Climatic attribution via neural network modeling

As well known, Global Climate Models (GCMs) are the main tool for performing attribution studies of the recent climate change. Their results show quite clearly that the trend of the recent warming at global scale has been mainly driven by changes in anthropogenic forcings. However, for changes occurred at other scales and/or related to different variables and limited periods (decades), attribution is much more uncertain. Recently, the attribution problem has been tackled by a distinct strategy, through neural network (NN) modeling. First of all, this change of perspective has permitted to find again (by a completely distinct method) the results obtained by GCMs about the fundamental relevance of changes in anthropogenic forcings for reconstructing the recent global warming, so corroborating our view of this phenomenon. Secondly, a clearer attribution of temperature and precipitation changes at regional scale has been obtained. Finally, at both global and regional levels, the role of internal climate variability has been studied by NN modeling and the relevance of some ocean-atmosphere oscillations has been recognized: in particular, this leads to a more accurate reconstruction of temperature changes at decadal scale. Here, recent studies of attribution via NN modeling are reviewed and the results of new investigations are presented.