PDF version
for printing

© 2006 Plant Management Network.
Accepted for publication 27 February 2006. Published 5 April 2006.

Attempt to Validate a Remote Sensing-Based Late-Season Corn Nitrogen Requirement Prediction System

Ravi P. Sripada, Department of Crop Science, North Carolina State University, Box 7620, Raleigh 27695-7620; Ronnie W. Heiniger, Department of Crop Science, Vernon James Research and Extension Center, North Carolina State University, 207 Research Road, Plymouth 27962; Jeffrey G. White, Department of Soil Science, North Carolina State University, Box 7619, Raleigh 27695-7619; and Carl R. Crozier, and Alan D. Meijer, Department of Soil Science, Vernon James Research and Extension Center, North Carolina State University, 207 Research Road, Plymouth 27962

Corresponding author: Ravi P. Sripada. Ravi_Sripada@ncsu.edu

Sripada, R. P., Heiniger, R. W., White, J. G., Crozier, C. R., and Meijer, A. D. 2006. Attempt to validate a remote sensing-based late-season corn nitrogen requirement prediction system. Online. Crop Management doi:10.1094/CM-2006-0405-01-RS.

Abstract

Recent research showed that corn (Zea mays L.) may respond profitably to N applied at tasseling (VT), and that aerial color-infrared (CIR) photography could be used to predict economic optimum N rates at VT using the green difference vegetation index (GDVI, near-infrared brightness minus green brightness) calculated relative to a high N reference strip (relative GDVI, RGDVI). This technique could be used by farmers to identify N stress and quantify N requirements. To validate this technique for practical application, experiments were conducted with different rates of N applied at planting and at VT on irrigated and non-irrigated sites in North Carolina during 2003. In both irrigated and non-irrigated systems, maximum yield potential was achieved with 200 lb of N per acre at planting. The difference between predicted and observed optimum N rate at VT ranged from -27 to 80 lb/acre. Greater differences between predicted and observed optimum N rate at VT occurred when N requirement was high, which was attributed to lower yield potential in 2003 compared to the years when the model was developed. Although the model tended to over-predict N rates, it did capture differences in N requirements across the range of conditions tested, indicating that this technique can be an effective tool to determine late-season corn N need. Differences between the estimated and actual N rates might be reduced by incorporating a method for determining yield potential.

Introduction

The traditional practice for N management in corn (Zea mays L.) in the southeast USA has been to apply some N at planting along with a sidedress application between the V3 and V7 growth stages (3,6). However, since approximately one-third of the total N uptake in corn occurs after tasseling (VT) (1,3,6), and research has demonstrated yield increases to N applied at VT (9), there is potential to both meet late-season N requirements and minimize N losses by applying N as late as VT. While most growers in the southeast USA do not apply N this late, they might adopt this practice given an accurate method of determining late-season N requirements with the potential to improve N-use efficiency and minimize N losses to the environment. Recent research showed that aerial color-infrared (CIR) photography can be used to determine in-season N requirement just prior to VT (9).

Traditional methods of estimating optimum N requirements for corn are based on soil testing (5), tissue N concentrations (10), and chlorophyll concentration (leaf greenness) (2). These methods are labor intensive, time consuming, and may not be economical for large fields. Remote sensing via aerial film or digital photography can provide valuable crop information both within fields and within seasons. The brightness of the green band from aerial color photographs, measured relative to a well fertilized reference strip (i.e., "relative brightness of green") has been used with some success to predict optimum sidedress N in corn at V6-V7 growth stage (8). From aerial color-infrared (CIR) photographs, the green difference vegetation index (GDVI), which is the near-infrared (NIR) brightness minus the green brightness, calculated relative to a well fertilized reference strip that provides a relative GDVI (RGDVI) has been shown to estimate economic optimum N rate at VT (9).

From a farmer’s perspective, the adoption of a remote sensing technique to predict in-season N requirements would depend in part on the accurate prediction of that requirement resulting in greater profit and/or environmental stewardship via a reduction in N used or an increase in N-use efficiency. Our objective was to validate the remote sensing-based late-season N requirement prediction model for corn developed by Sripada et al. (9) to determine if it could accurately predict late-season N rates for growers in the southeast USA.

Field Study Design and Validation Procedure

Five field studies were conducted in 2003 in North Carolina at each of two locations: the Peanut Belt Research Station (PBRS) near Lewiston-Woodville and the Tidewater Research Station (TRS) near Plymouth. Five of the field studies were irrigated and five were not, resulting in a total of ten sites at two locations to test the model. All experiments were planted with conventional tillage following a corn crop. At each site, a two-way factorial design was implemented in randomized complete block design with a split-plot treatment structure with two replications. The soil at PBRS is a Norfolk sandy loam (fine-loamy, siliceous, thermic Typic Paleudults) and at TRS is a Hyde loam (fine-silty, mixed, thermic Typic Umbraquults). Nitrogen applied at planting was the main plot factor and N applied at VT the sub-plot factor. The main plots were 30 ft by 84 ft with 3-ft row spacing. The subplots were four rows wide. ‘Pioneer 31G98’ was planted at approximately 24,000 seeds per acre. Urea-ammonium nitrate solution (UAN, 30% N) was surface applied at planting and VT using a CO₂-pressurized backpack sprayer. The N rates at planting were 0, 50, 100, and 200 lb/acre, and N rates at VT were 0, 50, 100, 150, 200, and 250 lb/acre.

Fertilizer rates other than N were based on North Carolina Department of Agriculture (NCDA) soil test results and recommendations (4). Conventional tillage was used and herbicides applied based on weeds present. Excellent weed control was obtained at all sites. From planting to R5 (dent stage), if precipitation over a five-day period was less than 0.5 inch, then irrigated plots received overhead sprinkler irrigation at the rate of 1.0 inch/week.

Grain was harvested from the center two rows of each plot using a Gleaner (AGCO Corp., Duluth, GA) combine. Moisture content and yield were recorded using a HarvestMaster Grain Gauge (Juniper Systems, Inc., Logan, UT). Grain yield was adjusted to moisture content of 15.5%. Grain yield responses to irrigation, N applied at planting and N applied at VT were analyzed using PROC GLM in SAS version 8 (SAS Institute Inc., Cary, NC).

Image Acquisition and Conversion to Color Values

Aerial CIR photographs were taken at VT at each site from altitudes (~2500 to 3000 ft) such that the entire experimental field was covered in a single image under conditions as cloud-free as possible. The spectral properties of the CIR film and the procedure to obtain the images are described by Sripada et al. (9). The ground resolution of images was 1.4 to 1.8 ft. Using georeferenced targets, images were georegistered in ERDAS Imagine version 8.7 (Leica Geosystems, Atlanta, GA). Areas of interest corresponding to individual plots were identified and included approximately the same number of pixels. The areas of interest included both corn plants and any soil visible between rows; there was no separation of soil and crop pixels. The mean digital number representing the brightness for each spectral band for each plot was determined from the areas of interest. Using the digital number for the individual bands, spectral indices GDVI (i.e., NIR – G) and RGDVI (i.e., [NIR – G]_plot/[NIR – G]_{reference plot}, where a reference plot was the plot that received the highest N rate at planting at a site) were calculated. To avoid working with negative values, 255 and 1 were added to GDVI and RGDVI, respectively.

Predicted and Observed Economic Optimum N Rates at VT

The predicted optimum N rate at VT was calculated using the linear-plateau function of RGDVI illustrated in Fig. 1 (9):

Economic  Optimum N rate at VT (lb/acre) =
165 + [-3105 × (RGDVI – 0.9563)]

Fig. 1. Model showing the relationship between economic optimum Nvt rate and RGDVI (9).

Grain yield response to N applied at VT was modeled as a quadratic-plateau function using PROC NLIN in SAS Version 8 (SAS Institute Inc., Cary, NC). The observed economic optimum N rates at VT (hereafter, "optimum N_vt") were calculated using the first derivative of the quadratic-plateau model and a price ratio of 4:1, the ratio of the price per pound of N to the price per pound of corn. If a response did not fit a quadratic-plateau function as determined by the model significance (alpha = 0.05), treatment means were compared using Fisher’s protected LSD to determine the optimum N_vt. Where the yield response to fertilizer N was not significant by either of these methods, the optimum N_vt was set equal to zero. The NIR, red, and green digital number and the spectral indices calculated from the aerial photograph were regressed against the optimum N_vt using four different models: linear and quadratic models were fit using PROC REG; linear-plateau and quadratic-plateau models were fit using PROC NLIN in SAS Version 8.

To test the accuracy of the N requirement prediction model, the observed optimum N_vt were plotted against the predicted optimum N_vt. The difference between the predicted and observed optimum N_vt rates were calculated and plotted to quantify the under-and/or over-prediction of optimum N_vt.

Grain Yield Response to N

Across sites and N treatments, split-plot grain yield ranged from 33 to 205 bu/acre (Table 1). The main effect of N rate at planting was significant as illustrated by the different intercepts for the grain yield response to N rate at VT at different N rates at planting (Fig. 2). The corn plots receiving zero N at planting lost yield potential, which was not regained even after applying high N rates at VT , with or without irrigation. These resulted in a lower maximum yield and thus lower optimum N_vt compared to treatment with 50 lb of N per acre at planting (Fig. 2a and b). When the plots were irrigated, N rates of 50 and even 100 lb/acre at planting lost yield potential that was not recoverable. This illustrates that when sidedress N is to be applied at or near VT adequate N must be applied at planting and/or at layby (e.g., V7-8) to maintain yield potential (3). In both irrigated and non-irrigated systems, maximum yield potential was achieved with 200 lb of N per acre at planting. One of the reasons for the irrigated plots yielding lower (Table 1) could be the high intensity precipitation events that occurred immediately following the N applications at planting, resulting in lower utilization of N applied. Overall, during the 2003 growing season, there were moderate temperatures (0.9°F warmer than 30-year average) and high rainfall (8.7 inches more precipitation than 30-year average).

Table 1. Split-plot minimum, maximum, mean, and standard deviation for corn grain yields obtained at different experimental sites.

 ^x PBRS = Peanut Belt Research Station; TRS = Tidewater Research Station.

 ^y IR = Irrigated; NI = Non-irrigated.

Fig. 2. Quadratic-plateau fit of corn grain yield response to N applied at VT for different rates of N applied at planting at (a) an irrigated site at the Peanut Belt Research Station (PBRS) and (b) a non-irrigated site at the Tidewater Research Station (TRS) during 2003. Each point is the mean of two replicates.

Predicting Optimum N Rate at Tasseling from Spectral Data

The different N rates at planting created a range of spectral variability among plots that was evident in the aerial CIR photographs and resulted in a wide range of optimum N_vt. The range of optimum N_vt was 0 to 134 lb/acre with a mean of 61 lb/acre. This was a lower mean and narrower range of optimum N_vt compared to the previous trials from which the model was developed (0 to 196 lb/acre, mean = 93 lb/acre) (9). This was probably due to the fact that mean yield levels (Table 1) obtained in this study were lower than those obtained in earlier studies (9).

Consistent with earlier studies, a better prediction of optimum N_vt was observed with relative indices (Table 2) than with individual spectral bands or absolute indices. Overall in the present study, a linear-plateau model using RGDVI was the best predictor (r² = 0.81; Table 2 and Fig. 3) of optimum N_vt. The narrow range of RGDVI (0.9524 to 1.02) that was responsive to optimum N_vt was of particular importance because even small variations in RGDVI could change the predicted optimum N_vt substantially (Fig. 3). That said, it is important to realize that RGDVI is a relative measure, the full range of which we have observed to be about 0.84 to 1.02, that is, about 0.18 "RGDVI" units. The optimum N_vt responsive range in RGDVI is thus ~38% of the observed range. These results were consistent with those obtained by Sripada et al. (9) when developing the model being validated in this study.

Table 2. Regression analysis of economic optimum N rate (lb/acre) at tasseling versus near-infrared (NIR), red (R), green (G), the Green Difference Vegetation Index (GDVI), and relative GDVI (RGDVI) across all response trials in this study. The model significance and the coefficient of determination (r² or R²) for the linear, linear-plateau, quadratic, and quadratic-plateau models are given.

*, **, NS Significant at the 0.05 and 0.01 probability levels and not significant, respectively.

Fig. 3. Comparison of the remote sensing-based in-season N-requirement prediction model (9) to the model obtained in this validation study.

Comparison of the In-season N Requirement Prediction Model to the Model Obtained in this Validation Study

Statistical tests showed that there was no significant difference (not shown) in the slope (b) and inflection point (x₀) between the linear-plateau model parameters of the model developed by Sripada et al. (9) and those obtained in this study. However, the intercept a, which represents the maximum optimum N_vt, was significantly different between the two models, reflecting the maximum optimum N_vt of 115 and 165 lb/acre for the model from this study versus the developed model, respectively (Fig. 3). This might have been due to greater loss of yield potential caused by delaying the N application until VT that occurred in this study compared to the previous trials from which the model was developed (9).

Validation of Predicted Optimum N Rates at VT

A linear regression of the predicted versus observed optimum N_vt indicated that the model accounted for 85% of the variability in observed optimum N rates (Fig. 4). In general, the model overestimated optimum N_vt. When 200 lb of N per acre was applied at planting the model in this study over-predicted N rate at VT by 29 lb of N per acre consistently (Fig. 5). When N at 0 to 100 lb/acre was applied at planting, the model in this study under or over estimated optimum N_vt by -27 to +80 lb of N per acre, with overestimation predominating (Fig. 5). There was a narrower range of optimum N_vt observed in this study (0 to 134 lb/acre, mean = 61 lb/acre) compared to the range in optimum N_vt found in the previous studies in which the model was developed (0 to 196 lb/acre, mean = 93 lb/acre) (9). The lower yields observed in this study and the tendency of the model to over predict optimum N_vt rates suggest that the in-season N prediction model might be improved if it could be adjusted for yield potential in a given season. Shanahan et al. (7) suggested that corn Green Normalized Difference Vegetation Index (GNDVI) measured during mid–grain-filling period was highly correlated with grain yield. We are exploring our data from this and other studies to determine if CIR at VT can provide some indication of yield potential. While the model currently does not predict optimum N_vt accurately enough for use by farmers, the model does highlight the variability associated with yield response to N applied at VT.

Fig. 4. Relationship between the predicted economic optimum N rates at VT (Nvt) using the model developed by Sripada et al. (9) and the observed economic optimum N rates during the 2003 growing season.

Fig. 5. Scatter plot of the residuals of the predicted economic optimum N rates at VT (Nvt) using the model developed by Sripada et al. (9) and the observed economic optimum N rates during 2003.

Summary and Conclusions

At present, most corn growers do not apply sidedress N as late as VT. This study and the previous study (9) demonstrated significant increases in yield in response to N applications at VT, indicating that growers could maintain yield potential yet delay final N applications to better match fertilization timing with crop uptake. However, this study also demonstrated that maximum yields were obtained when N was applied at 200 lb/acre at planting with no N applied at VT. This study shows that:

(i) a linear-plateau function similar to that developed by Sripada et al. (9) resulted in the best fit for the relationship between optimum N_vt and RGDVI;

(ii) the remote sensing-based model was reasonably successful (r² = 0.85) in predicting the optimum N_vt;

(iii) when optimum N_vt is estimated using this linear-plateau function, the maximum optimum N_vt rate can vary slightly from year to year due to changes in yield potential caused by changes in weather and/or management.

This remote sensing-based in-season N prediction model can likely be enhanced by an adjustment based on estimated yield potential. The obstacles to the adoption of this technique include the unpredictability of yield response to N at VT in different environments, necessity for high-N reference strips in the field (9), and need for high-clearance applicators to apply N at VT. If these obstacles can be overcome, the remote sensing based technique offers promise to adjust N rates during the growing season.

Acknowledgments

We thank Jared Williams, Dianne Farrer, and Brian Roberts for their excellent field support and Dr. Marcia L. Gumpertz for insights on the statistical analyses. We thank the support staff at the Peanut Belt Research Station, Lewiston-Woodville and Tidewater Research Station, Plymouth, NC for their help in field work. This project was supported in part by Initiative for Future Agriculture and Food Systems Grant no. 00-52103-9644 from the USDA Cooperative State Research, Education, and Extension Service.

Literature Cited

1. Bigeriego, M., Hauck, R. D., and Olson, R. 1979. Uptake, translocation, and utilization of ¹⁵N-depleted fertilizer in irrigated corn. Soil Sci. Soc. Am. J. 43:528-533.

2. Blackmer, T. M., and Schepers, J. S. 1995. Use of a chlorophyll meter to monitor nitrogen status and schedule fertigation of corn. J. Prod. Agric. 8:56-40.

3. Crozier, C. R. 2002. Fertilizer and lime management. Pages 30-36 in: Corn Production Guide 2002-2003. R. W. Heiniger, ed. North Carolina Coop. Ext. Serv., Coll. of Agric. and Life Sci. North Carolina State Univ., Raleigh, NC.

4. Hardy, D. H., Tucker, M. R., and Stokes, C. E. 2005. Crop fertilization based on North Carolina soil tests. Raleigh (NC): North Carolina Department of Agriculture and Consumer Services, Agronomic Division. Circular No. 1.

5. Magdoff, F. 1991. Understanding the Magdoff Pre-Sidedress Nitrate Test for corn. J. Prod. Agric. 4:297-305.

6. Ritchie, S. W., Hanway, J. J., and Benson, G. O. 1993. How a corn plant develops. Spec. Rep. 48. Iowa State Univ., Ames.

7. Shanahan, J. F., Schepers, J. S., Francis, D. D., Varvel, G. E., Wilhelm, W. W., Tringe, J. M., Schlemmer, M. R., and Major, D. J. 2001. Use of remote-sensing imagery to estimate corn grain yield. Agron. J. 93:583–589.

8. Scharf, P. C., and Lory, J. A. 2002. Calibrating corn color from aerial photographs to predict sidedress nitrogen need. Agron. J. 94:397-404.

9. Sripada, R. P., Heiniger, R. W., White, J. G., and Weisz, R. 2005. Aerial color infrared photography for determining late-season nitrogen requirements in corn. Agron. J. 97:1443-1451.

10. Tyner, E. H., and Webb, J. W. 1946. The relation of corn yields to nutrient balance as revealed by leaf analysis. J. Am. Soc. Agron. 38:173-185.