Mechanism-based clustering reveals distinct controls on phytoplankton community composition across Korean marginal...

Phytoplankton regulate the marine food web, biogeochemical processes, and climate-relevant carbon export to depth in the oceans, but the mechanisms controlling their distributions vary across space and time, limiting inference from pooled analyses. We developed a mechanism-based clustering framework that groups stations by covariate-community response fingerprints inferred from a multivariate generalized linear mixed model (GLMM). Station-specific random effects across major taxa were treated as fingerprints and clustered to define mechanism-consistent groups; Geostationary Ocean Color Imager (GOCI) satellite time series were analyzed with dynamic factor analysis (DFA) to provide complementary dynamical context. Applied to spring observations in the Korean marginal seas, GLMM-fingerprint clustering identified four clusters and increased redundancy analysis explanatory power relative to both the pooled data set (adjusted R² = 0.221) and a composition-based clustering baseline. Phytoplankton community composition was most closely related to stratification in the Yellow Sea, mesoscale eddy/upwelling forcing in the East Sea, and contrasting warm-current versus freshwater influences in the South Sea. The Yellow Sea cluster showed strong coherence with stratification-related controls (adjusted R² = 0.847). Explanatory power remained lower in the East Sea cluster (adjusted R² = 0.272), consistent with mesoscale forcing that is difficult to capture using field covariates alone, whereas DFA-based satellite diagnostics supported eddy/upwelling-related variability. The South Sea separated into two clusters (adjusted R² = 0.492 and 0.424), reflecting contrasting warm-current versus freshwater influences. This framework offers an interpretable pathway for mechanism-oriented regionalization in dynamically complex coastal seas.