Agile Postprocessing: Towards User-Centered Ensemble Information Extraction and Visualization

This presentation summarizes an effort to post-process and visualize the NSSL Experimental Warn-On-Forecast System for ensembles (NEWS-e) tornado hazard guidance in the 2016 & 2017 Probabilistic Hazard Information (PHI) prototype experiments in the Hazardous Weather Testbed. Object based post processing was used on 4-D updraft and vorticity fields. Active updrafts were used to locate storm areas and vorticity vertical profiles were data mined to detect mid- and low-level mesocyclones. Features were labeled spatiotemporally, every 5 minutes, to meet the needs of the forecasters at scales that meshed with underlying gridded fields (e.g., radar data). The visualization allowed the forecasters to query the data, in real-time, matching the post processed data to the task at hand. For example, a forecaster issuing a 45-minute warning could match the warning length (spatially) with data length and manipulate the vorticity thresholds or number of ensemble members to effectively create probability information tailored to each storm. This data mining/user centered extraction of information presents new and exciting challenges in post-processing that can alleviate probabilistic information overload and over production of post-processed variables. These components are viewed as vital for maximizing usage and improving forecast quality under the time sensitive windows inherent to warning decision-making.