Reflectance Variation within the In-Chlorophyll Centre Waveband for Robust Retrieval of Leaf Chlorophyll Content

The in-chlorophyll centre waveband (ICCW) (640–680 nm) is the specific chlorophyll (Chl) absorption band, but the reflectance in this band has not been used as an optimal index for non-destructive determination of plant Chl content in recent decades. This study develops a new spectral index based solely on the ICCW for robust retrieval of leaf Chl content for the first time. A glasshouse experiment for solution-culture of one chlorophyll-deficient rice mutant and six wild types of rice genotypes was conducted, and the leaf reflectance (400–900 nm) was measured with a high spectral resolution (1 nm) spectrophotometer and the contents of chlorophyll a (Chla), chlorophyll b (Chlb) and chlorophyll a+b (Chlt) of the rice leaves were determined. It was found that the reflectance curves from 640 nm to 674 nm and from 675 nm to 680 nm of the low-chlorophyll mutant leaf were drastically steeper than that of the wild types in the ICCW. The new index based on the reflectance variation within ICCW, the difference of the first derivative sum within the ICCW (DFDS_ICCW), was highly sensitive (r = −0.77, n = 93, P<0.01) to Chlt while the mean reflectance (R_ICCW) in the ICCW became insensitive (r = −0.12, n = 93, P>0.05) to Chlt when the leaf Chlt was higher than 200 mg/m2. The best equations of R-ICCW and DFDS_ICCW yielded an RMSE of 78.7, 32.9 and 107.3 mg/m2, and an RMSE of 37.4, 16.0 and 45.3 mg/m−2, respectively, for predicting Chla, Chlb and Chlt. The new index could rank in the top 10 for prediction of Chla and Chlt as compared with the 55 existing indices. Additionally, most of the 55 existing Chl-related VIs performed robustly or strongly in simultaneous prediction of leaf Chla, Chlb and Chlt.


Introduction
Chlorophyll (Chl) a and Chl b are major constituents of the photosynthetic apparatus in higher plants. Chl a and Chl b are interconverted in the chlorophyll cycle [1]. Leaf Chl a concentration (Chla) and Chl b concentration (Chlb) indicate a plant's photosynthetic capacity and health status, and determination of Chla, Chlb and ratios of Chla to Chlb are also helpful for understanding the light acclimation mechanisms in higher plants [2]. Conventionally, leaf Chla and Chlb are determined with a traditional wet extraction analysis based on measuring the extinction of the extract at the major red absorption maxima of Chl a (,664 nm) and b (,647 nm) in the in-chlorophyll centre waveband (640-680 nm), and by inserting these values into simultaneous equations [2,3]. In recent decades, there has been an increasing interest in non-destructively determining leaf and canopy Chl content by measuring leaf and canopy spectral reflectance. Particular efforts have been devoted to the development of robust algorithms for Chlt determination from the leaf to canopy scale [4][5][6][7][8][9][10]. Contrastingly, studies conducted for determination of individual Chla or individual Chlb with spectral vegetation indices (VIs) are much less frequent [4,6,11]. Reflec-tance in the ICCW had been used for a long time as an indicator of chlorophyll content of leaves, but has not been used as an optimal index since Thomas and Gausman (1977) [12] found that reflectance near 675 nm became saturated at medium to high chlorophyll concentrations [6]. In recent decades, many studies have found that reflectance in the green and red-edge spectral regions was optimal for non-destructive estimation of leaf Chl content in a wide range of its variation [13][14][15][16]. The results of Féret et al. (2011) [17] showed that the reflectance in the red-edge and near infrared spectral regions simulated with the Prospect 5 radiative transfer model provided an accurate estimation of leaf Chl content. Recently, Main et al. (2011) [11] assessed the performance of 73 published VIs for leaf Chl estimation and also found that the indices using off-chlorophyll absorption centre wavebands (OCCW) performed better than those using ICCW. To our best knowledge, no VIs based solely on ICCW for Chl estimation have been developed since Thomas and Gausman (1977) [12] found the saturated reflection of plant leaves. Plant leaves have a reflectance minima around 675 nm, and there are substantial differences in reflectance among different wavelengths in the ICCW. Is the reflectance difference within the ICCW

Plant materials and growth conditions
A pot experiment was conducted in a greenhouse with natural light (mean daily photosynthetically active radiation 130 mmol m 22 s 21 during the whole growth period) and controlled temperature (daily maximum 27.6uC, daily minimum 16.2uC during the rice growing period) and humidity (24.5-85.1% average daily relative humidity, RH, throughout the whole rice growing period) at Zhejiang University Experimental Farm, Hangzhou, China (30u149 N, 120u109 E). Six wild types of rice genotypes (IG1, IG23, IG24, DJ, NIP and ZH11) and one chlorophyll-deficient mutant (IG20) were solution-cultured according to the IRRI prescription [18], but the nitrogen level was designed as 1/5640 mg l 21 (low N) and 40 mg l 21 (normal N), respectively, for two nitrogen treatments. The mutant 'IG 20' is an isogenic line of the recurrent parent ''Zhefu 802'' bred by China National Rice Research Institute. A completely random design with four replications was used. Each pot contained a 6.0-L nutrient solution and three seedlings. The nutrient solution was

Chlorophyll meter and spectral measurements
The second uppermost leaves of each treatment were measured in situ with a SPAD 502 model chlorophyll meter (Konica Minota Inc., Japan) around the midpoint at tillering, booting and heading. After the measurement of the chlorophyll meter, the leaves were immediately sampled and stored in an ice box, and transported to the lab for leaf reflectance measurements. The reflectance of the single leaf was measured with an integrating sphere (model LISR-3100, Shimadzu Scientific Instruments Inc., Japan) coupled to a UV-3600 UV-VIS-NIR spectrophotometer (Shimadzu Scientific Instruments Inc., Japan) in the wavelength range of 400-900 nm around the midpoint of each leaf. The spectral meter has a 1-nm resolution in the region of 400-900 nm.

Determination of leaf Chl contents
After spectral measurements, 15 leaf discs of 0.5 cm 2 from each leaf were sampled for determination of leaf Chl content. The Chl a and Chl b contents per unit area were measured spectrophotometrically using a solution of alcohol, acetone and water (4.5:4.5:1, V/V/V) as a solvent, employing the equations of Lichtenhaler and Wellburn (1983) [19]. The total Chl content was calculated as Chla plus Chlb. The leaves that appeared evidently desiccative were not used in this study. We measured a total of 108 leaves across tillering, booting and heading stages, which included 12 leaves of the mutant and 96 leaves of the wild types.

Data analysis
The scatterplots of the reflectance and the first derivative (FD) reflectance vs Chla, Chlb and Chlt were plotted, and the curves were visually analysed for extraction of spectral signatures of interest including shape, peak position, trough position and inflection point. FD was calculated with the following equation: where FD(l), R(l) and R(l+1) represent the first derivative reflectance at wavelength l (nm), reflectance at l and reflectance at l+1, respectively. The existing published Chl-related VIs selected in this study and their formulations were summarized in Table 1 [4][5][6][7]. Only leaf-scale indices were collected. Among the 55 indices, none were solely based on the ICCW, although 21 indices used the ICCW.
The sensitivity of the VIs to Chl contents were tested with the correlation coefficients between the VIs and the Chl content, and the correlation coefficients were computed with Excel 10.0 (Microsoft).
The relationship between the VIs and the Chl content (Chla or Chlb or Chlt) were fitted with linear, power, exponential, logarithmic and polynomial equations and the equation with the highest determination coefficients (R 2 ) was selected as the best equation. The root mean square error (RMSE) was computed for each best equation, and the predictive performance of the VIs was assessed by ranking the RMSE values in ascending order. The relationships were fitted with Excel.

Rice leaf Chl content
All the leaves of both the normal N treatment and the low N treatment of the mutant 'IG 20' were yellow-green in color during the whole growth period. The leaves of the wild types were green in colour, although the low N treatments were shallower in leaf colour than the normal N treatments. The means and ranges of Chl content (mg/m 2 ) for the 96 leaf samples of the conventional genotypes as well as Chla/Chlb were 260. 5

Leaf spectral reflectance signatures and construction of the new VI
As shown in Figure 1A, a profound difference in leaf spectral reflectance was observed between the conventional rice genotypes and the mutant. The reflectance curves from 640 nm to 674 nm and from 675 nm to 680 nm of the mutant leaf of a low Chl content were drastically steeper than those of the wild types in the ICCW. For both the wild types and the mutant, the inflection point of the reflectance spectra in the ICCW was 645 nm, where the FD value of reflectance started to be positive ( Figure 1B). Additionally, the reflectance trough around 620 nm became evident, and the green peak around 550 nm was broadened and deformed in the reflectance spectra of the mutant as compared with that of the wild types. The reflectance spectra of all the leaves of the mutant were visually similar in shape and reflection band positions.
Based on the spectral signatures in the ICCW we observed, we found that the reflectance variation within the ICCW was sensitive to the Chl content, and constructed a new VI-the difference of where the sum of FD 675-680 and the sum of FD 640-674 represent the sum of the first derivative reflectance between R675 and R680 and that between R640 and R674, respectively. R640, R674, R675 and R680 are the reflectance at 640 nm, 674 nm, 675 nm and 680 nm, respectively.

Sensitivity of the VIs to Chla, Chlb and Chlt
Of the 55 VIs tested (Table 2), 24 were robustly sensitive to the leaf Chlt (r2 §0.81, n = 108), 19 were strong (0.49!r2,0.81, n = 108), 5 were moderate (0.25!R2,0.49) and 5 were weak (0.04!R2,0.25). Only 2 indices, SR4 A and SR5 A , were insignificantly (P.0.05, n = 108) related to the leaf Chla, Chlb and Chlt. Generally, the sensitivity of the indices to Chlt was similar to that of Chla, and the sensitivity of the indices to Chlb was slightly lower than Chlt or Chla. The results showed that most of the tested indices were highly sensitive to Chla, Chlb and Chlt.
The mean reflectance in the ICCW (R-ICCW) was significantly (P,0.05) related to Chla, Chlb and Chlt with a low correlation strength, yielding an r (n = 108) of 20.45, 20.40 and 20.45, respectively. In contrast, the new VI, DFDS_ICCW, had an r (n = 108) of 20.86, 20.76 and 20.85 as correlated with Chla, Chlb and Chlt, respectively, indicating that this index was highly sensitive to Chlt, Chla and Chlb. When leaf Chlt was higher than 200 mg/m 2 , the r value was 20.77 (n = 93, P,0.01) and 20.12 (n = 93, P.0.05) respectively between DFDS-ICCW and Chlt and between R_ICCW and Chlt. The results demonstrated that DFDS-ICCW was still highly sensitive, but R_ICCW became insensitive to Chlt when Chlt was at medium and high levels. As shown in Figure 2C, the R-ICCW tended to be saturated when leaf Chlt.200 mg/m 2 . Contrastingly, DFDS_ICCW decreased sensitively with the Chlt even when Chlt was higher than 200 mg/ m 2 ( Figure 3C). This result confirmed the saturated reflection of the leaves at medium to high Chl content.

Prediction of Chla, Chlb and Chlt with the best-fit equations
The best equations of R-ICCW ( Figure 2 Chlt. The results indicated that DFDS_ICCW had a drastically higher prediction accuracy for Chla, Chlb and Chlt than R_ICCW. The prediction accuracy of DFDS_ICCW was slightly lower for Chlb than Chla or Chlt. The prediction performance with a best prediction equation for all of the 55 existing indices are presented in Table 2. Interestingly, none of the best equations were linear; they were exponential, polynomial and power. The RMSE (mg/m 2 ) ranged from 33.8 to 85.8 for Chla, from 12.2 to 37.9 for Chlb and from 38.6 to 120. 3 for Chlt, which demonstrated that there was a large difference of prediction accuracy between the best index and the last index. However, the RMSE (mg/m 2 ) of the top 30 indices ranged from 33.8 to 49.6 for Chla, from 12.2 to 17.1 for Chlb and from 38.6 to 60.3 for Chlt, indicating that the differences in the RMSE were not large in the top 30 indices. An index of high predictive ability for Chlt (e.g. Green Model) generally also performed well for prediction of Chla or Chlb, although the prediction accuracy for Chlb was generally and slightly lower than that for Chla or Chlt, and an index of low predictive ability for Chlt (e.g. PSRI) was also weak for prediction of Chla or Chlb. The SPAD reading ranked 13 th , 35 th and 16 th among the 55 indices, respectively for prediction of Chla, Chlb and Chlt with the polynomial equations, which indicated that it was also a strong index for predicting the leaf Chl contents.
The prediction results of the best VI, Green Model, together with the SPAD reading are also presented in Figure 4 and Figure 5, which confirm their high accuracy for prediction of Chla, Chlb and Chlt.
The results in this study demonstrated that most of the existing indices could be used for simultaneous retrieval of Chla, Chlb and Chlt.
As compared with the 55 indices, the prediction accuracy of DFDS_ICCW was similar to Datt2 A , ranking 7 th for Chla prediction, similar to SR6 A ranking 28 th for Chlb prediction and similar to Carter4 A ranking 7 th for Chlt prediction. The results indicated that DFDS_ICCW could simultaneously and robustly predict Chla, Chlb and Chlt.

Discussion
Most of the existing VIs as well as the SPAD reading were simultaneously and robustly or strongly related to Chla, Chlb and Chlt, and achieved a high accuracy for Chla, Chlb and Chlt prediction. As most of the indices were originally sought for prediction of Chlt, the results in this study suggested that the indices could be extended for simultaneous retrieval of Chla, Chlb and Chlt. None of the best-fit equations for prediction of Chla, Chlb and Chlt were linear equations; therefore, the ranking of the existing indices in this study was not in agreement with that of Main et al. (2011) [11], who used a linear equation for all indices. The VIs based on red edge (e.g. REP_LE A and REP_Li A ) ranked high for leaf Chl prediction in the previous study, but ranked low in this study. The indices excluding the ICCW generally performed better than those including the ICCW in this study, which is consistent with the previous study [11]. Particularly, both the best index for Chla and Chlt, SR3 A , and the best index for Chlb, Vogelmann2 A , did not use ICCW. The simple ratio indices-SR4 A based on 670 nm in the ICCW and 700 nm and S5 A based on 675 nm in the ICCW and 700 nm-were the only indices that were insignificantly (P.0.05) related to the Chl contents. In contrast, another simple ratio index, SR3 A based on 550 nm and 750 nm in the OCCW, was the best index for prediction of Chla and Chlt. In the ICCW, the reflectance curves from 640 nm to 674 nm and from 675 nm to 680 nm of the mutant leaf of a low Chl content were drastically steeper than that of the wild types of medium to high Chl content. This spectral signature could enlighten us to use the reflectance variation within the ICCW for retrieval of plant Chl content, although further studies are needed for understanding the mechanisms causing this signature. The successful detection of the reflectance variation within the ICCW in this study could be attributed to the high spectral resolution (1 nm) of the spectral photometer, as the current widely-used spectral meter, the Field Spec spectroradiometer (Analytical Spectral Devices, Boulder, CO, USA), has a spectral resolution of 3 nm in the red band.
Plant leaves tend to have saturated reflectance in the ICCW [6,12] when leaf Chlt is medium to high, which has limited the use of this spectral region for non-destructive determination of leaf Chl. The results in this study also showed that the R-ICCW tended to be saturated when leaf Chlt was higher than 200 mg/ m 2 . However, the new spectral index based on the reflectance variation within the ICCW decreased sensitively with the Chlt even when Chlt was greater than 200 mg/m 2 . The new index could rank in the top 10 for prediction of Chla and Chlt as compared with the 55 tested indices, and also achieved a promising accuracy for Chlb prediction. Therefore, the results suggested that ICCW could also be used for development of robust VIs for retrieval of plant Chl contents. Unlike the existing VIs, the new index is solely based on the specific Chl adsorption band. Therefore, the retrieval of Chl by using this index may not be confounded by non-Chl factors, e.g. other pigments and leaf structure. Further studies are needed for confirmation of this finding at different scales (e.g. canopy and region) and for different plant species.

Conclusions
Most of the 55 existing VIs could robustly or strongly and simultaneously predict Chla, Chlb and Chlt in the rice leaves of a large variation of ratios of Chla to Chlb in this study. It was found that the reflectance curves from 640 nm to 674 nm and from 675 nm to 680 nm of the mutant leaf were drastically steeper than those of the wild types in the ICCW, which implied that the reflectance variation within ICCW could be used for retrieval of Chl content. The new index based solely on the reflectance variation within the ICCW were simultaneously and strongly sensitive to Chla, Chlb and Chlt and achieved a high accuracy for prediction of Chla, Chlb and Chlt. The results suggested that ICCW could also be of potential for development of robust VIs for retrieval of plant Chl content with non-destructive reflectance measurement approaches.