Chlorophyll-a (Chl-a) concentration is one of the most important parameters for the analysis of inland water quality. Remote sensing data with the advantages of wide spatial area and multi-temporal monitoring has been applied as a reliable source of Chl-a concentration. However, as optical characteristics of inland water bodies are complex with high spatial and temporal (diurnal) variations, there are still limitations to estimate Chl-a concentration with traditional remote sensing data and single model. In the proposed solution, the first geostationary ocean color satellite sensor, Geostationary Ocean Color Imager (GOCI), which provides an image per hour (eight images per day from 8:16 to 15:16), was used as a data source of Taihu Lake. Based on hierarchical clustering method, water types were identified from in situ normalized spectral reflectance collected in Taihu Lake (216 samples in different seasons from 2010 to 2012). Then eight GOCI images which were obtained on May 6th, 2012 were classified separately according to different water types by calculating spectral angle distance between each spectrum in GOCI images and the classified spectra. According to the classified remote sensing images and the spectral bands of GOCI data, classed-based models were subsequently developed for the estimation of Chl-a concentration. The results indicated that four water types (Type 1 to Type 4) were identified based on the in situ normalized spectral reflectance in Taihu Lake. The spectra of Type 1 mainly represented the characteristics of floating algae. This type had little significance to in estimating Chl-a concentration because sensors could only receive signal of floating algae. Then Type 1 was usually used as the evidence of algal blooms. Meanwhile, two-band semi-analytical algorithms were established for Type 2—Type 4 waters which were separately dominated by Chl-a concentration, high suspended solid, low Chl-a and low suspended solid. Comparing with the two-band algorithms, band 7 and band 6 combination was more suitable for Type 2 and Type 3 while the correlation between Chl-a concentration and b7/b5 was higher than that between b7/b6 for Type 4. The accuracies of classification models (Type 2—Type 4) were higher than that of the overall model, with the reduced average relative errors of 7%, 12.3% and 15.9%, respectively. Moreover, the inversion results of GOCI data not only reflected the spatial distribution of Chl-a, but also showed the diurnal variation of the Chl-a concentration of Taihu Lake. This study has demonstrated great potential for dynamic monitoring of eutrophication pollution with GOCI data. In addition, the results suggested that optical classification algorithm can improve the accuracy of Chl-a concentration and the application performance of semi-analytical model. GOCI data and the class-based algorithm provide a basis for accurate estimation of diurnal and spatial variation of Chl-a concentration.