Crop Science 41:774-777 (2001)
© 2001 Crop Science Society of America
CROP ECOLOGY, PRODUCTION & MANAGEMENT
Estimation of Fall Dormancy in Alfalfa by Near Infrared Reflectance Spectroscopy
R.L. Kallenbach*a,
C.A. Robertsa,
L.R. Teuberb,
G.J. Bishop-Hurleya and
H.R. Benedicta
a Dep. of Agronomy Univ. of Missouri, Columbia, MO 65211
b Dep. of Agronomy and Range Science, Univ. of California, Davis, CA 95616
* Corresponding author (KallenbachR{at}missouri.edu)
 |
ABSTRACT
|
|---|
Estimating fall dormancy (FD) in alfalfa (Medicago sativa L.) is time-consuming and expensive. The objective of this study was to estimate the FD class of alfalfa using near infrared reflectance (NIR) spectroscopy. Eleven alfalfa cultivars ranging in FD from 1 to 11 were grown as spaced plants at four diverse locations. In early autumn, 55 to 65 plants were randomly selected at each location and FD determined for each plant. For each sample, NIR spectra were collected and FD values regressed against first and second derivative transformations of spectra by modified partial least squares regression. The optimum equation had a calibration R2 of 0.90 and a mean and standard error of 7.93 ± 1.14. Blind validation samples showed that this equation could accurately predict the FD class of individual plants (r2 = 0.85) as well as entire cultivars (r2 = 0.94). We concluded NIR spectroscopy has the potential to estimate FD classes of alfalfa.
|
INTRODUCTION
|
|---|
FALL DORMANCY IN ALFALFA is related to several important traits including initiation of spring growth, growth habit, root growth, winter hardiness, freezing tolerance, herbage yield, and stand persistence (Kohel and Davis, 1960; Smith, 1961; Barnes et al., 1978; Perry et al., 1987; Schwab et al., 1996; Johnson, et al., 1998). Because of the influence of FD on the adaptation and productivity of alfalfa cultivars, it must be determined before any new cultivar is approved by either the National Alfalfa Variety Review Board or the Plant Variety Protection Office of the USDA. Fall dormancy is also frequently utilized in settling legal disputes over the identity of a cultivar. However, fall dormancy is regulated by a complex set of genetic, morphological, physiological, and environmental factors and is time consuming to measure accurately (McKensie et al., 1988; Fairey et al., 1996; Cunningham and Volenec, 1998).
At present, FD in alfalfa is determined by planting new or unknown cultivars with a set of standard check cultivars in field nurseries in the spring, maintaining plots over the summer, and then clipping plots in September or October depending on location. Then, approximately 25 d after the final clipping, individual plant heights are measured, a regression equation developed with the standard check cultivars, and a FD rating assigned to the new cultivar by this equation (Barnes et al., 1978; Teuber et al., 1998). Although FD is a relatively stable trait, genotype x environment interactions do exist (Fairey et al., 1996; Teuber et al., 1998). Therefore, it is desirable to repeat these measurements in several locations or over multiple years for each cultivar (Teuber et al., 1998).
If a quick, reliable method could be developed to measure FD of an alfalfa cultivar, especially its average dormancy over several environments, it could allow plant breeders to screen experimental lines and process samples efficiently. Potentially, this could be accomplished without establishing separate experiments for this purpose. Schnieder (1984) suggested an alternative method of determining FD using mean unifoliate internode length. Unfortunately, this method did not provide sufficient accuracy to separate cultivars with similar FD ratings. It may be possible to use NIR spectroscopy to accurately measure FD in alfalfa. Prior reports have demonstrated that NIR spectroscopy can accurately quantify protein, fiber, and other chemical constituents in forages (Norris et al., 1976). Reports that are more recent show that this technology can estimate nonchemical characteristics as well, particularly in alfalfa. Such nonchemical characteristics have included leaf:stem ratio (Hill et al., 1988), mold content (Roberts et al., 1987), and legume proportion in grass-legume mixtures (Moore et al., 1990).
As it is used in forage crop analysis, NIR reflectance spectroscopy estimates constituents in whole plant tissue via an empirical approach known as calibration (Shenk and Westerhaus, 1995). Calibration involves identification of a sample set, collecting data on these samples, obtaining spectra, then regressing known data against derivatives of infrared reflectance spectra. After calibration, the regression equation permits accurate analysis of many other samples by prediction of data solely on the basis of the spectra.
Because NIR technology can accurately estimate both chemical and nonchemical traits in forage plants, we hypothesized that it could estimate traits associated with FD in alfalfa. The objective of this study was to determine the feasibility of estimating the FD class of individual alfalfa plants and cultivars by NIR spectroscopy.
|
MATERIALS AND METHODS
|
|---|
Field Methods
Eleven alfalfa cultivars ranging in FD class from 1 to 11 were grown as spaced plants at four locations in California. The locations and their diverse latitudes were: the Intermountain Research and Extension Center at Tulelake, CA, 41°53'N; (mean temperature 6.8°C), the Agronomy and Range Science Field Research Facility at Davis, CA, 38°32'N; (mean temperature 15.7°C), the Kearney Agricultural Center near Parlier, CA, 36°35'N; (mean temperature 17.2°C), and the Desert Research and Extension Center near El Centro, CA, 32°48'N; (mean temperature 22.6°C).
The four locations also had diverse soil types. The soil type at Tulelake was an Osborn silty-clay loam (fine-mixed, mesic, Andaqueptic Haplaquoll) with 47 mg kg-1 available N and 100 g kg-1 organic matter. At Davis the soil series was Reiff (coarse-loamy, mixed, nonacid, thermic Mollic Xerofluvent) with 25 mg kg-1 available N and 13.5 g kg-1 organic matter. The soil at Parlier was a Hanford sandy-loam (coarse-loamy, mixed, nonacid, thermic, Typic Xerorthent) with 2 mg kg-1 available N and 8 g kg-1 organic matter. At El Centro the soil type was an Imperial clay (fine, calcarious montmorillionic, hyperthermic, Vertic Torrifluvent) with 4 mg kg-1 available N and 3 g kg-1 organic matter.
Eleven different cultivars representing a wide range in FD were seeded in 1999 on 20 May, at Tulelake, 24 May, at Davis, 7 May, at Parlier, and 11 May, at El Centro. The cultivars planted at each location were Maverick, Vernal, Pioneer 5246, Legend, Archer, ABI 700, Dona Ana, Pierce, CUF 101, UC-1887, and UC-1465. Each plot was a single row 7.6 m in length with 0.76 m between rows. Within each row, plants were spaced 0.46 m apart (approximately 17 plants per plot). Each plot was replicated four times at each location by means of a randomized complete block design. The plots were clipped in 1999 to a stubble height of 5 cm between 1 July and 15 July at all locations and again on 7 September at Tulelake, 7 October at Davis and Parlier, and 23 October in El Centro. All forage from these clippings was discarded.
Three and one half weeks after the last clipping, 55 to 65 plants at each location were scored by measuring natural plant height (NPH) by the 1 to n scale outlined by Teuber et al. (1998). Each increment on the scale was equal to 5 cm of growth. Each plant was given a score and scores square root transformed to remove any heterogeneity of variance. Transformed height data were converted to standard FD classes by the following equation reported by Taggard et al. (2000):
The values generated from this equation were used as the reference data for NIR spectroscopy calibration.
Immediately after scoring for NPH, plants were clipped to a stubble height of 5 cm, plant tissue placed in a paper bag, and dried at 40 to 50°C for 96 h. After drying, samples were ground to pass a 1-mm screen with a cyclone type grinder and then stored at 21°C until spectral analysis.
Spectral Analysis
Ground samples were scanned from 1110 to 2490 nm on a NIRSystems scanning monochromator, model 5000 (Silver Spring, Maryland), and log 1/reflectance (log 1/R) was recorded at 2-nm intervals. After scanning, log 1/R data were transformed to first and second derivative spectra, and prediction equations for FD were developed by regressing field data against derived spectra. The procedure for NIR spectroscopy calibration was modified partial least squares regression as performed by software obtained from Infrasoft International (version 1.02, Port Matilda, PA). Equations were validated with four cross-validation groups, and outliers were eliminated in two outlier passes (Shenk and Westerhaus, 1991, 1995). Except for wavelength selection, an optimum equation was chosen according to the criteria outlined by Windham et al. (1989). Optimum equations were identified as those with high coefficients of determination for calibration (R2) and 1 - variance ratio (1 - VR), a low standard error of calibration (SEC), and a low standard error for cross validation (SECV).
After the equation was developed and cross validated, it was tested in two ways. First, it was tested on samples collected from 28 individual alfalfa plants, which were independent of those used in calibration and validation. These samples were collected from the same four locations and FD class estimated by the procedure described above. Second, it was tested for predicting the average FD class for each of nine alfalfa cultivars. This was achieved by scanning five plants from each cultivar and comparing the mean FD generated by NIR spectroscopy to the long-term FD rating for each cultivar reported by Taggard et al. (2000). Plants from these cultivars were independent of those used in calibration and validation. For both tests, predicted FD class was regressed against actual FD class by least squares regression (Neter et al., 1989).
|
RESULTS AND DISCUSSION
|
|---|
Near infrared reflectance spectroscopy accurately predicted the FD class of alfalfa (Table 1). The R2 of 0.90 calculated in equation development and 1 - VR of 0.85 for cross validation indicate that an acceptable portion of the variance was explained by the model. The mean and standard error of calibration (7.93 ± 1.14) demonstrate the potential of this empirical procedure to predict FD. Although the error associated with this calibration equation is acceptable (Shenk and Westerhaus, 1995), achieving a lower standard error would be possible with more accurate reference data. This became obvious during calibration development as every outlier identified in the regression was a T outlier while none was identified as an H outlier. In NIR spectroscopy, a T outlier identifies a sample with a relationship between its reference value and its spectra that is different from the relationship for other samples in the population; the H outlier identifies a sample that is spectrally different from other samples in the population (Shenk and Westerhaus, 1995). A T outlier, therefore, indicates extreme reference values, not spectral values; it usually implies an error in the regression that could be improved by a reduction of the error in the reference data.
View this table:
[in this window]
[in a new window]
|
Table 1. Regression statistics for estimation of fall dormancy in alfalfa determined by near infrared reflectance (NIR) spectroscopy.
|
|
The opportunity to improve the reference data by more precise field measurements is intuitive because actual height could be recorded and used in the calculation of FD class. As mentioned earlier, the standard protocol for collecting FD data involves recording regrowth plant height on a scale of 1 to n with each increment equaling 5 cm. Reducing this increment to 1 cm, though perhaps impractical for field measurements, would likely provide calculations of FD class that would more precisely reflect the spectra associated with FD. This likely would lead to a lower standard error of calibration for NIR spectroscopy.
When tested for accuracy on samples not part of calibration development, the equation was able to predict FD of individual plants with an r2 of 0.85 (Fig. 1). Because this was achieved with independent samples, the successful prediction of FD demonstrates the usefulness of this equation. In addition, because it was achieved with 28 random samples collected from multiple locations, that in total contained nearly 3000 plants, the good correlation demonstrates the robust nature of this calibration equation.
View larger version (13K):
[in this window]
[in a new window]
|
Fig. 1. Predicted versus actual fall dormancy rating of individual alfalfa plants (n = 28). Fall dormancy was predicted by NIR spectroscopy and regressed against fall dormancy calculated from actual field data on a plant-by-plant basis.
|
|
While NIR spectroscopy proved to be a reliable method to predict the FD class of individual plants, plant breeders are often more interested in the mean FD class of an entire alfalfa cultivar. When used to predict the mean FD class of nine alfalfa cultivars grown near the present study, the values produced by NIR spectroscopy showed an r2 of 0.94 (Fig. 2). From a practical sense, this shows the utility of NIR spectroscopy for identifying the FD class of a single cultivar for plant breeding purposes. Seemingly, relatively few plants were needed to accurately predict the mean FD class of a cultivar, suggesting that the spectral qualities associated with FD are unique and stable. This further suggests that NIR spectroscopy might be used to develop a universal equation that could predict FD for a range of cultivars grown in different environments.
View larger version (12K):
[in this window]
[in a new window]
|
Fig. 2. Predicted versus actual fall dormancy rating of nine alfalfa cultivars. Fall dormancy was predicted by NIR spectroscopy on samples collected from five plants from each cultivar and the mean values regressed against their standard fall dormancy class reported by Taggard et al. (2000).
|
|
We concluded that NIR spectroscopy could estimate the FD class of alfalfa. The equation reported in this study, though accurate, only demonstrates its potential use. The development of a universal equation would require further testing to determine if similar accuracy could be achieved with additional cultivars collected across multiple years and more environments. In addition, FD predictions may be improved by increasing the precision of the reference data.
|
NOTES
|
|---|
This research was a joint contribution from the Univ. of Missouri Agric. Exp. Stn. and the Univ. of California Exp. Stn., paper no. 13059. Mention of trade name or proprietary product does not constitute endorsement by the Univ. of Missouri or the Univ. of California over products of other manufacturers that may also be suitable.
Received for publication August 18, 2000.
|
REFERENCES
|
|---|
- Barnes, D.K., D.M. Smith, R.E. Stucker, and L.J. Elling. 1978. Fall dormancy in alfalfa: A valuable prediction tool. p. 34. In D.K. Barnes (ed.) Proc. 26th N. Am. Alfalfa Improvement. Conf., Brookings, SD. 6–8 June 1978. USDA, Beltsville, MD.
- Cunningham, S.M., and J.J. Volenec. 1998. Seasonal carbohydrate and nitrogen metabolism in roots of contrasting alfalfa (Medicago sativa L.) cultivars. J. Plant. Physiol. 153:220–225.
- Fairey, D.T., L.P. Lefkovitch, and N.A. Fairey. 1996. The relationship between fall dormancy and germplasm source in North American alfalfa cultivars. Can. J. Plant Sci. 76:429–433.
- Hill, N.S., J.C. Petersen, J.A. Stuedemann, and F.E. Barton. 1988. Prediction of percentage in leaf in stratified canopies of alfalfa with near IR reflectance spectroscopy. Crop Sci. 28:354–358.[Abstract/Free Full Text]
- Johnson, L.D., J.J. Marquez-Ortiz, J.F.S. Lamb, and D.K. Barnes. 1998. Root morphology of alfalfa plant introductions and cultivars. Crop Sci. 38:497–502.[Abstract/Free Full Text]
- Kohel, R.J., and R.L. Davis. 1960. The inheritance of cold resistance and its relation to fall growth in F2 and BC1 generations of alfalfa. Agron. J. 52:234–237.[Abstract/Free Full Text]
- McKensie, J.S., R. Paquin, and S.H. Duke. 1988. Cold and heat tolerance. p. 259–302. In A.A. Hanson et al. (ed.) Alfalfa and alfalfa improvement. Agron. Monogr. 29. ASA, CSSA, and SSSA, Madison, WI.
- Moore, K.J., C.A. Roberts, and J.O. Fritz. 1990. Indirect estimation of the botanical composition of alfalfa-smooth bromegrass mixtures at three stages of maturity. Agron. J. 82:287–290.[Abstract/Free Full Text]
- Neter, J., W. Wasserman, and M.H. Kutner. 1989. Applied linear regression models. 2nd ed. Irwin, Inc. Homewood, IL.
- Norris, K.H., R.F Barnes, J.E. Moore, and J.S. Shenk. 1976. Predicting forage quality by infrared reflectance spectroscopy. J. Anim. Sci. 43:889–897.[Abstract/Free Full Text]
- Perry, M.C., M.S. Mcintosh, W.J. Wiebold, and M. Welterlen. 1987. Genetic analysis of cold hardiness and dormancy in alfalfa. Genome 29:144–149.
- Roberts, C.A., K.J. Moore, D.W. Graffis, H.W. Kirby, and R.P. Walgenbach. 1987. Quantification of mold in hay using near infrared reflectance spectroscopy. J. Dairy Sci. 70:2560–2564.[Abstract/Free Full Text]
- Schnieder, M. 1984. Relationship between unifoliate internode length and fall dormancy in alfalfa. M.S. thesis. Univ. of California, Davis, CA.
- Schwab, P.M., D.K. Barnes, and C.C. Sheaffer. 1996. The relationship between field winter injury and fall growth score for 251 alfalfa cultivars. Crop Sci. 36:418–426.[Abstract/Free Full Text]
- Shenk, J.S., and M.O. Westerhaus. 1991. Population structuring of near infrared spectra and modified partial least squares regression. Crop Sci. 31:1548–1555.[Abstract/Free Full Text]
- Shenk, J.S., and M.O. Westerhaus. 1995. Forage analysis by near infrared spectroscopy. p. 111–120 In R.F. Barnes et al. (ed.) Forages. Vol. II. The science of grassland agriculture. 5th ed. Iowa State University Press, Ames, IA.
- Smith, D. 1961. Association of fall growth habit and winter survival in alfalfa. Can. J. Plant Sci. 41:224–251.
- Taggard, K.L., L.R. Teuber, and L.K. Gibbs. 2000. Notice of release of alfalfa germplasm selected under field conditions for fall growth. Univ. of Calif. Agron. Prog. Rep. 268.
- Teuber, L.R., K.L. Taggard, L.K. Gibbs, S.B. Orloff, S.C. Mueller, C.A. Frate, D.H. Putnam, and J.J. Volenec. 1998. Check cultivars, locations, and management of fall dormancy evaluation. p. 25. In Proc. 36th N. Am. Alfalfa Imp. Conf. Bozeman, MT. 2–6 Aug. 1998. North Amercian Alfalfa Improvement Conference Committee, Beltsville, MD.
- Windham, W.R., D.R. Mertens, and F.E. Barton II. 1989. 1. Protocol for NIRS calibration: Sample selection and equation development and validation. p. 96–103. In G.C. Marten et al. (ed.) Near infrared reflectance spectroscopy (NIRS): Analysis of forage quality. USDA Agric. Handb. 643 (revised). U.S. Gov. Print. Office, Washington, DC.
This article has been cited by other articles:
|
|
|
|
 
J. L. Halgerson, C. C. Sheaffer, N. P. Martin, P. R. Peterson, and S. J. Weston
Near-Infrared Reflectance Spectroscopy Prediction of Leaf and Mineral Concentrations in Alfalfa
Agron. J.,
March 1, 2004;
96(2):
344 - 351.
[Abstract]
[Full Text]
[PDF]
|
|
|
TOP
ABSTRACT
INTRODUCTION
