Abstract

Large-scale meteorological patterns (LSMPs) associated with temperature extremes are evaluated in a suite of regional climate model (RCM) simulations contributing to the North American Regional Climate Change Assessment Program. LSMPs are characterized through composites of surface air temperatur... Show moreLarge-scale meteorological patterns (LSMPs) associated with temperature extremes are evaluated in a suite of regional climate model (RCM) simulations contributing to the North American Regional Climate Change Assessment Program. LSMPs are characterized through composites of surface air temperature, sea level pressure, and 500 hPa geopotential height anomalies concurrent with extreme temperature days. Six of the seventeen RCM simulations are driven by boundary conditions from reanalysis while the other eleven are driven by one of four global climate models (GCMs). Four illustrative case studies are analyzed in detail. Model fidelity in LSMP spatial representation is high for cold winter extremes near Chicago. Winter warm extremes are captured by most RCMs in northern California, with some notable exceptions. Model fidelity is lower for cool summer days near Houston and extreme summer heat events in the Ohio Valley. Physical interpretation of these patterns and identification of well-simulated cases, such as for Chicago, boosts confidence in the ability of these models to simulate days in the tails of the temperature distribution. Results appear consistent with the expectation that the ability of an RCM to reproduce a realistically shaped frequency distribution for temperature, especially at the tails, is related to its fidelity in simulating LMSPs. Each ensemble member is ranked for its ability to reproduce LSMPs associated with observed warm and cold extremes, identifying systematically high performing RCMs and the GCMs that provide superior boundary forcing. The methodology developed here provides a framework for identifying regions where further process-based evaluation would improve the understanding of simulation error and help guide future model improvement and downscaling efforts. Show less