Ted Bohn - Multi-Model Ensemble Forecasting Methods

Incorporation of multi-model ensemble framework into UW Experimental Surface Water Monitor

The experimental surface water monitor has already proven useful in monitoring soil moisture conditions across the continental US, using just one land surface model, VIC. We have extended this system to include an ensemble of 4 land surface models (VIC, NOAH, SAC/SNOW17, and CLM). Results of the individual models are expressed as percentiles of their historical (1920-2003) distributions, averaged together, to get a "raw" multi-model result, and then this multi-model result is expressed as a percentile of its historical distribution.

Wang A., T.J. Bohn, S.P. Mahanama, R.D. Koster, and D.P. Lettenmaier, 2008: Multimodel reconstruction of drought over the continental United States, J. CLimate (submitted)

Incorporation of multi-model ensemble framework into UW West-Wide Seasonal Hydrological Forecast System

The west-wide seasonal forecast system has already proven successful in long lead-time stream flow prediction, using just one land surface model, VIC. Now we are attempting to improve performance further, using an ensemble of 3 land surface models (VIC, NOAH, and SAC/SNOW17). We are investigating several methods of combining the model results: step-wise linear regression, Bayesian model averaging, and principal component analysis. We are evaluating these methods in terms of performance in deterministic and probabilistic forecasts, as a function of lead time.

(paper by Bohn, Sonessa, et al in preparation)
Bohn, T.J., A.W. Wood, D.P. Lettenmaier, European Geosciences Union Spring Meeting, April 2006, Talk, A multimodel streamflow forecasting system for the western U.S.
Bohn, T.J., A.W. Wood, A. Akanda, D.P. Lettenmaier, American Geophysical Union Fall Meeting, Dec 2005, Poster, A multi-model hydrologic ensemble for seasonal streamflow forecasting in the western U.S.
Bohn, T.J. UBC-UW Hydrology Symposium, Sept. 2005, Machiavellian Forecasting: Does the end justify the means?