Abstract While the Department of Defense (DoD) infrastructure is no stranger to extremes, recent events have been unprecedented, with climate change acting as a growing risk multiplier. To assess the level of exposure of DoD installations to extreme weather and climate events, site-specific climate information is needed. One way to bridge the scale gap between outputs from existing global climate models (GCMs) and sites is climate downscaling. This makes the information more relevant for impact assessment at the DoD installation and facility scale. However, downscaling GCMs is beset by a myriad of challenges and sources of uncertainty, and downscaling methods were not designed with specific infrastructure planning and design needs in mind. Here, we evaluate state-of-the-science dynamical downscaling and statistical downscaling and bias correction for climate variables (i.e., temperature and precipitation) at the daily scale. We also combine downscaling approaches in novel ways to optimize computational efficiency and reduce uncertainty. Furthermore, we examine the sensitivity of the downscaled outputs to the choice of reference data and quantify the relative uncertainty related to downscaling approach, reference data, and other factors across the climate variables and aggregation scales. Results show that empirical quantile mapping (EQM), a statistical downscaling, consistently performs well and has less sensitivity to the choice of reference data. Moreover, the hybrid downscaling that leverages EQM improves the performance of dynamical downscaling. Our findings highlight that the choice of reference data dominates uncertainties in temperature downscaling, while their role is more muted for precipitation but still non-negligible.