The use of calibrated language in climate assessments has evolved over time and is an important tool for better characterizing climate risk. Despite improvements in communication, likelihood and confidence terms continue to be misinterpreted by nontechnical audiences. Public understanding of scientific certainty in climate impacts, particularly at the extreme tails of likelihood distribution, is inadequate. This study evaluates the use of calibrated language in the four most recent U.S. climate assessments. Across the assessments, there is inconsistent use, definition, location, and formatting designating a term as calibrated. Authors include additional, undefined certainty categories bridging qualitative confidence categories (e.g., "medium to high") and likelihood terms with no associated statistical value (e.g., "highly likely"). Confidence and likelihood levels are more frequently reported for terms at the high end of certainty distributions, with at least 70% of all instances of confidence terms reporting high or very high confidence in three of the assessments. Almost all (>98%) likelihood language reported is in the likely (>66%) category or above and no likelihood terms below 50% are reported in any of the assessment's Key Messages. There is significant room for improvement in the representation of likelihood, as the word “likely” is frequently used with ambiguous intent; at least half of all instances of likelihood terms are used with noncalibrated, potentially colloquial intent in three of the assessments. The study concludes with a set of recommendations for improving the use of calibrated language and communication of risk in the development of future assessment products.

中文翻译：

---

在美国气候评估中改进校准语言的使用

随着时间的推移，校准语言在气候评估中的使用已经发展，并且是更好地表征气候风险的重要工具。尽管交流得到了改善，但非技术性受众仍然对可能性和置信度术语进行了误解。公众对气候影响的科学确定性的理解不足，尤其是在可能性分布的最末端。这项研究评估了美国最近四次气候评估中校准语言的使用。在评估中，用法，定义，位置和格式不一致，表示已校准的术语。作者包括其他未定义的确定性类别，将定性置信度类别桥接在一起（例如，“中到高”）和没有相关统计值的可能性项（例如“高度可能”）。在确定性分布的最高端，经常会报告置信度和可能性级别，至少有70％的置信度实例在三个评估中报告具有高或非常高的置信度。几乎所有的（> 98％）之间的相似度语言报道是在可能（> 66％）类别或更高类别，并且在评估的任何关键信息中均未报告低于50％的可能性项。由于“可能”一词的使用经常含糊不清，因此存在很大的改进可能性。在所有评估中，至少有一半的可能性项实例使用了未经校准的，可能是口语的意图。该研究最后提出了一系列建议，以改善将来评估产品的开发中使用校准语言和风险交流的能力。