Report summarising review of existing approaches for communicating confidence levels. Summary

A review of existing approaches to communicating confidence and uncertainty points to a need to systematically test how representations are interpreted and address potential misunderstandings. This review was undertaken with two primary goals: 1) to examine general issues pertaining to the communication of confidence and uncertainty, with a view to informing best practice in the context of seasonal-to-decadal climate predictions (S2D); and 2) to discuss current and proposed methods of presenting uncertainty in climate predictions. With respect to the former goal, this article identified several factors likely to influence end users’ interpretation of information concerning confidence and uncertainty in climate predictions. These include: ‘ambiguity aversion’, trust in information providers, institutional protocol, technical and statistical expertise, the type of visualisation tool(s) used, and cognitive biases. The importance of clarifying what is meant by the terms ‘confidence’ and ‘uncertainty’ is also stressed. We go on to explore numeric, verbal and visual methods of representing uncertainty and highlight the uses and potential pitfalls of various approaches. Numeric presentations permit uncertainty to be formally represented as ranges and confidence limits; but those users with less experience of using statistical information may struggle to extract appropriate meaning from them. The employment of verbal descriptors and evaluative categories may enhance the ability of these users to interpret statistical uncertainty information , but providing these without accompanying numeric ranges can lead to high variability in the way in which 'uncertainty language' is interpreted. Visualisations provide a versatile way to display uncertainty information at varying levels of complexity, and various maps and graphs already exist for the purpose of conveying probability, dispersion and reliability in the context of climate predictions. However, as there has to date been relatively little systematic testing of their efficacy amongst a varied user base, more research is needed to identify which visualisations best suit which context. Key considerations identified with respect to the development of visualisations for use in communicating uncertainty in the context of S2D thus include: 1) user expertise; 2) the extent to which user training will be provided; 3) potential for misunderstanding; and 4) the importance of ensuring that information regarding reliability is salient to users. We conclude by reiterating the importance of systematically testing which communications suit which user, and examining whether new forms of representation (or adjustment to existing ones) can increase ease of understanding and reduce misinterpretation.