Published online 30 July 2007
Published in Crop Sci 47:1647-1651 (2007)
© 2007 Crop Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
FORAGE & GRAZINGLANDS
Estimation of Forage Production of Nilegrass Using Vegetation Reflectance
Chwen-Ming Yanga,*,
Yuh-Jyuan Leea,
Kuo-Yuan Hongb and
Fu-Hsing Hsub
a Crop Science Division, Taiwan Agricultural Research Institute, Wufeng, Taichung Hsien 41301, Taiwan ROC
b Forage Crops Division, Taiwan Livestock Research Institute, Hsinhua, Tainan Hsien 71246, Taiwan ROC. This work was supported by the research grants (92AS-8.1.1-CI-C1, 93AS-8.1.1-CI- C1, and 94AS-8.2.1.-CI-C1) from Council of Agriculture, Executive Yuan, Taiwan ROC
* Corresponding author (cmyang{at}wufeng.tari.gov.tw).
 |
ABSTRACT
|
|---|
Changes in reflectance spectrum of a crop are known to follow the morphological development of vegetation, and thus spectral models combining spectral characteristics correlated with biomass production may be used for yield estimation. Field experiments were conducted to validate use of reflectance spectra (350–2400 nm) to estimate forage production (i.e., aboveground fresh weight) of nilegrass (Acroceras macrum Stapf) vegetation from June 2002 to May 2004. Correlation coefficients (r) between spectral reflectance and forage production varied across the spectral range of measurements. A linear relationship (P < 0.010) was found for several wavebands, with the highest r value located at 891 nm (r = 0.671; P < 0.010). Of the examined spectral indices, forage production was found to be best correlated with RGREEN/RNIR ratio (R2 = 0.654, P < 0.001) where RGREEN was reflectance of green light (490–560 nm) maximum and RNIR was reflectance of the near-infrared (740–1300 nm) peak. Assessment of forage production was further improved by using a multiple linear regression (MLR) model. The best five-variable linear regression equation provided the best fit (R2 = 0.726, P < 0.001, Mallows' Cp criterion = 6.000). When validating the MLR model with other datasets from different growing seasons, the model gave reasonable prediction values (r = 0.833; P < 0.001) with a slope of 1.086 and root mean square error of 3.891 (N = 21).
Abbreviations: LAI, leaf area index • MLR, multiple linear regression • NDVI, normalized difference vegetation index • SC, spectral characteristic
|
INTRODUCTION
|
|---|
A NUMBER OF FORAGE grasses have been selected and commercially released from the Taiwan Livestock Research Institute to meet the local demand for livestock forage supply for animal feeds (Buu et al., 1993; Shaug et al., 1999). Nilegrass is a C3 perennial grass that, because it is well adapted to the soil and climate in the region, is increasingly being used as a forage crop in Taiwan (Rhind and Goodenough, 1976, 1979; Lippke, 1980; Rout et al., 1990). Under normal growing conditions, sward canopy height may reach 0.9 to 1.2 m at the time of harvest, and dry matter production can be 25 to 30 Mg ha–1 yr–1 (Shaug et al., 2002). By using a suitable means of monitoring plant growth and understanding the optimal timing of harvesting, there is potential for this grass to further expand its cultivated area and production.
Spectroradiometry is a commonly used remote sensing technique for a wide range of industrial applications. In agriculture, the technique has been adopted to study changes in morphology, coloration, internal tissue structure, and biochemistry in relation to the optical nature of the canopy so as to establish algorithms for evaluating growth status of a plant population (Gausman et al., 1969; Gausman, 1982; Masoni et al., 1996; Blackburn, 1998, 1999; Su and Yang, 1999; Yang, 2001; Fassio and Cozzolino, 2003; Yang and Chen, 2004). Moreover, the temporal and spatial variation of the desired vegetation characters within a field can also be assessed and monitored under normal conditions or in response to stresses (Kanemasu et al., 1985; Clarke, 1997; Yang and Su, 2000; Yang and Cheng, 2001). Furthermore, this technology has potential use in large-scale evaluation of grasslands and in developing precision agriculture technologies for forage-based livestock production.
Knowledge about vegetation or canopy reflectance in the visible, near-infrared, and thermal infrared wavebands has greatly improved in the past decades. Applications of spectral data to varied aspects of agriculture have been extensively explored, especially in the areas of crop growth modeling and yield estimation and prediction. Spectral models or algorithms which incorporate multiple wavebands into a single value assessment are the most commonly used (Heilman et al., 1977; Bouman, 1995; Anderson et al., 2000; Thenkabail et al., 2000; Diker and Bausch, 2003; Yang and Chen, 2004; Chen and Yang, 2005). Generally the spectral parameters related to growth or production processes were identified to constitute the proper functional models empirically or mechanistically.
The objectives of the present research were to take near-ground high-resolution reflectance spectra of nilegrass vegetation in conjunction with aboveground growth measurements to determine suitable spectral characteristics, and to use varied numbers of characteristics to estimate forage production.
|
MATERIALS AND METHODS
|
|---|
Field evaluation was conducted in the experimental forage farm of the Taiwan Livestock Research Institute (23°04' N, 120°26' E, elevation of 31 m) at Hsinhua headquarters for eight growing periods from June 2002 to May 2004 (Table 1). Datasets acquired from Periods 1, 2, 3, 4, 5, and 7 were selected to develop a multiple regression model, and datasets from Periods 6 and 8 were used for validation. A 4-yr-old stand of nilegrass (cultivar Taishigrass No. 1) was used for the experiment. Pasture size was 3.6 ha and average land slope was less than 3°. The soil was an acidic sandy loam with a pH of 4.8 and organic matter less than 0.8%. The experimental field was divided into four plots, with each plot having an area of 0.9 ha. Information on fertilization management during the eight growing periods is listed in Table 1. To prevent interference with animal health, no pesticides were used and weed infestation was controlled by hand weeding at 1 to 2 wk after cutting. The experimental plots received no irrigation during the growing periods.
View this table:
[in this window]
[in a new window]
|
Table 1. Growing periods and associated fertilization management for nilegrass (Acroceras macrum Stapf) grown at the experimental forage farm of Taiwan Livestock Research Institute (Hsinhua) from June 2002 to May 2004.
|
|
On the days of spectral measurements, three 1-m2 areas per plot were harvested to obtain fresh weights of the aboveground plant parts (fresh-based forage production) and the plot average was used to couple with spectral data. Reflectance spectrum of nilegrass vegetation periodically was taken with a portable spectroradiometer (model GER-2600, Geophysical & Environmental Research Corp., Millbrook, NY) on near cloudless moments between 1030 and 1230 h local time, which followed recommendations by Yang and Chen (2004) to minimize the effect of solar radiance on spectra. There were 24 random quadrats selected for spectral measurements per plot on each sampling date, and each measurement was the average of four full-range (330–2600 nm) spectral scans. The mean reflectance spectrum of each plot was calculated from those 24 measurements and was used to correlate with the forage production (aboveground fresh weight) of the plot. The spectroradiometer had a 10° field-of-view lens and was held by a tripod at a nadir of 1.7 m over nilegrass vegetation to acquire the reflected radiance. The reflectance spectrum was calculated by comparing the radiance spectrum of the target vegetation with the radiance spectrum of a standard reference panel of known spectral property, the so-called "Spectralon" (Labsphere, Inc., North Sutton, NH). Each pair of measurements took less than 1 min. Care was taken to minimize the influence of shadow and background. Only reflectance data in the 350- to 2400-nm range were used to avoid severe noise at both ends of the spectrum, reducing the spectral channels to 537. All spectral measurements and calibrations for the GER-2600 are controlled by a simple menu-driven software supplied by GER on a notebook computer. Radiance, percent reflectance (correct for reflectance of the Spectralon panel), and other display modes are available for interpretation of the data by the software.
Three approaches were used to establish spectral models for estimation of forage production of nilegrass and the results of production variability were compared (Yang and Chen, 2004). In the first approach, linear correlation between spectral reflectance of the measured wavebands and forage production was analyzed to identify the narrow band of the maximum value of correlation coefficient (r), and the best-fit function at this single spectral characteristic (SC) was determined. The second approach was to use spectral indices of two SCs. The normalized difference vegetation index (NDVI) was calculated by the formula (RNIR – RRED)/(RNIR + RRED), where RNIR was the reflectance of the near-infrared (740–1300 nm) peak and RRED was the reflectance of red light (640–740 nm) minimum. The ratios of RRED/RNIR, RGREEN/RNIR, and RRED/RGREEN also were computed, where RGREEN was the reflectance of green light (490–560 nm) maximum. The third approach was to combine more than two SCs in the multiple linear regression (MLR) analysis. The Collinearity Diagnostics and the R-square selection methods of Statistical Analysis System version 8.1 (SAS Institute, 1998) were used in combination to select suitable SCs, which were then modeled by the step-wise selection procedure. The optimum number of wavebands can be selected based on the desired determination coefficient (R2) and Cp value. For example, the best five-variable MLR model was in the form of Y5 = a + a1R
1 + a2R2 + a3R3 + a4R4 + a5R5, where a is the five-variable MLR constant, dependent variable Y5 was the forage production, and independent variables R1, R2, R3, R4, and R5 were reflectance of the characteristic narrow bands 1, 2, 3, 4, and 5 selected from the reflectance spectrum. The root mean square error (RMSE) was used for accuracy assessment and was calculated as
 |
where Xi was the measured value and 
i was the estimated value.
|
RESULTS AND DISCUSSION
|
|---|
Linear correlation between spectral reflectance and forage production across the analyzed spectral domain (350–2400 nm) is plotted in Fig. 1
, providing an overview of the relationship. The spectral reflectance in the range of 350 to 2400 nm had varied correlation coefficients with aboveground fresh weight of harvested forage (fresh-based forage production) of nilegrass so that variation in r along the spectral range of measurements was observed. A positive r value was found in wavebands from the near-infrared (740–1300 nm) to the shorter wavelengths (1300–1400 nm) in the shortwave infrared region (1300–1800 nm), while reflectance of the longer wavebands showed a negative value of r. Green vegetation absorbs large amounts of radiation in the visible light and reflects a greater proportion of radiation in the near-infrared region (Knipling, 1970; Tucker, 1979). A significant linear relationship (P < 0.010) was found in several wavebands, with the highest r value located at 891 nm (r = 0.671; P < 0.010), where the best fit for the reflectance–forage production relationship was with a quadratic function (R2 = 0.510, P < 0.001) (Fig. 2
).
View larger version (16K):
[in this window]
[in a new window]
|
Figure 1. Trends in linear correlation coefficients between spectral reflectance in the range of 350 to 2400 nm and forage production of nilegrass (Acroceras macrum Stapf).
|
|
View larger version (18K):
[in this window]
[in a new window]
|
Figure 2. The quadratic relationship between forage fresh weight and spectral reflectance at 891 nm of vegetation spectra of nilegrass (Acroceras macrum Stapf).
|
|
In selecting a single SC to correlate with alfalfa (Medicago sativa L.) vegetation, Guan and Nutter (2002) pointed out that reflectance at 810 nm had a positive linear correlation with leaf area index (LAI) and yield. Yang et al. (2003) reported that reflectance in the red wavebands was negatively correlated with aboveground fresh weight and leaf area of Amaranthus mangostanus L., a dicot that can be used as a vegetable or as animal feed. A reverse correlation was shown in the visible and near-infrared wavebands and the maximum absolute values of r were located at 767 nm (r = 0.741, P < 0.010) and 574 nm (r = –0.628, P < 0.010), respectively.
Three SCs in the narrow bands of the near-infrared peak (RNIR), the red light minimum (RRED), and the green light maximum (RGREEN) were further identified to calculate the spectral indices. Of the examined four spectral indices, forage production was found best correlated with RGREEN/RNIR ratio (R2 = 0.654, P < 0.001) (Fig. 3
). A negative exponential function was determined, with increases in forage production being combined with a decrease in the RGREEN/RNIR ratio. Results imply that the proportional increase in RNIR due to an increasing forage biomass is smaller relative to the decrease of RGREEN. On the other hand, the commonly used spectral index NDVI had a positive exponential relationship (R2 = 0.602, P < 0.001) with forage production. The increase of NDVI followed an increment of forage production. In cereal crops such as rice (Oryza sativa L.) (Yang and Su, 1998; Yang and Chen, 2004) and corn (Zea mays L.) (Gilabert et al., 1996), both leaf fresh weight and LAI were found exponentially related to NDVI, and there also were scaling problems due to the saturation of reflectance factors. In the case of A. mangostanus, aboveground fresh weight showed a negative linear correlation with RGREEN/RNIR ratio but a positive linear correlation with NDVI (Yang et al., 2003).
View larger version (21K):
[in this window]
[in a new window]
|
Figure 3. The quadratic relationships between forage fresh weight and spectral indices (normalized difference vegetation index [NDVI] and RGREEN/RNIR ratio) calculated from vegetation spectra of nilegrass (Acroceras macrum Stapf).
|
|
Results of the present experiment are consistent with reports by Yang et al. (2003) and Yang and Chen (2004) that showing modeling of SC to forage production is improved by incorporating multiple reflectance parameters into the MLR model. The R2 was increased from 0.510 of the single SC (891 nm) to 0.654 of RGREEN/RNIR ratio, and then to 0.726 (P < 0.001, Cp = 6.000) with the best five-variable linear regression model (Table 2). When validating the MLR model with datasets from two other growing seasons, the model gave reasonable prediction values (r = 0.833; P < 0.001) with a slope of 1.086 and RMSE of 3.891 (Fig. 4
).
View this table:
[in this window]
[in a new window]
|
Table 2. The best five-variable multiple linear regression equation (Y = a + a1R439 + a2R891 + a3R1709 + a4R1909 + a5R2255) established from spectral characteristics of vegetation for estimation of aboveground fresh weight (fresh-based forage production) of nilegrass (Acroceras macrum Stapf).
|
|
View larger version (15K):
[in this window]
[in a new window]
|
Figure 4. Comparisons between the predicted values of forage fresh weight from the multiple linear regression model and the measured values of forage fresh weight of nilegrass (Acroceras macrum Stapf).
|
|
Results suggest that forage production of nilegrass across its phenological development may be assessed and monitored by models established from vegetation high-resolution reflectance data. Changes in chemical or physical characteristics of nilegrass as it increases in biomass were reasonably associated with reflectance of green and near-infrared wavelengths. Spectral models incorporated with multiple SCs resulted in a higher sensitivity and better accuracy in estimating forage production. To expand its applications, however, more data containing the variability caused by cultivation management, soil background, and climatic variation should be included.
|
NOTES
|
|---|
All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Permission for printing and for reprinting the material contained herein has been obtained by the publisher.
Received for publication October 25, 2006.
|
REFERENCES
|
|---|
- Anderson, J., S. Deloach, D. Waters, and P. Howard. 2000. Comparative field and image-derived metrics for forecasting yields in cotton. p. 1–130. Proc. Int. Conf. on Geospatial Information in Agriculture and Forestry, 2nd, Lake Buena Vista, FL. 10–12 Jan. 2000. ERIM International, Ann Arbor, MI.
- Blackburn, G.A. 1998. Spectral indices for estimating photosynthetic pigment concentrations: A test using senescent tree leaves. Int. J. Remote Sens. 19:657–675.[CrossRef]
- Blackburn, G.A. 1999. Towards the remote sensing of matorral vegetation physiology: Relationships between spectral reflectance, pigment, and biophysical characteristics of semiarid bushland canopies. Remote Sens. Environ. 70:278–292.[CrossRef]
- Bouman, B.A.M. 1995. Crop modeling and remote sensing for yield prediction. Neth. J. Agric. Sci. 43:143–161.
- Buu, R.-H., Y.-M. Shy, C.-P. Chen, and M.-C. Chen. 1993. Effects of different cutting stages on forage yield, chemical composition and nutritive value of pangolagrass. (In Chinese, with English abstract.) J. Chin. Soc. Anim. Sci. 22:373–386.
- Chen, R.-K., and C.-M. Yang. 2005. Determining optimal timing for using LAI and NDVI to predict rice yield. J. Photogramm. Remote Sens. 10(3):239–254.
- Clarke, T.R. 1997. An empirical approach for detecting crop water stress using multispectral airborne sensors. HortTechnology 7(1):1–16.
- Diker, K., and W.C. Bausch. 2003. Potential use of nitrogen reflectance index to estimate plant parameters and yield of maize. Biosys. Eng. 85:437–447.[CrossRef]
- Fassio, A., and D. Cozzolino. 2003. Non-destructive prediction of chemical composition in sunflower seeds by near infrared spectroscopy. Ind. Crops Prod. 20:321–329.[CrossRef]
- Gausman, H.W. 1982. Visible light reflectance and absorptance of differently pigmented cotton leaves. Remote Sens. Environ. 13:233–238.[CrossRef]
- Gausman, H.W., W.A. Allen, V.I. Myers, and R. Cardenas. 1969. Reflectance and internal structure of cotton leaves, Gossypium hirsutum L. Agron. J. 61:374–376.[Abstract/Free Full Text]
- Gilabert, M.A., S. Gandia, and J. Melia. 1996. Analyses of spectral-biophysical relationships for a corn canopy. Remote Sens. Environ. 55:11–20.[CrossRef]
- Guan, J., and F.W. Nutter, Jr. 2002. Relationships between percentage defoliation, dry weight, percentage reflectance, leaf-to-stem ratio, and green leaf area index in the alfalfa leaf spot pathosystem. Crop Sci. 42:1264–1273.[Abstract/Free Full Text]
- Heilman, J.L., E.T. Kanemasu, J.O. Bagley, and V.P. Rasmussen. 1977. Evaluating soil moisture and yield of winter wheat in the Great Plains using Lansat data. Remote Sens. Environ. 6:315–326.[CrossRef]
- Kanemasu, E.T., G. Asrar, and M. Yoshida. 1985. Remote sensing techniques for assessing water deficits and modeling crop response. HortScience 20:1043–1046.[Web of Science]
- 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.[CrossRef]
- Lippke, H. 1980. Forage characteristics related to intake, digestibility and gain by ruminants. J. Anim. Sci. 50:952–961.[Abstract/Free Full Text]
- Masoni, A., L. Ercoli, and M. Mariotti. 1996. Spectral properties of leaves deficient in iron, sulfur, magnesium, and manganese. Agron. J. 88:937–943.[Abstract/Free Full Text]
- Rhind, J., and D.C.W. Goodenough. 1976. The assessment and breeding of Acroceras macrum Stapf. Proc. Grassland Soc. South Afr. 11:115–117.
- Rhind, J., and D.C.W. Goodenough. 1979. Acroceras macrum Stapf. (Nile grass): A review. Proc. Grassland Soc. South Afr. 14:27–36.
- Rout, C.J., L.G. Howe, and L.P. du Toit. 1990. The yield of Paspalum dilatatum and Acroceras macrum under irrigation in the Dohne Sourveld. S. Afr. J. Plant Soil 7:240–242.
- SAS Institute. 1998. SAS/STAT user's guide, version 8.1. SAS Inst., Cary, NC.
- Shaug, S.-P., J.-B. Lin, W.-W. Jin, W. Chen, Y.-Y. Chen, S.-C. Chang, and S.-F. Yan. 2002. Breeding of nilegrass Taishi No. 1. (In Chinese, with English abstract.) J. Taiwan Livestock Res. 35:91–100.
- Shaug, S.-P., K.-D. Lo, and J.-B. Lin. 1999. Effects of cutting stages on forage yield and quality of nilegrass. (In Chinese, with English abstract.) J. Taiwan Livestock Res. 32:219–226.
- Su, M.-R., and C.-M. Yang. 1999. Estimation of rice growth from reflectance spectra of vegetative cover. J. Photogramm. Remote Sens. 4:13–23.
- Thenkabail, P.S., R.B. Smith, and E. De Pauw. 2000. Hyperspectral vegetation indices and their relationships with agricultural crops. Remote Sens. Environ. 71:158–182.[CrossRef]
- Tucker, C.J. 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 8:127–150.[CrossRef]
- Yang, C.-M. 2001. Estimation of leaf nitrogen content from spectral characteristics of rice canopy. TheScientificWorld 1(S2):81–89.[CrossRef]
- Yang, C.-M., and R.-K. Chen. 2004. Modeling rice growth using hyperspectral reflectance data. Crop Sci. 44:1283–1290.[Abstract/Free Full Text]
- Yang, C.-M., and C.-H. Cheng. 2001. Spectral characteristics of rice plants infested by brown planthopper. Proc. Nat. Sci. Counc. ROC B 25:180–186.
- Yang, C.-M., and M.-R. Su. 1998. Seasonal variations of reflectance spectrum and vegetation index in rice vegetation cover. p. 574–593. In Proc. Asian Crop Sci. Conf., 3rd. 27 Apr.–2 May 1998. Chin. Soc. Agron., Taichung, Taiwan.
- Yang, C.-M., and M.-R. Su. 2000. Analysis of spectral characteristics of rice canopy under water deficiency. p. 13–18. In Proc. Asian Conf. on Remote Sensing, 21st. 4–8 Dec. 2000. Chin. Soc. Photogrammetry and Remote Sensing, Taipei, Taiwan.
- Yang, C.-M., J.-T. Wu, B.-K. Shen, J.-Z. Yu, C.-T. Lo, and Y. Shen. 2003. Estimating growth and nitrogen status in plants of Amaranthus mangostanus with canopy spectral characteristics. (In Chinese, with English abstract.) J. Agric. Res. China 52:268–290.
TOP
ABSTRACT
INTRODUCTION
