Soil total carbon analysis in Hawaiian soils with visible, near-infrared and mid-infrared diffuse reflectance spectroscopy

Accurate assessment of total carbon (Ct) content is important for fertility and nutrient management of soils, as well as for carbon sequestration studies. The non-destructive analysis of soils by diffuse reflectance spectroscopy (DRS) is a potential supplement or alternative to the traditional time-consuming and costly combustion method of Ct analysis, especially in spatial or temporal studies where sample numbers are large. This alternative technique has been utilized in several locations in the United States and elsewhere, but has not been tested on the unique and diverse tropical soils in Hawaii. This study investigated the feasibility of DRS for Ct prediction of Hawaiian agricultural soils by creating visible, near-infrared (VNIR) and mid-infrared (MIR) spectral libraries and developing chemometric models with partial least squares regression (PLSR) and random forests (RF) ensemble tree regression. The sample set contained 305 soils from across the five main Hawaiian Islands, representing 10 soil orders and over 100 soil series. The Ct of these samples measured by dry combustion ranged from < 1% to 56%, greater than that of most previously published studies. The VNIR spectra of the soils commonly exhibited features associated with OH− and H2O, iron oxides, phyllosilicates, and organic molecules. The numerous features in the MIR spectra can be attributed to OH−, organic molecules, and a variety of silicate minerals, such as phyllosilicates and quartz. Excellent results were obtained from both PLSR models using VNIR and MIR spectra. The models had R2 values of 0.95/0.94 (VNIR/MIR), root mean squared error (RMSE) values of 2.80%/3.08%, residual prediction deviation (RPD) values of 4.25/3.91, and ratio of performance to inter-quartile distance (RPIQ) values of 3.66/3.38. The RF VNIR and MIR models were also of high quality, with R2 = 0.95/0.96 (VNIR/MIR), RMSE = 2.82%/2.28%, RPD = 4.27/5.28, and RPIQ = 3.68/4.56. An evaluation of the different wavelength ranges and chemometric methods determined that for this sample set all were capable of robust predictions of Ct. These results indicate that DRS is an alternative technique for analysis of Ct in Hawaiian soils.

► We tested diffuse reflectance spectroscopy to predict total carbon in Hawaiian soils. ► Visible near‐infrared and mid-infrared soil spectra were collected and characterized. ► Both of these wavelength ranges produced robust total carbon prediction models. ► Both partial least squares and random forests regression produced robust models. ► This technique is a promising alternative for soil carbon analysis in Hawaii.