Micro- and nanoscale surface roughness within subsurface rocks is ubiquitous due to weathering and diagenesis. To assess the impacts of surface roughness on non-wetting gas flooding operations, we conducted immiscible drainage experiments in glass micromodels representing single-scale rock matrices with identical pore topology but varying degrees of surface roughness. The experiments were performed in capillary-dominated regimes where air was the non-wetting phase and crude oil was the wetting phase, emulating an enhanced oil recovery or non-aqueous phase liquid phase remediation process. We find that, for approximately constant pore-space topology, surface roughness (quantified with the average hillock height to pore depth ratio, Omega) has minor impact on sweep efficiency when Omega < 5.5%. However, once a critical threshold value of Omega > 12.5% is reached, recovery becomes consistently 10% higher and gas breakthrough occurs at later times. In addition, the roughest micromodel displays the highest repeatability of dendrite pathways validating diagenetic controls on gas flood sweep efficiency. We also find that surface roughness does not considerably affect the morphology of nonwetting phase dendrites and phase topology versus air saturation curves. Sub-pore scale visualizations in the micromodels indicate that contact-line pinning, the phenomenon responsible for higher sweep and greater dendrite tortuosity, only occurs in the roughest micromodel with Omega > 12.5%, whereas pendular rings and grain-lining thin films occur in all of the micromodels. Non-local snapoff occurs in micromodels regardless of roughness. The effects of isolated fracture surface roughness on capillary trapping saturations is found to be negligible for both drainage and imbibition saturation cycles. However, the presence of an isolated fracture diverts gas dendrites from sweeping the matrix and therefore increases oil trapping by approximately 10-30% compared to the micromodels without fractures. The workflow and experimental results in this paper provide benchmarking opportunities for direct numerical simulation algorithms for porous geomaterials. In addition, the paper aims to highlight the nontrivial implications of surface roughness for reservoir quality assessment and various subsurface operations.