|
Abstract:
|
Ensemble weather forecasts are an attempt to sample forecast distribution and characterize uncertainty in forecasts. Several methods exist for generating ensembles, including accounting for uncertainty inherent in initial conditions and making a collection of smaller changes within a single modeling system. An alternative is to vary individual parameters, thereby introducing variability with minimal impact on the model structure. A multiparameter approach is appealing because the forecast sample is drawn from a single distribution and can be interpreted easily. We propose Latin hypercube sampling to generate random, independent samples of four key parameters. Initial results from a month of 48-h forecasts that the multiparameter approach is competitive with approaches that vary entire closures, and may offer a viable alternative to more common, yet potentially problematic, methods.
|
