Surface roughness controls the mechanical performance and durability (e.g., wear and corrosion resistance) of laser powder bed fusion (LPBF) components. The evolution mechanisms of surface roughness during LPBF are not well understood due to a lack of in situ characterisation methods. Here, we quantified key processes and defect dynamics using synchrotron X-ray imaging and ex situ optical imaging and explained the evolution mechanisms of side-skin and top-skin roughness during multi-layer LPBF of Ti-6AI-4V (where down-skin roughness was out of the project scope). We found that the average surface roughness alone is not an accurate representation of surface topology of an LPBF component and that the surface topology is multimodal (e.g., containing both roughness and waviness) and multiscale (e.g., from 25 µm sintered powder features to 250 µm molten pool wavelength). Both roughness and topology are significantly affected by the formation of pre-layer humping, spatter, and rippling defects. We developed a surface topology matrix that accurately describes surface features by combining 8 different metrics: average roughness, root mean square roughness, maximum profile peak height, maximum profile valley height, mean height, mean width, skewness, and melt pool size ratio. This matrix provides a guide to determine the appropriate linear energy density to achieve the optimum surface finish of Ti-6AI-4V thin-wall builds. This work lays a foundation for surface texture control which is critical for build design, metrology, and performance in LPBF.