A CP conditioned hybrid weather generator

Long continuous time series of meteorological variables such as temperature and precipitation are required for applications such as derived flood frequency analyses. Observed time series are however generally too short, too sparse in space, or incomplete, especially at the sub-daily timestep. Stochastic weather generators allow an alternative to using observations, being able to generate time series of arbitrary length which are then used as input to hydrological models.

A hybrid hourly space-time weather generator has been developed based on a stochastic alternating renewal rainfall model. Modelling of non-rainfall climate variables is achieved using a non-parametric k-nearest neighbour (k-NN) resampling approach, which is coupled to the space-time rainfall model via rainfall state.

Circulation pattern (CP) or weather pattern classifications can be useful as a conditioning variable for stochastic rainfall models and weather generators. One primary use is the downscaling of future climate scenarios. Furthermore, CP conditioned models may better simulate rainfall and other climate variables through a better partitioning of observations into distinct rainfall and weather types.

Previous research has shown that the point rainfall model performs better, particularly regarding extremes, if conditioned on an optimised fuzzy-rule based objective weather pattern classification. Appropriate model revisions have now been made to allow the full hybrid space-time weather generator to also be conditioned on this classification.

This study assesses the performance of the weather pattern conditioned hybrid weather generator compared to the previous seasonal (summer-winter) conditioned model. For testing, 400 meso-scale catchments across Germany were selected.