Analysis and forecast error correlations explored in an observing system simulation experiment framework

Correlations between the error fields of forecasts and verifying analyses can affect the usefulness of analyses for validation of forecast skill. In practice, these correlations are difficult to estimate, particularly during the initial forecast period due to the lack of a suitable verification dataset for error calculation. We attempt to characterize these correlations using an observing system simulation experiment framework in which the “true” state of the atmosphere is known in order to directly calculate correlations between the forecast and analysis errors. Both self‐analysis correlations and correlations between forecast errors and analysis errors from a “twin” analysis cycled from a slightly modified initial state are computed for up to 10‐day forecasts. A significant drop in error correlations is found during the initial 48‐hr forecast, with somewhat lower correlations for the twin analysis compared with self‐analysis. This finding is in line with previous work that has shown the greatest correlations between verifying analyses and forecasts occur during the first 2 days of lead time.