Nitrogen prediction in grasses: effect of bandwidth and plant material state on absorption feature selection

Abstract

We analysed stability and predictive capabilities of known nitrogen absorption features between plant material prepared for NIRS (dried) and RS (fresh) studies. Grass spectra were taken of the plant canopy, and again after the grass sample was dried and ground. Models were derived using stepwise multiple linear regression (sMLR). Regression values (adj.r ²) produced using the dried material were greater than those produced using canopy material. For dried material only wavebands from the SWIR region were selected. Wavebands selected by sMLR on canopy material were located in both the VNIR and SWIR regions. Using wavebands selected for dried material models produced low adj.r ² values when applied to canopy plant material; differences in adj.r ² values are smaller when wavebands selected in canopy material models are applied to dried material. Widening of nitrogen features produced higher adj.r ² values for both dried and canopy material. This work shows that obtaining models with high predictive capabilities for nitrogen concentration is possible, but waveband selection should not be limited to features identified by NIRS studies. To accommodate for variability in absorption features, and instrument errors, absorption features should be widened.

Acknowledgements

The greenhouse experiment was made possible through collaboration with the Resource Ecology Group, Wageningen University. We would like to thank the staff of UNIFARM, Wageningen University for their assistance in the greenhouse experiment, Dr David Rossiter, ITC, for his advice on the greenhouse setup and nutrient applications, Dr Harald van der Werff, ITC, for assistance with the spectral measurement setup, Anne-Marie van Driesche, Resource Ecology Group, Wageningen University, for her help in the chemical analysis, and the rest of the TEMBO team for their support. We wish to thank the reviewers for their constructive criticism that we feel has helped us strengthen this paper. This research was made possible through WOTRO, NWO funding (W. 01.53.2004).