DRS-based PLSR Model for Predicting Soil Organic Matter under Different Moisture Conditions in Saline and Non-saline Paddy Soils

Soil organic matter (SOM) significantly influences crop growth and productivity through nutrient retention, water holding capacity, and microbial activity. Traditional SOM analysis methods involve labor-intensive sampling and chemical analysis, limiting rapid field assessments. To overcome this, diffuse reflectance spectroscopy (DRS) is increasingly used as a non-destructive technique, providing rapid estimations of SOM using spectral data, particularly effective at 500-600 nm and 2100-2300 nm. However, under field conditions, strong moisture absorption peaks around 1400 nm and 1900 nm distort spectral signals, reducing prediction accuracy and necessitating effective moisture correction strategies. This study analyzed 210 paddy soil samples from Korea, obtaining spectral measurements (350-2500 nm) in both wet and dried states using an ASD Fieldspec4 spectrometer, excluding noisy wavelengths (350-399 nm, 2451-2500 nm). Partial least squares regression (PLSR) models evaluated with 5-fold cross-validation (R², RMSE, RPD) showed slightly improved accuracy for dried soils (R²=0.58, RMSE=4.17, RPD=1.55) compared to wet soils (R²=0.56, RMSE=4.30, RPD=1.50). Drying minimized moisture-related spectral distortions, enhancing clarity of SOM-specific wavelengths. Results highlight the critical need for quantifying moisture effects and employing appropriate preprocessing to improve the reliability of DRS-based models for rapid field-scale SOM assessment.