Satellite thermal images are frequently used in the study of surface heat exchange dynamics over lakes. Such investigations can depict large-scale thermal patterns with a resolution of O(1 km2), but not meso- or small-scale processes, requiring O(1 m) and O(1 cm) resolution. Using an airborne platform, the effect of meso-scale spatial heterogeneity of Lake Surface Water Temperature (LSWT) on the surface cooling estimation at the sub-pixel satellite scale O(1 km2) was investigated in Lake Geneva. The measurement system consists of a tethered balloon launched imaging and monitoring platform, equipped with an uncooled infrared camera that records the LSWT, and an autonomous catamaran that simultaneously measures in situ surface/near surface temperatures. Cold season data did not show significant LSWT heterogeneity, and hence surface cooling spatial variability. However, based on three selected daytime meso-scale maps, a LSWT contrast of > 2°C in spring and > 3.5°C in summer is observed, corresponding to the spatial surface cooling range of > 20 Wm-2 and > 40 Wm-2, respectively. Due to the nonlinear relationship between turbulent surface heat fluxes and LSWT, negatively skewed LSWT spatial distributions resulted in negatively skewed surface cooling patterns under very stable or predominantly unstable Atmospheric Boundary Layer (ABL) conditions and positively skewed surface cooling patterns under predominantly stable ABL conditions. Implementing a mean spatial filter, the effect of the area-averaged LSWT on the surface cooling estimation of a typical satellite pixel was assessed. The effect of the averaging filter size on the mean spatial surface cooling values was negligible, except for the predominantly stable ABL conditions when a reduction of ~3.5 Wm-2 from high O(1 m) to low O(1 km) pixel resolution was obtained. The results revealed that the bias in the sampling of meteorological parameters, particularly wind speed, could affect both the mean and the range of spatial surface cooling. Errors in air temperature may also alter the surface cooling distribution from negatively to positively skewed, and hence affect the area-averaged estimates.