We employ a multi scale approach combining experiments and modelling to elucidate the impacts of small scale (sub-seismic resolution <10m) capillary pressure heterogeneities in field scale CO2 flow and trapping. We analyse 48 rock cores (~3cm length, 4cm diameter) covering the entire 100m interval of the Captain D sandstone in the Goldeneye field, UK North Sea. We experimentally measure porosity, capillary pressure, absolute permeability, relative permeability and trapping characteristics for the cores, which are used to create 3D numerical models with heterogeneities defined at the mm scale. These models are validated by predicting experimental X-Ray CT observations of saturation (mm scale) and pressure measurements (cm scale) at various flow rates and fractional flows of gas-water. The intrinsic, core scale properties are then used to populate 2D meso scale numerical simulations (50m x 10m size), with heterogeneities defined at cm scale using a geostatistical representation. We vary the correlation length and variance of the fields within the bounds of the experimental observations to investigate the impacts of small scale heterogeneity on vertical and lateral CO2 plume migration under different Capillary, Bond and Gravity numbers. At low flow potential, layered capillary pressure heterogeneities can speed up lateral plume migration by up to 20%, with gravitational segregation significantly enhancing the migration (See Fig. 1). In the vertical case, layered heterogeneities can significantly increase CO2 trapping, which is further enhanced with the inclusion of capillary pressure hysteresis. Finally, we derive capillary limit, upscaled equivalent properties from the meso scale simulations, which incorporate the impacts of small scale heterogeneity. These are used to represent the grid block properties in a full field 3D numerical model of the Goldeneye field (lateral ~5 km, depth 200m). We analyse the impact of varying relative permeabilities (with anisotropy) on the field scale plume migration and trapping. We see large differences compared to cases using intrinsic rock properties, indicating that proper inclusion of small scale heterogeneity is needed in upscaling workflows when predicting and assessing uncertainty in low flow potential CO2 plume migration at the field scale.