Efforts to understand the spatial and temporal distribution of soil properties across landscapes require the analysis of many soil samples. These analyses, in some cases, can be costly and time intensive. Diffuse reflectance spectroscopy has shown success to sense base soil properties rapidly and cost-effectively, however, there is limited knowledge of how visible-near infrared spectra (VNIRS) relate to micro- and macro-nutrients. The latter are profoundly important in smallholder agricultural settings in south-east Asia where cop yield and livelihood depend on soil health and sustainable nutrient levels. VNIRS has the potential to transform soil management in smallholder farm communities, though has rarely been utilized.

Our objectives were to (i) use chemometrics to infer base soil properties and macro- and micronutrients in rainfed and irrigated smallholder farms in southern India. We utilized VNIRS as a means of rapid soil analysis. We collected 255 soil samples of Veritsols and vertic Inceptisols and analyzed them for soil texture, soil organic carbon, pH, electrical conductivity, and soil macro- and micronutrients. Sieved soil subsamples were also scanned in the VNIRS range (350 – 2,500 nm) at 1 nm resolution. We applied several spectral processing techniques, including Savitzky-Golay smoothers and first and second derivatives, to reduce noise and enhance signals. Then we employed several multivariate regression methods including partial least squares regression analysis and ensemble regression trees and used rigorous validation analysis to identify the best performing model for each soil property. Results confirm our postulated expectations that VNIRS-soil prediction models offer an alternative to traditional soil mapping in this region. Once spectral soil models are successfully developed and validated for a soil geographic domain they can be applied in the future on new soil samples reducing the need for costly analytical analysis.