The eutrophication of lakes causes serious problems, such as water-bloom consisting of the harmful blue-green algae. To prevent an unusual degree of growth of phytoplankton, various restoration methods have been tried. “Biomanipulation”, which uses zooplankton to prey on the phytoplankton, is a promising method because it can, in principle, create a desirable ecosystem. However, there are many equivocal points in the long-term stability of a newly developed ecosystem after biomanipulations. The fate prediction of a water ecosystem using a lake mathematical model is thus essential for implementing biomanipulations. There remain, however, many uncertain points in the dynamic parameters of the lake models, especially the parameters concerned with zooplankton and fish. In this study, we established a lake mathematical model based on the carbon, nitrogen and phosphorus flows among the biotic and abiotic substances in a water ecosystem. In this model, two types of phytoplankton (non blue-green algae and blue-green algae), three types of zooplankton (rotatoria, cladocera and copepoda), planktivorous fish (crucian carp), detritus and dissolved matters were considered. The dynamic parameters were estimated with a set membership estimation algorithm using random scanning based on mesocosm experiments, which can provide more rigid data than actual lake observations can. The model calculations showed good agreement with the observation data for both the non-filamentous blue-green algae dominant case and the filamentous blue-green algae dominant case. Using this model, we calculated the long-term stability of the effects of biomanipulations in a hypothetical lake. The calculation results clarified the blue-green algal species specificity in the long-term effects of biomanipulations. In the non-filamentous blue-green algal dominant case, the long-term stable suppression of algal growth was predicted under a certain threshold fish biomass, while in the filamentous blue-green algal dominant case, the suppression was not sustainable even in a no-fish-coexisting series. Model simulation is very important to the design of biomanipulations, and the model proposed in this study is suitable for this purpose.