Meteorological Event Identification Using National Weather Service Forecast Discussions

Scientific advancement is predicated upon the publication and distribution of new findings. One of the primary challenges of scientific advancement in the field of weather is the identification of significant meteorological events from existing large, multi-variate data sources. Thus, cognitive computing efforts focused on data mining and predictive classification are imperative for promoting scientific discovery in Earth science as data volumes increase. Here, we present the feasibility of using text extraction techniques to glean meaningful scientific information from the National Weather Service (NWS) area forecast discussions. NWS offices issue area forecast discussions approximately four times per day. Each forecast discussion summarizes short and long term weather patterns relevant to a specific warning forecasting office (WFO). The Iowa Environmental Mesonet webpage hosted by Iowa State University hosts archived area forecast discussions from 2001 - present. From this archive, we retrieve more than 2 million NWS forecast discussions over the continental US from 2007 - 2017 and store the text files in a database. A subset of 20 terms is selected from the American Meteorological Society (AMS) Glossary of Meteorology and extracted from the text forecast database. This presentation will demonstrate use cases for terms extracted from the NWS forecasts, including: 1) Case identification using spatial and temporal anomalies and 2) Trend analysis of term usage for climatological applications.