On the development of polygon recommenders for IDSS within the Aviation Weather Center

An emergent decision-support capability for the NWS are so-called recommenders. These are polygons derived using artificial intelligence meant to expedite the work flow of forecasters and lend heightened situational awareness in ambiguous or complex weather scenarios. While the concept is simple and the idea seems, on its surface, easy to implement, the actual process of developing software that mimics human decision making in creating polygons is quite challenging, especially in complex or ambiguous situations. The myriad of challenges faced on the road from here to there are presented in this talk by using our attempts to produce an automated convective SIGMET recommender for the Aviation Weather Center. These challenges include (1) non-specific guidance in NWS directives and the resulting subjective decision making by human forecasters, (2) development of a methodology that has the flexibility to be tuned so that it can perform optimally in a variety of different weather scenarios, especially high-impact, but rare events, (3) addressing uncertainty in the input data, and (4) creating output that fits within the tight constraints of the communication format for SIGMETs. The algorithm’s performance in different scenarios will also be highlighted.