Abstract

High-resolution soil texture mapping is critical for precision agriculture, soil management, and environmental monitoring, yet traditional field sampling methods are time-intensive and spatially limited. This study evaluates airborne hyperspectral imaging for detailed soil texture characterization and rock fragment analysis across 45 study sites covering 12,340 hectares in diverse agricultural and semi-natural landscapes. We employed the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS-NG) with 5-meter spatial resolution and 425 spectral bands (380-2510 nm) to map clay, silt, sand fractions and rock fragment content. Machine learning algorithms including partial least squares regression (PLSR), support vector regression (SVR), and random forest (RF) were calibrated using 2,847 field samples analyzed through laser diffraction and sieving methods. Results demonstrate exceptional accuracy for clay content prediction (R² = 0.89, RMSE = 3.2%), moderate accuracy for sand fraction (R² = 0.76, RMSE = 8.7%), and good performance for silt estimation (R² = 0.72, RMSE = 6.1%). Rock fragment detection achieved 91% classification accuracy with fragments >2 cm diameter reliably identified. Spectral feature analysis revealed clay absorption features at 2200 nm and 2350 nm as primary predictors, while sand content correlated with visible/near-infrared reflectance patterns. Iron oxide absorption (870 nm) and carbonate features (2340 nm) provided additional texture discrimination. Spatial analysis demonstrated high-resolution mapping capability with texture boundaries detected at 10-meter precision, enabling identification of field-scale variability patterns. Validation across soil types showed consistent performance in Mollisols (R² = 0.87) and Alfisols (R² = 0.84), with reduced accuracy in Vertisols (R² = 0.69) due to complex clay mineralogy. Economic analysis indicates cost reduction of 78% compared to conventional grid sampling while providing complete spatial coverage. However, vegetation cover >40% significantly reduced accuracy, and atmospheric conditions affected spectral quality. These findings demonstrate that airborne hyperspectral remote sensing enables operational high-resolution soil texture mapping with substantial advantages for precision agriculture applications, soil carbon modeling, and land management optimization.