Impacts from extreme weather events on critical infrastructure have been increasing in recent years.   Power system operators are under growing pressure to conduct long-term planning for reliability and resilience, and to include future extreme weather in the considerations.  However, there are too many plausible scenarios of future extreme events, exceeding the ability of system planners and their tools to manage.  In this presentation, we demonstrate how statistical machine learning tools can be combined with multi-sector dynamic models to identify common patterns among large scenario sets.   We apply this method to a coupled water-energy-economic model of the western U.S. to consider future scenarios of water temperature stress on the electric power system.