Predicting soil nitrogen content using narrow-band indices from <i>Eucalyptus grandis</i> canopies as proxy

Abstract

Optimal fertiliser applications for sustainable forest stand productivity management, whilst protecting the environment, is vital. This study estimated soil nitrogen content using leaf-level narrow-band vegetation indices derived from a hand-held 350–2 500 nm spectroradiometer. Leaf-level spectral data were collected and subjected to continuum removal spectral transformations, in addition to using raw reflectance spectra. These leaf spectral indices were used to explain the variance of soil nitrogen status in the forest soils of compartments under Eucalyptus grandis canopies. Soil samples were collected at depths of 0.3–0.8 m and analysed for nitrogen. Results indicated significant correlations (0.37 ≤ r ≤ 0.80; p < 0.05) between leaf spectral indices and soil nitrogen. The ANOVA results for spectral indices–site interactions showed that significant differences were only observed between good–medium and good–poor sites. The predictive capability of the models developed for soil nitrogen showed that continuum removed spectra had high adjusted R² values (R² = 0. 85; p < 0.05) and low PRESS statistic values (0.05) when compared with approaches based on raw spectra (R² = 0.77; p < 0.05; PRESS = 0.07). The results indicate a potential for forest managers to monitor the status of soil nitrogen in commercial forestry compartments and determine how much fertiliser is required to optimise tree growth.

Keywords: hyperspectral remote sensing, leaf spectral indices, soil nitrogen, statistical analysis