|
Ensembles of numerical weather prediction models provide useful information about the
forecast uncertainty for a number of variables, including precipitation. The ensemble predictive
distribution typically does not have the desired coverage, but this can be remedied through
statistical post-processing. The ensemble forecasts are used as predictors in generalized linear
models that describe the distribution of observed precipitation. In addition, models can be
combined using Bayesian Model Averaging (BMA), which incorporates model uncertainty by
using a weighted average of models, favoring models that have superior predictive performance.
This study applies the methodology to an ensemble of 16 models to develop and assess probabilistic
precipitation forecasts for lead times of one to five days.
|
