The study evaluated whether simple site-specific empirical corrections can improve the agreement between Landsat 8 and ECOSTRESS thermal products and ground-based surface temperature logger observations acquired in Kraków, Poland. Three field campaigns conducted in 2023–2024 were paired with Landsat 8 and, when available, ECOSTRESS acquisitions. Two correction approaches, linear regression and additive bias correction, were tested for eight calibration subset configurations representing different urban surface contexts, including paved surfaces, urban parks, heterogeneous urban sites, waterfront locations, spatially dispersed sites, vegetation-covered sites, and a subset defined by low initial mismatch. Performance was assessed using RMSE and the standard deviation of residuals relative to the reference measurements. Both correction approaches reduced disagreement relative to the uncorrected data, but the magnitude of improvement depended strongly on acquisition date, product type, and calibration subset composition. The results show that simple local correction can increase the practical utility of satellite thermal data in a heterogeneous urban setting; however, the findings should be interpreted with caution because contact-based logger measurements are not strictly equivalent to radiometric land surface temperature, and some satellite ground pairs were not fully synchronous.