Real-time Monitoring and Flagging of Extreme Value Forecasts - A Practical Example

Forecasts of record values are usually avoided unless expected to occur with great confidence within less than 48 hours, or in association with an extreme event such as a hurricane. Otherwise the risk of a high visibility false alarm outweighs the benefit of a correct early hit . Yet automated forecasts may occasionally include record values beyond day 2, which forecasters may choose to downplay, or not. In Canada, forecasters keep their focus on high impact weather for days 1 and 2, so that forecasts for day 3 and beyond are mostly automated and usually released after a quick glance. So a process was designed to bring up cases where automated temperature forecasts exceed known records for a number of sites, with the sole purpose of alerting the forecasters who may decide whether or not modifications are needed before release. We will discuss the origin of the process, its logics, its current status, interesting by-producst, and possible improvements.