Using pluralist research strategies can be a profitable way to study complex systems. This contribution focuses on the approaches for studying the climate that make use of multiple different models, aiming to increase the reliability (in terms of robustness) of attribution results. It is examined how the traditional approach, which is based on ensemble runs of Global Climate Models, allows only partially the application of a robustness scheme, owing to the difficulty to match or evaluate the conditions required for robustness (i.e., independence or heterogeneity among models). An alternative "multi-approach" strategy is advanced, beyond dynamical modelling but still preserving the idea of model pluralism. Such a strategy, which uses a set of ensembles of different model-types by combining dynamical modelling with data-driven methodological approaches (i.e., neural networks and Granger causality), seems to better match the condition of independence. In addition, neural networks and Granger causality lead to achievements in attribution studies that can complement those obtained by dynamical modelling.
Climate model pluralism beyond dynamical ensembles
Mazzocchi F;Pasini A
2017
Abstract
Using pluralist research strategies can be a profitable way to study complex systems. This contribution focuses on the approaches for studying the climate that make use of multiple different models, aiming to increase the reliability (in terms of robustness) of attribution results. It is examined how the traditional approach, which is based on ensemble runs of Global Climate Models, allows only partially the application of a robustness scheme, owing to the difficulty to match or evaluate the conditions required for robustness (i.e., independence or heterogeneity among models). An alternative "multi-approach" strategy is advanced, beyond dynamical modelling but still preserving the idea of model pluralism. Such a strategy, which uses a set of ensembles of different model-types by combining dynamical modelling with data-driven methodological approaches (i.e., neural networks and Granger causality), seems to better match the condition of independence. In addition, neural networks and Granger causality lead to achievements in attribution studies that can complement those obtained by dynamical modelling.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.