Prospects for Dynamical Prediction of Meteorological Drought

The prospects for U.S. seasonal drought prediction are assessed by diagnosing simulation and hindcast skill of drought indicators for the period 1982-2008. The 6-month standardized precipitation index (SPI6) is used as the primary predictand for drought. The skill of unconditioned drought forecasts, reflecting the statistical persistence of the drought indicator alone, serves as the baseline against which the performance of dynamical methods are evaluated . Predictions that are conditioned on the state of global sea surface temperatures (SST) are assessed using atmospheric climate simulations conducted in so-called AMIP mode in which observed SSTs are specified. Predictions that are conditioned on the initial states of global ocean, atmosphere, and land surfaces are next analyzed using the NOAA Climate Forecast System (CFS). The results indicate that the inherent persistence of the drought indicator (SPI6), yields considerable seasonal skill, and there is critical prognostic information linked to how drought likelihoods relate to a region's annual cycle of precipitation. Dynamical models forced by the particular conditions of the observed global SSTs yield increased skill relative to this unconditional baseline skill, though improvements are principally realized in winter and are geographically confined to the southern U.S. where drought probabilities are sensitive to the phases of the El Niño Southern Oscillation. Fully coupled initialized model hindcasts are shown to yield little additional skill in seasonal drought predictions compared to AMIP simulations during winter, apparently due to the preponderance of the ENSO skill source. However, we show that initialized models appreciably improve on the AMIP skill alone during the spring and summer, presumably due to the increased importance of land surface feedbacks, the environment for which are initialized in the CFS.