Secchi disk depth (SDD) is a widely critical indicator of water transparency. However, existing retrieval models often suffer from limited transferability and biased predictions when applied to optically diverse waters. Here, we compiled a dataset of 6218 paired in situ SDD and remote sensing reflectance (R_rs) measurements to evaluate model generalization. We benchmarked nine machine learning (ML) models (RF, KNN, SVM, XGB, LGBM, CAT, RealMLP, BNN-MCD, and MDN) under three validation scenarios with progressively decreasing training-test overlap: Random, Waterbody, and Cross-Optical Water Type (Cross-OWT). Furthermore, SHAP analysis was employed to interpret feature contributions and relate model behaviors to optical properties. Results revealed a distinct scenario-dependent generalization gradient. Random splits yielded minimal bias. In contrast, Waterbody transfer consistently shifted predictions toward underestimation (SSPB: −16.9% to −3.8%). Notably, Cross-OWT extrapolation caused significant error inflation and a bias reversal toward overestimation (SSPB: 10.7% to 88.6%). Among all models, the Mixture Density Network (MDN) demonstrated superior robustness with the lowest overestimation (SSPB = 10.7%) under the Cross-OWT scenario. SHAP interpretation indicated that engineered indices, particularly NSMI, functioned as regime separators, with substantial shifts in feature attribution occurring at NSMI values between 0.4 and 0.6. Accordingly, feature sensitivity analysis showed that removing band ratios and indices improved Cross-OWT robustness for several classical ML models. For instance, KNN exhibited a significant reduction in Median Symmetric Accuracy (MdSA) from 96% to 40% after feature reduction. These findings highlight that model applicability must be evaluated under scenario-specific conditions, and feature engineering strategies require rigorous testing against optical regime shifts to ensure generalization.

Coastal waters exhibit pronounced spatial and temporal variability in optical properties and water quality, which cannot be fully characterized by observations from discrete monitoring stations. Satellite Optical Water Type (OWT) classification provides spatially continuous information on optical conditions. However, comparisons with water quality observations have generally relied on the assigned OWT, and it remains unclear whether membership provides additional information within the same class. This study examined the relationships of assigned OWTs and membership with monthly water quality conditions in Guangdong coastal waters. Sentinel-3 OLCI observations acquired from 2017 to 2024 were matched with station observations of dissolved inorganic nitrogen (DIN), dissolved inorganic phosphorus (DIP), chemical oxygen demand (COD), dissolved oxygen (DO), and eutrophication classes. The analysis first compared water quality among assigned OWTs and then evaluated the secondary OWT and membership values within OWT 4a, the class with the largest sample size. Assigned OWTs distinguished broad differences in DIN, DIP, COD, and eutrophication classes, whereas DO did not differ significantly among OWTs. Within OWT 4a, the class associated with the second highest membership separated records with different nutrient, COD, and eutrophication conditions. Continuous membership values provided further information, but their relationships with water quality depended on the secondary OWT and were most consistent for DIN. OWT occurrence and station eutrophication patterns showed partial spatial correspondence, particularly around the Pearl River Estuary and Huangmao Bay. These findings show that assigned OWTs identify broad water quality differences, while secondary OWTs and membership values distinguish additional variation within OWT 4a. The mapped optical patterns provide complementary spatial context for interpreting station-based water quality observations.