Prediction of wood density and carbon-nitrogen content in tropical agroforestry tree species in western Kenya using infrared spectroscopy

Abstract

The global debate on climate change needs to be furnished with accurate and precise measurement of biomass in agricultural landscapes. Wood density is a supporting parameter for biomass estimation; however, empirical methods for wood density determination are destructive and complex, as are conventional wet chemistry analyses of carbon and nitrogen. Thus a low cost and non-destructive method of estimation is required. Infrared Spectroscopy coupled with chemometrics multivariate techniques offers a fast and non-destructive alternative for obtaining reliable results without complex sample pre-treatments. This study sought to develop a prediction model for estimation of wood density, carbon and nitrogen across species using Infrared Spectroscopy. Empirical data for determination of these parameters were obtained from coring 77 trees sampled from three benchmark sites (Lower, Middle and Upper Yala blocks) along Yala basin in Western Kenya. Samples from cored holes in the tree (branch, stem and roots) were used to estimate wood biovolume and density. Models for estimation of these parameters were derived from scanning 404 cores using diffuse reflectance Infrared Spectroscopy and reference values for carbon and nitrogen obtained using a Carbon-Nitrogen analyzer. Partial least squares regression, using first derivative spectra pre-treatment, was used to develop a model based on different calibrations sets. Models were compared on the basis of the accuracy of prediction using the coefficient of determination (R2), Standard Error of Calibration (SEC) and Standard Error of Prediction (SEP). Calculated wood density range was 0.20-0.95gcm-3 with the mean being 0.59 gcm-3, while IR predicted 0.25-0.95 gcm-3 (mean 0.53 gcm-3) in the Near Infrared Region (NIR) and 0.32-0.86 gcm-3 (mean 0.53 gcm-3) in the Mid Infrared Region (MIR). Measured carbon range was 40%- 52% (mean 48%), while IR predicted 44%-51% (mean 48%) in NIR region and 46%-51% (mean 48%) in MIR region. Measured nitrogen range was 0.09-0.48% (mean 0.28%), while IR predicted 0.18%-0.47% (mean 0.24%) in NIR region and 0.18%-0.38% (mean 0.24%) in MIR region. Values of SEC were low relative to laboratory analytical errors. Interactions between densities with tree species and tree parts showed significant effect (p0.57 for all the parameters. This suggests large variations within species that cannot be predicted using IR. The data generated here on densities were comparable with those given in a global wood density database. On the other hand, carbon content varied among species but not between the sites, an indication that the often assumed default value of 50% carbon in wood is over estimation of tree carbon and would lead to over estimation of the total carbon stocks. NIR region gave better predictions than MIR, although the prediction performance was insufficient to recommend Infrared Spectroscopy as a practical method for direct determination of wood density and carbon content across species when different percentages were used