Examining the Utility of Twitter Hail Reports to Verify Spatiotemporally Dense Forecasts

Forecasting techniques of high impact weather events have improved dramatically over the last decade, but verification techniques have remained largely the same. Reports gathered from the National Weather Service are quality controlled and verified after the event is over, which does not help in providing real time, storm specific information to forecasters while they are making warning decisions. Twitter has become increasingly popular in forecast operation settings because it can provide real time updates on specific storms. However, there is often a large amount of noise within Twitter reports. This is particularly a problem with hail events because “hail” is a culturally more popular word that other weather events like “tornado” or “thunderstorm”. This work analyzes the accuracy and usefulness of sorting algorithms in identifying hail reports on Twitter. Results indicate that while some algorithms are better at filtering out non-reports, they can also be too stringent in identifying accurate reports such that they miss many of them. There ultimately needs to be a balance between properly identifying reports while also filtering out noise if these reports are going to be used in real time or as a supplement to local storms reports.