Early detection of canopy nitrogen deficiency in winter wheat (Triticum aestivum L.) based on hyperspectral measurement of canopy chlorophyll status

Abstract

A spectroscopic method was developed to measure the nitrogen status of winter wheat (Triticum aestivum L.) canopies. Two years of field experiments, including a range of cultivars grown with differing levels of nitrogen fertilization, were conducted and ground-based hyperspectral data were collected to develop and validate an empirical model for early detection of low canopy chlorophyll content. Canopy reflectance was measured with a spectrometer, fitted with a 25° field of view fibre-optic adaptor. Canopy chlorophyll density (CCD), representing the total amount of chlorophyll present in the canopy per unit ground area, was combined according to the contribution of winter wheat leaves in different layers of the canopy and related to canopy reflectance. Combined canopy chlorophyll density (CCCD) calculated with both layers 1 and 2 and with layers 1, 2 and 3 were better related to difference vegetation index (DVI=R_NIR−R_RED, where R_NIR and R_RED were reflectance at 890 nm and 670 nm, respectively) than CCD in any individual layer. Statistical prediction models of canopy chlorophyll status in winter wheat were developed. The CCCD₁₊₂ model demonstrated lower root mean square errors and higher modelling efficiencies than those of the CCCD₁ and CCCD₁₊₂₊₃ models. Chlorophyll status in the two uppermost layers of the wheat canopy could be quantified using DVI. Therefore, early detection of canopy nitrogen deficiency in winter wheat was achieved.

Keywords:

• canopy nitrogen deficiency
• early detection
• vegetation index
• winter wheat
• remote sensing

Introduction

Chlorophyll largely determines the amount of photosynthetically active radiation (PAR) absorbed by the leaf, photosynthetic rate and primary production. High chlorophyll content of plant communities is related to a large leaf area index (LAI), increasing radiant energy absorption, and total dry matter production (Anderson 1967). A precise estimation of plant chlorophyll status is important for nitrogen (N) fertilization recommendations because much of leaf N is incorporated in chlorophyll (Filella et al. 1995). The determination of chlorophyll status with traditional laboratory methods can be labour intensive, time consuming and expensive, especially when a large number of samples are involved.

Rapid and non-destructive chlorophyll status detection can be made using a portable leaf chlorophyll meter (e.g. SPAD-502, Soil Plant Analysis Development, Minolta Camera Co Ltd, Japan) through measurements of leaf chlorophyll based on the spectral transmittance properties of leaves in wheat (Reeves et al. 1993), sweet pepper (Madeira et al. 2003), peace lily (Wang et al. 2004), butterhead lettuce (León et al. 2007) and grapevine (Steele et al. 2008). Previous research has shown that canopy reflectance at wavebands around 550 nm (green region) and 670 nm (red region) was strongly related to leaf chlorophyll concentration (Thomas & Gausman 1977). Schlemmer et al. (2005) reported that reflectance of corn leaves in the 630 nm region increased (the leaf turned from green to a more yellowish green colour) as chlorophyll content decreased. Gao et al. (2008) reported that the optimal band for estimation of chlorophyll content was in the 698–715 nm range. Vegetation indices involving green and red wavebands, such as Normalized Pigment Chlorophyll-a Index (=(R₆₈₀−R₄₃₀)/(R₆₈₀+R₄₃₀)) (Peñuelas et al. 1993), Green Normalized Difference Vegetation Index (=(R₇₅₀-R₅₅₀)/(R₇₅₀+R₅₅₀)) (Gitelson et al. 1996) and R₇₅₀/R₅₅₀ (Gitelson & Merzlyak 1996), have been developed and used to estimate chlorophyll status on leaf and canopy levels of various plant species. Recently, Xue & Yang (2009) stated that the normalized derivative difference ratio [=(D₇₂₂–D₇₀₀)/(D₇₂₂+D₇₀₀), (Le Maire et al. 2004)] gave a good estimation of chlorophyll content in vegetable leaves.

In general, estimation of chlorophyll status has been based on the relationships between chlorophyll content of leaves in the upper layer or whole canopy and spectral parameters (Serrano et al. 2000; Wright et al. 2004; Vuolo et al. 2010), whereas contributions of chlorophyll and/or spectral reflectance from specific vertical layers in the canopy has not been considered. As the most mobile element of the plant and a very important constituent of chlorophyll, N content declines with leaf aging (Monks & Efford 2006) and is always transferred from aged leaves to new ones (Charles-Edwards et al. 1987). Yamasaki et al. (1996) reported that the content of chlorophyll per leaf area was highest in the uppermost young leaves and decreased with decreasing height level, indicating that there was a vertical gradient of chlorophyll abundance per leaf area through the plant canopy. Poorter et al. (1995) stated that chlorophyll content per unit leaf weight changed with canopy layer. Zhou et al. (2010) studied the relationship between the vegetation index, as a modified simple ratio (mSR705), and the chlorophyll content of the uppermost leaf and the third uppermost leaf of rice plants and concluded that use of the difference of the modified simple ratio between layer 1 and layer 3, i.e. mSR705L1–mSR705L3, greatly reduced the influence of the growth stage and genotype in assessing the N status using reflectance data. If leaf chlorophyll status in the middle and bottom canopy layers could be detected, corresponding N treatments may be applied in a timely manner due to the strong relationship between chlorophyll and N.

The objectives of this study were, first, to develop a spectral reflectance model for early detection of declining chlorophyll content in winter wheat canopies and, second, to validate the precision and accuracy of the resulting model.

Materials and methods

Experimental design

Modelling experiment

The modelling experiment was conducted during the 2002–03 growing season at the China National Experimental Station for Precision Agriculture (40°11′N, 116°27′′E), located in Changping district, Beijing, China. The soil was a silt clay loam. Pre-sowing soil (0–0.2 m) tests indicated organic matter at 14.2–14.8 g kg⁻¹, total N at 0.8–1.0 g kg⁻¹, available phosphorus (P) at 20.1–55.4 mg kg⁻¹ and available potassium (K) at 117.6–129.1 mg kg⁻¹. Soil was tested and fertilized to ensure no nutrient limitation.

The investigated wheat cultivars included eight erectophile plant types—4P3, 9128, 95128, 98-100, CA16, Chaoyou66 (CY66 for short), Jing411 (J411) and Lunxuan201 (LX201)—and seven planophile plant types—9428, 9507, Lunai17 (LA17), Linkang2 (LK2), Linkang502(LK502), Nongda3214 (ND3214) and Zhongmai9 (ZM9). All the cultivars are popularly planted in northern China. The model developed with these cultivars is anticipated to be suitable for early wheat N deficiency detection in northern China. A randomized complete block design with three replicates of each cultivar was employed. Each plot (3×5 m) received 180 kg ha⁻¹ urea, 225 kg ha⁻¹ (NH₄)₂HPO₄ and 150 kg ha⁻¹ K₂SO₄ before sowing. Topdressing N with an amount of 280 kg ha⁻¹ urea was applied with two splits, 50% at Feekes 5.0 and 50% at Feekes 7.0. Sowing was on 21 October 2003 at a rate of 225 seeds m⁻². The between-row space was 0.23 m. At elongation, the average plant density was 476.6 stems m⁻² for the erectphile type and 434.4 stems m⁻² for the planophile type, and the corresponding figures at anthesis were 409.8 stems m⁻² and 381.9 stems m⁻², respectively, because weaker stems died after elongation.

Validation experiment

The validation experiment was carried out at a site near the modelling experiment during the 2001–02 growing season at the same research station and on the same soil type as the modelling experiment. The nutrient contents of soil from 0–0.2 m depth were: organic matter 12.1–13.2 g kg⁻¹, total N 0.9–1.2 g kg⁻¹, available P 25.2–48.3 mg kg⁻¹ and available K 96.6–128.8 mg kg⁻¹.

An erectophile plant type, Jingdong8, and a planophile plant type, 9507, were investigated in this experiment. A randomized complete block design with four N fertilizer treatments (N0, N1, N2 and N3) and three replicates was used. Each treatment received 300 kg ha⁻¹ (NH₄)₂HPO₄ before sowing. No additional N was applied to N0. N1, N2 and N3 were supplied with an additional 150, 300 and 450 kg N ha⁻¹, respectively. The N source was urea, which was applied with three splits: at pre-sowing (50% of the total amount), Feekes 5.0 (25% of the total amount) and Feekes 7.0 (25% of the total amount). Sowing was on 16 October 2002 at a rate of 225 seeds m⁻². The between-row space was 0.23 m. At elongation, the average plant density was 554.9 stems m⁻² for Jingdong8 and 596.1 stems m⁻² for 9507, and the corresponding figures at anthesis were 426.3 stems m⁻² and 453.4 stems m⁻², respectively, with more weaker stems dying after elongation compared with the modelling experiment.

Field and laboratory measurements

In each plot, an area (1 m²) of the wheat canopy was selected for canopy spectral reflectance measurements, physiological and biochemical analyses. Measurements were performed at elongation and anthesis in both modelling and validation experiments.

Wheat canopy radiance measurements were taken from a height of 1.3 m above ground (the height of the wheat was 0.9±0.05 m at maturity), under clear sky conditions between 10:00 and 14:00, using an ASD FieldSpec Pro spectrometer (Analytical Spectral Devices, Boulder, CO, US) fitted with 25^o field of view (FOV) fibre optics, operating in the 350–2500 nm spectral region with a sampling interval of 1.4 nm between 350 and 1050 nm, and 2 nm between 1050 and 2500 nm, and with spectral resolution of 3 nm at 700 nm, 10 nm at 1400 nm. The 1.3 m measurement height coupled with 25^o FOV equated to 0.26 m² area on the ground, which gives the appropriate ground sample area with less soil background influence under our experimental design (the between-row space was 0.23 m). A 0.4×0.4 m BaSO₄ calibration panel was used as a reference (Herrmann et al. 2010). Twenty scans were recorded for each plot, and the canopy radiance measurements were taken by averaging these at an interval of 1 sec, with a dark current correction with every spectral measurement. Reference panel radiance measurements were taken before and after the canopy radiance measurement. The wheat canopy radiance was calculated using the equation below:

1

where ρ is the spectral reflectance, DN _Target is the mean (of 20 scans) raw digital number measured over the wheat canopy, DN _Panel is the mean raw digital number measured over the reference panel, F _Panel is the calibration factor of the reference panel (provided by the manufacturer), and λ is the wavelength.

Following canopy spectral measurements, the PAR at the top of every layer and at the bottom of the whole canopy was measured with a line quantum sensor (LI-191SA, LI-Cor Inc, Lincoln, NE, US), which measures PAR under spatially non-uniform canopies. Plants in the 1 m² area were divided into three to four layers of equal vertical thickness (three in the modelling experiment, four in the validation experiment, Fig. 1). The height was standardized as the actual height of the plant at elongation, and the average height from the top of the flag leaf to the ground at anthesis. The line quantum sensor was held with both ends between two rows and parallel with the row direction.

Figure 1  Sketch map of canopy layers at elongation (left) and anthesis (right) stages. At anthesis, the unbroken line is for the modelling experiment, and the broken line for the validation experiment.

Leaf area index (LAI) and chlorophyll analysis were performed at elongation and anthesis for both modelling and validation experiments. Whole-plant samples were taken immediately after the spectral and PAR measurements, placed in a cooled black plastic bag and transported to the laboratory for subsequent analysis. The plants in the 1 m² area were cut into three or four layers of equal vertical thickness according to the standard used for PAR determination. Leaves in each layer were collected together for LAI, leaf mass per unit area (LMA, g m⁻²), and chlorophyll content (mg g⁻¹ fresh leaf weight) measurements. The LAI was determined by a dry weight method (Zhu et al. 1998). Approximately 60 cm² reference leaf segment was cut from the central part of 20–30 leaves in each individual layer. The reference leaf segments and the rest of the sample leaves were oven-dried at 70 ^oC to constant weight and weighed, respectively. Leaf area index was calculated as:

2

where S _r (m²) is the area of the reference leaf segments, W _t (g) is the total dry weight of the leaves in the measured layer, S _l is the ground area of the sample (m²), and the W _r (g) is the dry weight of the reference leaf segments. Before determining LAI by dry weight method, the area of all the leaves in each layer was measured using a CI-203 Portable Laser Leaf Area Meter (CID Inc, US). LAIs from both methods were averaged to assess the LAI of the specific layer. LMA was derived from the fresh weight and the area of the reference leaf segments.

Chlorophyll was extracted with 80% acetone immediately after the samples were transported to the laboratory. The absorption of the extracts at wavelengths of 645 nm and 663 nm were measured with a Helios spectrophotometer (Thermo Electron Company, Cambridge, UK). The chlorophyll content was calculated using the formula of Arnon (1949) as total chlorophyll (mg g⁻¹)=20.3×D₆₄₅+8.04×D₆₆₃, where D₆₄₅ and D₆₆₃ are chlorophyll absorbance at 645 and 663 nm, respectively.

Data analysis

Hinzman et al. (1986) reported that canopy chlorophyll density (CCD)—the total amount of chlorophyll present in the canopy per unit of ground area—was a sensitive indicator of N deficiencies in wheat. Referring to their concept, we calculated CCD (g m⁻²) as in equation (3) below:

3

Because leaves in layers 2 and 3 (or 2–4 for the validation experiment) were not fully exposed to sunlight (Woolley 1971), we calculated the combined canopy chlorophyll density (CCCD) from each individual layer using the attenuation degree (AD) of PAR in each layer as a weighting factor. A series of calculations were necessary, beginning with the calculation of actual radiation (RAD) received by a given layer as in equation (4):

4

where i represents the i^th layer of the canopy (Fig. 1). RAD_ib is the actual RAD at the bottom of the i^th layer, which is the same as RAD_(i+1)t, being the actual RAD at the top of layer i+1. PAR_ib and PAR_it are the observed PAR at the bottom (ib) and top (it) of the i^th layer, and τ is the transmittance coefficient for RAD penetration of a single leaf, which was measured by the ASD FieldSpec Pro 2500 spectrometer coupled with an integrating sphere (LI-Cor Inc, Lincoln, NE, US) on 20 leaves from each treatment. The AD of RAD of the i^th layer was calculated as in equation (5):

5

The combined weight (w) of RAD in the i^th layer was calculated as in equation (6):

6

and the CCCD_n, the weighted sum of individual layer CCD _i , as in equation (7):

7

where CCCD_n is the CCCD of three kinds of combined layers (n=1, 1+2, or 1+2+3).

Difference vegetation index (DVI) (Jordan 1969), calculated from the reflectance in red (670 nm) and NIR (890 nm) wavelength regions, was used to estimate CCD. Filella et al. (1995) and Serrano et al. (2000) pointed out that increases in red reflectance were related to the decreases in chlorophyll content resulting from lower N supply. Soudani et al. (2006) reported that reflectance in the NIR region correlated significantly with LAI.

The leaf chlorophyll content and CCD in each layer across all the 15 cultivars, and the averages of both leaf chlorophyll content and CCD in each layer across erectophile and planophile plant types in the modelling experiment were subjected to analysis of variance (ANOVA). The least significant difference test was performed using the GLM procedure in SAS 9.1 (SAS Institute 2003). The observed and estimated CCCD_n were subjected to linear regression analysis to evaluate the value of the model. We calculated both root mean squared error (RMSE), as in equation (8), and modelling efficiency (MEF) (Loague & Green 1991), as in equation (9), to quantify the degree of fit between observations and simulations.

8

9

where n is the number of the samples, E _i is the estimated CCCD_n, O _i is the observed CCCD_n, Ō and is the average of all the observed CCCD_n. The RMSE is defined as the standard deviation of the model residuals, which was the difference between model estimations and actual observations. The MEF measures how well the model predicts relative to the average of the observations. A value near 1 indicates a close match between model estimations and observations (Loague & Green 1991).

Results and discussion

Leaf chlorophyll content and canopy chlorophyll density in different layers in the modelling experiment

Leaf chlorophyll content decreased with canopy layer at elongation and anthesis in both erectophile and planophile plant types (Fig. 2), confirming the results of previous studies (Yamasaki et al. 1996; Wang et al. 2005). ANOVA showed that there was no significant difference in leaf chlorophyll content among cultivars in each layer (data not shown). Compared with canopy layer 1, layers 2 and 3 had 9–21% less chlorophyll at elongation. However, at anthesis, layers 2 and 3 had as much as 16–20% less chlorophyll. Differences among layers in both erectophile and planophile plant types were significant at both elongation (P<0.01) and anthesis (P<0.0001) (Fig. 2).

Figure 2  Leaf chlorophyll content (mg g⁻¹) in different canopy layers at elongation and anthesis of both erectophile and planophile plant types. Data for erectophile and planophile plant types are averaged from eight erectophile plant type cultivars and seven planophile plant type cultivars, respectively. Letters A, B and C following lines at elongation indicates significance at P < 0.01, and letters A, B and C following lines at anthesis indicates significance at P < 0.0001.

Table 1  Canopy chlorophyll density (CCD, g m⁻²) in different layers at elongation and anthesis.

		Erectophile plant type								Planophile plant type
		4P3	9128	95128	98-100	CA16	CY66	J411	LX201	9428	9507	LA17	LK2	LK502	ND3214	ZM9
	Layer 1	0.71±0.09	1.00 ±0.15	0.93 ±0.09	0.95 ±0.15	1.06±0.08	0.97±0.03	0.79±0.03	0.95±0.01	1.35±0.10	1.08±0.12	1.04±0.06	1.21±0.08	1.07±0.03	1.15± 0.06	1.13±0.04
Elongation	Layer 2	1.02±0.14	0.95±0.09	1.06±0.07	1.06±0.15	1.33±0.08	1.23±0.08	0.97±0.06	1.28±0.06	1.11±0.03	1.08±0.12	1.29±0.03	1.43±0.19	1.02±0.02	1.38± 0.12	1.05±0.14
	Layer 3	0.39±0.05	0.66±0.12	0.47±0.02	0.74±0.13	0.72±0.06	0.72±0.01	0.57±0.03	0.55±0.03	0.63±0.06	0.45±0.02	0.50±0.02	0.70±0.08	0.45±0.05	0.57± 0.11	0.47±0.07
	Layer 1	1.19±0.19	0.96±0.12	1.21±0.11	1.19±0.10	1.18±0.09	1.30±0.28	1.32±0.05	0.91±0.06	1.48±0.18	1.34±0.18	1.42±0.11	1.75±0.29	1.74±0.25	1.56± 0.25	1.65±0.29
Anthesis	Layer 2	1.35±0.04	1.03±0.12	1.32±0.11	1.34±0.17	1.47±0.09	1.62±0.23	1.33±0.02	1.13±0.16	1.64±0.08	1.49 ±0.20	1.47±0.16	1.79±0.33	1.76±0.05	1.86± 0.11	1.68±0.17
	Layer 3	0.88±0.06	0.89±0.05	0.77±0.06	0.80±0.03	0.88±0.07	0.95±0.11	0.78±0.07	0.80±0.06	0.80±0.17	0.80±0.13	0.78±0.20	0.91±0.14	0.72±0.10	0.73± 0.15	0.85±0.23
The number after±represents standard error.

Filella et al. (1995) noted that LAI values are related to the amount of leaf tissue in the field of view of the reflectance sensor. Stanhill et al. (1972) stated that biomass was the main factor in the reflectance response of N-deficient wheat canopies. Hinzman et al. (1986) suggested that chlorophyll density in the canopy was related to both leaf chlorophyll content and LAI, which correlated with spectral reflectance at red and NIR wavelength regions (Gitelson et al. 1996; Soudani et al. 2006), respectively. Therefore, canopy chlorophyll status in this study was expressed as an integrated measure, i.e. CCD, of leaf chlorophyll content, LAI and LMA. In our study, the CCD in layer 2 was generally greater than that in layer 1 at both elongation and anthesis stages, while the CCD value in layer 3 was the lowest among the three layers (Table 1 and Fig. 3). At elongation, the percentage CCD in layers 1 and 2 accounts for 77.0% of the CCD of the whole canopy for the erectophile plant type and 81.3% for the planophile plant type. At anthesis, the corresponding figures were 75.1% and 80.6% (Table 2). For the erectophile plant type, highly significant differences in CCD were observed among the three layers at elongation (P<0.0001), and between layers 1 and 2, on one hand, and layer 3, on the other hand, at anthesis (P<0.01). In the planophile plant type, there was no significant difference in CCD between layer 1 and layer 2, while CCD in layer 3 was significantly different from those in both layer 1 and layer 2 (P<0.0001 at elongation, and P<0.01 at anthesis) (Fig. 3). The results indicated that the planophile plant type had higher CCD in the upper two layers, and lower CCD in the bottom layer in comparison with the erectophile plant type. No significant difference in the CCD among all the 15 investigated cultivars was observed within each layer (data not shown).

Figure 3  Canopy chlorophyll density (CCD, g m⁻²) in different layers at elongation and anthesis of both erectophile plant type and planophile plant type. Data for erectophile and planophile plant types are averaged from eight erectophile plant type cultivars and seven planophile plant type cultivars, respectively. Letters A, B and C following lines at elongation indicates significance at P < 0.0001, and letters A and B following lines at anthesis indicates significance at P < 0.01. The same letters between layers show no significant difference.

Table 2  The percentage of CCD (canopy chlorophyll density, g m⁻²) in each layer of the whole canopy of erectophile and planophile plant types at elongation and anthesis.

	Erectophile plant type		Planophile plant type
	Elongation	Anthesis	Elongation	Anthesis
Layer 1	34.9%	35.4%	40.0%	38.1%
Layer 2	42.1%	39.7%	41.3%	42.5%
Layer 3	23.0%	24.9%	18.7%	19.4%

Model development

The coefficients of determination (R²) between CCD in layers 1, 2 and 3 and DVI were 0.59, 0.71 and 0.30 at elongation and 0.55, 0.68 and 0.52 at anthesis, respectively, in the modelling experiment (Table 3). Since each layer had a different contribution to the canopy reflectance, the degree of contribution of each individual layer in the canopy was calculated as a combined weight (w). The CCCD₁, CCCD₁₊₂ and CCCD₁₊₂₊₃, which included CCD from layer 1, layers 1 and 2, and all three layers, respectively, were calculated with the w value. The R² between CCCD_{1 + 2} and CCCD_{1 + 2+3} and DVI were 0.73 and 0.80 at elongation, and 0.70 and 0.74 at anthesis, respectively (Table 3). The relationship between chlorophyll status and DVI was strengthened by CCCD_{1 + 2} and CCCD_{1 + 2+3} when compared with layer 1 alone. Therefore, the information in layers 1 and 2 or layers 1, 2 and 3, instead of the information in layer 1 alone, should be taken into consideration when assessing canopy chlorophyll status. The layers involved should be combined according to their contribution to the canopy reflectance. The models developed with the regression method based on the relationship between DVI and CCCD_n are presented in Table 4.

Table 3  The coefficient of determination (R²) between CCD (canopy chlorophyll density, g m⁻²) and combined CCD (CCCD, g m⁻²) and DVI (difference vegetation index) in the modelling experiment.

	DVI
	Elongation	Anthesis
CCD in layer 1 (or CCCD1)	0.59**	0.55**
CCD in layer 2	0.71**	0.68**
CCD in layer 3	0.30**	0.52**
CCCD1 + 2	0.73**	0.70**
CCCD1 + 2+3	0.80**	0.74**
** indicates significant regression at P < 0.01.

Model validation

Data collected during the elongation phase of the validation experiment were used to validate the three developed models (Fig. 4). Regression residual plots showed that all three models were well supported (Fig. 4). The RMSE of estimated vs. observed CCD of models CCCD₁, CCCD_{1 + 2} and CCCD_{1 + 2+3} were 0.07, 0.06 and 0.13, respectively, and the MEF of CCCD₁, CCCD_{1+ 2} and CCCD_{1 + 2+3} were 0.54, 0.66 and −1.39, respectively. Model CCCD_{1 + 2} had relatively higher MEF and lower RMSE. However, the MEF of model CCCD_{1 + 2+3} was negative with an absolute value larger than 1, indicating that chlorophyll status was being overestimated (all the points produced by estimations and observations in Fig. 4C were above the y=x line). Overestimation using the CCCD₁₊₂₊₃ may have been due to an inflated value for w. From Fig. 4, the R² and slope of estimated vs. observed CCCD for the CCCD_{1 + 2} and CCCD_{1 + 2+3} models were higher than that of CCCD₁, and their intercepts were lower.

Figure 4  Comparison of: A, observed CCCD₁ and estimated CCCD₁; B, observed CCCD_{1 + 2} and estimated CCCD_{1 + 2}; and C, observed CCCD_{1 + 2+3} and estimated CCCD_{1 + 2+3} at elongation of the validation experiment. CCCD is combined canopy chlorophyll density. F_0.0(1,22)=0.515.

Unfortunately, the models developed based on the data at anthesis could not be validated. Data collected at anthesis were not used for model validation because layers at anthesis in the validation experiment were not equal to those in the modelling experiment. Because radiation in the NIR region had strong transmittance (Woolley 1971) and CCCD_{1 + 2+3}contained the chlorophyll information in the whole canopy, the CCCD_{1 + 2+3} model was supposed to yield higher accuracy and precision. However, this hypothesis was not supported in this study. This could also be attributed to the w value used in the CCCD_{1 + 2+3} model. An alternative method to define the contribution of layer 3 to whole canopy reflectance should be developed in a future study.

Table 4  Models established based on the relationship between DVI and CCCD_n. DVI is difference vegetation index; CCCD (g m⁻²) is combined canopy chlorophyll density. The n in CCCD is 1, 1 + 2 and 1 + 2+3, respectively. Data are from the modelling experiment, and data from both planophile and erectophile are used together to develop the model. Chl, chlorophyll; N, nitrogen. The units of leaf N content, leaf chlorophyll concentration are mg g⁻¹ dry leaf weight and mg g⁻¹ fresh leaf weight, respectively.

Elongation		Anthesis
Equations	R^2	Equations	R^2
CCCD1=0.031DVI+0.051(58.3≥Ntop≥41.6, 3.49≥Chltop≥2.12, 43.4≥DVI≥24.0, and 1.53≥CCCD1≥0.59)	0.59**	CCCD1=0.055DVI-0.449(55.3≥Ntop≥46.7, 5.27≥Chltop≥3.04, 45.3≥DVI≥23.5, and 2.23 ≥ CCCD1≥0.72)	0.55**
CCCD1 + 2=0.038DVI-0.071(55.3≥Nmid≥36.2, 3.29≥Chlmid≥1.21, 43.4≥DVI≥24.0, and 1.50≥CCCD1 + 2≥0.72)	0.73**	CCCD1 + 2=0.064DVI-0.636(52.7≥Nmid≥43.2, 4.49≥Chlmid≥2.18, 45.3≥DVI≥23.5, and 2.32≥CCCD1 + 2≥0.79)	0.70**
CCCD1 + 2+3=0.036DVI-0.117(47.4≥Nlow≥32.5, 2.78≥Chllow≥0.89, 43.4≥DVI≥24.0, and 1.41≥CCCD1 + 2+3 ≥ 0.64)	0.80**	CCCD1 + 2+3=0.062DVI-0.710(43.2≥Nlow≥34.2, 3.99≥Chllow≥1.29, 45.3≥DVI≥23.5, and 2.28 ≥ CCCD1 + 2+3≥ 0.80)	0.74**
** indicates significant regression at P < 0.01.

Conclusion

Remotely-sensed chlorophyll status offers the possibility of rapidly estimating crop N status for improved N management strategies. A method based on the relationship between canopy vertical chlorophyll distribution and the characteristics of canopy reflectance was developed to detect chlorophyll status in lower layers in winter wheat canopy. Both CCCD_{1 + 2} (CCD combined with both layers 1 and 2) and CCCD_{1 + 2+3} (CCD combined with layers 1, 2 and 3) were better related to DVI than CCD in any individual layer. A model based on CCCD_{1 + 2} explained the majority of the variability in chlorophyll status (R²=0.73 and 0.70 at elongation and anthesis, respectively) and showed lowest RMSE and highest MEF when predicting winter wheat canopy chlorophyll status. Therefore, chlorophyll status in the two uppermost layers of the wheat canopy could be quantified using the DVI, and early detection of chlorophyll deficiency was achieved.

Acknowledgements

This work was supported by National Natural Science Foundation of China (40571118, 40701119, 40701120) and the National High Tech R&D Program of China (2006AA10A302, 2006AA12Z138). Thanks to Dr Zhihong Ma, Ms Hong Chang, Mr Beihong Wang, Mr Weiguo Li and Mr Yanbo Wang for their contributions in sample analysis and hyperspectral data obtaining, and to Dr Akira Hirano (JIRCAS, Japan International Research Center for Agricultural Sciences) for his suggestions in writing.