MAPPING PARTICLE SIZE AND SOIL ORGANIC MATTER IN TROPICAL SOILS USING THE AISAFENIX HYPERSPECTRAL SENSOR

Brazil has a wide range of soils, which result from varying topography, climate, biota, and parent materials. Agriculture is a key economic driver, making soil classification and management essential. However, detailed soil maps are scarce, covering only 0.25–1% of Brazil at appropriate scales. Traditional mapping requires extensive fieldwork, but hyperspectral sensors offer a cost-effective alternative for soil property estimation. Digital soil mapping (DSM) enhances efficiency but must align with measured attributes. Clay content and soil organic matter (SOM) are crucial for fertility, influencing ion exchange, water retention, and soil structure. Most cultivated soils in Brazil, primarily Oxisols and Ultisols, have low natural fertility due to weathering. Conventional laboratory analyses for soil properties are accurate but costly and time-consuming. Remote sensing, particularly Vis–NIR–SWIR spectroscopy, enables rapid, non-destructive soil assessment. In this study, we evaluated the use of airborne hyperspectral imaging in the Vis–NIR–SWIR spectral range to assess particle size and soil organic matter (SOM) in the surface layer of tropical soils (Oxisols, Ultisols, and Entisols). The study was conducted in a 135-hectare sugarcane cultivation area near Piracicaba, São Paulo, Brazil. Hyperspectral imaging was performed in April 2016 using the AisaFENIX airborne sensor, which has a spectral resolution of 3.5 nm between 380 and 970 nm and 12 nm between 970 and 2500 nm. A total of 66 surface soil samples were collected and analyzed for particle size distribution and SOM content. To model the relationship between spectral data and soil properties, Partial Least Squares Regression (PLSR) was applied to predict clay, silt, sand, and SOM. The resulting PLSR models were used to generate prediction maps for clay, sand, and SOM based on the hyperspectral imagery. The imaging sensor demonstrated strong predictive capability in the cross-validation step, achieving R² values of 0.62, 0.66, and 0.67 for clay, sand, and SOM, respectively. However, silt content was not well predicted. A Pearson correlation greater than 0.849 was observed between measured and predicted values for clay, sand, and SOM. Overall, our study successfully generated large-scale, high-resolution soil property maps using hyperspectral imaging, highlighting its potential for improved soil management in tropical regions.