δ13CPOM ~ glm[f(class) + f(layer) + Lat + T + S + Chl + Nitrate] (3). The r2 value of the least AIC δ13CPOM model was found to be 0.659. ANOVA indicated that all the remaining explanatory variables were significant (χ2 ≥ 7.53, p ≤ 0.006). The responses of latitude, salinity, and nitrate concentration were significantly negative (p < 0.001), whereas those of temperature and chlorophyll-a concentration were significantly positive (p < 0.001) (Fig. 8). The lsmean δ13CPOM with ANOVA suggested that classes III (lsmean ± SE: –25.6 ± 0.25‰) and IV (–20.6 ± 0.19‰) were significantly higher and lower than those of other classes, respectively (pair-wise test with Tukey’s adjacent, p < 0.001). The difference between classes I (–23.0 ± 0.36‰) and II (–23.4 ± 0.10‰) was insignificant (pair-wise test with Tukey’s adjacent, p = 0.542) (Fig. 8a). The δ13CPOM of the surface (lsmean ± SE: –23.4 ± 0.11‰) was higher than that of the subsurface (–23.7 ± 0.13‰, pair-wise test with Tukey’s adjacent, p = 0.0154). Moreover, δ13CPOM at the 100-m depth (–22.4 ± 0.34‰) was significantly higher than those of the surface and subsurface (p ≤ 0.001) (Fig. 8).