The data provided below represents the degree of uncertainty or variation between 8 individual general circulation models (GCM) for three metrics commonly used to assess the intensity of exposure to climate change. The three exposure metrics (forward and backward climatic velocity and local climatic dissimilarity) were calculated based on the first two principal components (PC) scores derived from 11 different climate variables. 

Frameworks and heuristics supporting climate adaptation for conservation often rely on projections of climate change or climate exposure. However, projections of climate change vary among alternative GCM outputs, different emissions scenarios, and different future time periods. The potential for these model predictions to vary geographically presents a source of uncertainty in assigning climate-informed conservation strategies to landscapes. Regions with high agreement among predictions could be more confidently assigned a climate-informed strategy, whereas regions with less agreement among predictions may require a more cautious approach. 

In a 2018 paper, Belote and colleagues evaluated agreement among 18 different climate change predictions and three climate exposure metrics. The 18 different predictions of climate exposure were generated from 8 GCMs, including one ensemble developed from 15 GCMs, plus two different emissions scenarios (RCP 4.5 and 8.5). 

Belote and colleagues classified each map of climate exposure into quartiles producing 18 different maps for each of the three metrics. Next, the researchers calculated the most frequently assigned integer quartile value for every location (i.e., the mode among the 18 alternatives) and counted the frequency of this quartile mode. This essentially maps the most common relative climate prediction from low (1) to high (4), and its frequency among predictions (high frequency = higher agreement among predictions). (Note that the data layers provided below differ from those used in Belote et al.'s study in that they encompass North America and represent between-GCM variation within each combination of RCP and time period).

Figure 1. Mode of quartile in classified data of three climate vulnerability metrics and frequency of mode among 18 different projections for forward velocity (A,B), backward velocity (C,D), and climate dissimilarity (E,F). Maps B,D,F present an index of inter-simulation uncertainty for each climate metric, with areas of red indicating lower intermodel agreement and higher uncertainty (figure from Belote et al. 2018).

Belote and colleagues then combined each alternative classified exposure map with the wildland conservation value of Belote et al. 2017a into bivariate maps of “value and vulnerability” proposed in the heuristic of Belote et al. 2017b. This allowed the researchers to evaluate which areas would be classified into the same category of value and vulnerability irrespective of climate prediction, as well as how assignment of areas to these classes varies with the degree of agreement among alternatives. 

Figure 2. Maps of locations occupying the four corners of the conceptual framework of Belote et al.11, based on forward velocity (top row), backward velocity (middle row), and climate dissimilarity (bottom row), with columns showing the winnowing that happens with increasingly stringent thresholds for climate model consensus, ranging from low at left to high at right: =5 (left-hand column), >10 (second column from left), >16 (third column from left), or 18 (right-hand column)(figure from Belote et al. 2018).

Belote and colleagues found that areas where conservation strategies would be confidently assigned based on high agreement among climate projections varied substantially across regions. In general, there was more agreement in forward and backward velocity estimates among alternative projections than agreement in estimates of local dissimilarity. Consensus of climate predictions resulted in the same conservation recommendation assignments in a few areas, but patterns varied by climate exposure metric. However, despite the variability that was observed among climate predictions in combination with conservation values, the researchers observed agreement in the direction of many individual climate variables used to calculate the multivariate PC scores, even though the degree of change varied. An important exception, however, was found in predictions of the direction of precipitation change (both annual and seasonal). This work demonstrates an approach for explicitly evaluating alternative predictions in geographic patterns of climate change.