Localized variants of multiplicative scatter correction for enhancing visible and near infrared spectral modelling of chlorophyll in sugarcane

Abstract

Conventional multiplicative scatter correction (MSC) assumes that inter‐sample variability can be modelled as a uniform multiplicative effect across the full spectrum, a constraint that limits its effectiveness for broad ultraviolet-visible and near infrared (UV-Vis–NIR) spectroscopy of heterogeneous plant materials. In sugarcane leaves, where localized spectral distortions arise from variable pigment distributions and tissue structures, such a global approach can leave residual baseline and slope artifacts, reducing predictive accuracy. As a proposed alternative, this study addresses this limitation by comparing and evaluating two localized MSC variants, namely piecewise MSC (PMSC) and segmented MSC (SMSC), for total chlorophyll content estimation. UV-Vis–NIR reflectance spectra (200–1400 nm) were collected from 166 sugarcane leaf samples during tillering stages and the corresponding chlorophyll contents were measured. The MSC, PMSC and SMSC corrected spectra were used to establish partial least squares (PLS) regression models on both the full spectrum and reduced spectra derived from five established wavelength selection approaches. Experimental results showed that across all preprocessing–selection combinations, PMSC and SMSC consistently outperformed MSC. Particularly, the PLS regression model employing PMSC preprocessing in combination with competitive adaptive reweighted sampling wavelength selection achieved the highest predictive performance (R²_cal = 0.81, r²_pred = 0.72, RMSEC = 0.15 mg g^-1, RMSEP = 0.17 mg g^-1), whereas SMSC exhibited a modest accuracy reduction of 2–19% while achieving an approximately 28-fold decrease in computational time. These findings demonstrate that localized scatter‐correction strategies can enhance predictive modelling accuracy, offering a robust alternative to conventional MSC for agronomic trait estimation.