Reducing Sample Quantity and Maintaining High Prediction Quality of Grassland Biomass Properties with near Infrared Reflectance Spectroscopy

Abstract

Broad-scale ecological research often suffers from insufficient spatial and temporal replication. Near infrared (NIR) reflectance spectroscopy offers the opportunity for rapid and cheap measurements of many chemical constituents in organic materials. However, standard NIR instrumentation requires a certain amount of sample material which strongly restricts the fields of application for the NIR technique. Therefore, we tested if reliable predictions from NIR spectra can be obtained utilising a device that reduces the amount of required sample material by more than 95% compared to standard equipment. For large and small sample quantities, we present two sets of calibration models for C, N, P, K, Ca and Mg concentrations as well as fibre components such as neutral detergent fibre (NDF), acid detergent fibre (ADF) and acid detergent lignin (ADL) in above-ground grassland community biomass. Coefficients of multiple determination ( R2) of calibration models based on spectral data derived from standard equipment for C, N, P, K, Ca, Mg, NDF, ADF and ADL were 0.78, 0.98, 0.78, 0.92, 0.87, 0.89, 0.95, 0.94 and 0.87, respectively. Except for C and P, the ratio of standard deviation of the reference values to the standard error of cross validation and ratio of performance deviation indicated acceptable to high model precision. The application of NIR spectroscopy for C and P measurements was limited due to low variation in concentrations and/or low concentrations in the analysed above-ground grassland biomass. As compared to the deviation of duplicate reference measurements, the standard error of prediction was less than two times higher for C, N, NDF, ADF, ADL and K and up to three times higher for P, Ca and Mg. Prediction models based on the spectral data recorded with a small sample cell (volume of sample material less than 0.25 cm3) were of similar precision. The significant reduction of sample material required for NIR analysis and, at the same time, maintaining (high) precision of calibration models is an important advance towards the wider adoption of the NIR technique in ecological research.