In optically complex coastal and eutrophic waters, suspended-particle scattering, colored dissolved organic matter (CDOM), and changes in phytoplankton assemblages can alter the relationship between in vivo fluorescence signals and extracted chlorophyll a (Chl a) reference concentrations. This limits the transferability of Chl a estimates from commercial single-wavelength fluorometers. We developed a framework for estimating Chl a in situ using 15-channel multi-wavelength excitation fluorescence signals measured with a Phytoplankton Optical Detector (POD). The framework combines a multi-wavelength physical baseline with residual correction. The physical baseline was derived from laboratory monoculture samples to represent major group-related bio-optical responses. The residual network was trained with synthetic algae–turbidity samples and field samples to compensate for deviations not explained by the linear baseline. Independent testing showed that the POD-based framework reduced overall bias across the full independent test set. Compared with the multi-wavelength physical baseline, the residual correction model decreased MAPE from 58.13% to 30.36%, reduced mean bias across the full independent test set from 0.5628 to 0.1011 µg L⁻¹, and achieved an R² of 0.9575. Cross-group comparisons showed further error reduction in diatom- and dinoflagellate-dominated natural samples, but the added benefit was limited for the cyanobacteria-dominated mixed samples used in this study. Retaining raw multi-wavelength excitation fluorescence signals, together with a physical baseline and residual correction, improved the stability and interpretability of in situ Chl a estimation. This framework provides a practical strategy for high-frequency Chl a monitoring in optically complex coastal and eutrophic waters. Its transferability to cyanobacterial bloom conditions requires further validation.