Abstract
To evaluate the ability of field remote sensing for predicting pasture macronutrients, hyperspectral reflectance data between 350 and 2500 nm were acquired from a number of dairy and sheep pasture canopies in New Zealand. Reflectance factor, absorbance, derivatives, and continuum-removal data were regressed against pasture nitrogen (N), phosphorus (P), and potassium (K) concentrations using partial least squares regression (PLSR). Overall, more accurate predictions were achieved using the first derivative data. The accuracy of the PLSR calibration models to predict pasture N, P, and K concentrations increased with the separation of the pasture samples by season. Predictions with reasonable accuracy (coefficient of determination, R 2 > 0.74, and the ratio of standard deviation (SD) of the nutrients measured to the root mean square error of cross-validation (RMSECV) ≥ 2.0) were obtained for N during winter (RMSECV ≤ 0.23%), autumn (RMSECV ≤ 0.36%), and summer (RMSECV ≤ 0.43%) seasons; P during autumn (RMSECV = 0.05%); and K during summer (RMSECV = 0.33%).
Acknowledgments
I.D. Sanches was supported by a PhD scholarship from the Brazilian Government (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – CAPES). The authors thank Roger Parfitt (Landcare Research) for access to the plots and data from the biodiversity study and to P21 Feeds (C10X0604) for access to the plots and data from the N × P × irrigation interaction study. Dr P. Loganathan, Mr M. Bretherton, and Joy and Clem Smith of Alfredton are thanked for making their trial data and farm available for this study, respectively. We thank Martin Hawke for assisting with the trial work at Tokoroa, Atiamuri, and Manawahe.