Quantification of aboveground biomass (AGB) from local and regional to global scales is significant for sustainable ecosystem management, and object-based analysis technique is attractive in AGB studies. This study extended the AGB inversion researches on short vegetation species at a relatively fine scale using a homogenous vegetation response unit (HVRU)-oriented analysis approach, and explored a convenient scheme to determine diagnostic indicators. In the meantime, spectral, structural and geographic indicators related to AGB were derived from multisource data to generate HVRUs using multiresolution segmentation technology; the utilities of different variable types, regression algorithms and research scales were further evaluated in HVRU-based AGB modeling to determine the diagnostic indicators. Results showed that, inversion accuracies based on mean variables were much higher than those based on texture variables. Partial least squares regression (PLSR) and support vector regression (SVR) provided similar accuracies, in general, SVR performed better in digesting texture variables and time consumption, while PLSR was more suitable for mean variables. Compared to spectral indicators alone, the introduction of non-spectral indicators weakened accuracies at the scale of 50; at the scale of 100, 150 and 200, integrating spectral and structural indicators significantly improved accuracies on the whole; additional geographic indicators showed adverse impacts in most cases. The optimal inversion accuracy with R²_cv of 0.83, RMSE_cv of 0.20, RPD of 2.34 was realized based on the mean variables of spectral and structural indicators at the scale of 150 using PLSR, furthermore, FVC, OSAVI and (1/R_742nm)'' contributed much more than other indicators in modeling. Effective indicators excavated in this study could be directly applied for AGB survey in Yancheng National Nature Reserve and similar areas in China’s coastal mudflat. The HVRU-based analysis approach fusing the advantages of multisource data and multiresolution segmentation technology has broad application prospects especially for research objects which are hard to quantify.