Uncertainty refers to epistemic situations in which there is incomplete or unknown information. It applies to predictions of future events, existing physical measurements, and to what is not known. Uncertainty arises in partially observable or stochastic randomly distributed environments, as well incomplete knowledge. A relevant example of uncertainty is the worldwide climate projections provided by general circulation models (GCMs).
The Intergovernmental Panel on Climate Change (IPCC) publishes assessment reports (ARs) every 5–8 years. AR1 to the most recent AR6 (2021, 2022) include GCMs which present the historical past and projected climatic worldwide conditions, including meteorological parameters such as precipitation and air temperature (up to 2100); however, there is inherent uncertainty in the data used.
The most recent assessment report, AR6, includes more than 30 GCMs, each of which is tested according to different emission scenarios or atmospheric boundary conditions projected up to year 2100; these conditions define possible climate futures. All these alternatives (+100) are currently applicable and are considered valid projections.
There is no consensus in the literature related to model selection or even to recommendations for following specific emission scenarios. These GCMs and emission scenarios display possible paths, and while our future may not follow any of these projections exactly, these projections highlight expected ranges which consider the best known to science (by IPCC) and therefore should be evaluated and considered.
Explaining this uncertainty is one of the most challenging issues for climate change projects. This is contrasted with the civil design approaches where structures are ‘sized’ for a specific event, such as 1:100 years, but this fixed value definition loses significance in climate change concepts where most of the design values are in fact ranges defined by every GCM and every possible meteorological projection.
Figure 1 presents some alternatives about how a single meteorological parameter, such as mean annual air temperature, can be presented. In this example, the one parameter (1D) display variability is represented by valid points ranging in number from 30 to more than 100.
To simplify the point-cloud, one of the most frequently used approaches (although not necessarily the most appropriate) is to consider only the point-cloud median (1 point), or maximum, median and minimum values (3 points), or even a boxplot
representation (5 points). However, none of these simplifications may always be representative of the parameter variability; in this case, as a multimodal distribution (more than 1 peak), most of the points are aligned around the boxplots extreme
values, which differ significantly from the point-cloud median.
Other approaches are based on kernel density distributions, where values are represented by a single smooth density estimation. Using this interpretation, quantile intervals can be also seen (as with the boxplot) and highest density intervals (HDIs) highlight the most credible values (even in multimodal distributions).
Today’s GCMs involve an impressive variability of results which contrast with the typical single-value design required by civil engineers. The simplification to 1D representation of GCM values (from point-cloud to median/HDI) typically depends on: 1) relevance of the meteorological parameter; 2) actual impact of the variability on the design; and 3) the status of our own knowledge as project managers/clients to understand this uncertainty problem. The modeler should be included in these decisions/representations/simplifications to assess and present the uncertainty
related to GCMs.
