Mock Observations: Morphological Analysis of Galaxies in TNG100 Simulations

In this study, we investigate the morphology of galaxies in the TNG100 simulation by applying mock observation techniques and compare the results with the observational data from the Sloan Digital Sky Survey (SDSS). By employing a hierarchical Convolutional Neural Network (CNN) approach, we classify galaxies into four morphological categories (Ellipticals, S0/a, Sab/Sb, and Sc/Sd/Irregulars). Our findings show that the morphological characteristics of the mock-observed galaxy samples closely match those observed in the SDSS, successfully reproducing key features such as distinct parameter distributions for different types. However, some discrepancies are identified: notably, a significant lack of early-type galaxies (ETGs) in the dwarf galaxy regime ($M_* < 10^{10} M_{\odot}$) and minimal distinction between Sab/Sb and Sc/Sd/Irregular galaxies in the mock-observed samples, unlike the clear differences seen in actual observations. These divergences may stem from simulation properties such as elevated star formation efficiency at low mass end or resolution limits. Observational effects, including the impact of the Point Spread Function, sky background, and instrumental noise, can independently cause approximately 7.87\% morphological misclassifications by our CNN model. Compared to previous studies using gravity-based definitions of galaxies that failed to clearly distinguish the parameter distributions of ETGs versus Late-Type Galaxies, our brightness-based sample definition method better recovers the observed morphological parameter distributions, especially their distinct differences. Our study underscores that, alongside mock observations, employing galaxy segmentation methods consistent with observational practices is crucial for appropriately recovering realistic morphological parameters from simulations and enabling fair comparisons.