Optimal Spatial Resolution Identification in UAV Hyperspectral Imagery for Soil Pb Content Estimation

Soil heavy metal concentrations are strongly influenced by soil physicochemical properties, and both the concentrations of heavy metal pollutants and the physicochemical properties of soils exhibit spatial heterogeneity. Determining an optimal image resolution to capture the spatially varying signals of physicochemical properties and pollutant content to improve the estimation accuracy is the difficulty in monitoring soil heavy metal content with unmanned aerial vehicle (UAV) hyperspectral imagery. In this study, soil pH, soil organic matter (SOM), and soil moisture content (SMC) were incorporated alongside spectral features as predictive factors for modeling soil Pb content, with the optimal combination of these factors being identified. Subsequently, the original hyperspectral images were resampled at 0.6-, 1.0-, 1.4-, and 1.8-m resolutions, and the optimal parameter combination model was used to re-estimate the Pb content and to determine the effect of image resolution on the estimation accuracy. The results showed that: 1) the fusion of spectra and three physicochemical parameters improved the model performance, with the $R^{2}$ of the validation set increasing from 0.51 to 0.58, the root mean square error (RMSE) decreasing from 83.4 to 78.4 mg/kg, and the relative percentage difference (RPD) increasing from 1.63 to 1.73; and 2) the 0.6-m image achieves an optimal balance between spectral information richness and noise suppression, and the $R^{2}$ of the validation set is improved to 0.66, the RMSE is reduced to 61.23 mg/kg, and the RPD is improved to 2.13, which is a small improvement over the 0.2-m resolution model. This study holistically addresses the impacts of spatial resolution variation and soil component heterogeneity on estimation accuracy, establishing both a theoretical basis and technical support for UAV-based monitoring of heavy metal contamination in agricultural soils.