Beschreibung:
<jats:p>Skilful predictions of the frequency of flood events over long lead times (e.g., from 1 to 10 years ahead) are essential for governments and institutions making near‐term flood risk plans. However, little is known about current flood prediction capabilities over annual to decadal timescales. Here we address this knowledge gap at 286 U.S. Geological Survey gaging stations across the U.S. Midwest using precipitation and temperature decadal predictions from the Coupled Model Intercomparison Project (CMIP) phase 5 models. We use the 1–10‐year predictions of precipitation and temperature as inputs to statistical models that have significant skill in reproducing inter‐annual and decadal changes in the observed frequency of flood events. Our results indicate that the limited skill of basin‐averaged precipitation predictions suppresses the skill of flood event frequency predictions, even at the shortest lead time, but downscaling and bias correction improves predictions across all lead times and especially in spring.</jats:p>