Multimodel ensembles of streamflow forecasts: Role of predictor

6 [1] A new approach for developing multimodel streamflow forecasts is presented. The 7 methodology combines streamflow forecasts from individual models by evaluating 8 their skill, represented by rank probability score (RPS), contingent on the predictor state. 9 Using average RPS estimated over the chosen neighbors in the predictor state space, 10 the methodology assigns higher weights for a model that has better predictability under 11 similar predictor conditions. We assess the performance of the proposed algorithm by 12 developing multimodel streamflow forecasts for Falls Lake Reservoir in Neuse River 13 Basin, North Carolina (NC), through combining streamflow forecasts developed from two 14 low-dimensional statistical models that use sea-surface temperature conditions as 15 underlying predictors. To evaluate the proposed scheme thoroughly, we consider a total of 16 seven multimodels that include existing multimodel combination techniques such as 17 combining based on long-term predictability of individual models and by simple pooling 18 of ensembles. Detailed nonparametric hypothesis tests comparing the performance of 19 seven multimodels with two individual models show that the reduced RPS from 20 multimodel forecasts developed based on the proposed algorithm is statistically significant 21 from the RPSs of individual models and from the RPSs of existing multimodel 22 techniques. The study also shows that adding climatological ensembles improves the 23 multimodel performance resulting in reduced average RPS. Contingency analyses on 24 categorical (tercile) forecasts show that the proposed multimodel combination technique 25 reduces average Brier score and total number of false alarms, resulting in improved 26 reliability of forecasts. However, adding multiple models with climatology also increases 27 the number of missed targets (in comparison to individual models’ forecasts) which 28 primarily results from the reduction of increased resolution that is exhibited in the 29 individual models’ forecasts under various forecast probabilities.