This study presents a comprehensive validation of the Geostationary Environmental Monitoring Spectrometer (GEMS) v3.0 tropospheric NO2 vertical column density (TrVCD) product over Ulsan, South Korea. The evaluation is based on three years of Pandora TrVCDs retrieved from the sky-scan mode measurements at UNIST site (August 2021-July 2024) and two MAX-DOAS instruments operated near emission sources during and after the SIJAQ (Satellite Integrated Joint monitoring of Air Quality) 2022 campaign. Compared with Pandora, GEMS showed fair agreement with a slope of 0.74, a correlation coefficient of 0.55, and a mean relative difference of +46\%, generally overestimating at low NO2 levels but underestimating under high-NO2 conditions (\>1 × 1016 molec. cm−2). To account for spatial representativeness mismatches between GEMS and Pandora, we introduced the Tropospheric Dominance Index (TDI), defined as the ratio between TrVCDs from sky-scan and direct-sun observations. The high-TDI cases (TDI \> 85th percentile) exhibited a pronounced negative bias between GEMS and Pandora (MD = –2.67 × 1015 molec. cm−2, MRD = –13.9\%), indicating that they were generally associated with GEMS underestimation in high-NO2 conditions. Seasonal analyses of diurnal variation at UNIST revealed that GEMS exhibited a systematic positive bias but captured similar daily patterns to Pandora in autumn and winter, whereas in spring and summer, the diurnal patterns were less consistent with Pandora. These discrepancies were linked to near-surface advection, with distinctly lower agreement under polluted inflow conditions. Comparisons with MAX-DOAS near emission sources showed lower agreements, with a slope of 0.81 and a correlation coefficient of 0.43, but agreement improved under clean inflow conditions (slope = 1.06, r = 0.59). These results highlight the importance of considering site characteristics and transport conditions in satellite validation and demonstrate the potential of TDI as an auxiliary indicator to identify cases of high NO2 inhomogeneity and improve future validation strategies for geostationary satellite products.