Agronomy Journal Grow Your Career With ASA
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online 19 October 2005
Published in Agron J 97:1515-1523 (2005)
DOI: 10.2134/agronj2005.0040
© 2005 American Society of Agronomy
677 S. Segoe Rd., Madison, WI 53711 USA
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
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 Google Scholar
Google Scholar
Right arrow Articles by Bélanger, M-C.
Right arrow Articles by Chamberland, M.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Bélanger, M-C.
Right arrow Articles by Chamberland, M.
Agricola
Right arrow Articles by Bélanger, M-C.
Right arrow Articles by Chamberland, M.
Related Collections
Right arrow Potato
Right arrow Remote Sensing
Right arrow Nutrient Management
Right arrow Nitrogen

Remote Sensing

Determination of a Multivariate Indicator of Nitrogen Imbalance (MINI) in Potato Using Reflectance and Fluorescence Spectroscopy

M-C. Bélangera,*, A. A. Viaua, G. Samsonb and M. Chamberlandc

a Laboratoire de Géomatique Agricole et d‘Agriculture de Précision, Local 3731-A, Pavillon Casault, Université Laval, QC, G1K 7P4, Canada
b Université du Québec à Trois-Rivières, Case Postale 500, Trois-Rivières, QC, G9A 5H7, Canada
c Telops, 100-2600, Avenue Saint-Jean-Baptiste, Québec, QC, Canada, G2E 6J5

* Corresponding author (marie-christine.belanger.1{at}ulaval.ca)

Received for publication February 2, 2005.

   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
In this study, we evaluated the potential of reflectance and fluorescence for the detection of the Compositional Nutrient Diagnosis (CND) N index (IN). The CND reflects nutrient deficiencies as well as nutrient interactions, contrary to conventional methods of nutrient stress detection. Potato plants (Solanum tuberosum L. cv. Superior) were grown in a greenhouse, and three different nutrient deficiencies (K, Mg, and N) were induced at three levels and compared with a control receiving a complete nutrient solution. Nitrogen deficiency induced a significant biomass reduction compared with control plants whereas no significant effect was observed for K or Mg treatments. Foliar analyses were realized to compute the CND_r2 using the CND. The ANOVA conducted on CND_r2 and IN showed significant differences only between N-deficient and control plants. Using a canonical discriminant analysis over reflectance and fluorescence indices, it was possible to correctly classify 96.6% of potato plants in its corresponding IN class. A new Multivariate Indicator of Nitrogen Imbalance (MINI) was developed using the canonical variable computed from reflectance and fluorescence indices. The MINI can detect almost 70% of the N-deficient plants and more than 90% of the N-sufficient plants. This indicator allows a rapid data acquisition and handling and provides deficiency detection within the time-window for plant response to recovery fertilization.

Abbreviations: CND, Compositional Nutrient Diagnosis • CND_r2, Compositional Nutrient Diagnosis nutrient imbalance index • DAE, day after emergence • IN, nitrogen index • MINI, Multivariate Indicator of Nitrogen Imbalance


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
THE POTATO CROP is cultivated in more than 120 countries worldwide with an annual production of 310 million Mg (FAO, 2004). Potato crops are generally grown on sandy soils and can suffer from nutrient stresses resulting from nutrient leaching or inadequate fertilization. To assure nutrient balance and improve nutrient management during the growing season, N and other nutrient deficiencies must be diagnosed precisely. Nutrient stresses may be identified by comparing the leaves' content in nutritive elements with the critical threshold values for each nutrient. However, the detection of nutrient deficiencies one by one might not reflect the real nutritional status of the plant. Actually, there are nutrient interactions in plants that may interfere with the interpretation of the results (Munson and Nelson, 1990). In 1973, Beaufils developed the Diagnosis and Recommendation Integrated System (DRIS) to estimate the nutritional status of plants by using many dual ratios (N/P, N/K, N/Ca...X/Y) that are computed using the nutrient concentration in the leaves (Beaufils, 1973). Those dual ratios represent the minimum expression for nutrient interactions. Parent and Dafir (1992) developed a new indicator of plant nutritional status: the CND. The CND not only evaluates the foliar composition in nutrients, but it also estimates the interactions that might appear between any nutrient included in the plant (Parent and Dafir, 1992). The CND is based on the Compositional Data Analysis (Aitchison, 1986), and its first assumption is to consider the analyzed leaf as a whole or a simplex; therefore summing up all its constituents will equal 100%. The computation of the CND_r2 of a diagnosed specimen involves using the geometrical mean of all nutrients: each nutrient content is compared with all others, thus providing information on nutrient interactions and leading to the identification of nutrient imbalances. The CND_r2 has a chi-square distribution (Khiari et al., 2001a).

Up to now, nutrient diagnosis techniques have been based on destructive analyses of leaves' mineral contents. Remote sensing techniques such as fluorescence and reflectance could lower the costs and accelerate the acquisition, handling, and analysis of plant tissues and do it in a nondestructive manner. Those two techniques respond to plants' structural and biochemical properties that are sensitive to plant nutrient status, notably the content in photosynthetic pigments and phenolic compounds (Cerovic et al., 1999; Read et al., 2002; Schuerger et al., 2003). Indeed, several studies have demonstrated the possibility of detecting plant mineral deficiencies, especially for N, by reflectance (Adams et al., 1993; Gamon et al., 1997; Haboudane et al., 2002; Vouillot et al., 1998; Zhao et al., 2003) and by UV-induced fluorescence (Apostol et al., 2003; Chappelle et al., 1984; Corp et al., 2003; McMurthey et al., 1994; Mercure et al., 2004; Tartachnyk and Rademacher, 2003). To evaluate the efficiency of reflectance and/or fluorescence to detect plant nutrient deficiencies, most studies relate the remote sensing indices to the leaves' pigment content (chlorophyll), the treatment induced, or occasionally to biomass or nutrient content. As mentioned in Bélanger (2005), it is important to measure growth parameters to evaluate the changes induced on fluorescence or reflectance indices. The combination of reflectance and fluorescence indices was suggested by McMurthey et al. (1994) to improve the detection of nutrient deficiencies and was confirmed in a previous experiment conducted on N-, K-, and Mg-deficient potato plants (Bélanger, 2005). By determining the relation between the CND indices and the remote sensing indices, we will have an indication of the nutrient imbalance and interactions occurring in the plant without destroying it.

The objectives of this study are (i) to determine if there is a relation among reflectance, fluorescence, and nutrient imbalance (CND_r2) in potato and (ii) to develop an indicator of the N imbalance in potato plants combining reflectance and fluorescence indices using canonical discriminant analysis.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
The processes of data acquisition and treatment are schematized in Fig. 1 . Optical measurements were taken once a week between the 15th day after emergence (DAE) and the 44th and were used to compute a canonical discriminant variable. Chemical analyses of leaf tissues were realized to calculate the N imbalance using the CND. A MINI was computed using the canonical variable and tested on a validation data set. Its efficiency was evaluated by estimating errors {alpha} and ß as well as its predictive value (yield).



View larger version (26K):
[in this window]
[in a new window]
  		Fig. 1. Schematic representation of data acquisition and treatment.

 
Plant Material
Three experiments (A, B, and C) took place in 2002 (A, B) and 2003 (C) in an experimental greenhouse located on Laval University Campus (Quebec City, Canada). The experimental plan was set as a nested design including five blocks and three treatments (N, K, and Mg) applied at four levels (15, 30, 60, and 100% of nutrient in mineral solution). A total of 170 potato plants were individually grown in 3-L containers under a 16-h photoperiod. The growing medium was a mix of 0.8 L L–1 vermiculite (Fafard et Frères Inc., Quebec, Canada) and 0.2 L L–1 quartz sand (Unimin Corp., New Canaan, CT, USA), washed with demineralized water before planting (Tukaki and Mahler, 1990). From emergence to the end of experiment, control plants were fertilized using a complete mineral solution made of Ca(NO3)2·4H2O, MgSO4·7H2O, CaCl2, KH2PO4, KNO3, Fe chelate (13.3%), Mn chelate, borax, Zn chelate, Cu chelate, and Na2MoO4·2H2O and corresponding to 100% of nutrient in mineral solution (160 mg N L–1, 29 mg P L–1, 234 mg K L–1, 160 mg Ca L–1, 48 mg Mg L–1, 62 mg S L–1, 1.83 mg Fe L–1, 0.5 mg Mn L–1, 0.5 mg B L–1, 0.05 mg Zn L–1, 0.02 mg Cu L–1, and 0.01 mg Mn L–1) (Tukaki and Mahler, 1990). To induce mineral deficiencies, other plants received modified mineral solutions providing 15, 30, or 60% of total N, K, or Mg concentrations received by the control treatment. In 2003, a LI-COR agrometeorological weather station (LI-COR Biosciences, Lincoln, NE, USA) installed in the greenhouse collected relative humidity (day: 29.9 ± 13.5%, night: 32.4 ± 12.5%), light intensity (day: 4.1 ± 3.2 klux, night: 0.1 ± 0.4 klux), and air temperature (day: 20.0 ± 1.1°C, night: 16.7 ± 2.1°C). In 2002, this instrument was not available, and we could only have access to minimum and maximum daily temperature (19.9 ± 1.1°C and 31.1 ± 2.3°C, respectively).

Compositional Nutrient Diagnosis
Foliar nutrient analyses for the CND computation were taken at the blooming stage because it corresponds to a period of high nutrient uptake during which potato crops are sensitive to nutrient imbalance (Parent and Dafir, 1992). On every plant, foliar analyses were conducted on the fourth fully expanded leaf taken from a main stem, at full bloom corresponding to the 37th DAE for Experiments A and C and to the 44th DAE for Experiment B. Plant tissues were dried (during 48 h at 65°C), ground, and digested following Parkinson and Allen (1975). Nutrient concentrations (P, K, Ca, and Mg) were determined using an Inductively Coupled Plasma (ICP) OPTIMA model 4300DV (PerkinElmer, Boston, MA). Nitrogen concentration was determined using the Quikchem method (Zellweger Analytic Inc., 2003) on a Flow Injection Analyzer (FIA), model Quikchem4000 (Lachat Instruments, Milwaukee, WI).

Our population of potato plants has been divided into a high- and a low-yield subpopulation. The yield threshold was set according to the biomass produced. Because only N deficiency had an impact on biomass production, the low-yield subpopulation was only composed of N-deficient plants. Other potato plants are included in the high-yield subpopulation and are considered to be "balanced." Compositional Nutrient Diagnosis computation has been done following Khiari et al. (2001b) and is presented here in four different steps.

Step 1: Determining the "Filling" Value (R5)
The compositional simplex (Sd) reflects the leaf composition and is bounded to 100% by summing up determined elements (N, P, K, Ca, Mg) and a "filling" value (R5) and may be defined as follows (Khiari et al., 2001b):

[1]
where N, P, K ... are the nutrient proportions (%) and R5, the filling value, represents the undetermined elemental composition as follows:

[2]

Step 2: Computing Log-Centered Ratio
Each nutrient concentration is divided by the geometrical mean of all nutrients' concentration to account for all nutrient interactions simultaneously, and the natural log is computed.


[3]
where

[4]
where d = number of analyzed nutrients (here 5). Note: the filling value does not account for an analyzed nutrient.

Step 3: Computing the CND Indices
The Vx mean (µVx) and standard deviation ({sigma}Vx) of the high-yield subpopulation are the CND norms to compute the CND indices as follows:

[5]
where µVn is the VN mean for the high-yield subpopulation and {sigma}Vn is the VN standard deviation for the high-yield subpopulation and so on...

Step 4: Computing the CND_r2
The CND_r2 is the sum of squared CND indices as follows:

[6]
where IN is the N index and IP is the P index, and so on...

The sum of d + 1 squared independent and unit-normal variables produces a new variable having a chi-square distribution with d + 1 degrees of freedom (Ross, 1987). To divide the population into unbalanced and balanced potato plants, a critical CND_r2 has to be determined. Figure 2 presents a chi-square distribution ({chi}2) having six degrees of freedom (df). The critical CND_r2 is of 12.6 for a {chi}2 having six degrees of freedom and a probability level {alpha} of 0.05. The critical CND_r2 can be validated by summing all critical squared CND indices that were independently determined by the Cate–Nelson partitioning procedure (Nelson and Anderson, 1977). In the Cate–Nelson procedure, an ANOVA is first computed to reach the highest sum of squares, and the threshold value is confirmed graphically by maximizing the number of points in the negative quadrants (Fig. 3) . The Cate–Nelson procedure is computed to determine each critical CND indices as well as critical CND_r2.



View larger version (12K):
[in this window]
[in a new window]
  		Fig. 2. Chi-square probability distribution function.

 


View larger version (15K):
[in this window]
[in a new window]
  		Fig. 3. Validation of IN2 using Cate–Nelson partitioning procedure.

 
Fluorescence
In 2002, fluorescence measurements were taken the 15, 23, 30, and 37th DAE (Experiment A) and the 44th DAE (Experiment B) and were made using a Xenon-Pulse-Amplitude-Modulation (Xe-PAM) fluorometer (Walz, Effeltrich, Germany). In 2003, fluorescence measurements were taken the 30 and 37th DAE (Experiment C) using a Fluorimètre pour Télédétection Embarquée (FLUTE) fluorometer (GAAP-TELOPS, Quebec, Canada). Both fluorometers can achieve UV-induced fluorescence measurements.

The Xe-PAM measurements were taken on a leaf disk placed on the instrument's sample holder. The FLUTE measurements were taken on an area of 720 mm2, 0.5 m from the top of the whole plant, under ambient light. The two fluorometers use a Xenon (Xe) flash lamp to induce excitation. Ultraviolet and blue excitations were obtained by placing filters in front of the Xe-excitation flash lamp, respectively, DG11 + UG11 (360 ± 20 nm) and BG39 + UV blocking (400 to 600 nm) (Schott Glass Technol., PA, USA) for the Xe-PAM and 360 ± 40 nm and 436 ± 20 nm (Chroma Technol. Corp., Rockingham, VT, USA) for the FLUTE. For the Xe-Pam, the detection was made at 440, 520, 690, and 750 ± 10 nm using Oriel filters of 2.54-cm diam. (Spectra-Physics, Stratford, CT, USA). For the FLUTE, detection was made at 440, 520, 690, and 740 nm ± 10 nm using 5.08 cm diameter filters (CVI Laser, Albuquerque, NM, USA). Intensities of UV-induced fluorescence were measured at the four emission bands whereas blue-induced fluorescence intensities were measured at 690 and 740 nm. Fluorescence intensities were calibrated against photodiode sensitivity and transmittance of the filters at the different wavelengths. Seven ratios of fluorescence intensities were computed for data analysis (Table 1).


View this table:
[in this window]
[in a new window]
  		Table 1. Name, formula, and source of the computed fluorescence indices.

 
Reflectance
Reflectance spectra ranging from 350 to 1500 nm (sampling interval of 1.4 nm) were collected using a Field Spec Pro spectroradiometer (Analytical Spectral Devices Inc., Boulder, CO, USA) having a 25° field of view. All reflectance spectra were taken between 1000 and 1400 h, 0.3 m directly above the plants, under ambient light in the greenhouse. Three spectra for each plant were collected at 15, 23, 30, and 37 DAE (Experiment A), at 44 DAE (Experiment B) in 2002, and at 30 and 37 DAE (Experiment C) in 2003. The instrument was regularly calibrated against the ambient light using a Spectralon plate (Labsphere, Inc., North Sutton, NH, USA) of known reflectance. For each plant, an average from three spectra (30 acquisitions) was computed; the averaged spectrum was smoothed by a Savitsky–Golay filter (Order 5, Window 21), and 17 reflectance indices were calculated (Table 2).


View this table:
[in this window]
[in a new window]
  		Table 2. Name, formula, and source of the computed reflectance indices.

 
Data Analyses
To identify outliers, a cluster analysis was performed using the FASTCLUS procedure from SAS software package (V8.2, SAS Inst., Cary, NC). An ANOVA was conducted on dry shoot biomass, CND_r2, and IN as dependant variables using the GLM procedure for a nested design (levels are nested within the treatments and treated as a random effect) and the LSMEANS option for simple means comparison. The CANDISC procedure was used across reflectance and fluorescence indices to generate canonical variables (linear combinations of quantitative variables) for inclusion into a discriminant function. The sum of raw canonical coefficients (a, b, c ... i) multiplied by the centered reflectance and fluorescence indices (Ax, Bx, Cx, ... Ix) produces a candisc value for each potato plant (Eq. [7]).


[7]
where Ax = A – µA = centered parameter A and i = raw canonical coefficient for parameter I.

Several discriminant analyses (depending on the classification variable) were conducted using the DISCRIM procedure to classify each plant into its corresponding treatment resulting in a classification percentage.

The MINI was developed using reflectance and fluorescence observations (total n = 313) taken on potato plants from Experiment A at 15 (n = 46), 23 (n = 59), 30 (n = 59), and 37 (n = 55) DAE and from Experiment C taken at 37 (n = 46) and 44 (n = 48) DAE. Once the canonical coefficients were determined, the MINI was tested on a validation data set. This data set corresponds to potato plants from the Experiment B (n = 60) that have not contributed to the previous development of MINI. Their reflectance and fluorescence measurements (n = 60) were taken at a different time (44 DAE) than the ones used in the computation data set (15, 23, 30, and 37 DAE). The IN for Experiment B was computed following the steps presented earlier (Eq. [1 to 6]).

To evaluate the efficiency of the new MINI, we estimated error {alpha}, error ß, and the predictive value. In terms of nutrient stress detection and fertilization, error {alpha} represents the probability of incorrectly concluding that plants should be fertilized, and error ß corresponds to the probability of incorrectly concluding that plants should not be fertilized. Increasing error {alpha} will increase fertilization cost, enhance leaching and environmental contamination risks, and reduce profits. By convention, error {alpha} is usually set at 5% whereas sometimes it can be set at a higher level (10 or 20%) depending on the variables tested and on the researchers' preference (Guénette, 2003; Irvine et al., 1999). Increasing error ß will reduce yield potential and hence profits. The power of a test, defined as 1 – ß, should be approximately 80% (Cohen, 1988). Both errors have been estimated using a Gaussian distribution of probabilities for a population X (Eq. [8–9]).


[8]
which probability distribution function is

[9]
The predictive value gives the probability of having a real positive subject when tested positive (Schork and Remington, 2000). The frequencies are denoted by the variables a, b, c, and d, and a corresponds, for example, to the number of subjects (here plants) tested positive when the reference is positive (Table 3). The predictive value can be computed using Eq. [10] and is useful to evaluate the odds of identifying a real positive subject, for instance.


View this table:
[in this window]
[in a new window]
  		Table 3. Generalized table of frequencies to compute the original Bayes' rule formulation of yield.{dagger}

 

[10]
where a, b, c, d, and N come from Table 3.


   RESULTS AND DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
CONCLUSIONS
 REFERENCES
 
The Results section is divided into four main sections presenting respectively the effects of the induced nutrient deficiencies on the dry shoot biomass, the IN, the discriminant analyses, and the development and computation of the MINI.

Dry Shoot Biomass
The effects of the induced nutrient deficiencies on the dry shoot biomass are presented in Table 4. The dry shoot biomass was significantly reduced by the N deficiency in Experiment A and C (respectively by 39.1 and 40.9%). However, no significant effects were observed on the dry shoot biomass for Mg and K treatments. Hence, the low-yield subpopulation for the computation of the compositional nutrient diagnosis will only be composed of plants receiving 15 or 30% of N.


View this table:
[in this window]
[in a new window]
  		Table 4. Dry shoot biomass for each induced mineral deficiency (K, Mg, or N) and for the control plants of Experiment A and C. Significant differences between the induced treatment and the control plants are shown by corresponding number of asterisks.

 
Compositional Nutrient Diagnosis
The high-yield subpopulation is composed with all plants receiving more than 30% of N, thus including K- and Mg-deficient plants and representing 81.3% of our population of potato plants.

The CND nutrient norms were computed as the mean and standard deviation of the high yield subpopulation (Table 5). The nutrient norms are similar to those obtained by Khiari et al. (2001c) and Parent et al. (1994). The CND indices were computed using the CND norms presented in Table 5.


View this table:
[in this window]
[in a new window]
  		Table 5. Compositional nutrient diagnosis (CND) norms.{dagger}

 
The critical squared IN has been independently determined using the Cate–Nelson partitioning procedure (Fig. 3) and corresponds to 4.5. In Fig. 3, the data in the upper left corner represent the potato plants being diagnosed as N balanced and having a high yield. The data in the lower right corner represent potato plants diagnosed as N unbalanced and having a low yield. The data in the lower left corner are N balanced but have a low yield that may be due to other deficiencies or to a random effect. The sufficiency ranges of CND indices can be computed as the square root of critical value and are as follows for IN:

The other critical squared nutrient indices were independently determined using the Cate–Nelson partitioning procedure and are respectively 1.7 for IP2, 1.2 for IK2, 1.7 for ICa2, 1.3 for IMg2, and 0.8 for IR52. The sum of the critical squared nutrient indices was 11.2. The CND_r2 obtained by the Cate–Nelson partitioning procedure was 11.5, meaning that plants having a CND_r2 lower than 11.5 will be considered as balanced, hence validating the critical CND_r2 obtained previously (11.2). For a chi-square distribution having six degrees of freedom, the CND_r2 of 11.2 corresponds to a probability level ({alpha}) of 0.08.

An ANOVA was performed to identify if nutrient treatments induced significant differences in the CND_r2 and in the IN. The results of simple mean comparisons are presented here: CND_r2 and IN from N-deficient potato plants (N15) were significantly different than the values obtained by the control plants (respectively, p = 0.0110 and p = <0.0001). Potassium- and Mg-deficient plants did not produce any significant effect on CND_r2 or IN compared with control plants (for CND_r2: p = 0.4055 and p = 0.4456; for IN: p = 0.7724 and p = 0.2699, respectively). Hence, N was the driving variable for diagnosing nutrient imbalance.

Discriminant Analyses
Discriminant analyses were conducted on fluorescence and reflectance indices (Tables 1–2) taken at 15, 23, 30, and 37 DAE from Experiment A in 2002 and at 30 and 37 DAE from Experiment C in 2003. Four different discriminant analyses were conducted on the same data set, using different classes: (1) nutrient treatments considering three treatments classes (N15, N30, and N_over30) corresponding to the amount of N applied; (2) leaf N content (g kg–1) based on Hochmuth et al. (2004) and Walworth and Muniz (1993) and considering three classes (deficient, sufficient, and excessive) depending on position (below, between, and over) compared with its sufficient level included between 0.30 and 0.45 g kg–1; (3) VN, divided into three classes (deficient, sufficient, and excessive) based on its confidence interval ({alpha} = 0.05); and (4) IN including two classes (deficient and sufficient) depending on position (below or over) compared with critical value of –2.13. An excessive class could have been added; however, no plants in our data set had an IN value exceeding 2.13.

The proportion of potato plants correctly classified to its specific class is presented in Table 6. The discriminant analysis conducted over IN classes provided the highest averaged classification percentage (94.9%) followed by leaf N content (88.6%). In general, the proportions of plants correctly classified according to discriminant analyses based on either fluorescence (84.3%) or reflectance (83.2%) indices were similar. The combination of fluorescence and reflectance indices enhanced the treatments' discrimination (86.3%).


View this table:
[in this window]
[in a new window]
  		Table 6. Reclassification percentage using the cross-validation option for each discriminatory analysis.

 
Those results support the development of a multivariate indicator able to correctly diagnose N imbalance. As the classes of the IN are the most easily detected over our data set, the indicator development will be based on the canonical discriminant variable calculated using reflectance and fluorescence indices over IN classes.

Coefficients to compute the canonical variable across reflectance and fluorescence indices for the detection of N imbalance (IN) are presented in Table 7.


View this table:
[in this window]
[in a new window]
  		Table 7. Coefficients to compute canonical variables.

 
Description of MINI
The first canonical variable of the model using both reflectance and fluorescence indices was positively correlated with the independent variable (0.561) and estimated more than 30% of the variance. The canonical variable computed using reflectance or fluorescence indices could explain respectively only 24 and 6% of the variation. The MINI was thus computed as a canonical variable across both reflectance and fluorescence indices, using the raw coefficients presented in Table 7.

The linear distribution of MINI values from the computation and the validation data sets is presented in Fig. 4 as their position according to their CAN1 value. The important discrimination to be made relates to deficient IN, indicating higher N requirements. Sufficient IN values indicate adequate growth condition. The MINI was developed using Can1: the deficient IN zone was delimited using the lower boundary from the confidence interval ({alpha} = 0.001) of the Can1 averaged value over all deficient IN. The threshold was set at Can1 > 0.94 across reflectance and fluorescence indices.



View larger version (5K):
[in this window]
[in a new window]
  		Fig. 4. Position of sufficient and deficient plants according to their CAN1 value.

 
Table 8 presents the frequencies of classification using the MINI over both computation and validation data sets. The proportion of N-sufficient potato plants identified as N sufficient by the indicator is 90.5%. The predictive value representing the probability that a plant be N sufficient when it is declared as N sufficient by the MINI is 97.5%. Even though the proportion of N-deficient plants represented only 6.9% of the total number of observations, there were 69.2% of N-deficient plants declared as N deficient by our indicator.


View this table:
[in this window]
[in a new window]
  		Table 8. Table of frequencies over N index (IN) classes using Candisc variable.

 
For the computation data set (Experiment A and C, n = 313), the significance level ({alpha}) is 9.7%, and the power of the indicator (1 – ß) is 76.9%. Both errors are in an acceptable range for remote sensing ({alpha} < 0.20 and 1 – ß {approx} 80%) (Guénette, 2003; Irvine et al., 1999; Cohen, 1988). For the validation data set (Experiment B, n = 60), the {alpha} is 15.2%, and the power is 62.7%. The power could be higher if the data set included more samples.

Considering its current format, the MINI can discriminate N sufficiency from N deficiency in potato plants. It may also be important to detect N excess to avoid fertilizer application on overfertilized plants. Reflectance and fluorescence measurements can detect N excess on plants (Railyan et al., 1990; Romanova et al., 1987) as well as the IN, when IN > 2.13. To adapt the indicator to N excess detection, additional data involving plants suffering from N excess should be aggregated to the database and the N excess zone determined by setting the indicator threshold. Moreover, by using kriging, for instance, a MINI map could be drawn over a complete field using data acquired only over a sampling grid, thus reducing acquisition time.


   CONCLUSIONS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
REFERENCES
 
Using greenhouse-grown plants and discriminant analysis, our results show that the detection of N imbalance (IN) by reflectance or fluorescence indices was possible. According to discriminant analyses, the proportion of plants correctly classified to its IN has enhanced from 93.3 to 96.6% when combining fluorescence and reflectance indices using a canonical variable (Table 6). Using this canonical variable, it was possible to combine reflectance and fluorescence indices and still reflect their complementary nature and the between-class variance.

The MINI was derived using the canonical variable across reflectance and fluorescence indices. This indicator can correctly detect almost 70% of the N-deficient plants and more than 90% of the N-sufficient plants. Contrary to the standard methods of deficiency detection (foliar analyses) that have been proven to be costly and time-consuming (Link et al., 2003), this indicator could allow a rapid data acquisition and handling and provide deficiency detection within the time window for plant response to a recovery fertilization. It is a first step toward the automation of diagnosing nutrient deficiencies using reflectance and fluorescence indices. Moreover, it could allow variable-rate fertilization, which has been proven to increase yield, reduce lodging, and achieve a more homogenous ripening of the crop, thus reducing harvesting costs and grain losses (Link et al., 2003). The net benefices of variable-rate fertilization can be as high as US$87 per hectare in corn crop (Lambert and Lowenberg-DeBoer, 2000). Since UV-exposed plants might have a different fluorescence response than greenhouse-grown plants, the indicator should now be tested in the field for N diagnosis and site-specific N fertilization using a tractor-mounted instrument (Belzile et al., 2003). Future works should also include testing for growth stage specificity and the development of the N-excess indicator, notably.


   ACKNOWLEDGMENTS
 
The authors thank the Fonds québécois de la recherche sur la nature et les technologies (FQRNT), the Canadian Foundation for Innovation (CFI), and the Natural Sciences and Engineering Research Council of Canada (NSERC) for their financial support; Léon-Étienne Parent, Ph.D., for appreciated review of this manuscript; Cultures Dolbec, Inc. for supplying the potatoes; Marie-Élaine Boivin, Laure Chandelier, Ludovic Béland, Marie-Amélie Bélanger, and Serge-Olivier Kotchi for their hard work in the greenhouse; and Charles Belzile, Simon Roy, Nelson Landry, and Stéfan Parmentier for their technical advice.


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





This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
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 Google Scholar
Google Scholar
Right arrow Articles by Bélanger, M-C.
Right arrow Articles by Chamberland, M.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Bélanger, M-C.
Right arrow Articles by Chamberland, M.
Agricola
Right arrow Articles by Bélanger, M-C.
Right arrow Articles by Chamberland, M.
Related Collections
Right arrow Potato
Right arrow Remote Sensing
Right arrow Nutrient Management
Right arrow Nitrogen

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Crop Science Vadose Zone Journal
Journal of Plant Registrations Soil Science Society of America Journal
Journal of Natural Resources
and Life Sciences Education		 Journal of
Environmental Quality