UCLA

Abstract A California “megaflood”—a weeks‐long cluster of storms with centuries‐long return times—would be catastrophic, yet its timing and magnitude under climate change remains uncertain. A central source of uncertainty is whether Earth System Model boundary conditions should be bias corrected before dynamical downscaling. Here, we analyze a large ensemble of downscaled simulations to quantify how this uncertainty shapes California's megaflood risk. Bias correction is necessary to reproduce observed storm statistics and strongly influences both the megafloods historical and future characteristics. Without correction, storm precipitation variability is inappropriately inflated, delaying the emergence of extreme precipitation signals by decades. Further, megaflood intensity estimates based on statistically extended observations are at least 10% drier than explicit computations derived from sufficiently sized storm populations. Our results show that bias correction choices crucially govern projections of rare, high‐impact events, with direct implications for infrastructure design/planning and water management in California. Plain Language Summary Infrastructure design and planning are often informed by the best available downscaled climate projections. However, downscaling as a technique can produce low‐quality outcomes if errors in driving climate models are not confronted and addressed (i.e., the garbage in, garbage out problem). Since billions in capital may be invested in infrastructure based on such information, it is critical to understand the consequences of climate model bias removal, termed bias correction, on high‐impact infrastructure‐defining events (e.g., the California megaflood). Here we show that with bias correction, coherent signals in the state's rarest 30‐day rainfall events emerge decades sooner than if climate model biases are not removed prior to downscaling. Our results indicate that policymakers and decision‐makers may potentially be making investment choices based on inappropriate time horizons in the 21st century when using non‐corrected climate data. Key Points Statistically extended observations under‐predict megaflood intensity by at least 10% 100‐ and 200‐year storms increase more rapidly over time—both in frequency and magnitude—than storms with shorter recurrence intervals When passed through dynamic downscaling, biases from ESMs amplify storm sequence variability, delaying emergent climate trends