TJ3.2 INVITED: Assessment of the National Multi-Model Ensemble System for the Prediction of Drought Over the NIDIS Test-beds

Monday, 7 January 2013: 1:45 PM
Ballroom F (Austin Convention Center)
Eric F. Wood, Princeton University, Princeton, NJ; and J. K. Roundy and X. Yuan

Extreme hydrologic events in the form of droughts are a significant source of social and economic damage in many parts of the world. Having sufficient warning of drought events allows managers to prepare for and potentially reduce the severity of their impacts to society. A hydrologic forecast system can give seasonal predictions that can be used by mangers to make better decisions; however hydrologic predictions rely on the skill of the climate model in order to predict drought extent and duration. Currently, most systems rely on a single climate model to make predictions. Recently, a National Multi-Model Ensemble (NMME) system was developed, which brings together several seasonal climate models in a consistent framework. The increase in skill in drought prediction due to a multi-model system is largely unknown. In this work we assess the improvement of skill in predicting drought by including a multi-climate model approach. To do this we downscale each model of the NMME and run hydrologic forecasts to produce baseline measures of predictability. Multi-model forecasts will be produced by combining the forecasts. It is unclear what the best procedure for combining the forecasts is but options the need to be considered range from equally weighing each ensemble member to optimally combining the forecasts in ways that consider complementary forecasts and model skill. Recent work by the authors has shown that an optimal clustering approach outperforms equal weights. Using the NMME Phase 1 reforecast monthly data set, we will combine the seasonal forecasts, downscale spatially and temporally, and use the Variable Infiltration Capacity (VIC) land surface model to generate hydrologic and drought forecasts. The increase in skill in using a multi-model approach is assessed by comparing the baseline runs with the multi-model combinations in terms of the predictability of the extent and duration of drought using drought indices such as soil moisture percentiles. We focus this work on two NIDIS test beds, the Apalachicola-Chattahoochee-Flint (ACF) River Basin in the South Eastern United States, and the Upper Colorado River Basin in the Western United States. Recommendations for application over CONUS are also discussed.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner