Improving urban topsoil copper (Cu) inversion in highly industrialized areas using compositionally nearby samples

Monitoring topsoil heavy metals in highly industrialized areas requires fast, low-cost, and environmentally friendly methods such as visible and near infrared (vis-NIR) spectral inversion. However, the influence of compositionally nearby samples on heavy metal inversion remains unclear. This study investigates how compositionally nearby samples affect the inversion of heavy metals-specifically, Cu-and explores the underlying mechanisms. We conducted experiments by selecting compositionally nearby samples to develop Cu estimation models using partial least squares regression. To thoroughly examine the effect, we gradually added different numbers of nearby samples to the model. Results show that compared with the traditional method (R_p² = 0.75, RMSEP = 8.56 mg kg^-1, RPD = 1.83), adding nearby samples yielded a significant enhancement in model performance (R_p² = 0.96, RMSEP = 3.21 mg kg^-1, RPD = 4.87). Increasing the number of nearby samples caused the model performance to first improve and then deteriorate, showing an inverted U-shaped trend. An optimal value of 125 nearby samples (approximately 62.5% of all samples) was identified. These findings suggest that including compositionally nearby sample can remarkably improve Cu inversion accuracy; however, the sample number must be carefully controlled to balance sample quantity and model complexity. This study proposes a novel approach to enhance heavy metal inversion and offers promising potential for effective soil monitoring in highly industrialized areas.