| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Dep. of Agron., 210 Waters Hall, Univ. of Missouri, Columbia, MO 65211
* Corresponding author (scharfp{at}missouri.edu)
Received for publication January 17, 2001.
 |
ABSTRACT |
|---|
 TOPNOTES ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Abbreviations: EONR, economic optimum nitrogen rate • NDVI, normalized difference vegetation index • NIR, near infrared
|
INTRODUCTION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Economic benefits would result from N fertilizer savings in areas that would otherwise be overfertilized and from yield increases in areas that would otherwise be underfertilized. The net benefit is not likely to be large after paying technology and management costs.
Environmental benefits would result from a reduction in the amount of surplus N in the production system. Overfertilization as a form of insurance is common. The difference between the applied N rate and the economic optimum N rate (EONR) is a strong predictor of soil water NO-3 concentration and residual soil NO-3 after harvest (Andraski et al., 2000). Net percolation in humid regions of the midwestern USA occurs primarily between harvest and the following spring; NO-3 or labile N present in the soil after harvest will be susceptible to movement out of the root zone with the percolating water. This N can then move to ground water, or it can move to surface water via drainage (Fenelon and Moore, 1998; David et al., 1997) or base flow (Steinheimer et al., 1998). Nitrogen rates in excess of crop need can radically increase postharvest NO3–N accumulation in soil (Schuman et al., 1975; Rice et al., 1995) and the amount of N entering streams via subsurface base flow (Burwell et al., 1976). Nitrate in drinking water is a human health concern, and N in surface water has been implicated in coastal and estuarine eutrophication, including a hypoxic zone in the Gulf of Mexico (Turner and Rabalais, 1994).
Corn color is related to its N status. Nitrogen-deficient corn reflects more light over the entire visible spectrum than N-sufficient corn (Al-Abbas et al., 1974; Blackmer et al., 1994) and usually reflects less near-infrared (NIR) radiation than N-sufficient corn (McMurtrey et al., 1994). Differences in reflectance are usually greatest for wavelengths 550 to 600 nm (Blackmer et al., 1994; McMurtrey et al., 1994). These reflectance differences are associated with differences in chlorophyll concentrations in leaves (Hong et al., 1997).
Minolta chlorophyll meters have been widely used to quantify corn color and N status (e.g., Piekielek and Fox, 1992; Sims et al., 1995). Research has shown that they can be used to translate corn color into N management decisions, including sidedress N rate recommendations (Piekielek et al., 1997; Blackmer and Schepers, 1995; Varvel et al., 1997). However, collecting enough data to adequately characterize field-scale or sub-field-scale N status is labor intensive. A large number of readings is required because each reading represents only a small leaf area (<2 mm diam.). These readings must be spatially well distributed over the field or subfield, and each reading requires hand-clamping the meter on a leaf.
Information about corn color from aerial photographs or satellite images is potentially cheaper, more robust (because the leaf area being sampled is many orders of magnitude greater), and more spatially detailed than information collected with chlorophyll meters. Pioneering research has shown that corn color (especially the red and green regions) measured in late-season aerial photographs is quantitatively related to N stress and yield (Blackmer et al., 1996; Blackmer and White, 1998; Blackmer and Schepers, 1996). Beatty et al. (2000) showed a high correlation (R2 > 0.90) between midseason (near silking) chlorophyll meter measurements of corn and airborne hyperspectral radiance measurements over a wide range of wavelengths, about 510 to 630 nm. This type of information is most useful if it can be obtained when there is still time to respond to deficiencies with a N fertilizer application—tasseling at the latest and preferably at a time when sidedress N applications can be made without requiring high-clearance equipment.
The large proportion of soil that is visible in a nadir (straight down) view of corn at normal times of N sidedressing presents a major technical obstacle to measuring corn color from aerial or satellite images. Walburg et al. (1982), using ground-based radiometers to measure canopy reflectance, found that they were able to clearly distinguish extreme N deficiencies by the eight-leaf stage but apparently no sooner. Although many indices have been developed to correct for soil color in mixed soil–plant views, their aim has almost universally been to accurately assess plant cover—with little concern for the color of the plants—despite variations in soil color. Only very recently have attempts been made to develop indices that are sensitive to plant color in a mixed soil–plant scene (Daughtry et al., 2000; Gitelson et al., 1996; Baret and Fourty, 1997; Clarke et al., 2000). We hypothesized that pixel sorting in very high resolution images (to discard soil pixels) would be a viable alternative to indices for accurately measuring corn color at stages V6 to V7 (growth stages described by Ritchie et al., 1993), despite the presence of a large proportion of soil in the images.
Our objective was to extend earlier work in three ways: (i) to calibrate corn color measured from aerial photographs not only to detect N stress, but also to predict the amount of N needed in response to the stress; (ii) to do so at a time when sidedressing can still be accomplished without high-clearance equipment; and (iii) to do so across a broad range of production environments. We anticipate that these calibrations could be used to transform aerial photographs into maps to direct variable-rate sidedress N applications.
|
MATERIALS AND METHODS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
|
Experiments also contained additional treatments that received 0, 112, or 224 kg N ha-1 at planting and additional N at times later than V7. Yield results from these treatments are not reported here. These treatments were also intended to be used to characterize the average V6–V7 color associated with these three planting N rates at each location. At V6 to V7, each location had 16 plots that had received zero N at planting, 10 plots that had received 112 kg N ha-1 at planting, and 10 plots that had received 224 kg N ha-1 at planting available for color measurements.
Minolta SPAD 502 chlorophyll meter readings were taken from the uppermost collared leaf at V6 or V7. Readings were taken from 10 plots with each of three planting N rates (0, 112, or 
224 kg N ha-1) at each site. Readings were taken on 20 plants per plot in 1997 and 10 plants per plot in 1998, except at Locations 1, 4, 8, and 9 where only five plants per plot were measured due to time constraints.
The center two rows of each plot were hand-harvested (1.5 by 6 m). Harvest population was also determined in this area. Grain was shelled and weighed, and grain moisture was determined with a hand-held moisture meter. Yields were corrected to a moisture of 150 g kg-1.
Determination of Economic Optimum Nitrogen Rate
When population effects on yield were observed, yields were corrected to the mean population for the location. Grain yield response to N rate was modeled as a quadratic-plateau function using PROC NLIN in SAS. An F-test was used to compare a single model vs. separate models for the different timing treatments at each location. Economic optimum N rate was calculated for each location using a N/grain price ratio of 5:1. PROC UNIVARIATE in SAS was used to compute the probability that EONR was distributed normally.
Photograph Acquisition and Color Measurements
Aerial photographs were taken from an altitude of approximately 150 m at growth stage V6 or V7. The low altitude of the photographs was selected to achieve very fine pixel size, which we anticipated would be critical for differentiating between soil and plant. We used both color-positive (Kodak Elite Chrome 100, Kodak, Rochester, NY) and color-infrared (Kodak Color-IR) film in 35-mm format. A yellow filter was used with the color-infrared film. Some photographs were taken in sunny weather and some in overcast conditions. Severe overexposure of color-infrared images for Locations 10, 11, and 16 precluded their use in our analyses.
All scanning and image analysis procedures were identical for both film types. Photographs were digitized with a slide scanner (Nikon LS-10E, Nikon, Tokyo) at 1064 cm-1 resolution. Pixel size of the digitized images was typically 2.5 to 3 cm. Each pixel was associated with 8-bit (i.e., ranging from 0–255) values for red, green, and blue. With the color-infrared film, these values corresponded to the intensity of NIR, red, and green light, respectively, striking the film. (In the Results section, reference will be made to the color of light exposing the film rather than to the color of the pigment developed on the film.)
Adobe Photoshop was used to extract color measurements from the digitized images. Median values for red, green, and blue were recorded for the center two rows of each measured plot and also for the plant pixels selected from the center two rows of each plot (Fig. 2) . Plant pixels were selected using the color range selection procedure in Adobe Photoshop. The input parameters for this procedure are typical plant color and fuzziness. Higher fuzziness values select pixels with colors farther from the chosen color, but we do not know the algorithm used by the Photoshop program to make this selection. Parameter values were selected independently for each image that successfully, in the judgement of the person analyzing the image, separated plant pixels from soil pixels over the range of plant and soil colors encountered in that image. Plant pixels selected in this way usually represented about 50% of the total pixels in a given area. We feel that this sorting was relatively conservative and erred on the side of retaining some nonplant pixels in preference to discarding any plant pixels. Our analysis of nadir-view photographs taken from about 2 m above the canopy usually scores only 25 to 40% of the pixels as plant material at V6 to V7.
|
224 kg N ha-1 at planting. We observed vignetting, or concentric gradients of brightness, in our aerial photographs. Vignetting was probably exacerbated by the low altitudes at which the photographs were taken and by the relatively short (50 mm) focal-length lenses used, resulting in different angles from camera to target to sun at different points in the photograph. We did not expect absolute color measured from the aerial photographs to be very useful due to variations in light conditions, camera settings, and film processing between locations; the observed vignetting is one more reason why absolute color would not be useful. In order to minimize the effect of vignetting on relative color measurements, readings were taken from closely grouped sets of three plots (0, 112, and 224 kg N ha-1 at planting). At most locations, we were able to identify about eight such sets of three plots from which to take readings. Relative color was then calculated as the color value from a 0 or 112 kg N ha-1 plot divided by the value for the same color from the nearby 224 kg N ha-1 plot. If a zero-N plot was in a brighter-than-average part of the photograph, the nearby high-N plot would also be brighter than average, and relative colors should be fairly accurate.
Relating Color Measurements to Economic Optimum Nitrogen Rate
Experiments were designed to determine the EONR for each experiment with zero N at planting and also, independently, with 112 kg N ha-1 at planting and to measure plant color at V6 on plots that had received these planting N rates. Regression analysis was used to identify color measurements that would be useful to predict optimum sidedress N fertilizer rate. Because EONR is a property of an experimental location, all color measurements used in these regressions were location averages. Thus, for a given color measurement, each nonmanured experiment produced two data points (one for zero N at planting and one for 112 kg N ha-1 at planting) relating color to optimum sidedress N rate, and each manured experiment produced only one data point (for zero N at planting because the treatments with 112 kg N ha-1 at planting followed by sidedress N rates were omitted, as were sidedress color measurements from plots receiving 112 kg N ha-1 at planting).
In addition to absolute and relative red, green, and blue values from color film, absolute and relative red/blue, green/blue, and red/green ratios were tested. The ratios with blue in the denominator were based on the idea that blue might serve as a normalization factor because blue reflectance of individual corn leaves has been reported to be relatively insensitive to leaf N status (McMurtrey et al., 1994; Blackmer et al., 1994). The red/green ratio was suggested by results reported by Daughtry et al. (2000). From color-infrared film, absolute and relative NIR, red, green, green/NIR (Bausch and Duke, 1996; Schepers et al., 1996), normalized difference vegetation index (NDVI; Tucker, 1979), and green NDVI (Gitelson et al., 1996) were tested along with an approach suggested by Daughtry et al. (2000) in which NIR/green is plotted against NIR/red and the slope is indicative of chlorophyll concentration. The best single-variable predictors of EONR were tested in combination with all other color variables from the same film type.
Relative colors from photographs were also regressed against relative SPAD chlorophyll meter readings as an indicator of the quality of the color measurements obtained from the photographs. Experimental unit was again location, and location averages were used for all color measurements in this analysis. Relative SPAD meter values were thus based on a minimum of 50, and more commonly 100 or 200, readings per location.
|
RESULTS AND DISCUSSION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
= 0.05) by separate models for the different timing treatments than by a single model across all timings; this would be expected by random chance with this and population size. Economic optimum N rate ranged from 0 to 336 kg N ha-1 in these experiments, with a mean of 156 kg N ha-1, a standard deviation of 96 kg N ha-1, and no evidence of nonnormality.
|
|
|
Removing soil pixels from the images also improved the correlations between relative corn color from aerial photographs and relative chlorophyll meter readings (Table 3). Though our pixel-sorting algorithm was poorly defined, this provides evidence that it did in fact result in better estimates of plant color.
|
Color measurements from color-infrared film were less useful for predicting EONR. For 15 locations for which we have data, relative green was the best predictor of EONR, with R2 = 0.50 when measured from the whole image with either 0 or 112 kg N ha-1 at planting. Unlike color film, discarding soil pixels or eliminating plots that had received 112 kg N ha-1 at planting did not improve the quality of prediction. However, the locations where planting N caused the most difficulties with predicting N need from color film (10 and 11) were missing from this analysis.
Implications for Use in Nitrogen Fertilizer Management
Timing is a critical question for N management based on corn color measurements. Chlorophyll meter readings have mainly, in nonirrigated production, been calibrated for corn growth stage V6 to maximize ease of N application. A majority of the nadir view of a corn field at this stage is soil—which, if corn color is to be remotely sensed, represents noise masking the information contained in crop color. We regard our results as proof of concept that pixel sorting in high-resolution images (Fig. 2) is a viable solution to this problem and can be used to obtain crop color measurements that are predictive for N need. Though color measurements were significantly related to N need without pixel sorting, the quality of that relationship was greatly improved by sorting.
Application of this concept to produce variable-rate N recommendation maps for whole fields would require much larger film size (to produce comparable resolution in whole-field images) and the development of a consistent and known pixel-sorting algorithm. Both acquisition and processing of extremely high resolution images at a commercial scale present substantial technical obstacles. The vignetting problems that we experienced might be minimized or eliminated with longer focal lengths.
The index approaches (Gitelson et al., 1996; Baret and Fourty, 1997; Clarke et al., 2000) to assessing plant color in a mixed scene have a great advantage over our approach in that they do not require ultra-high resolution images. However, these indexes have to date only been tested via laboratory measurements and computer simulation (Gitelson et al., 1996; Daughtry et al., 2000) or ground-based spectroradiometry (Baret and Fourty, 1997; Clarke et al., 2000). Green NDVI (Gitelson et al., 1996) did not prove to be well related to corn N need in our data set. Nor did there appear to be any relationship between optimum N rate and position on a NIR/red vs. NIR/green grid as proposed by Daughtry et al. (2000). We were unable to test the other two approaches due to the need for middle-infrared (Baret and Fourty, 1997) or far-red (Clarke et al., 2000) bands. Baret and Fourty (1997) acknowledge that "for low leaf area indices, the discrimination between different levels of chlorosis will be difficult."
Even when soil pixels were removed from images, variations in soil color within an experiment were reflected in the measured plant colors—plots with lighter soil gave lighter plant color measurements (data not shown). This may be due to transmission of soil-reflected light through the leaves or at the edges of leaves. Thus, large variations in soil color in a field are likely to reduce the quality of plant color–based sidedress N rate recommendations that can be produced from aerial photographs for that field.
One potential weakness in extrapolating from our results to field-scale recommendations is the distance between the high-N reference area and the area to be fertilized. In these experiments, the distance was very short, in particular because of the way that we compensated for vignetting in the photographs. With greater distances, the quality of the N rate recommendations could go down.
In addition to the technical obstacles to sidedressing N based on aerial photographs, there are major management and risk obstacles. Few producers currently use sidedress N management for corn. Those who do usually have finished applying N by growth stage V6, which is probably the earliest stage at which color can reliably predict N need. To wait until stage V6 to obtain photographs and then process the photographs to produce variable-rate application maps leaves a very narrow window of time to apply the N before the corn is too tall for conventional sidedressing equipment. Having high-clearance N application equipment available would be necessary to avoid an unacceptably high risk of not being able to apply N fertilizer to the field. Even with the capability for high-clearance N application, an extended wet period could preclude application or delay it enough to reduce yields. Insurance policies have been suggested as a possible way to manage this kind of risk (Green, 1999).
Making a modest N application at planting and then sidedressing based on corn color from aerial photographs would present several advantages over sidedress-only management: The amount of yield loss associated with being unable to sidedress due to weather would be much lower, and probably not all fields would need to be sidedressed. Unfortunately, when 112 kg N ha-1 had been applied at planting, we were not able to find any significant relationships between corn color in aerial photographs at stage V6–V7 and EONR.
Inability to relate corn color to EONR when N had been applied at planting might suggest that color would also poorly predict EONR for manured fields. However, the five manured site-years (of which two were N responsive) in this study followed very much the same relationship between color and EONR as the nonmanured experiments.
|
CONCLUSIONS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
NOTES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
REFERENCES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  R. P. Sripada, R. W. Heiniger, J. G. White, and A. D. Meijer Aerial Color Infrared Photography for Determining Early In-Season Nitrogen Requirements in Corn Agron. J., June 5, 2006; 98(4): 968 - 977. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. C. Scharf, S. M. Brouder, and R. G. Hoeft Chlorophyll Meter Readings Can Predict Nitrogen Need and Yield Response of Corn in the North-Central USA Agron. J., May 3, 2006; 98(3): 655 - 665. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. P. Sripada, R. W. Heiniger, J. G. White, and R. Weisz Aerial Color Infrared Photography for Determining Late-Season Nitrogen Requirements in Corn Agron. J., September 19, 2005; 97(5): 1443 - 1451. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
W. R. Raun, J. B. Solie, M. L. Stone, D. L. Zavodny, K. L. Martin, and K. W. Freeman AUTOMATED CALIBRATION STAMP TECHNOLOGY FOR IMPROVED IN-SEASON NITROGEN FERTILIZATION Agron. J., January 1, 2005; 97(1): 338 - 342. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. F. Bronson, T. T. Chua, J. D. Booker, J. W. Keeling, and R. J. Lascano In-Season Nitrogen Status Sensing in Irrigated Cotton: II. Leaf Nitrogen and Biomass Soil Sci. Soc. Am. J., September 1, 2003; 67(5): 1439 - 1448. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. A. Lory and P. C. Scharf Yield Goal versus Delta Yield for Predicting Fertilizer Nitrogen Need in Corn Agron. J., July 1, 2003; 95(4): 994 - 999. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 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 | |||
