Soil spectral signatures for sugarcane fertiliser recommendations through an adapted soil typology in Réunion

Characterization and study of soils are the basis of soil mapping and could lead to increases in crop productivity. In Reunion, the soil-specific nutrient management expert system, SERDAF, adapts its recommendations to the soil types found in sugarcane fields. The soil classification used in this expert system is based on morphopedological knowledge of soils and landscapes of Reunion, and laboratory measurements of soils constituents. The aim of this study was to develop an objective classification of Reunion volcanic sugarcane soils using near-infrared (NIR) spectra. More than 3000 soil samples NIR spectra (0-30 cm depth) were used to perform a non-supervised clustering, using K-means methodology. The spectral signatures (i.e. average spectrum of each cluster) allowed us to determine the mineralogical composition of the soil clusters and helped with a spectra principal component analysis (PCA). The spatial distribution of these clusters assisted with the development of soil maps. Organic and chemical properties and amorphous compounds allowed the description of soil clusters through their mean, standard deviation and ANOVA values. The spectral clustering identified eight clusters. Spectral signatures analysis showed the presence of three minerals in soils samples due to their reflectance peaks: allophane {amorphous mineral), halloysite and gibbsite {crystallized minerals). Contents of these minerals varied according to soil clusters as seen.on reflectance peaks intensity and PCA and allowed identification and development of a classification system according to a weathering sequence. The cluster spatialization of Reunion soils showed a high degree of fragmentation. This suggests that the high accuracy of NIR spectrometry could be the suitable basis for a soil classification tool. The use of the soils spectral signatures could increase the accuracy of the fertilizer recommendations proposed by SERDAF if it is directly used to classify soil samples in laboratory