Abstract
A nondestructive method for the determination of chlorophyll index for the tea plant based on reflectance spectral characteristics was investigated. Spectral data were collected from 184 samples with a spectroradiometer in a field experiment. Multivariate analysis techniques, including partial least squares (PLS) and multiple linear regression (MLR), were used for developing calibration models for the determination of chlorophyll index of the tea plant. The best calibration model was achieved using the PLS technique with a correlation coefficient (r) of 0.95, a se of prediction of 3.40, and a bias of 1.9e−06. When the model was used for predicting the unknown samples, good performance was also obtained with r of 0.91, se of calibration of 4.77, and bias of 0.02. Sensitive wavelengths were selected through loading analysis of latent variables in the optimal PLS model, and the validity of these wavelengths was proved by MLR and statistical analysis. Three fingerprint wavelengths (488, 695, and 931 nm) were determined and could potentially be used for developing a simple, low-cost, and efficient instrument for the measurement of chlorophyll index. The results proved the feasibility of reflectance spectra for measurement of chlorophyll index of the tea plant.
Nitrogen (N) is an important nutrient for tea plant growth to obtain high yield or high quality. An adequate amount of N can accelerate the growth of tea plant, whereas excess N fertilizer may contaminate surface water and groundwater. Accurate assessment of N content of the tea plant is necessary for N fertilization.
Traditional methods for chlorophyll and N determination include soil testing, plant tissue analysis, and long-term field trials. Although these methods are accurate, they are destructive, time-consuming, and expensive. Portable nondestructive meters have been successfully used to determine foliar chlorophyll or N of many plants (Abdelhamid et al., 2003; Castelli et al., 1996; Loh et al., 2002; Schaper and Chacko, 1991; Yamamoto et al., 2002). Chlorophyll meters (e.g., SPAD-502, Minolta, Osaka, Japan) have been used to assess plant N status by measuring transmittance radiation through a leaf at two wavelengths centered near 650 nm and 940 nm (Pinkard et al., 2006). Because the majority of foliar N is contained in chlorophyll molecules, a close relation between foliar chlorophyll content and foliar N content exists (Yoder and Pettigrew-Crosby, 1995). Foliar chlorophyll content is a good indicator of plant N status and photosynthesis activity (Chang and Robison, 2003). Leaf chlorophyll meter readings are generally linear with extractable chlorophyll contents for a wide variety of crops such as sugar maple leaves (van den Berg and Perkins, 2004), four species of hardwoods (Chang and Robison, 2003), and corn (Ulson et al., 2002). Furthermore, Song et al. (2002) studied the relationship between SPAD reading and chlorophyll content of fresh tea leaves and concluded that SPAD reading is correlated with chlorophyll content. Furthermore, high correlation exists between SPAD reading and ratio of chlorophyll a to chlorophyll b.
Although the chlorophyll meter provides a relative N assessment without complex laboratory analysis, 15 or more readings are needed and averaged to get one measurement. Collecting the data of N content for large fields with spatial variations is a time-consuming process using a chlorophyll meter (Daughtry et al., 2000). The measurement of leaf spectral reflectance using a spectrometer is a noncontact and nondestructive approach, and the scan time of one measurement can be less than 1 s; it is a promising method for fast sensing of N status (Bausch and Duke, 1996; Tumbo et al., 2002). Bausch and Duke (1996) used a mobile system to acquire spectral data of corn and found that the N reflectance index had a good correlation (r 2 = 0.81) with chlorophyll meter readings. Tumbo et al. (2002) assembled a spectrometer on a mobile tractor for fast measurement of chlorophyll index and found that spectral reflectance response patterns from a mobile sensor with speed of 0.6 km·h−1 could be used to predict chlorophyll in corn, and a strong relationship was obtained between near infrared/green (NIR/G) ratio and chlorophyll meter readings (in SPAD units). Several studies have shown that a direct relationship existed among spectral reflectance, chlorophyll content, and N status in green vegetation. Takebe et al. (1990) obtained a good correlation (r 2 = 0.90) between the chlorophyll meter and spectral reflectance of rice. The spectrophotometer has been used to detect N content of cabbage seedling based on the reflectance spectra of leaves, and a good correlation (rc 2 = 0.89) was obtained with significant wavelengths (566, 574, 1396, and 1530 nm) (Chen et al., 2004). Karimi et al. (2005) found a high accuracy of discrimination using hyperspectral reflectance spectra for plots with different N treatments. Min and Lee (2005) developed a N sensor for citrus trees based on diffuse reflectance of leaf samples; their calibration models showed a strong relationship between actual N concentration and reflectance spectra of citrus. For another economically important plant, tea, Hu et al. (2006) investigated the relationship between spectral reflectance and chlorophyll meter reading of fresh tea leaf and concluded that it was applicable to detect chlorophyll index based on reflectance spectra. However, the correlation coefficient was relatively low and the measurement se was large. At the same time, Ishikawa et al. (2006) focused on the determination of chlorophyll of the tea plant by spectral image; they concluded that spectral image made it possible to study differences of growth stage of the tea plant. However, spectral image analysis is more complex compared with the reflectance spectra technique.
In this study, the relationship between visible/near infrared (Vis/NIR) reflectance spectral characteristics of tea plant and chlorophyll index was studied, and more attention was focused on evaluating the performance of calibration models, selecting the proper data pretreatment method, and exploring sensitive wavelengths. The specific objectives were: 1) to evaluate the potential of reflectance spectral characteristics for determining the chlorophyll index of six tea cultivars with a portable spectrophotometer; 2) to explore an effective pretreatment method to enhance the signal-to-noise and reduce noise caused by field environment; and 3) to seek the optimum sensitive wavelengths that are strongly related with the chlorophyll index of the tea plant.
Materials and Methods
Six cultivars of tea plants were selected for this experiment, including C. sinensis cv. Jiukeng (JK), C. sinensis cv. Biyun (BY), C. sinensis cv. Juhuaxiang (JH), C. sinensis cv. Lvyafoshou (LY), C. sinensis cv. Meizhan (MZ), and C. sinensis cv. Zhenghedabaicha (ZH). The samples of the six cultivars were planted in the Seed Resource Garden of Tea Plant in Zhejiang University (long. 120.19° E, lat. 30.26° N), China. These cultivars were planted at two blocks with three cultivars in each block. Within each block, three cultivars were planted in adjacent rows. The physiological age of these plants is ≈30 years, the soil is red loam, and all the tea plants were treated with the same level of N fertilizer (90 kg·ha−1 urea). One hundred eighty-four leaves (samples) were randomly selected from ≈60 tea plants and approximately three samples, which were close to the top of the plant were pricked off from each plant. To add the diversity of samples, leaves with different physiological ages were taken into the experiment. The detailed information about the six cultivars is presented in Table 1.
Detailed information of the test plants.
On 20 April and 23 May in 2006, the field experiment was conducted. Reflectance spectra were acquired from 930 to 1630 under clear sky conditions using a field spectroradiometer (Fieldspec HandHeld; Analytical Spectral Devices, Boulder, CO) with 3.5-nm full-width-half-maximum spectral resolution and wavelength range from 325 to 1075 nm. The spectroradiometer was stabilized on a tripod with 45° angle between the spectroradiometer and the horizontal line. The spectroradiometer was set up at ≈100 mm above the surface of the leaf sample with a 10° field of view. Sun was the light source. The zenith and azimuth angle of the sun changed during the experiment period, from 0° to 90° and from –180° to 180°, respectively. To reduce the effect of solar radiation intensity variation, the spectroradiometer was calibrated every half hour by a 100-mm−2 optical reference standard panel with ≈100% reflectance across the entire spectrum. Reflectance was computed with measurements from both the leaf sample and a reference standard panel. For each sample, a mean spectrum was averaged by 60 scans. The reflectance spectra of all samples were transformed to absorbance [log (1/R)] values according to Beer-Lambert law (Williams and Norris, 2001). Typical absorbance spectra of samples were shown in Figure 1. The first 75 and the last 75 wavelengths values were cut off as a result of significant noise observed, and all analyses were based on the range of wavelengths between 400 and 1000 nm. After spectra measurement, the relative chlorophyll content was determined by a Minolta SPAD-502 (Minolta, Osaka, Japan) chlorophyll meter. The SPAD sensor was clipped on leaf mesophyll tissue instead of veins. Thirty measurements were taken per sample and averaged to provide a single chlorophyll index.
Typical absorbance spectra of fresh tea leaves.
Citation: HortScience horts 43, 5; 10.21273/HORTSCI.43.5.1586
To test the influences of pretreatment methods on the performance of the calibration models, several types of pretreatments were used, including smoothing, standard normal variate, multiplicative scatter correction, derivatives, and offset (Candolfi et al., 1999). These pretreatments and calculations were carried out using The Unscrambler V9.2 (Camo, Process, AS, Oslo, Norway), a statistical software package for multivariate calibration.
All 184 spectra were divided into training set and prediction set. To avoid bias in subset selection, all samples were sorted according to their respective Y-value (chlorophyll index), then one of three samples were selected as the prediction set, resulting in 124 spectra of training set and 60 spectra of prediction set with ≈2:1 division (Table 2). The range of Y-value in the training set covered the range of that in the prediction set; therefore, the distribution of the samples was appropriate in the training set and prediction set.
Statistical parameters of chlorophyll index (Y).
Partial least squares (PLS) is a very powerful and classical multivariate analysis technique in chemometrics (Gomez et al., 2006; Lammertyn et al., 1998; Min and Lee, 2005; Zou et al., 2007). Compared with multiple linear regression (MLR), the advantage of PLS is that it is a bilinear modeling method in which the original X (predictor variables) is projected onto a small number of orthogonal latent variables (LVs) to simplify the relationship between X and Y (response variables) (Lattin et al., 2003) and mitigate the colinearity problem (Helland, 2001; Min and Lee, 2005). In contrast to principal component regression (PCR), the selection of LVs in PLS is more rational. In detail, PCR chooses LVs with the maximum variance in X but does not consider the Y, but the PLS method balances the two aspects and seeks the LVs that explain both X and Y. The optimal number of LVs was generally chosen when the minimum of predicted residual error sum of squares (PRESS) value was obtained using full crossvalidation. The PRESS value was the sum of the squared difference between actual and predicted concentrations (during crossvalidation) (Madan et al., 2005). In this study, PLS regression was adopted on the full spectra (400 to 1000 nm) of all 184 samples to explore the relationship between reflectance spectral characteristic and chlorophyll index.
The MLR model is a linear combination of X variable that corresponds as closely as possible to the Y variable (Lattin et al., 2003). Comparing with PLS and PCR, the drawback of MLR is that the number of samples for MLR must be greater than the number of variables (Madan et al., 2005), and the possible correlation among X variables cannot be eliminated. The advantage of MLR is that a direct, simple, and clear relationship between X and Y can be obtained through MLR analysis. In this study, a small number of sensitive wavelengths from PLS analysis were set as predictor variables (X), and the size of these variables was much smaller than that of samples. The MLR model was used to build a direct, simple, and linear formula between X and Y.
The performances of the PLS and MLR models were evaluated by the se of calibration (SEC), the se of prediction (SEP), and the correlation coefficient (r) between the actual and predicted concentration. For SEC, a leave-one-sample-out crossvalidation was performed. A good model should have a low SEC, a low SEP, a high r, and also a small difference between SEC and SEP. A large difference indicates that too many LVs are used in the model and the noise is included to the model (Gomez et al., 2006; Min and Lee, 2005; Zou et al., 2007). SEC and SEP are calculated using Eqs. [1] and [2], respectively.
where ŷi -predicted value of the itch observation, yi measured the value of the ith observation, Ic = number of samples in the calibration set, n = number of independent variables in calibration, Ip = number of samples in the validation/prediction set, and bias = systematic difference between actual and predicted values calculated by formula
Results and Discussion
Figure 1 shows typical absorbance spectra for the fresh tea leaves. All the samples of six cultivars had similar characteristics in absorbance spectral curves. From 400 to 500 nm, the spectral curves were relatively flat and the absorbance values were at the highest level. After 500 nm, spectral curves began to decrease until a “V” shape appeared at 550 nm. From 550 to 675 nm, the absorbance values increased along with the wavelength with a peak at 675 nm. Then the spectral curve decreased again until the minimum was reached at 750 nm. From 750 to 1000 nm, the spectral curves were flat. It can be concluded that the leaf absorbed blue (400 to 500 nm) and red (680 nm) light in the visible range, as reported in previous literature (Datt, 1999) and reflected NIR light (750 to 1000 nm).
In Table 1, there are some differences in leaf color of these six cultivars. For example, the leaves of ‘Meizhan’ and ‘Zhenghedabaicha’ are greener than that of other cultivars. It can be found that the degree of greenness is generally consistent with the chlorophyll meter reading for cultivar; in other words, the cultivar with greener leaves usually has a higher chlorophyll reading. To develop a model with high adaptability, all the samples of the six cultivars were considered as a group for building the calibration model without building a model for each cultivar, although the diversity of samples from six cultivars provided some help for developing a model with good stability and high performance.
Considering the different spectral pretreatments, several calibration models were constructed based on the use of PLS. The results of these calibration models are summarized in Table 3. In model 9, r was very low and the values of SEC and SEP were much larger than that of the other models. This result indicated that the correlation between spectral characteristics and chlorophyll index was decreased by the second derivation spectral pretreatment. High-frequency noises in the spectra resulting from the reflectance mode in the field experiment could be magnified through the second derivation pretreatment (Candolfi et al., 1999; Zeaiter et al., 2005). The second derivation pretreatment was not suitable for these spectra.
Results of partial least squares models with different data pretreatment methods.
Model 6 processed by offset correction was evaluated as the best model compared with other models (Table 3). The correlation coefficient of model 6 is the highest, the SEC and SEP values were smallest, and the difference between SEC and SEP values was relative small. The plots of predicted chlorophyll index versus reference chlorophyll index and their statistical summaries are shown in Figure 2. Compared with a similar study by Hu et al. (2006), the same conclusion that reflectance spectra was strongly correlated with chlorophyll index was obtained, but our results were better than that by Hu et al. (2006); their calibration model was obtained with r of 0.7301 and root mean square error of prediction of 10.0781. The worse performance of model by Hu et al. (2006) may be caused by improper spectral pretreatment. The pretreatment method of second derivation used by Hu et al. (2006) had been proved to be an improper method for processing spectral data of tea leaves in this study.
Results of the optimum calibration model for determination of chlorophyll using partial least squares.
Citation: HortScience horts 43, 5; 10.21273/HORTSCI.43.5.1586
To better understand the relationship between diffuse reflectance spectra and chlorophyll index, the individual wavelengths were analyzed. The optimum calibration model 6 was obtained based on nine LVs of PLS that contributed to the formation of the model. Table 4 shows the variation of each LV (LV1 to LV9). The first nine LVs accounted for the most variation of the whole original data with 97.997%, and the good performance of model 6 proved that these LVs were closely correlated with the chlorophyll index. How the LV was constructed from the absorbance at entire individual wavelengths was demonstrated by the plot of LV loading in Figure 3, and the critical wavelengths for each LV could be detected. These wavelengths with small absolute loading values were less important than those with large values. The loading weight curves had strong peaks and valleys at certain wavelengths, including 435, 450, 488, 523, 550, 615, 650, 675, 695, 931, 970, and 989 nm (Fig. 3A–I). Those individual wavelengths might be of particular importance for determination of the chlorophyll index (Lammertyn et al., 1998). The wavelength 989 nm was not selected as a sensitive wavelength to avoid the noise because it was close to the edge of the wavelength region (1000 nm).
Percentage variation explained by the nine latent variables in model 6.
Loading weights of the first nine latent variables (A–I) of the optimum calibration model using partial least squares.
Citation: HortScience horts 43, 5; 10.21273/HORTSCI.43.5.1586
The MLR model was built based on these wavelengths to evaluate their validities [model 10, Eq. (4)]. All 11 wavelengths, including 435, 450, 488, 523, 550, 615, 650, 675, 695, 931, and 970 nm, were set as independent predictor variables and the chlorophyll index was set as the response variable in the MLR model.
The significance of the MLR model 10 was analyzed by analysis of variance at 99% confidence level (Table 5), and the variables x3, x9, and x10 were also significant for the model at 99% confidence level. So, a new MLR model was built based on these variables x3, x9, and x10 [model 11, Eq. (5)].
Analysis of variance table of the MLR model 10 based on all the 11 sensitive wavelengths.
MLR model 11 was statistically significant at 99% confidence level with three significant variables, x3, x9, and x10 (Table 6). The three corresponding wavelengths, 488, 695, and 931 nm, were considered as the characteristic fingerprint spectra, which were strongly correlated with the chlorophyll index. The wavelength of 695 nm was the most significantly associated with chlorophyll determination. This spectral region had been found to be closely related to chlorophyll content in many plant species such as Chinese cabbage (Min et al., 2006), maple, chestnut, cotoneaster, tobacco (Gitelson and Merzlyak, 1998), eucalyptus (Datt, 1998), and loblolly pine (Carter and Knapp, 2001). The sensitive wavelength 488 nm was very close to the wavelength of 480 nm adopted for the determination of total N by Card et al. (1988). Chlorophyll of fresh tea leaves consists of chlorophyll a and chlorophyll b (Tong, 2000), and chlorophyll a and chlorophyll b mainly absorb light in the range of 400 to 500 nm and 660 to 690 nm, respectively. These two ranges covered the two sensitive wavelengths (488 and 695 nm) found in this study. Among these three sensitive wavelengths, two wavelengths (488 and 695 nm) were in the visible band, and the lights in visible bands had been found to be effective for determination of chlorophyll by Yoder and Pettigrew-Crosby (1995). The wavelength of 931 nm was not reported as a sensitive wavelength for chlorophyll determination before, and it was contrary with the conclusion by Knipling (1970), in which they reported that reflectance in the NIR region was not correlated to leaf chlorophyll. However, the wavelength (931 nm) might actually be useful for chlorophyll determination of fresh tea leaves, because it was close to the third overtone absorptions band of methylene C-H stretching vibration and methane C-H stretching vibration (Cen and He, 2007).
Analysis of variance table of the MLR model 11.
The correlation coefficient (r) for the MLR model was 0.89 and SEC was 5.13 (Fig. 4A). When the model was used to predict the unknown samples, a reasonable result was obtained with r = 0.86 and SEP = 5.36 (Fig. 4B). The performance of the MLR model was slightly worse than that of the optimal PLS model (Fig. 2). This phenomenon was not consistent with the conclusion by Siesler et al. (2002), in which they reported that MLR model with a few selected spectral variables would frequently outperform the PLS model with full spectral region. This phenomena might be caused by the fact that these diffuse reflectance spectra obtained in this field experiment were interfered with by serious noise and colinearity, and PLS arithmetic can compress data and reduce colinearity between variables (Helland, 2001; Min and Lee, 2005), so the PLS method obtained better performance than the MLR model. However, the MLR model was only based on three individual wavelengths and showed reasonable validity for determination of chlorophyll index, so MLR was also an effective technique for multivariate analysis. In addition, the PLS model provided an approach for wavelength selection that led to a better understanding of diffuse reflectance spectra; the loadings and loading weights of LVs for the PLS model gave the indication of a sensitive wavelength for the predicted Y.
Results of the calibration model using multiple linear regression and three fingerprint wavelengths.
Citation: HortScience horts 43, 5; 10.21273/HORTSCI.43.5.1586
The results showed that it was feasible to determine the chlorophyll index of the tea plant using reflectance spectral characteristics. The multivariable analysis techniques (PLS and MLR) had the potential to develop regression models for assessing chlorophyll index based on spectral data. The proper pretreatment could enhance the quality of the model by filtering the noise. The optimal PLS model was built with r of 0.95, the SEP of 3.40, and a bias of 1.9e−06. The PLS model not only had high prediction accuracy, but also provided an approach for seeking the sensitive wavelengths. The MLR method was effective for selecting the most significant independent spectral variable using analysis of variance and achieved a reasonable result for determination of the chlorophyll index. Three individual sensitive wavelengths (488, 695, and 931 nm) were determined for the measurement of chlorophyll index and could potentially be used for developing a simple, low-cost, and efficient instrument.
Literature Cited
Abdelhamid, M. , Horiuchi, T. & Oba, S. 2003 Evaluation of the SPAD value in faba bean (Vicia faba L.) leaves in relation to different fertilizer applications Plant Prod. Sci. 6 185 189
Bausch, W.C. & Duke, H.R. 1996 Remote sensing of plant nitrogen status in corn Trans. ASAE 39 1869 1875
Candolfi, A. , De Maesschalck, R. , Jouan-Rimbaud, D. , Hailey, P.A. & Massart, D.L. 1999 The influence of data pre-processing in the pattern recognition of excipients near-infrared spectra J Pharm. 21 115 132
Card, D.H. , Peterson, D.L. , Matson, P.A. & Aber, J.D. 1988 Prediction of leaf chemistry by the use of visible and near-infrared reflectance spectroscopy Remote Sens. Environ. 26 123 147
Carter, G.A. & Knapp, A.K. 2001 Leaf optical properties in higher plants: Linking spectral characteristics to stress and chlorophyll concentration Amer. J. Bot. 88 677 684
Castelli, F. , Contillo, R. & Miceli, F. 1996 Non-destructive determination of leaf chlorophyll content in four crop species J. Agron. Crop Sci. 177 275 283
Cen, H. & He, Y. 2007 Theory and application of near infrared reflectance spectroscopy in determination of food quality Trends Food 18 72 83
Chang, S.X. & Robison, D.J. 2003 Nondestructive and rapid estimation of hardwood foliar nitrogen status using the SPAD-502 chlorophyll meter Forest Ecol. 181 331 338
Chen, C.T. , Chen, S. , Hsieh, K.W. , Yang, H.C. , Hsiao, S. & Yang, I.C. 2004 Estimation of Nitrogen content by spectral responses of cabbage seedlings using artificial neural network approach. ASAE/CSAE Annual International Meeting sponsored by ASAE/CSAE Fairmont Chateau Laurier The Westin, Government Centre Ottawa Ontario, Canada 1 4 Aug. 2004. Paper number: 041082.
Datt, B. 1998 Remote sensing of chlorophyll a, chlorophyll b, chlorophyll a + b, and total carotenoid content in eucalyptus leaves Remote Sens. Environ. 66 111 121
Datt, B. 1999 A new reflectance index for remote sensing of chlorophyll content in higher plants: Tests using eucalyptus leaves J. Plant Physiol. 154 30 36
Daughtry, C.S.T. , Walthall, C.L. , Kim, M.S. , de Colstoun, E.B. & McMurtrey, J.E. 2000 Estimating core leaf chlorophyll concentration from leaf and canopy reflectance Remote Sensing of Environment 74 229 239
Gitelson, A.A. & Merzlyak, M.N. 1998 Remote sensing of chlorophyll concentration in higher plant leaves Adv. Space Res. 22 689 692
Gomez, A.H. , He, Y. & Pereira, A.G. 2006 Non-destructive measurement of acidity, soluble solids and firmness of Satsuma mandarin using Vis/NIR-spectroscopy technique J. Food Eng. 77 313 319
Helland, I.S. 2001 Some theoretical aspects of partial least squares regression Chem. Intell. 58 97 107
Hu, Y. , Li, P. , Mao, H. , Chen, B. & Ding, Y. 2006 Spectral reflectance-based detection of nitrogen content in fresh tea leaves Proc. of SPIE. 6411 64110G1 64110G8
Ishikawa, D. , Etsuji, I. , Sekioka, S.I. , Okamoto, T. , Nesumi, A. & Oba, K. 2006 Development of yield and quality estimation methods for tea-plant by image processing Proc. of ASA-CSSA-SSSA Annual Meeting 43 44
Karimi, Y. , Prasher, S.O. , McNairn, H. , Bonnell, R.B. , Dutilleul, P. & Goel, P.K. 2005 Discriminant analysis of hyperspectral data for assessing water and nitrogen stresses in corn Trans. ASAE 48 805 813
Knipling, E.B. 1970 Physical and physiological basis for the reflectance of visible and near-infrared radiation from vegetation Remote Sens. Environ. 1 155 159
Lammertyn, J. , Nicolaï, B. , Ooms, K. , De Smedt, V. & De Baerdemaeker, J. 1998 Non-destructive measurement of acidity, soluble solids, and firmness of Jonagold apples using NIR-spectroscopy Trans. ASAE 41 1089 1094
Lattin, J.M. , Douglas, J. & Green, P.E. 2003 Analyzing multivariate data China Machine Press Beijing, China
Loh, F.C.W. , Grabosky, J.C. & Bassuk, N.L. 2002 Using the SPAD-502 meter to assess chlorophyll and nitrogen content of benjamin fig and cottonwood leaves HortTechnology 12 682 686
Madan, J. , Dwivedi, A.K. & Singh, S. 2005 Estimation of antitubercular drugs combination in pharmaceutical formulations using multivariate calibration Anal. Chim. Acta 538 345 353
Min, M. & Lee, W.S. 2005 Determination of significant wavelengths and prediction of nitrogen content for citrus Trans. ASAE 48 455 461
Min, M. , Lee, W.S. , Kim, Y.H. & Bucklin, R.A. 2006 Nondestructive detection of nitrogen in Chinese cabbage leaves using VIS-NIR spectroscopy HortScience 41 162 166
Pinkard, E.A. , Patel, V. & Mohammed, C. 2006 Chlorophyll and nitrogen determination for plantation-grown Eucalyptus nitens and E. globules using a non-destructive meter Forest Ecol. 223 211 217
Schaper, H. & Chacko, E.K. 1991 Relation between extractable chlorophyll and portable chlorophyll meter readings in leaves of eight tropical and subtropical fruit tree species J. Plant Physiol. 138 674 677
Siesler, H.W. , Ozaki, Y. , Kawata, S. & Michael Heise, H. 2002 Near-infrared spectroscopy: Principle, instrument, applications Wiley-VCH Weinheim, Germany
Song, X.Z. , Zhao, Y.J. , Zhang, J.F. , Liu, P. & Chen, J.H. 2002 Relationship between chlorophyll meter readings (SPAD readings) and chlorophyll content of tea leaves Shandong Forestry Science and Technology 6 10 12 (in Chinese)
Takebe, M. , Yoneyama, T. , Inada, K. & Murakami, T. 1990 Spectral reflectance ratio of rice canopy for estimating crop nitrogen status Plant Soil 122 295 297
Tong, Q.Q. 2000 Tea cultivation 3rd Ed Agriculture Publishers Beijing, China
Tumbo, S.D. , Wagner, D.G. & Heinemann, P.H. 2002 Hyperspectral characteristics of corn plants under different chlorophyll levels Trans. ASAE 45 815 823
Ulson, J.A.C. , Boas, R.L.V. , Godoy, L.J.G. & de Souza, A.N. 2002 A novel intelligent system to nitrogen content prediction in plants using indirect chlorophyll measurements 29 35 Zazueta F.S. & Xin J. Proc. of the World Congress of Computers in Agriculture and Natural Resources 13–15 March 2002 Iguacu Falls, Brazil
van den Berg, A.K. & Perkins, T.D. 2004 Evaluation of a portable chlorophyll meter to estimate chlorophyll and nitrogen contents in sugar maple leaves Forest Ecol. 200 113 117
Williams, P. & Norris, K. 2001 Near-infrared technology in the agricultural and food industries 2nd Ed American Association of Cereal Chemists St. Paul, MN
Yamamoto, A. , Nakamura, T. , Adu-Gyamfi, J.J. & Saiqusa, M. 2002 Relationship between chlorophyll content in leaves of sorghum and pigeon pea determined by extraction method and by chlorophyll meter (SPAD-502) J. Plant Nutr. 25 2295 2301
Yoder, B.J. & Pettigrew-Crosby, R.E. 1995 Predicting nitrogen and chlorophyll content and concentrations from reflectance spectra (400–2500 nm) at leaf and canopy scales Remote Sens. Environ. 53 199 211
Zeaiter, M. , Roger, J.M. & Bellon-Maurel, V. 2005 Robustness of models developed by multivariate calibration. Part II: The influence of pre-processing methods TRAC-Trends 24 437 445
Zou, X.B. , Zhao, J.W. , Huang, X.Y. & Li, Y.X. 2007 Use of FT-NIR spectrometry in non-invasive measurement of soluble solid contents (SSC) of ‘Fuji’ apple based on different PLS models Chem. Intell. 87 43 51