The Effect of Data Record Length on a Probabilistic Risk Assessment of Extreme Temperature Events

The expansion of surface-based observation networks in recent decades provides an increasing volume of environmental data that can be leveraged to understand dangerous high-impact heat events at local levels. Extreme heat is the leading cause of weather-related fatalities in the United States, contributing to or directly causing thousands of deaths worldwide annually, including hundreds of deaths in the United States alone. Heat waves also have significant societal and economic impact, contributing to agricultural losses, decreased economic productivity, and infrastructure damage – often disproportionately affecting overburdened communities. Increases in observation density present an opportunity to provide focused risk assessment and hazard information to stakeholders and decisionmakers in many of these communities. Unfortunately, the newest stations lack data records that satisfy the WMO standard of a 30-year climatology – but how much data is enough to begin reasonably characterizing the local risk of extreme temperature events?

To answer this question, this work analyzes the uncertainty associated with a probabilistic characterization of extreme heat events at several surface observing stations. Here, we use intensity-duration-frequency (IDF) curves to contextualize the severity of extreme heat events at a given location based on their probabilistic frequency of recurrence at a variety of given magnitudes and durations. IDF curves are ubiquitous in the hydrological community to evaluate flood risk, providing the estimated return period for rainfall of a given intensity and duration. Here, an objective fitting algorithm is used to construct IDF curves for temperature and heat index at several stations based on historical surface observations. In addition to IDF curves, the algorithm also calculates confidence intervals for the intensity at each duration/frequency pair of interest. Subsequent data denial experiments modify the length of the data record and the subset of years available to the algorithm, exploring the interplay between data record length and extreme event characterization. These relationships may help data producers and data consumers more resourcefully utilize nascent observational networks for risk assessment in their local communities.