The vertically generalized production model (VGPM), which was designed for open ocean waters (Behrenfeld and Falkowski, 1997a; henceforth BF), was evaluated using in situ measurements of primary productivity (PP) in the characteristically turbid coastal waters of Ariake Bay, southwestern Japan, to develop a regionally modified version of the model. The euphotic depth (Z eu )-integrated PP (IPP) calculated from the VGPM using in situ chlorophyll a (Chl a) and sea surface temperature (SST) was significantly overestimated (by factors of 2–3), but 52% of the observed variability was explained. The weak correlation could have partially resulted from overestimations by the sub-models embedded in the original VGPM model for estimation of Z eu (Morel and Berthon, 1989; henceforth MB) and the optimal Chl a-normalized PP (poptB). The sub-model estimates of poptB and Z eu with in situ poptB and Z eu showed significant improvement, accounting for 84% of the variability and causing less overestimation. Z eu was the most important parameter influencing the modeled IPP variation in Ariake Bay. Previous research suggested that the Z eu model, which was based on surface Chl a, overestimated in situ Z eu by a factor of 2–3, resulting in weak correlation between the modeled and in situ IPP. The Z eu sub-model was not accurate in the present study area because it was basically developed for clear (case 1) waters. A better estimation of Z eu could be obtained from the in situ remote sensing reflectance (R rs ) using a quasi-analytical algorithm (QAA) in this turbid water ecosystem. Among the parameters of PP models, poptB is conventionally considered the most important. However, in this study poptB was of secondary importance because the contribution of poptB to the variation in modeled IPP was less than the contribution of Z eu . The modeled and in situ poptB were weakly correlated with 50% of the data points that overestimated the in situ values. The estimation of Chl a was improved by optimizing the Chl a algorithm with in situ R rs data. Incorporating the QAA-based Z eu and the optimized Chl a and constant (median) poptB value led to improved performance of the VGPM for the study area. Thus, even though the VGPM is a global open ocean model, when coupled with turbid water algorithms for Z eu and Chl a and constant (median) poptB, it provided realistic estimates of IPP in the turbid water ecosystem of Ariake Bay.