Are NIR spectra useful for predicting site indices in sandy soils under Eucalyptus stands in Republic of Congo?

The objectives of this study were: (i) to test whether near infrared reflectance spectroscopy (NIRS) can be used to assess site fertility in sandy and ferralitic soils of the Congo; (ii) to evaluate the comparative advantages of this methodology with respect to classical regressions with chemical and physical analysis of soils; and (iii) to assess the ability of NIRS to predict concentrations of individual nutrients.Sixty-two plots were sampled in 20 selected forest stands with varying wood production. Site indices (Hlim) were estimated from the measured dominant height in the plots. Soil samples (N = 353) were analysed by NIRS, while physical and chemical analyses were carried out on 210 subsamples. Partial Least Squares regression (PLS) was used to model the site index from NIRS, while Variable Importance in Projection (VIP) allowed relating the main explicative frequencies to nutrient availability. A validation was performed based on an independent field trial (not used in the calibration set) in which soils were collected and analysed by NIRS. The measured dominant height growth was compared with the dominant height growth simulated from the site indices predicted using NIR spectra.Multiple regressions between site indices (Hlim) and soil properties failed to explain the variability in Hlim (R2 = 0.36). This variability was better predicted by NIR spectra, with an R2 = 0.64 and a predicted Root Mean Square Error of 7% for the mean value of Hlim. Based on analysis of VIP frequencies, the most important factors determining the site indices were organic matter, phosphorus, CEC, clay content, and pH. The comparison between measured and simulated dominant height in the validation set gave similar R2 (0.61) than in the calibration set.These results demonstrate that mapping of site indices could be performed at an affordable cost in this region. The methodology developed here, which links PLS and VIP to assess the main factors determining soil fertility, is not specific to this study and could be used elsewhere.

► The site index of sandy soils was analysed based on NIRS spectra and soil properties.
► The site index was poorly related to soil chemical analyses.
► Models based on NIRS spectra performed twofold better.
► The factors driving the site index were identified through Variable Importance in Projection.
► The site index could be mapped at a reasonable cost and precision in this region.