You are on page 1of 25

Paper No.

99 – 3034
An ASAE Meeting Presentation

ASSESSING NITROGEN STRESS IN CORN VARIETIES OF VARYING COLOR

by

Wonsuk Lee1 Stephen W. Searcy1 Takashi Kataoka2
Postdoctoral Research Associate Professor Visiting scholar

1 2
Department of Agricultural Engineering Department of Agricultural Engineering
Texas A&M University Iwate University
College Station, TX77843, USA Morioka, Iwate 020-8550, Japan
wlee@agen.tamu.edu & s-searcy@tamu.edu tkataoka@iwate-u.ac.jp

Written for Presentation at the
1999 ASAE Annual International Meeting
Sponsored by ASAE

Sheraton Centre Toronto Hotel
Toronto, Ontario Canada
July 18-21, 1999

Summary: Nitrogen (N) is an essential element for crop growth and its accurate assessment in plants is a key to
nutrient management. This research was conducted to explore the feasibility of using spectral reflectance to predict
nitrogen content in corn plant varieties with varying color. A problem identified by previous researchers is the
variability of spectral responses from varieties with different canopy colors. Reflectance was measured for samples
from subplots of three different corn varieties (selected for varying leaf color) and 5 N treatments. Multivariate
statistical analysis was performed on the spectral responses for each variety and N treatment to determine the ability
to predict the actual leaf N content. Partial Least Square (PLS) regression, Principal Component Regression (PCR)
and Multiple Linear Regression (MLR) were used to build prediction models. The smallest SEP for the ear leaf was
0.15% for a PCR model, and the smallest SEP for younger leaves was 0.25% for a PLS model.

Keywords: Nitrogen stress, Multispectral response, SPAD meter, Greenness, PLS.

The author(s) is solely responsible for the content of this technical presentation. The technical presentation does not necessarily reflect
the official position of ASAE, and its printing and distribution does not constitute an endorsement of views which may be expressed.

Technical presentations are not subject to the formal peer review process by ASAE editorial committees; therefore, they are not to be
presented as refereed publications.

Quotation from this work should state that it is from a presentation made by (name of author) at the (listed) ASAE meeting.

EXAMPLE - From Author's Last Name, Initials. "Title of Presentation." Presented at the Date and Title of meeting. Paper No. X.
ASAE, 2950 Niles Rd., St. Joseph, MI 49085-9659 USA.

For information about securing permission to reprint or reproduce a technical presentation, please address inquiries to ASAE.

ASAE, 2950 Niles Rd., St. Joseph, MI 49085-9659 USA
Voice: 616.429.0300 FAX: 616.429.3852 E-Mail:<hq@asae.org>

ASSESSING NITROGEN STRESS IN CORN VARIETIES OF VARYING COLOR

Wonsuk Lee Stephen W. Searcy Takashi Kataoka
Postdoctoral Research Associate Professor Visiting scholar

ABSTRACT

Nitrogen (N) is an essential element for crop growth and its accurate assessment in plants
is a key to nutrient management. This research was conducted to explore the feasibility of using
spectral reflectance to predict nitrogen content in corn plant varieties with varying color. A
problem identified by previous researchers is the variability of spectral responses from varieties
with different canopy colors. Reflectance was measured for samples from subplots of three
different corn varieties (selected for varying leaf color) and 5 N treatments (0, 67.0, 134.1, 201.1,
and 268.1 kg/ha). Multivariate statistical analyses (stepwise discriminant analysis, canonical
discriminant analysis and principal component analysis) were performed on the spectral
responses for each variety and N treatment to determine the ability to predict the actual leaf N
content. It was found that the leaf greenness did not have any effect to canopy spectral
reflectance to assess nitrogen status. Partial Least Square (PLS) regression, Principal Component
Regression (PCR) and Multiple Linear Regression (MLR) were conducted to build prediction
models. The standard errors of calibration (SEC) for ear leaf were 0.08%, 0.09%, and 0.0% for
PLS, PCR, and MLR, respectively and the standard errors of prediction (SEP) for ear leaf were
0.16%, 0.15%, and 0.20% for PLS, PCR, and MLR, respectively. For the younger leaf samples,
the SEC were 0.18%, 0.18%, and 0.0% for PLS, PCR, and MLR, respectively, and the SEP were
0.25%, 0.25%, and 0.90% for PLS, PCR, and MLR, respectively.

INTRODUCTION

The success of U.S. agriculture is attributable to the effective use of agricultural chemicals.
In 1996, 98% of over 283 million ha received nitrogen treatment for corn plants. However, this
heavy reliance on chemicals raises many environmental and economic concerns. For an effective
use of agricultural chemicals to get maximum yield without raising any environmental concern,
farmers need to find a way of applying optimum amount of agricultural chemicals. In order
accomplish this task, the status of plant nutrition should be assessed accurately.
Among plant nutrients, nitrogen (N) is the most important and essential for growing crops
and is also most concerning nutrient element for maintaining healthy environment. N is an
integral part of chlorophyll, which is the primary absorber of light energy needed for
photosynthesis. If N is used properly in conjunction with other necessary soil fertility nutrients,
it can speed the crop growth such as corn and small grains. Plants normally contain between 1
and 5% N by weight (Tisdale et al., 1993). If N is deficient in crop plants, the leaf color changes
to light green or yellow-green and the leaf dies starting at the tip.

1

[450 nm. 760 nm. Even when two different corn plants have same amount of nitrogen in their canopies. Masoni et al. Ma et al. A new commercial nitrogen sensor has just started being marketed (model: Hydro N-Sensor. The GPS and GIS technology has been a basis for precision farming and site-specific application of agricultural chemicals and nutrients has been one of the major concerns in precision farming. The objectives of this research were to determine the effect of leaf greenness in assessing nitrogen status using the spectral reflectance of corn leaves. and to develop a model that can predict nitrogen content from leaf reflectance measurement. (1996). (1994). They tested the sensor with winter wheat and found that reasonable correlations were obtained between N uptake and NDVI (Normalized Difference Vegetation Index). One of the problems in developing a nutrient sensor would be variability of spectral response resulted from different canopy colors. thus can be used to detect N deficiency of crop plants. (1982)). It measures crop reflectance with 4 light 2 . BACKGROUND Many researchers have tried to utilize visible and near infrared spectral response from plant canopies to detect plant status such as moisture content (Rigney and Brusewitz. While these studies relied on laboratory work to assess nutrient status of the plants. The ultimate goal of this research is to develop a real-time multispectral sensor that can detect nitrogen deficiency in corn plant using spectral response from plant canopies. With the development of the nitrogen sensor system. the leaf characteristic greenness was studied for three different varieties. 690 nm. (1996). (1998)). This information could form the basis for development of a plant nitrogen sensor. 760 nm and 900 nm] (Walberg et al. (1998)). In this project. red and near infrared light. precision agriculture has been used widely as a management tool to maximize yield and to minimize cost. (1996). Yoder and Pettigrew-Crosby (1995). Reflectance ratios at other wavelengths were also studied for their capability to distinguish N status such as [550 nm and 675 nm] (Thomas and Oerther (1972)). 630 nm. farmers could optimize yield while minimizing the cost. From the last few years. Miles Opti-Crop division. (1995). and Bausch et al. Al-Abbas et al. 690 nm. (1996)). They reported that the interference of solar angle needed to be evaluated. [630 nm. their spectral response would be different if their canopy colors are different. as a preliminary step for developing a nitrogen sensor. Stone et al. Walberg et al. (1998) developed a spectral reflectance sensor to detect nitrogen status of cotton plants with 4 spectral bands of blue. 1997) and nutrient stress (Thomas and Oerther (1972). Sui et al. Blackmer et al. Owensboro. 800 nm. green. KY). Blackmer et al. fertilizer could be accurately applied to specific locations only where needed. (1996) developed a sensor for nitrogen detection and weed detection with photodiode sensing elements. 710 nm. They tested the sensor in 2 situations: with an artificial illumination and with natural illumination. and [702 nm] (Bausch et al. (1974). They reported the preliminary test results for diagnosing nitrogen status in cotton was promising. The most common fact found by those studies is that reflectance near 550 nm shows good separation of leaf nitrogen concentration. and 900 nm] (Blackmer et al. (1982). some researchers have tried to implement a system for a real-time field use. By applying nitrogen only where it is needed. Filella et al.

five different nitrogen treatment levels were selected.89 ~ 0. They found that the canopy reflectance was well correlated with the applied nitrogen and that the yield could be predicted well from the canopy reflectance in the red channel. (1992) conducted an experiment for feasibility of using field chlorophyll measurements for evaluation of corn nitrogen status. soil samples were collected for examining soil fertility from 6 sampling locations at three different depths (0. Table 1. and Piekielek et al. however also reported that this trend was questionable since it was largely related to differences between sites instead of differences due to N applied within sites. Schepers et al. MATERIALS AND METHODS Field planting of corns The main objectives of this research were to examine the effect of leaf characteristic greenness on the spectral reflectance of plant leaves with similar nitrogen status. (1985) studied a feasibility of using plant canopy reflectance as a means of distinguishing weeds from crops. In order to observe the nitrogen effect to corn growth. One widely used chlorophyll meter is The SPAD-502 (Minolta. (1998) used spectral video images of bush bean plants taken with two bandpass filters to assess nitrogen status of the plants. 0. They found that a statistically significant linear correlation existed between grain yield and the chlorophyll meter reading at R4 (dough stage) and R5 (dent stage).22 m.82 ~ 0.sensors (@gInnovator. Thai et al.45 m and 0. they successfully classified some crop plants and weed species with efficiency of 68% ~ 100%. Blackmer and Schepers (1995) tried to evaluate the ability of SPAD meter as a nitrogen assessment tool for corn plants. 3 . Tracy et al. Wood et al. Co. Another way of assessing nitrogen stress is to use a chlorophyll meter since nitrogen is closely related to chlorophyll in the leaf. They found that significant curvilinear relationships between SPAD readings and tissue N concentrations at the V10 and midsilk stages (R2 = 0. three corn varieties with a wide range of characteristic color were chosen with the assistance of the local Asgrow Seed company representative. (1992). Using color infrared (CIR) aerial photography. All soil samples showed very low in nitrogen (1 ~ 19 ppm). Blackmer and Schepers (1994). GopalaPillai et al. and to determine if leaf reflectance can predict nitrogen status. From the field trials for wheat. To accomplish these objectives. (1995)). Utilization of reflectance has also been used as a basis for remote sensing technology with the fact that different object has different reflectance at a given wavelength. the sensor reportedly increased the yield by 471 kg/ha (7 bu/ac). Smeal and Zhang (1994).91) and between SPAD readings and grain yield (R2 = 0. (1998) used high resolution color infrared aerial images for detecting nitrogen stress in corn.90 m). A fifth sensor was used for compensating the sunlight change. (1992).88). 1999). The SPAD meter measures light transmittance at 650 nm as a source of chlorophyll concentration and at 950 nm as a reference to compensate such effect as leaf moisture content and thickness (Blackmer and Schepers (1995)). They used neural network to distinguish different nitrogen treatments and reported that the two selected bands could be used to distinguish different nitrogen treatments.). Prior to planting. Menges et al. Most of the studies related to the chlorophyll meter was to evaluate its feasibility for assessing nitrogen status of crop plants (Piekielek and Fox (1992).

The soil contained enough moisture for the plants to maintain their vigor with 4 days’ rain in the previous week of leaf sampling. CVI Laser Corp.0 (N2) RX901W (V2.300 seeds/ha on March 16. There were 4 data sets of reflectance measurement. Figure 1. They were quickly put in a plastic bag and transported from the field to the laboratory (15 min. lighter color) 67. Variety (Asgrow) N treatment (kg/ha) 0 (N1) RX897 (V1. and lighting system (model: AS220. 4 .1 (N3) RX938 (V3.). an integrating sphere (model: IS040SL.4 m long and was arranged as a split-plot design with 4 replications. the system was turned on for an hour prior to testing. The plants were at R3 (milk stage) to R5 (dent stage) growth stages (Ritchie and Hanway (1982)). Corn varieties and nitrogen treatment used in the experiment. 1999 in the Texas A&M University experimental farm located in Brazos River bottom near College Station. Table 2. respectively. Table 1.76-m wide rows and 30. Prior to measuring reflectance. avoiding any boundary effects from the adjacent rows of different nitrogen treatment.1 (N4) 268. Corp. medium color) 134. The field was composed of 48 rows of 0.). Reflectance of a white reference was measured for every 5 samples using a reference cap painted with barium sulfate (BaSO4).).5 m long and had one variety and one nitrogen treatment. Each data set from Field B1 and B4 consisted of 150 reflectance spectra (75 spectra for ear leaf and 75 spectra for younger leaf) with 2 dependent variables (variety and nitrogen treatment) and 176 independent variables (spectral bands). CVI Laser Inc. A 2. Five plants were randomly selected from the middle two rows in each subplot for sampling from Field B1 and B4. The YoungerB1 and YoungerB4 were obtained as the same manner. 32% N) on March 26. darker color) 201. driving distance). The EarB4 and EarB1 are reflectance measured from ear leaves in Field B4 and B1. The field was composed of 60 subplots. All other cultural practices were managed conventionally.1 (N5) The corn was planted at the seeding rate of 59. a detector (model: AD120. The split-plot design was selected since it was more practical for a field experiment and it was difficult to plant different variety in a small area.76 m wide and 213.)) was set up in a laboratory. Each sample was measured its reflectance from 400 nm to 1100 nm with 4 nm increment. Reflectance measurement The reflectance measurement system (a monochromator (model: CM110. Nitrogen was applied in the form of UAN (Urea Ammonium Nitrate. Labsphere Inc. Each subplot contained 4 0.5-cm diameter leaf disk was obtained from the middle portion of an ear leaf and a fully developed younger leaf from each plant using a specially designed circular knife. CVI Laser.

June 29-30. Reflectance measurement data set. June 17-18. 1999 EarB1 Ear leaf reflectance from Field B1. A set of five leaf disks from a subplot was submitted as one sample for nitrogen analysis. These samples were analyzed for individual nitrogen concentration to see whether the samples of uniform reflectance had actually homogeneous nitrogen concentration and whether the samples of varying reflectance had different nitrogen concentration in each sample. 1999 Nitrogen analysis and chlorophyll measurement The nitrogen content was analyzed for the leaf samples using LECO method (Sheldrick. 4 sample sets of 5 leaf disks were selected such that reflectance in one set had uniform reflectance and the other set had varying reflectance for ear leaf and younger leaf. For each sample 5 . 1999 YoungerB1 Younger leaf reflectance from Field B1. Data set Description EarB4 Ear leaf reflectance from Field B4. June 17-18. Additionally. Field corn planting and N treatment layout. June 29-30. 1999 YoungerB4 Younger leaf reflectance from Field B4. 1986). Figure 1. Table 2.

Ver 2. The MATLAB (The Mathworks. PLS finds factors that capture the greatest amount of variance in the predictor variables. Ver 5. the SAS procedure PRINCOMP was used. principal component analysis was executed with the selected wavelengths by the STEPDISC procedure. A regression analysis was conducted to correlate the SPAD measurements with actual nitrogen content of samples. Canonical discriminant analysis is a dimension-reduction technique related to principal component analysis and canonical correlation. the SAS procedure CANDISC was used. 6 . Since the stepwise discriminant analysis does not select the best subset of wavelengths due to multi-colinearity.. For canonical analysis. WA) was used for the analysis. the result from this analysis was used for further analysis. The EarB1 and YoungerB1 were used for building calibration model and the EarB4 and YoungerB4 were used to validate the model. PLS. The standard error of prediction for calibration data (SEC) and the standard error of prediction for validation data (SEP) are calculated as follows (Williams and Norris. Given a classification variable and several quantitative variables. Inc.5-cm leaf disk). this analysis derives canonical variables (linear combination of the quantitative variables) that summarize between-class variation in much the same way that principal components summarize total variation. a stepwise discriminant analysis was conducted for each data set. Principal component analysis is generally used to maximize the variance of a linear combination of the variables and is used when highly correlated independent variables may produce unstable estimates. The SAS procedure CORR (SAS. First. 1990) was used for this step. With the wavelength bands selected by the STEPDISC. Minolta Co. PLS is a quantitative spectral decomposition technique that is closely related to Principal Component Regression (PCR).(2. Manson. Inc. Prediction of nitrogen content Another commonly used method to analyze spectral data is Partial Least Square (PLS) regression. Additionally. PCR and MLR (multiple linear regression) were conducted to build a calibration model and compare their performance.15. The SAS procedure STEPDISC was used for this process with stepwise selection method and significance level of 0.). a correlation analysis was conducted between nitrogen amount and reflectance at each wavelength to look at how much correlation there would be. in order to reduce the number of variables. a canonical discriminant analysis was conducted.0. 1987). A series of multivariate statistical analysis was conducted. For this step. Then. Multivariate statistical analysis The objective of the statistical analysis was to find the wavelengths which best discriminate nitrogen treatment and variety.. The PCR method combines the Principal Component Analysis (PCA) spectral decomposition with an Inverse Least Squares (ILS) regression to crate a quantitative model for complex samples.0) PLS_Toolbox (Eigenvector Research. In this research. its chlorophyll content was measured from five different locations with a chlorophyll meter (model: Spad-502.

The more nitrogen was applied to the plants. 7 . These pictures were taken on May 14. 1999. 63 days after their planting. y = actual nitrogen content (%). The line in the middle indicates an intersection of the two different varieties or two different nitrogen treatments. n = number of samples.2 m tall) (2.8 m tall) (1. The leaf color of three different varieties was also examined with the plant leaves of no N applied. (a) V3N4 & V3N1 (b) V2N2 & V3N5 (1. The order of the greenness from lighter to darker green was RX897 (V1). This order confirmed the leaf characteristics originally obtained from the seed company. 1/ 2  ∑ e2  SEC =   n − 2 1/ 2  ( ∑ e 2 − ( ∑ e) 2 / n )  SEP =    n −1  where. y’ = predicted nitrogen content (%). Examples of nitrogen treatment effects on corn growth.0 m tall) Figure 2. These figures show clear effect of different nitrogen treatment. RX901W (V2) and RX938 (V3).5 m tall) (1. RESULTS AND DSCUSSION N fertilization effect Figure 2 shows some examples of the nitrogen treatment and varietal effects on their growth. the taller the plants were. e = y – y’.

Figures 4 . The samples from the low N treatment subplots had also high reflectance in near infrared range.1 kg/ha N). Another important feature is that the reflectance between 400 nm and 500 nm was nearly identical for most of the samples regardless of their variety and N treatment levels.01). closer examination of the nitrogen analyses results show that the differences in the reflectance curves for the zero nitrogen treatment may be explained by nitrogen status variations. medium and highest nitrogen applications. In Figure 4. 1.6 show typical reflectance spectra of the samples. The first noticeable characteristic of the reflectance spectra is that the samples from the lower N treatment subplots (0 and 67.0 kg/ha N) showed higher reflectance near 550 nm than those from high N treatment subplots (134.Leaf reflectance measurement Figure 3 . which absorbed more light around 550 nm. while figures 5 and 6 show little difference in reflectance. The variety effect to the reflectance at 552 nm was not significant at significance level 0. This fact is especially useful when a ratio would be used to distinguish different nitrogen amount from the leaves (e. However.01. corresponding to a water absorption band. For the ear leaf. 201. The ear leaf data was taken from EarB1 and EarB4 and the younger leaf data was from YoungerB1 and YoungerB4. RX901W and RX938. The stronger impact of nitrogen content is suspected since the highest reflectance was the medium colored variety and the lowest N percent.01 except the Waller test for both the ear leaf and younger leaf samples.95. the nitrogen concentrations were 1. The result is shown in Table 3. This is because the samples from the high N treatment subplots had more nitrogen and chlorophyll in their leaves. a means test was conducted with a reflectance value at 552 nm using the SAS procedure GLM. The significance level was 0.1 and 268.1. the more nitrogen was applied to the plants. respectively. ratio of 450 nm to 550 nm). All spectra had a dip around 980 nm.01. differences appear to relate to varietal differences. Test method Waller Duncan Tukey Ear leaf V2 V3 V1 V2 V3 V1 V2 V3 V1 by variety Ear leaf N1 N2 N3 N4 N5 N1 N2 N3 N4 N5 N1 N2 N3 N4 N5 by N treatment Younger leaf V2 V1 V3 V2 V1 V3 V2 V1 V3 by variety Younger leaf N1 N2 N3 N4 N5 N1 N2 N3 N4 N5 N1 N2 N3 N4 N5 by N treatment 8 . Means test with reflectance at 552 nm (α = 0.37 percent for the RX897.6 show the reflectance spectra of the three varieties for the zero. Underlined treatments are not significantly different at α = 0. At 550 nm. To statistically evaluate the impact of variety and N treatment on the reflectance difference.g. Table 3. Figure 3 shows N treatment effect on one variety for the ear leaf and the younger leaf.25 and 2. the lower reflectance they had.

Reflectance of ear leaf and fully developed younger leaf of V1 (RX897) with 5 different nitrogen treatments. Figure 3. (b) Younger leaf. (a) Ear leaf. 9 .

Reflectance of ear leaf and fully developed younger leaf of V1 (solid).0 kg/ha N treatment. Figure 4.42%. RX938-2. V2 (dot). (a) Ear leaf. N percents were RX897-1. RX901W-1.36%.25%.37%. (b) Younger leaf. and V3 (x) with 0.95%. RX901W-1. 10 .11%. N percents were RX897-2. RX938-2.

(a) Ear leaf. Reflectance of ear leaf and fully developed younger leaf of V1 (solid). Figure 5. (b) Younger leaf. 11 .1 kg/ha N treatment. and V3 (x) with 134. V2 (dot).

(a) Ear leaf.1 kg/ha N treatment. 12 . Reflectance of ear leaf and fully developed younger leaf of V1 (solid). and V3 (x) with 268. (b) Younger leaf. Figure 6. V2 (dot).

36 3.73 2.22 2. In fact.96 2. This result was supported by the sample nitrogen content analysis shown below.14 2.. The samples of uniform reflectance showed lower standard deviations than those of varying spectra.55 1. and [N3. it can be concluded that the reflectance difference at 550 nm was not affected by variety.36 V1N4 3.75 to 2.50 V1N2 2. and N5 subplots showed higher N content.93 V3N3 3.79 2.37 V3N5 3.38 3. N4.51 3. but by N treatment and that the excessive amount of N treatment over N3 (134.46 3. Lab analyzed nitrogen content (%dry matter basis). All the N treatments showed that the reflectance of the samples from the subplots of [N1]. according to the established critical N levels in the corn ear leaf at anthesis of 2. Variety & N treatment EarB4 YoungerB4 EarB1 YoungerB1 V1N1 1. N4.50 2. Table 5 shows the N content analysis result of the samples of uniform and varying reflectance spectra.73 2. In fact.48 2.66 3. and N5] were significantly different at α = 0.94 V1N3 3.47 2.05 2. it was adequate for the corn plants in this growing season.31 1. with little difference from each other.01. 13 .34 V3N4 3.32 2.36 1. 1992). Lab analyzed nitrogen content The samples were submitted for their nitrogen content analysis and these results are shown in Table 4.03 2.67 V2N5 3.82 1.50 1.55 V1N5 3.51 For most of the samples.10 3. The general trend is that the actual N content for samples from N1 subplots were very low and that samples from N3.73 2.42 1.80% (Schepers et al.97 2.03 3.28 1.00 1.15 2. [N2].72 2.11 1.95 2.41 V2N1 1.01 2. Table 4.35 1.54 V3N2 2.46 V3N1 2.08 2.50 3.37 2. the corn plants in the subplots with N1 and N2 treatments showed severe N deficiency symptoms on their mature growth stages (R1 stage and beyond). Therefore.40 V2N2 2.19 V2N4 3.09 2. If the N was more than or equal to the N3 treatment level.92 2.52 3. the N content of the ear leaf samples was higher than that of the younger leaf samples.67 V2N3 3.59 2.1 kg/ha) was not consumed by corn plants.36 1. especially those on N1 subplots.31 2.25 1. no nitrogen deficiency was experienced in any of the three larger application treatments.

1 4.2 2.1 4.20 1.4 1.668 Std. 0.6 3.2 2.2 2.53 37.9 2.03 57.3 1.1 1.8 4.1 3.2 2.5 2.80 V1N4 61.0 3.34 53.07 59.82 53. Table 6.089 0.9 3.5 4.20 2.48 51.3 1.17 1.33 44.41 32.16 1.70 55. Variety & EarB4 YoungerB4 EarB1 YoungerB1 N-treatment Avg Stdev Avg Stdev Avg Stdev Avg Stdev V1N1 43.6 3.85 43.3 1.96 49.10 V3N4 63.3 2.67 14 .4 2.81 V1N5 67. Figure 7 shows Spadmeter readings of EarB4 and YoungerB4 sample set in bar graphs.2 2.16 36.308 2.40 48. Stdev = standard deviation of 5 replicates).36 43. The result showed basically the same trend as the N content result.74 Average 3.1 1.5 2.2 2.8 3.108 0.08 V1N3 64.46 30.17 1.24 49.188 0.32 62.15 1.9 2. the ear leaves had higher SPAD values than the younger leaves did.01 56.).0 1.57 V2N2 51.5 2.77 50.59 3.41 57.82 48. Samples of uniform reflectance Samples of varying reflectance Sample 1 Sample 2 Sample 1 Sample 2 3.6 1.6 2.39 36.48 1.54 28.66 V1N2 53.06 54.20 1. Table 5.62 43.312 1. The result also showed that the SPAD readings increased as the nitrogen application rate increased.18 51.92 54.05 35.4 2.6 2.7 1.74 38.2 4.84 V2N1 34. Most of the time.7 2.39 2. Chlorophyll measurement results in SPAD units (Avg = average of 5 replicates.82 47.9 4.6 3.28 V2N5 61.77 45.67 V3N1 51.0 4.9 2. Note that the measurement result showed high variation since it measures very small area (2 mm x 3mm).5 1.76 3.55 1.50 V3N2 51.67 V2N3 62. which indicates that many measurements would be required in order to extract a proper management decision. Minolta Co.5 3.0 1.85 56.18 2.11 V2N4 61.27 30. Table 6.188 1.3 2.4 3.9 3. dev.3 4.40 2.09 1.117 SPAD meter measurements The chlorophyll contents were measured for the same leaf samples using a chlorophyll meter (model: Spad-502.10 56.75 3.4 3.95 53.33 1.99 55.62 57.2 3.8 2.6 2.9 2.83 49.86 V3N5 62.41 59.37 2.3 4.6 2.30 59.3 2.4 2.04 49.32 42.50 3.62 V3N3 59. N content (%) of the samples of uniform and varying spectra.99 34.3 1.3 3.22 45.8 3.35 57.

808 Statistical analysis To find out how well actual N content would relate to reflectance.0 30.0 V1N1 V1N2 V1N3 V1N4 V1N5 V2N1 V2N2 V2N3 V2N4 V2N5 V3N1 V3N2 V3N3 V3N4 V3N5 Variety and N treatment Figure 7. which could be considered as source of errors.405 0.0 50.576 0. Regression result between SPAD reading (X) and actual N content (Y. and 708 – 760 nm in the NIR range.0 RX901W RX938 RX897 60. Generally the SPAD reading shows good relationship with the actual N content with high R2 values except YoungerB1 data. Ear Leaf. Table 8.0583X – 0.928 EarB1 Y = 0.0558X – 0.0 Spad reading 40.0 0.0788X – 1. Data set Regression equation R2 EarB4 Y = 0.962 YoungerB4 Y = 0.544 0. Table 7. The YoungerB1 samples had many small degraded spots on their leaves.0789X – 1. Average of 5 samples 70.261 0. The commonly selected wavelength ranges were 540 – 556 nm in the visible range.0 10. a correlation analysis (SAS CORR) was conducted.0 20. SPAD meter measurement for EarB4 and YoungerB4 samples.942 YoungerB1 Y = 0. 15 . This result was used as supplementary information for further analysis. A regression analysis was conducted between SPAD reading and actual N content of the samples. %). Average of 5 samples Younger Leaf. Table 7.

73 and 21 wavelength bands for EarB4. it could be concluded that the calculated canonical variables derived from the wavelengths.166 0. 47. The CANDISC analysis from the other sets showed also very good separation among varieties and among N treatment levels with the calculated canonical variables. given by the squared canonical correlation for EarB4 data set.864 5. 708 – 868 | r | > 0.9 YoungerB1 528 – 576.000 2.1 0.646 8.9 A stepwise discriminant analysis (SAS STEPDISC) was conducted to reduce the number of variables. selected by the stepwise discriminant analysis. Correlation analysis result (r: correlation coefficient). 16 . In both cases. Another way to examine the data set is Principal Component Analysis (PCA).5 YoungerB4 528 – 620. 696 – 1100 | r | > 0.509 0. The separation was not as good as canonical variables. variety and N treatment levels could be separated very well with the calculated canonical variables. The R2.953 3. were able to identify variety and N treatment levels.652 0.0001 CAN3 2.955 0.0001 CAN4 1. 704 – 760 | r | > 0. 86.0001 CAN2 7. conducted with N as a class variable. Table 8.877 0.0020 Figures 8 and 9 show canonical variables calculated with N as a class variable and with variety as a class variable.151 0. Canonical Canonical Squared variable Eigenvalues correlation canonical Cumulative Approx F Pr > F correlation CAN1 21. With CAN1 and CAN2. respectively.7% of the total variance of EarB4 data could be explained and 99. Therefore. EarB1. was 0.936 0.6% of the total variance could be explained with the first 10 PCs. With the PRIN1.4% of the total variance could be explained.778 0. Table 10 shows one of the results. Canonical discriminant analysis for EarB4 with N as a class variable.863 0.605 1. 700 – 1100 | r | > 0. with the EarB4 data set.5 EarB1 540 – 556.5 0. and YoungerB1. The analysis yielded a total of 29. With these selected variables. Figure 10 shows the first two principal components of EarB4 data set.745 0. Table 9 shows one of the results for EarB4 data set.955 between CAN1 and treatment N and 0. PCA was conducted using the SAS procedure PRINCOMP.977 0. a canonical discriminant analysis (SAS CANDISC) was executed.9 0. respectively.2 0.877 between CAN2 and treatment N. YoungerB4. Table 9. 66. Data set Selected wavelength range Selection criteria EarB4 520 – 604.

001 0.210 0.0 kg/ha 67.1 kg/ha 201.3307 13.954 PRIN5 0.877 PRIN3 1.1 kg/ha 2 CAN2 0 -2 -4 -6 -8 -6 -4 -2 0 2 4 6 8 10 CAN1 Figure 8.2568 0.6847 0.0 kg/ha 6 134. Table 10.029 0.983 PRIN10 0. Canonical discriminant analysis of EarB4 data set with variety as a class variable. 5 RX897 4 RX901W RX938 3 2 1 CAN2 0 -1 -2 -3 -4 -5 -6 -4 -2 0 2 4 6 8 CAN1 Figure 9.2859 0.667 PRIN2 6.043 0.1352 0.2300 0.9709 0.667 0. Canonical discriminant analysis of EarB4 data set with N as a class variable. Principal component analysis result with EarB4 data.1006 4.8439 0.1 kg/ha 4 268. 8 0. 0.033 0.996 17 . Principal Eigenvalue Difference Proportion Cumulative component PRIN1 19.920 PRIN4 0.8357 0.0319 .

Figure 12. Another way of estimating the prediction error is to look at the predicted residual error sum of squares (PRESS). In the cross-validation process. both PLS and PCR had the minimum PRESS at the 5th latent variable (LV) or principal component (PC) for EarB4 calibration (Figure 12) and 1 LV and 1PC for YoungerB4 calibration (data not shown). 12 and Figure 11. The prediction models by PLS and PCR performed similar each other. which significantly decreased its performance.435). The overall performance of the prediction model could be presented as SEP or R2. the prediction model would be insufficient. it is important to find the optimum number of the parameters for the prediction model.1 kg/ha 4 201. Either PLS or PCA could be the method to build a good prediction model in this analysis. 18 . however they were better than the MLR model. the model would be over fitted. On the other hand.1 kg/ha 268. Principal component analysis for EarB4 data set.0 kg/ha 134. if there were too many parameters. 8 0 kg/ha 6 67. N content prediction The prediction results of the N content (%) by PLS. If there were too few parameters. The MLR model produced an outlier (-0. PCR and MLR are summarized in Tables 11.1 kg/ha 2 PRIN2 0 -2 -4 -6 -10 -5 0 5 10 15 20 PRIN1 Figure 10. For this calculation. the MATLAB PLS_Toolbox was used. which would not work for other data sets. In this analysis.

420 1.608 3.830 1.698 -0.144 1.080 0.308 3.279 2.000 R2 between 0.466 2.961 0. Calibration (EarB1) Validation (EarB4) Actual PLS PCA MLR Actual PLS PCA MLR 1.030 3.730 2.359 3.370 1.861 2.002 3.500 1.820 2.296 2.433 2.282 3.667 1.252 2.832 0.316 2. Calibration (YoungerB1) Validation (YoungerB4) Actual PLS PCA MLR Actual PLS PCA MLR 1.243 2.838 1.703 2.209 3.331 2.780 2.328 1.540 1.673 2. Table 11.760 1.078 2.159 2.580 3.434 2.940 2.790 3.360 3.354 2.254 1.670 1.987 0.460 3.342 3.363 2.065 2.550 2.479 1. PCR.036 2.249 0.400 1.135 2.523 2.661 1.650 2.661 1.970 3.268 2.370 2.775 2.562 SEC 0.244 3.350 1.500 2.483 2.341 2.240 2.373 2.000 2.434 2.725 2.378 3.180 0.370 3.862 1.603 2.662 3.820 1.561 1.119 2.098 2.920 2.358 3.720 1.688 2.223 3.234 2.350 2.629 1.284 2.030 3.000 R2 between 0.560 2.346 1.978 2.702 2.940 1.252 3.181 3.000 SEP 0.645 1.611 2.309 1.670 2.876 1.549 2.029 Actual Actual 19 .460 2.410 2.950 0.292 2.094 2.484 2.308 3.128 3.039 3.047 3.924 Actual Actual Table 12.148 0.564 2.550 1.467 2.190 2.535 3.970 3.799 2.162 0.585 3.426 SEC 0.330 2.250 2.088 2.899 R2 between 0.964 2.835 1.137 3.697 2.633 2.250 0.759 1.905 0.552 1.930 2.306 3.550 3.321 1.700 3.387 1.715 2.324 2.400 2.207 2.500 1. and MLR for Ear leaf samples.150 2.908 0. and MLR for Younger leaf samples.658 3.190 2.524 2.510 2.480 3.540 2.881 1.867 1.363 2.334 1.591 2.360 2.402 2.633 2.725 2.478 1.538 2.299 2.340 2.670 2.546 3.649 2.583 3.465 2.093 0.455 2.022 1.340 2.843 2.320 1.312 1.217 3.778 2. PCR.182 0.005 2.474 1.790 2.280 2.513 3.242 3.503 3.543 2.199 2.490 1.920 3.410 1.269 2.436 2.480 2.038 3.435 2.479 2.410 3.930 2.983 1.585 3.655 2.590 2.703 2.114 2.550 2.131 2.141 2.101 3.721 1.523 3.905 2.580 2.461 3. N prediction result (%) by PLS.675 2.000 2.341 2.565 2.728 2.730 3.120 1.513 3.312 1.948 1.533 1.500 3.0 SEP 0.510 3.864 2.426 2.400 1.325 2.486 1.190 3.592 2.500 2.202 2.590 3.660 1.558 2.682 1.500 1.864 1.670 2.023 2.477 1. N prediction result (%) by PLS.202 R2 between 0.026 2.538 2.

3. R2 = 0.5 3.0 1. and MLR 3.029 2.899.924 3.0 PLS SEP =0.961 Predicted N (%) by PLS.202. R2 =0.5 PLS SEP = 0. 4. Figure 11.5 1.249.5 2.0 1.5 2. Actual N (%) and predicted N (%) by PLS.5 2.162.5 1. R2 =0. and MLR for corn leaf samples.0 PCA MLR SEP = 0.5 2.950 3. R2 =0. PCR.0 1.0 2.0 1.0 2.905 Predicted N(%) by PLS.148.5 Actual N (%) (b) Younger leaf.0 1. PCR. PCR.5 3. R2 = 0.0 1. 20 .5 PCA SEP =0.250.0 Actual N (%) (a) Ear leaf.5 4. and MLR MLR SEP =0.908 SEP = 0.0 2.0 3.0 3. R2 = 0.

respectively. and 0.18%. 0.0% for PLS. The standard errors of calibration (SEC) for ear leaf were 0. The canonical discriminant analysis (CANDISC) with the wavelengths selected by the stepwise discriminant analysis showed promising result to distinguish different variety and N treatment.08%.0% for PLS. using spectral reflectance. a means test showed that the variety effect to the reflectance at 552 nm was not significantly different at α = 0. reflectance was measured to assess nitrogen content. The leaf chlorophyll content was measured with SPAD-502 chlorophyll meter and the readings showed good relationship with the actual N content (%). and 0. 0. With the well known wavelength band for distinguishing nitrogen status of corn plants (i. Using three varieties of different leaf characteristic color and five different N treatments. 0. and the SEP were 0. For the younger leaf samples. PCR. However.01. 21 . the measurement result also showed high variation since it measures only small portion of leaf.. and MLR.25%.20% for PLS.18%. 550 nm). the SEC were 0. and 0. and MLR.25%. PCR.09%. which acquires many measurements for correct N assessment. This study was conducted to find effect of variety and nitrogen treatment to leaf spectral reflectance. PCR. and 0. however the N treatment effect to the reflectance at 552 nm was significantly different. Predicted residual error sum of squares (PRESS) for N content (%) prediction for EarB4 data set.e. PCR. respectively. Figure 12. 0. One of the challenges in developing a multispectral sensor was expected to be the variability of spectral responses resulted from different canopy colors.90% for PLS. CONCLUSIONS This research was conducted as a preliminary step for developing a real-time multispectral nitrogen sensor for corn plants for site-specific variable rate application.15%. respectively and the standard errors of prediction (SEP) for ear leaf were 0. respectively. N prediction models built by PLS and PCR performed better than the one by MLR. while the performance of PLS and PCR was similar. and MLR.16%. and MLR.

Ma.. M. Blackmer. M. 1998. The authors would like to thank Mr. Serra. K. J. Use of a chlorophyll meter to monitor nitrogen status and schedule fertigation for corn. Laron Peters (Asgrow Seed Co. Schepers. Spectra of normal and nutrient-deficient maize leaves. GopalaPillai. ASAE Paper no. Goetz. A.. and E. 66: 16-20. Curtis. E. Agronomy Journal. and J. Walter-Shea. ACKNOLEDGEMENT This research was supported by the Binational Agricultural Research and Development Foundation (BARD) and USDA/CSREES and the Texas Agricultural Experiment Station. L. Beal. Blackmer. H. J. L. and G. 1996. Joseph. Agronomy Journal. 1995. Al-Abbas.. F. 88: 915-920. ASAE Paper No. Communications in soil science and plant analysis. 6(8): 2. Penuelas. T. 1996. Journal of Production Agriculture. M. Tian.) for providing corn seeds for the experiment. Dwyer. Varvel. 1994. Joseph. Crop Science. T. 8(1): 56-60. Barr. A. Evaluating wheat nitrogen status with canopy reflectance indices and discriminant analysis. R. F. 86: 934-938. and J. Light reflectance compared with other nitrogen stress measurements in corn leaves. Schepers. M. Filella. J. Blackmer. Agronomy Journal. Morrison. H. 2950 Niles Rd. 1974. Diker.. ASAE. St. 1994.. Serrano. and B. W. Baumgardner. REFERENCES @gInnovator. J. Bausch.. S. 1998.. T. I. Techniques for monitoring crop nitrogen status in corn. and M. E. Schepers. 22 . A. 25: 1791-1800. L. Blackmer. 1999. Canopy light reflectance and field greenness to assess nitrogen fertilization and yield of maize.. 35: 1400-1405. ASAE. Schepers. and L. L. F. S. S. GPS & VRA. St. 1995. Hall. and J. Hyperspectral characteristics of nitrogen deficient corn.. M. MI49085-9659 USA. MI49085-9659 USA. 2950 Niles Rd. Detection of nitrogen stress in corn using digital aerial imaging. and J. D. 983061. 88: 1-5. C. S. Varvel. M. Crane. 983030. B. J. S. G. T. Agronomy Journal. Nitrogen deficiency detection using reflected shortwave radiation from irrigated corn canopies.

R. L.. Warrendale PA. SAS institute. W. Solie. ASAE. and A. D. Sheldrick. D. M. J. How a corn plant develops. R. Weed Science. J. W. Iowa State University of Science and Technology. and M. S. and G. ASAE Paper No. Agronomy Journal. Spectral properties of leaves deficient in iron. 1997. M. SAS/STAT User’s Guide. P. 87: 403-408. and J. R. J. Rigney. SAE Technical paper series. SAE Paper No. Ustin. L. and S. Nixon. B. J. Sensors for detection of nitrogen in winter wheat. 4th Ed. Ercoli. MI49085-9659 USA... Cooperative Extension Service. Stone. Multivariate statistical classification of soil spectra. S. Taylor and J. L. H. W. Schepers. Whitney. Ringer. 84: 59-65. Piekielek. Lees. Zhang.. Use of a chlorophyll meter at the early dent stage of corn to evaluate nitrogen sufficiency. D. Use of a chlorophyll meter to predict sidedress nitrogen requirements for maize. Raun and H. Communications in Soil Science and Plant Analysis. Test of the LECO CHN-600 determinator for soil carbon and nitrogen analysis. L. W.. E. 1994. Brusewitz. and H. 2950 Niles Rd. H. Fox. 1996. L. 23 . and manganese. R. W. Optical sensor based field element size and sensing strategy for nitrogen application. Solie. St. D. W. Transactions of the ASAE 39(5): 1623-1631. Solie. J. W.. Agronomy Journal. M. L. W. B. J. R. D. Hanway. Raun. Piekielek. P. R. L. Communications in soil science and plant analysis. A. Light reflectance and remote sensing of weeds in agronomic and horticultural crops.. 1982. J. 25: 1495-1503. Stone and J. Menges. E. Francis. Use of spectral radiance for correcting in-season fertilizer nitrogen deficiencies in winter wheat. R. M. Chlorophyll meter evaluation for nitrogen management in corn. R. W. Mariotti. Below. D. 1& 2. 1996. 976056. M. 88: 937-943. Remote sensing of environment. R. P. Richardson. 1996. 1985. and R. Smeal. Version 6. Toth. 1996. Joseph.Masoni. Nondestructive peach moisture measurement from near-infrared spectral reflectance. Agronomy Journal. M. 33: 569-581. A. Whitney. Cary. B. S.. H. Ringer. and F. Whitney. Macneal. 1995. Vol. H. Vigil. Ritchie. 57: 108-118. B. Inc. Fox. Canadian Journal of Soil Science 66: 543-545. Transactions of the ASAE 39(6): 1983-1992. NC. SAE. Stone. 23 (17-20): 2173-2187. and K. 1992. 1986. 1992. magnesium. Palacios-Orueta. sulfur. 1990. P. 1996. Comparison of corn leaf nitrogen concentration and chlorophyll meter readings. Raun. 961757.

.. Effects of nitrogen nutrition on the growth . Thomas. 74: 677-683. Minnesota. R. S. W. Housley. B. 1972. Walberg. J. Vol. and K. Inc. Wood and K. Wilkerson. D. D. Duffield. S. 3: 1099-1100. 2950 Niles Rd. Oerther. Howard. C. and reflectance characteristics of corn canopies.. 1987. D. R. yield. MI49085-9659 USA. ASAE. Reeves. Norris. Integration of neural networks with a spectral reflectance sensor to detect nitrogen deficiency in cotton. Theisen.. Thai. Joseph. Paul. Wood. Halvin. Beaton. C. G. MI49085-9659 USA. and R. Edmisten. American Association of Cereal Chemists. and G. Agronomy Journal. F. 1998. P. D. Predicting nitrogen and chlorophyll content and concentrations from reflectance spectra (400-2500 nm) at leaf and canopy scales. Evans. L. 1992. NJ 07458. E. 24 . W. 64: 11-13. Estimating nitrogen content of sweet pepper leaves by reflectance measurements.. St. Visible & nir imaging of bush beans grown under different nitrogen treatments. Daughtry. 1992 Proceedings Beltwide Cotton Conferences. E. 1993. W. Hart and D. 983104. Williams. Tracy. 1982. J. Nelson. L. C. Near-infrared technology in the agricultural and food industries.. W. X. Edmisten. Yoder. and K. Upper Saddle River. 2950 Niles Rd. Deng. Agronomy Journal.. 15(4): 487-500. ASAE Paper No. L. S. R. St. Joseph.Sui. USA. W. Soil Fertility and Fertilizers. J. E. M. Theory behind the use of instantaneous leaf chlorophyll measurement for determining mid-season cotton nitrogen recommendations. P. C. Bauer. and T. and J. Tisdale. ASAE Paper No. Pettigrew-Crosby. G. Field chlorophyll measurements for evaluation of corn nitrogen status. Prentice Hall. R. 983074. 1995. 1998. ASAE. N. L. Remote Sensing of Environment 53: 199-211. Journal of Plant Nutrition. Hefner. L. T. B. M. L. W. J. 1992. and A.. St. F.