Communicating uncertainty in seasonal climate forecasts: Lessons from the EUPORIAS project

14:00 Tuesday 28 May

SS007 • OC040

Room S2


Andrea Taylor (United Kingdom) 1; Suraje Dessai (United Kingdom) 1

1 - University of Leeds

In communicating seasonal climate forecasts it is important to convey both the probabilistic nature of forecasts, and how well the forecasting system performs (i.e. skill). Failing to do this may lead false perceptions of certainty, maladaptive decision making, and loss of trust in providers. However, forecast users re diverse; varying in both expertise and the types of decision made using forecast information. Forecast providers must there find effective ways to communicate information about forecast probability and skill to their users.

We report on a programme of work, undertaken as one part of the EUPORIAS project, to identify more effective strategies for communicating confidence in seasonal forecasts. In a preliminary survey (n=50), users’ perceptions of current information provision and preferences for receiving probabilistic information were explored. A series of communication formats were then developed for statistical experts (Bubble Map, Violin Plot, Bar Graph, Quantitative Table) and statistical novices (Confidence Index, Bar Graph, Qualitative Table). These were tested in two studies with decision makers from relevant sectors (n=264 and n=56), where we measured objective understanding, preference, perceived familiarity, and subjective interpretation of ‘No Skill’ and ‘Higher Skill’ forecasts in each format.

We find, first, that when forecasts are worse than climatology (i.e. no skill), forecast probabilities have an undue influence on subjective expectations about future conditions. Second, using qualitative categories to represent skill can aid understanding of this information, especially amongst those who are novices at using statistical information. Third, preference for particular formats is linked to perceived familiarity, but not better objective understanding. Fourth, when presenting information about ‘likelihood of tercile’ tabular representations are better understood than visualisations, but less useful when users require information about the ensemble distribution.

Based upon these findings, we make the following recommendations:

  1. forecasts should not be presented to end-users by default if skill is worse than climatology;
  2. users should be provided with frameworks – such as qualitative categories – to help them understand information about forecast skill;
  3. communication formats should be tested with intended user groups, in order to identify areas where preferred formats may be misinterpreted and address these misunderstandings.

At a broader level, our finding further highlight the benefits of tailoring climate services to meet the needs of specific user groups.