Near infrared spectroscopy of plantation forest soil nutrients in Sabah, Malaysia, and the potential for microsite assessment

• Main Authors: Ahmad Jani, Hamzad Fahmi, Meder, Roger, Abdul Hamid, Hazandy, Mohd Razali, Sheriza, Mohd Yusoff, Khairul Hafiz
• Format: Article
• Published: SAGE Publications 2021
• Online Access: http://psasir.upm.edu.my/id/eprint/94341/; https://journals.sagepub.com/doi/10.1177/09670335211007673

Knowledge of soil physical and chemical properties is vital to the optimal growing performance of agricultural crops, including plantation forest trees. Near infrared (NIR) spectroscopy has been shown to be a tool that enables rapid and low-cost assessment of soils, however its use in forest plantations has been slow to develop. This study shows the development of calibrations for total organic carbon, total nitrogen and soil pH using a handheld NIR spectrometer for soils at three sites in Sabah, Malaysia. Soil samples were collected, dried, milled and scanned after which they were analysed using standard chemical methods to obtain total organic carbon (TOC) and total nitrogen (TN). Partial least squares regression was used to develop calibrations between reference data and NIR spectra and validated using an independent sample set. The calibration of soil pH is made using a subset of samples across A- and B-horizons for samples from two of the three sites. The most effective spectral pre-treatment was the standard normal variate for TOC and TN while the Savitzky-Golay first derivative was the best pre-treatment for predicting soil pH. Principal component analysis was performed on the raw NIR spectra of all samples to confirm that the samples from different sites were able to be used in a single regression analysis. Kennard-Stone selection was used to create calibration sets and validation sets from the combined spectra from all sites and both soil horizons. Calibrations were also developed independently on the A- and B-horizon samples, but there were insufficient sample numbers to utilize an independent validation set. The coefficients of determination for the validation set (r2p) were 0.77 and 53 for TOC and TN respectively while the root mean square error of prediction (RMSEP) was 0.44 g 100 g−1 for TOC and 0.051 g 100 g−1 for TN. In addition, it showcases the application of these calibrations to provide spatial assessment of two differing micro-sites within a single Eucalyptus pellita progeny breeding trial. Combined with the potential to monitor foliar nutrients, the ability to obtain high spatial details of soil composition will assist tree plantation growers and also other agricultural producers, such as oil palm plantation managers, to better manage their soil and fertiliser regimes.