Drought Prediction over the United States Using the North American Multi Model Ensemble

Wednesday, 17 December 2014: 5:20 PM
Kingtse C Mo and Dennis P Lettenmaier, NOAA Science Center, College Park, MD, United States
We analyzed the skill of drought forecasts over the United States based on drought indices derived from the hydroclimate forecasts from the North American Multi model ensemble (NMME). The test period is from 1982-2010 and forecasts are initialized from the beginning of January, April, July and October. We analyzed the forecast skill of drought indices such as the 6-month standardized precipitation index (SPI6), monthly mean soil moisture percentiles (SMP) and the 3-month standardized runoff index (SRI3). The soil moisture and runoff were obtained by drive the Variable Infiltration Capacity land model with forcing derived from the NMME members (NMME_VIC). Drought indices from each member were computed and they were put into percentiles determined from all members in the training period at a given lead. We then formed an ensemble grand mean by averaging all indices together and determined the concurrence measure which is the extent to which all different members agree.
We find that : 1) The grand mean has higher skill than individual member; 2) During winter, forecasts are skillful in the regimes where the initial conditions dominant contributions to skill, the agreement between the grand mean and members are above 70-80% . At high leads, the concurrence measure drops to 50-60%, even when forecasts are unskillful. 3) During summer, forecast skill is low and concurrence measure drops to 10-30%, 4). The skill of drought forecasts is regionally and seasonally dependent. The NMME_VIC forecasts tend to over forecast drought events with large false alarm rate. After lead-1, the low thread score indicates no skill. The forecast errors will be analyzed to determine the origin of forecast skill.