Skip to main content
ARS Home » Research » Publications at this Location » Publication #243683
Title: Prediction of forage quality from remotely sensed data: comparison of cultivar-specific and cultivar-independent equations using three methods of calibration

Author
item Starks, Patrick
item Brown, Michael

Submitted to: Crop Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/19/2010
Publication Date: 8/31/2010
Citation: Starks, P.J., Brown, M.A. 2010. Prediction of forage quality from remotely sensed data: comparison of cultivar-specific and cultivar-independent equations using three methods of calibration. Crop Science. 50:2159-2170.

Interpretive Summary: Remote sensing of pasture canopy reflectance could reduce laborious field sampling and sample processing procedures and improve the efficiency of forage utilization and quality evaluation. Furthermore, it would permit mapping of the pasture nutritional landscape and provide timely information to ranch and livestock managers. Three approaches were used to develop equations to predict concentrations of nitrogen (N), neutral detergent fiber (NDF), and acid detergent fiber (ADF) from remotely sensed reflectance data collected from six cultivars of live, standing bermudagrass canopies. The reflectance data were collected over the visible through near-infrared portion of the spectrum. The main goal of the research was to evaluate the three calibration methods for the development of robust equations to predict N, NDF, and ADF independent of bermudagrass cultivar. The study findings suggested that a cultivar-specific equation was not necessary for prediction of N concentration in the bermudagrass cultivars used in this study. Application of cultivar-independent calibration equations to the validation data set revealed poor to moderate performance of the equations for both NDF and ADF. Generally, cultivar-specific equations also performed poorly for prediction of NDF and ADF. It is anticipated that inclusion of reflectance data in the shortwave portion of the spectrum will improve predictions of N, NDF, and ADF.

Technical Abstract: Remote sensing of pasture canopy reflectance could reduce laborious field sampling and sample processing procedures and improve the efficiency of forage utilization and quality evaluation. Furthermore, it would permit mapping of the pasture nutritional landscape and provide timely information to ranch and livestock managers. Three approaches were used to develop equations to predict concentrations of nitrogen (N), neutral detergent fiber (NDF), and acid detergent fiber (ADF) from hyperspectral remotely sensed reflectance data collected from six cultivars of live, standing bermudagrass canopies. The reflectance data were collected over the visible through near-infrared (400 – 1000 nm) portion of the spectrum. These equation calibration approaches were partial least squares regression (PLS), multiple linear regression with maximum r2 improvement (MAXR), and artificial neural networks (ANN). The main goal of the research was to evaluate the three calibration methods for the development of robust equations to predict N, NDF, and ADF independent of bermudagrass cultivar. Three specific objectives of the study were to: i) develop both cultivar-independent and cultivar-specific equations to predict N, NDF, and ADF from remotely sensed canopy reflectance, ii) validate the accuracy of these equations on a independent data set, and iii) compare the results from the cultivar-specific and cultivar-independent equations to determine how well equations developed across cultivar perform relative to equations developed for a specific cultivar of bermudagrass. The study findings suggested that a cultivar-specific equation was not necessary for prediction of N concentration in the bermudagrass cultivars used in this study. Application of cultivar-independent calibration equations to the validation data set revealed poor to moderate performance of the equations for both NDF and ADF. Generally, cultivar-specific equations also performed poorly for prediction of NDF and ADF. It is anticipated that inclusion of reflectance data in the shortwave portion of the spectrum will improve predictions of N, NDF, and ADF. Comparison of PLS, MAXR, and ANN calibration and validation results revealed that ANN rarely outperformed both PLS and MAXR. Additionally, a number of subjective decisions had to be made in order to design and run the neural network, and much trial and error is typically associated with this approach. For these reasons, equation calibration via PLS and MAXR is preferred.