Use of NIRS for the rapid prediction of total N, minerals, sugars and starch in tropical root and tuber crops
The objective of the present study was to test the robustness of near infrared spectroscopy (NIRS) for the prediction of total N content in underground storage organs across a diverse range of root and tuber crop varieties. Overall, 1096 accessions (acc.) from five different species (cassava = 112 acc., cocoyam =117 acc., sweet potato = 225 acc., taro = 306 acc. and yams = 266 acc.) were chemically analysed for total N and minerals, as well as starch, sugars and cellulose. For validation of the models, 178 samples composed of the same five different species were collected in farmers' fields, at random. All spectra were taken over the wavelength range of 350-2500 nm. Partial least-squares (PLS1) regression technique was used to develop predictive models. Their comparison with the chemical values allowed the establishment of equations of calibration. In terms of predictive performance, the equation for total N should be considered as very good with a r 2 pred of 0.93 (SEP = 0.87). Minerals presented low r 2 pred of 0.62 (SEP = 1.05). Starch and sugars presented r 2 pred of 0.77 and 0.86, respectively (SEP = 3.2 and 1.82). Cellulose could not be satisfactorily predicted with a low r 2 cv (0.57) for the calibration. The r 2 pred values of total N, starch and sugars are high enough to allow good estimates of their contents, confirming the interest of NIRS for predicting rapidly these major compounds. Potential applications are discussed.
Summary: | The objective of the present study was to test the robustness of near infrared spectroscopy (NIRS) for the prediction of total N content in underground storage organs across a diverse range of root and tuber crop varieties. Overall, 1096 accessions (acc.) from five different species (cassava = 112 acc., cocoyam =117 acc., sweet potato = 225 acc., taro = 306 acc. and yams = 266 acc.) were chemically analysed for total N and minerals, as well as starch, sugars and cellulose. For validation of the models, 178 samples composed of the same five different species were collected in farmers' fields, at random. All spectra were taken over the wavelength range of 350-2500 nm. Partial least-squares (PLS1) regression technique was used to develop predictive models. Their comparison with the chemical values allowed the establishment of equations of calibration. In terms of predictive performance, the equation for total N should be considered as very good with a r 2 pred of 0.93 (SEP = 0.87). Minerals presented low r 2 pred of 0.62 (SEP = 1.05). Starch and sugars presented r 2 pred of 0.77 and 0.86, respectively (SEP = 3.2 and 1.82). Cellulose could not be satisfactorily predicted with a low r 2 cv (0.57) for the calibration. The r 2 pred values of total N, starch and sugars are high enough to allow good estimates of their contents, confirming the interest of NIRS for predicting rapidly these major compounds. Potential applications are discussed. |
---|