Collaborative correction for the coupled influence of soil particle size and moisture content on total nitrogen prediction using visible-near infrared spectroscopy

The rapid, in-situ prediction of soil total nitrogen (STN) using visible-near infrared (VIS-NIR) spectroscopy is critical for precision agriculture. However, it remains challenging due to interference from soil particle size and moisture. This study investigated their coupled interference on spectral STN prediction and developed a strategy for their simultaneous correction. A total of 720 soil samples with four particle sizes (0.3, 0.45, 0.9, and 2 mm) and three moisture levels (2%, 6.5%, and 11%) were analyzed using spectral measurements and chemically determined STN values. Under low moisture (2%), scattering dominated and finer particle size improved prediction accuracy. At intermediate moisture (6.5%), absorption and scattering interacted, which diminished the effect of particle size. Under high moisture (11%), strong absorption masked spectral differences due to particle size. To correct the interference of particle size and moisture simultaneously, a weighted fusion method encoded sample states, and variable importance in the projection (VIP) analysis identified ten critical interference bands. A weighted average correction model was applied to correct the bands exhibiting significant coupling interference, producing a corrected full-spectrum. A successive projection algorithm (SPA) was used to select STN-related wavelengths, followed by the development of partial least squares regression (PLSR) and one-dimensional convolutional neural network (1D-CNN) models. The collaborative correction consistently outperformed uncorrected spectra and single-factor corrections in both models. These results demonstrate its potential for improving spectroscopic STN prediction and provide a basis for future validation under more diverse soil conditions.