Monitoring soil salinization in Arid cotton fields using Unmanned Aerial Vehicle hyperspectral imagery

• Fractional-order differentiation enhances contrast between soil and cotton canopy spectra. • 2D/3D indices with cross-band info enhance correlation and spatial expression of soil salinity. • PSO-GPR model accurately predicts soil salinity during cotton emergence and squaring stages. • UAV hyperspectral imagery enables accurate mapping of soil salinity spatial variation in cotton fields. Soil salinization is one of the main causes of low yield in cotton fields, leading to soil compaction and significantly affecting crop growth and nutrient absorption. There is an urgent need for rapid, non-destructive, and efficient methods to extract soil salinity spatial distribution information under cover conditions. These methods are essential for optimizing irrigation techniques, controlling salinization, and promoting sustainable agriculture. To address these issues, this study proposes a method combining unmanned aerial vehicle (UAV) hyperspectral remote sensing data with a particle swarm optimization Gaussian process regression (PSO-GPR) model. Fractional-order differentiation (FOD) technology was used for spectral preprocessing, combined with the random frog algorithm (RF), uninformative variable elimination (UVE), and bootstrap soft shrinkage (BOSS) selection algorithms to optimize one-dimensional spectral bands. By integrating it with the optimal two-dimensional and three-dimensional indices, a PSO-GPR model will be developed to accurately predict soil salinity during the cotton emergence and squaring periods. The results indicate: (1) SG smoothing and FOD technology effectively enhanced the spectral characteristics of cotton field soils and crop canopies. (2) The soil salinity prediction model constructed using 0.7-order FOD combined with the BOSS algorithm yields the best performance, with an R 2 of 0.92, RMSE of 0.15 dS m −1 , and RPD of 3.54 on the test set. (3) The soil salinity distribution map generated from the optimal model clearly revealed the spatial distribution characteristics of salinity in the 0–10 cm topsoil layer, with significant differences observed under different cover conditions. This study presents a novel method for rapidly and accurately extracting soil salinity information in covered cotton fields. It supports scientific irrigation and salinity management, mitigates crop yield losses caused by salt accumulation, fosters sustainable development of the cotton industry, and provides a scientific foundation for soil salinity monitoring and management in precision agriculture.