Keywords --- IPCC, signal to noise, precipitation, CMIP5, CMIP6, climate change, projectionsIntroductionClimate models are the key tool to characterise future climate projections and hence inform on the impacts that might result from increasing greenhouse gas concentrations, changing aerosols and land use. Ensemble of coordinated climate model experiments are now routinely run as part of the Coupled Model Inter-Comparison Project (CMIP) in order to better characterise the uncertainty and the robustness in the projected climate change. As managing climate change is increasingly recognised as a problem in risk assessment, assessing the potential for large future changes needs to take a central part in the communication of future climate projections to stakeholders and policy makers. In the fifth Assessment Report (AR5) of the Inter-governmental Panel on Climate Change (IPCC), climate projections are primarily communicated via spatial maps presenting the multi-model mean projection for the climate field of interest, such as temperature or precipitation (see Fig 1a). The discrimination between large and small changes is based the signal-to-noise of the multi-model mean response, where by signal it is meant the forced climate change response and by noise the unforced variability in the climate system. In particular, where the mean response is large compared to the internal variability on the response itself, and at least 90\% of models agree on the direction of change, the maps is stippled to indicate "a large change and high model agreement". On the other hand, if the mean response is smaller than the internal variability, the maps are hatched to indicate "small signal or low model agreement". This ambiguity in the interpretation of the hatching resulted because models can either genuinely agree on a small response or project large changes of opposite sign; In both cases, the response would be small - hence hatched - in the mean. Furthermore, a number of regions remain neither hatched nor stippled (Fig 2a). There is therefore some additional ambiguity on whether these unmarked regions reflect consistent changes of intermediate amplitude, or potential large changes in which less than 90\% of models agree on the direction of change. Overall, this suggests that the method is not specifically informative for the purpose of risk assessment.A number of more advanced statistical methods - of varying degree of complexity - have been proposed to fully discriminate the spread in the model responses, hence separating cases in which there is agreement on a small change, large changes of opposite sign (lack of agreement), and agreement on a significant change. Additional methods have also been developed to account for the uncertainty in the magnitude of the change, since a large inter-model spread can still be present in regions where projections agree on the direction of change. Nonetheless, while acknowledging all these different and more informative approaches, the IPCC still preferred to use the simpler - though more ambiguous - diagnostic described above. This reflects how no single method is necessarily better than the others when having to balance the need to be informative with that of keeping a simple message. Furthermore, while all methods adopt objective criteria, establishing model agreement ultimately requires the application of some subjective thresholds on the imposed criteria.Here, we propose a novel diagnostic to summarise climate information that is as simple as the standard IPCC approach, but more suitable to provide relevant information for risk assessment. Within this context, a priority is to identify the regions where a large change is plausible, including those regions where not all models necessarily agree on the change. This is achieved by quantifying the mean of the signal-to-noise of the individual model responses, rather than signal-to-noise of the mean response. This change of approach avoids to introduce compensation in the magnitude of the projected signal to noise in regions where models disagree on the direction of change, while keeping it otherwise similar to that previously used in the IPCC. The impact of internal variability on the individual model responses is properly accounted for in order to not inflate the mean signal to noise. While the previously proposed methods may provide a richer description of robustness and model uncertainty, the present approach follows the standard IPCC approach, but it offers a less ambiguous interpretation that is directly useful to inform risk-assessment.