Addendum to Bayesian ensemble forecast of river stages and ensemble size requirements

作者：Herr Henry D^*; Krzysztofowicz Roman

来源：Journal of Hydrology, 2014, 515: 304-306.

DOI：10.1016/j.jhydrol.2014.04.065

• Forecasting
• Stochastic processes
• Probability
• Ensemble
• Estimation errors
• Rivers

Ensemble size requirements were analyzed using Bayesian ensemble forecasts (Herr and Krzysztofowicz, 2010). These forecasts were generated using output from an analytic-numerical Bayesian forecasting system (BFS) within a Monte-Carlo algorithm. Empirical distributions were estimated from the ensemble forecasts and compared with the analytic distributions output by the BFS. Using the maximum absolute difference as a measure of the quality of fit, it was shown that a single graph described the relationship between the ensemble size and the expected maximum absolute difference. Within the scope of the experiments, the graph was insensitive to forecast point, precipitation event (rain or no rain), and lead time. However, the graph only applied to probabilistic river stage forecast and probabilistic flood forecast, but not to probabilistic stage transition forecast, which is specified by a family of conditional predictive one-step transition distributions. Herein, further mathematical analysis of the estimation algorithm shows that, with a small modification, the single graph can be applied to all three types of forecasts. Published by Elsevier B.V.

• 出版日期2014-7-16