The safety of large and aging water infrastructures is gaining attention in water management given the accelerated rate of change in landscape, climate and society. In current engineering practice, such safety is ensured by the design of infrastructure for the Probable Maximum Precipitation (PMP). Recently, several numerical modeling approaches have been proposed to modernize the conventional and ad hoc PMP estimation approach. However, the underlying physics have not been investigated and thus differing PMP estimates are obtained without clarity on their interpretation. In this study, we present a hybrid approach that takes advantage of both traditional engineering practice and modern climate science to estimate PMP for current and future climate conditions. The traditional PMP approach is improved and applied to five statistically downscaled CMIP5 model outputs, producing an ensemble of PMP estimates in the Pacific Northwest (PNW) during the historical (1970-2016) and future (2050-2099) time periods. The new historical PMP estimates are verified against the traditional estimates. PMP in the PNW will increase by 50%±30% of the current level by 2099 under the RCP8.5 scenario. Most of the increase is caused by warming, which mainly affects moisture availability through increased sea surface temperature, with minor contributions from changes in storm efficiency in the future. Moist track change tends to reduce the future PMP. Compared with extreme precipitation, PMP exhibits higher internal variability. Thus long-time records of high-quality data in both precipitation and related meteorological fields (temperature, wind fields) are required to reduce uncertainties in the ensemble PMP estimates.