Much of the post-processing and graphics generation for the ensemble leverages the python scientific computing and visualization packages. Using these packages, an in-house library was developed to compute and plot common ensemble diagnostic fields (e.g. ensemble mean, maximum, and spread), as well as many fields unique to convective-scale ensemble forecasting (e.g. neighborhood probabilities, probability matched mean, paintball plots, and probabilities derived via kernel density estimation). The forecast fields and plot properties are controlled via a python dictionary, allowing for rapid development and modification of new and existing products without requiring deep knowledge of the underlying framework. The library is used in parallel on NCAR supercomputing resources, enabling the timely creation of over 80,000 graphics each day.
The early feedback from the ensemble forecasting system has been extremely positive, in part due to the success of the python-based ensemble post-processing system. This presentation will summarize the design of the system, highlight challenges encountered post-processing and visualizing large ensemble datasets, and examine the tradeoffs of using python versus other languages commonly used in the geosciences for visualizing information from ensembles (e.g. NCL).