The paucity of research studies related to the evaluation of short-fuse hazard warnings in the US is rather troubling, especially as urban populations continue to rise in hazard prone regions (Ashley, 2007). Nevertheless, the majority of evaluative studies have assessed the quality of tornado warning detection, with the use of statistical measures, such as POD/FAR (Brooks, 2004). For instance, as described herein, past studies of tornadoes in the contiguous US have shown phenomenal improvements in tornado detection, especially after the adoption of the NEXRAD program (Simmons and Sutter, 2005). However, this evaluation has been more recently criticised, with Brotzge et al. (2011) arguing that the value provided by these statistical measures is rather ambiguous, with POD largely dependent upon the level of verification, alongside FAR failing to account for close calls and varying with parameters such as tornado order, distance from radar and climatology. Furthermore, the use of radar has been critically evaluated by a number of studies (e.g.Heinselman et al., 2012; Mahale et al., 2012), with a notable root cause study of 146 unwarned tornadoes between 2004-2009 reporting ‘radar sampling’, ‘no radar signature’ and ‘radar use’ as three of the top ten reasons for the failure to detect (Quoetone et al., 2009).
Additionally, with the plethora of new sensors (e.g.GOES-R), evaluative studies are demonstrating the problems of ‘data overload’, with ‘workload’ cited in one-third of all missing warnings (Quoetone et al., 2009). However, the use of the NOAA HWT to evaluate the accuracy and operational utility of new science and technology has rapidly helped to recently mitigate these challenges, acting as the initial stepping stone towards ‘warn on forecast’ (Calhoun et al., 2014). This new paradigm development will enhance warning lead times, whilst providing more accurate detections through the establishment of the NearCast model, which uses a Lagrangian approach to project full-resolution 10km GOES sounder moisture/temperature retrievals forward in time and space (Gravelle et al., 2016). The concept of incorporating GOES-R capabilities into AWIPS was rigorously evaluated by Gravelle et al. (2016), concluding that the system will allow users to quickly diagnose a more complete physical depiction of short-term changes in the atmosphere, in turn allowing meteorologists to make better/more well informed decisions on whether to issue warnings.
Furthermore, the evaluation of warning communication by Coleman et al. (2011), demonstrated a clear technological progression since 2007, with the establishment of the ‘storm-based warnings’. The further analytical evaluation by Sutter and Erickson (2010), reported that these newly established warnings reduced the number of warned person-hours by 30%-50%, subsequently saving between $500 million and $1.9 billion annually. However, the US is likely to become a ‘minority majority’ by the middle of the century. This greater diversity among the population presents additional challenges, with a number of evaluative studies acknowledging the clear presence of a ‘digital divide’ (Meyers, 2014; Donner et al., 2007). Therefore, despite the attempts made by FACETs to incorporate improved usable outputs (e.g.PHI) and advancements in social media tools (Rothfusz et al., 2014), the small number of studies evaluating communication all demonstrate a failure to provide equal information access for all actors in society, alongside the inherent limitations of an ageing communication network of sirens/NWR (Brotzge and Donner, 2013).
Finally, there remain too few evaluative studies to reach a robust conclusion with regards to the public response/use of tornado warnings. Nevertheless, the research literature reveals concerning trends among certain groups in society to misunderstand the instructions, alongside a failure of some to effectively receive warnings (Schumann et al., 2018; Donner et al., 2007). Therefore, most evaluative studies recommend enhancements in public education programs, alongside the adoption of more diverse approaches, for example CAP, which consider the rapidly changing demographics of the US (Schumann et al., 2018). Additionally, in terms of the protective action intentions, a chasm of knowledge remains within the scientific community; more research is required to evaluate the use of protective procedures (e.g. sheltering and evacuation), with only tenuous conclusions by Brotzge and Donner (2013) eluding to the necessity of a continuum of lead times that consider more vulnerable populations. Thus, Schumann et al. (2018) suggests that efforts to understand how the public uses/responds to warnings, to explore new warning dissemination methods/formats and to educate the public on the additional warning guidance potentially provided by a ‘warn-on-forecast’ system, are strongly desired.
Conclusively, the transformative changes in weather detection technology clearly eclipse social science research on warning response, threatening to render the integrated tornado warning system obsolete. Thus, to further reduce severe weather deaths, it is imperative that social science research keeps pace with technological innovations, especially as the NWS are on the cusp of initiating the early stages of a ‘warn on forecast’ system.