Surface-enhanced Raman scattering (SERS) enables ultrasensitive molecular detection, yet systematic variations in spectral shape (not just signal intensity) often limit reproducibility and interpretation. Here, we present a combined experimental and chemometric study that elucidates how substrate optical response and molecular adsorption orientation jointly govern SERS spectral variability. Using 1,2-bis(4-pyridyl) ethylene (BPE) on oblique-angle-deposited silver nanorod (AgNR) substrates as a model system, we constructed a comprehensive spectral data set spanning six controlled experimental conditions, including defect mapping, batch-to-batch variation, nanorod length tuning, concentration-dependent drop-casting, static immersion, and real-time immersion measurements. Hierarchical cluster analysis (HCA) partitions the spectra into seven reproducible clusters with distinct average spectral shapes, separating low- and high-signal-to-noise regimes and revealing systematic evolution of relative intensities among the five characteristic BPE modes. By correlating cluster membership with experimental metadata, we show that specific spectral shapes are strongly associated with defined physical conditions, including surface defects, nanorod geometry, analyte concentration, and adsorption dynamics. To interpret the cluster-dependent spectral shapes, we introduce intensity web plots under three normalization strategies that isolate different governing physics: absolute intensities emphasize overall electromagnetic enhancement and analyte coverage; normalization by the sum of the five peak intensities suppresses global scaling and highlights substrate-dependent optical reweighting of Raman bands; and normalization to the 1015 cm–1 mode provides an internal reference that accentuates orientation-selective enhancement. Together, these results establish a physics-informed, data-driven framework for better understanding the origins of SERS spectral shape changes under complex experimental conditions.
Kumar et al. (Tue,) studied this question.