus, for using NDVI to estimate tiller density for GS-25 N recommendations in Virginia, we would only be concerned with the linear portion of the curve illustrated in Fig. 3, and saturation at higher densities is not problematic.



View larger version (13K):
[in this window]
[in a new window]
  		Fig. 2. Relationship between normalized difference vegetation index (NDVI) measurements collected from the same areas traveling either across or with the planted wheat rows using a hand-held spectral reflectance sensor.

 


View larger version (29K):
[in this window]
[in a new window]
  		Fig. 3. Relationship between winter wheat tiller density and normalized difference vegetation index (NDVI) measurements collected using a hand-held spectral reflectance sensor for 22 site-years in Virginia, 2000 to 2002.

 
Tiller densities above the critical point of 1053 tillers m–2 (based on Eq. [2]) were omitted, and independent linear regression models were generated for each site (Fig. 4) . Coefficients of determination at each location ranged from 0.67 to 0.99 (Table 5). Of the 22 regression models developed, four (Chatham 2001, Culpeper 2001, Charles City 2002, and New Kent 2002) had either a slope or intercept that was significantly different from at least one other site (p < 0.05; Table 5 and Fig. 4). The reasons why the data from these sites were different from the others are not clear. At Chatham 2001, the only unique characteristic was soil type. The color of the soil at Chatham was much darker than most other soil series in this study and was wet at the time of sensing. According to Huete (1987), both of these factors can influence the contribution of soil background to the overall spectral response. Flowers et al. (2003) also reported soil type to significantly influence the relationship between GS-25 tiller density and NIR digital counts and that this effect presents a problem for remote sensing because soil type differences often occur within a single field. Typically, darker soils result in lower NIR reflectance than lighter soils, which would result in lower NDVI at a constant level of vegetation (i.e., tiller density). However, the bare soil NDVI at Chatham (0.18) was not different from that at Suffolk 2001 (0.20), which included one of the lightest soils in the study. Further, if a soil influence were the sole source of variability at Chatham, it would have been expected that the deviation of NDVI values from those collected at other locations would have been greater at the lower tiller densities. In fact, the deviation became greater as tiller density increased, resulting in a significantly different slope rather than intercept (Fig. 4).



View larger version (38K):
[in this window]
[in a new window]
  		Fig. 4. Linear regression trendlines relating winter wheat tiller density and normalized difference vegetation index (NDVI) measurements collected using a hand-held spectral reflectance sensor for 22 site-years in Virginia, 2000 to 2002.

 

View this table:
[in this window]
[in a new window]
  		Table 5. Slope, intercept, and coefficient-of-determination values for the linear regression of normalized difference vegetation index (NDVI) and winter wheat tiller density at 22 experimental sites{dagger} in Virginia, 2000 through 2002.

 
Conversely at New Kent 2002, the intercept was significantly different from the other locations, but the slope was not (p < 0.05; Table 5). New Kent 2002 had been in continuous no-till since 1994. This resulted in an organic residue background rather than bare soil, which could have biased NDVI upward across tiller densities. However, other sites were also part of continuous no-till rotations (Table 1) but did not exhibit the same bias in NDVI measurements as New Kent 2002. For example, Charles City 2002 (no-till since 1991) had a fairly high intercept (non-soil background reflectance value), but the slope, like Chatham 2001, was significantly more shallow than many other sites (p < 0.05; Table 5 and Fig. 4). Culpeper, another continuous no-till site, had a significantly shallow slope and a median intercept compared with other sites (Table 5 and Fig. 4).

Flowers et al. (2003) documented one site out of 10 that had slope and intercept values that were significantly different from the other nine sites and their original model (Flowers et al., 2001). They reported that the differences at this site compared with the others were a substantial weed population that may have degraded the relationship between relative NIR and relative tiller density and extremely low tiller densities, which may have made it difficult to reliably detect differences using remote sensing (Flowers et al., 2003). Culpeper was the only site in our study with considerable weed pressure, which may have affected the spectral measurements. While site-specific calibration techniques might account for some of the weed pressure at a particular location, we agree with Flowers et al. (2003) that weedy fields are not prime candidates for remote sensing and should probably be avoided in applied situations. For the 18 other sites in our study, no significant differences in slopes or intercepts existed (p < 0.05; Table 5.)

The four inconsistent sites were omitted, and a single regression equation across locations was derived (Fig. 5) . The linear relationship had a coefficient of determination of 0.73 (Fig. 5). This result is similar to those reported by Flowers et al. (2001)( 2003), who regressed relative NIR digital counts against relative tiller densities for four sites in North Carolina (r2 = 0.77) and later validated the relationship at 10 more sites (r2 = 0.76). The significant relationship across 18 diverse locations in Virginia suggests that NDVI measurements collected using on-the-go, optical sensors might be able to be used to predict tiller density without having to physically count tillers (Fig. 5). However, there remains the issue of the four sites that had significantly different slopes (Chatham 2001, Culpeper 2001, and Charles City 2002) or intercepts (Chatham 2001 and New Kent 2002; Table 5). If the points from these locations were plotted on Fig. 5, only five of the 40 observations (three from Chatham 2001 and two from New Kent 2002) would fall outside the 95% confidence intervals within the range of tiller densities that would require N fertilization (data not shown). This result suggests that although the slopes and intercepts from individual sites were different from the average across locations, the majority of point estimates were within a range that could reasonably be represented by a single equation. However, to ensure the accuracy of indirect estimates of tiller density and maximize fertilizer N use efficiency (NUE), a two-point calibration (described previously) should be conducted for each major soil type at each site.



View larger version (28K):
[in this window]
[in a new window]
  		Fig. 5. Relationship between winter wheat tiller density and normalized difference vegetation index (NDVI) measurements collected using a hand-held spectral reflectance sensor with 95% confidence intervals for 18 site-years in Virginia, 2000 to 2002.

 
Current sensor-based fertilization strategies already require operators to make inputs into the software system (Raun et al., 2002); thus, this calibration can easily be done in the field at the time of fertilization using commercially available equipment (www.ntechindustries.com). A procedure similar to that proposed by Flowers et al. (2001) could be reliably used to generate a relative relationship that could be used across all environments in Virginia. However, the site-specific calibrations using raw NDVI values result in higher coefficients of determination on average (Table 5) than a single equation across locations (r2 = 0.73) or the relationship based on relative values that was developed and validated in North Carolina (r2 = 0.76; Flowers et al., 2003).

Grain Yield Response to Tiller Density–Based, Variable-Rate Nitrogen Applications
Grain yield responses to GS-25 tiller density–based N applications for six locations are reported in Table 6. For all site-years, a significant increase in grain yield (p < 0.05) occurred as a result of GS-25 N applications (Table 6). The standard N rate recommendation, calculated using Eq. [2] and the average tiller density in the field, averaged 54 kg N ha–1 across the six locations, ranging from 34 to 67 kg N ha–1. Rather than using a common equation for all sites, site-specific linear regression equations to estimate tiller density for the sensor-based treatments were developed at each location using the two-point calibration procedure described previously. Regression parameters for the calibration used at each site are reported in Table 7. The N rate recommendations resulting from sensor-based estimates of tiller density were not different across locations compared with standard recommendations, averaging 51 kg N ha–1 and ranging from 34 to 56 kg N ha–1 (Table 6). While the recommended N rate was not different for the two procedures when averaged across locations, there was some variation at individual sites. At Chatham 2001 and Painter 2001, the sensor-based recommendations were 11 kg N ha–1 lower than those resulting from the standard method. However, no difference in grain yield was observed at either location as a result of the method used to determine GS-25 N application rate (p < 0.05; Table 6). Neither N rate recommendation nor grain yield at the other four sites differed between the two methods.


View this table:
[in this window]
[in a new window]
  		Table 6. Winter wheat grain yield as affected by various methods used to determine Zadoks Growth Stage 25 (GS-25) N application rate based on tiller density at six locations in Virginia, 2001 and 2002.

 

View this table:
[in this window]
[in a new window]
  		Table 7. Linear regression parameters for the two-point calibration technique used to make sensor-based estimates of Zadoks Growth Stage 25 (GS-25) tiller densities for determining variable N application rates at six locations in Virginia, 2001 and 2002.

 
In addition to effectively determining GS-25 N rate compared with the standard method, the range of recommended N rates applied at a 1-m2 resolution using the sensor-based strategy substantiated the considerable variability in tiller density that existed in these fields (Table 6). Only two of the sites (Chatham 2001 and Charles City 2002) had mean N rate recommendations that varied greatly from the median of the range. This difference suggests that the large ranges of N rates applied at these sites were most likely due to a few poor areas within the plot while the ranges of N rates applied at the other sites were more representative of the actual variability that existed (Table 6).

Considering the magnitude of variability in these plots, it is interesting that grain yield was not improved by the variable N rate treatments compared with a fixed N rate as has been the case in other studies using optical sensors at a 1-m2 resolution (Raun et al., 2002). The lack of yield response in this study is further verified by the fact that there were no yield differences between the variable sensor-based treatments (N applied at a 1-m2 resolution) and the fixed sensor-based treatments (single N rate equivalent to the average of all variable rates; Table 6). Based on this result, one might argue that while the need for numerous, physical tiller counts is eliminated, there appears to be no economic benefit to making variable-rate N applications at a 1-m2 resolution at GS 25. A possible explanation for these results is that the fixed N rates applied at GS 30 (Table 6) masked any yield benefits that might have been observed following the variable-rate treatments at GS 25. Another possibility is that small-scale, spatial variability in tiller density has less impact on final grain yield than other variables that are remotely sensed at later growth stages (Raun et al., 2002). Whatever the reason yield increases did not occur following the sensor-based treatment, it is important to note that lower recommended N rates using the sensor did not result in lower grain yields at two locations (Table 6). This result would subsequently increase NUE at these sites compared with the standard method of determining GS-25 N rate. An equally noteworthy point is that the practical application of this technique would be as part of a split-spring, variable-rate fertilization strategy. The second N application at GS 30 would include variable rates determined using procedures similar to those proposed by Raun et al. (2002). Additional work in Virginia has indicated that applying variable N rates at GS 25 at a 1-m2 resolution significantly reduced the variability in remotely sensed grain yield potential estimates at GS 30 (Phillips and Mullins, unpublished, 2003). Future work will be focused on identifying the most efficient strategy for incorporating sensor-based fertilization technology into Virginia's winter wheat production systems.


   CONCLUSIONS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
REFERENCES
 
Previous work has suggested that canopy reflectance can be used to accurately predict winter wheat tiller density for GS-25 N applications (Flowers et al., 2001, 2003). This study evaluated the potential for using commercially available, optical sensor–based fertilization equipment (www.ntechindustries.com) to make these estimates at a 1-m2 resolution while traveling across the field. Across diverse production regions in Virginia, NDVI proved to be a reliable predictor of tiller density, with 18 of 22 sites having slopes and intercepts that were not different from one another, suggesting that a common relationship might be used to predict tiller density across locations (Table 5 and Fig. 5). However, given that current sensor-based applicators allow for producer inputs in the field, we suggest that a site-specific, two-point calibration be performed for each major soil type at each location to maximize accuracy.

Nitrogen fertilizer rates and grain yields resulting from using sensor-based estimates of tiller density were not different from those when using the standard practice at four out of six locations in Virginia (Table 6). At two sites, sensor-based N recommendations were 11 kg N ha–1 lower than standard recommendations with no effect on grain yield, resulting in higher NUE at these locations (Table 6). These results show that on-the-go, optical sensor technology can be used to accurately estimate winter wheat tiller density for determining and applying appropriate N fertilization rates at a 1-m2 resolution with minimal ground truthing required (one physical tiller count for each major soil type in a field).


   ACKNOWLEDGMENTS
 
The authors appreciate the financial support provided by the Virginia Agricultural Council and are extremely grateful to all the growers and Virginia Cooperative Extension agents who made this research possible. We also thank J.T. Custis and R.D. Ashburn for their assistance with field operations and W.R. Raun and G.V. Johnson at Oklahoma State University for providing the spectral reflectance sensor used in these studies.


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




This article has been cited by other articles:


Home page
 Agron. J.Home page
E. M. Pena-Yewtukhiw, G. J. Schwab, and L. W. Murdock
Univariate Distribution Analysis to Evaluate Variable Rate Fertilization
Agron. J., April 11, 2006; 98(3): 554 - 561.
[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 ISI 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 ISI Web of Science (2)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Phillips, S. B.
Right arrow Articles by Mullins, G. L.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Phillips, S. B.
Right arrow Articles by Mullins, G. L.
Agricola
Right arrow Articles by Phillips, S. B.
Right arrow Articles by Mullins, G. L.
Related Collections
Right arrow Nitrogen
Right arrow Wheat

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