Leaf sample preparation shapes NPK prediction accuracy from Vis-NIR-SWIR spectra in sugarcane

• Leaf sample preparation affects NPK prediction using Vis-NIR-SWIR spectra. • Fresh leaves provide better predictions for phosphorus and potassium contents. • Dried leaves improve nitrogen estimation by reducing water spectral interference. • Nonlinear RF and SVR outperformed linear PLS across all nutrients. • Identified red-edge and SWIR regions support model interpretability and field use. Leaf sample preparation is a critical but often overlooked factor influencing the predictive performance of spectral models for plant nutrient analysis. This study investigated the impact of fresh versus dried and ground leaf preparation on the accuracy of nitrogen (N), phosphorus (P), and potassium (K) predictions in sugarcane ( Saccharum officinarum L.) using Vis-NIR-SWIR spectroscopy (400–2400 nm) coupled with Partial Least Squares Regression (PLS), Support Vector Regression (SVR), and Random Forest (RF) algorithms. Spectral data were acquired under both conditions using an ASD FieldSpec 3 spectroradiometer, followed by preprocessing with Standard Normal Variate (SNV) and Savitzky–Golay derivatives. Results showed that model performance varied according to nutrient and preprocessing combination. RF and SVR yielded the highest predictive accuracies, while PLS exhibited consistent but lower performance due to its linear nature. Drying and grinding improved N prediction across all modeling approaches (RF, SVR, and PLS) by reducing water-induced spectral interference, achieving R 2 p > 0.67 and RPD p > 1.76, whereas fresh leaves produced superior accuracy for P and K, with R 2 p ≈ 0.74 and RPD p ≈ 2.00 for P (RF and SVR models), and R 2 p = 0.84; RPD p = 2.54–2.57 for K (RF and SVR models). The most relevant spectral regions included the red-edge (680–720 nm) and SWIR bands near 1450 and 1950 nm, associated with chlorophyll and water absorption features. These findings underscore the need to tailor sample preparation protocols to the nutrient of interest and confirm the feasibility of non-destructive, field based spectral assessment for rapid NPK monitoring in sugarcane. The study provides a methodological foundation for advancing precision agriculture through proximal sensing technologies.