Climate Model Diversity: Future Climate Predictions from the CMIP5 Multi-Model Ensemble

Efforts to understand how Earth’s climate system responds to external radiative or anthropogenic forcing are crucial for informing present and future human behaviour. Climate models are our greatest tool for understanding how and why the climate changes in response to these driving forces. Simulations of the climate under different socioeconomic and emission scenarios yield quantitative predictions of our future. There are two good reasons why the scientific community has and continues to develop multiple independent climate models. Firstly, it is a critical truth that the mean of a collection of models has smaller error, ie. is more accurate, than any one individual model (Flato 2011; Annan & Hargreaves 2010; Lambert & Boer 2001). All individual models are outperformed by both the mean and the median of the ensemble. The second reason is the fact that no single model outperforms the others in all respects (Flato 2011). One model may perform best for the temperature climate variable, but may not hold the same title with respect to precipitation. As expected, model diversity yields an equally diverse range of future climate predictions. Even small differences in the predicted value of one climate variable, for example global average temperature, can translate to extreme changes to the way of life for ecosystem inhabitants. It is therefore very important to understand model diversity and isolate the aspects of climate models from which it arises. What is it about different models that lead them to predict different futures? On the flip side, model diversity can and should be taken advantage of in efforts to quantify the uncertainty of future climate predictions. The ensemble mean benefits from “cancellation of errors” (Flato 2011), enabling more accurate quantification of uncertainty. International collaborations and standardized experiments, such as the Coupled Model Intercomparison Project (CMIP; Taylor et al. 2012; Meehl et al. 2009; Hibbard et al. 2007) are key for making steady progress towards understanding climate variability, climate change, and garnering robust predictions of the future. In this work, we obtain the future climate predictions of the CMIP Phase 5 (CMIP5) multi-model ensemble under four radiative or anthropogenic forcing scenarios, and analyze the diversity in this collection of models both qualitatively and quantitatively. We focus on two specific large-scale climate variables: global mean temperature and average monthly precipitation. In order to quantify the diversity in the range of future climate predictions, we invoke the concept of the “envelope,” which we define as the region within one standard deviation of the ensemble mean. To understand the source of the differences, we discuss the inner workings of overarching model classes and universal concepts in climate modelling.