Collective Intelligence 2015 May 31 – June 2, 2015 Santa Clara, CA Collective Intelligence for Rain Prediction Ofer Arazy, University of Alberta and University of Haifa Noam Halfon, Israel Meteorological Service Dan Malkinson, University of Haifa Collective intelligence is shared (or group) intelligence that emerges from the collective efforts of many individuals (Woolley et al., 2010). It is the aggregate of individual contributions: from simple collective decision making (using mechanisms such as a group’s average, or voting) to more sophisticated aggregations such as in crowdsourcing and peer-production systems (e.g. Wikipedia) (Lévy and Bonomo, 1999). In particular, collective intelligence could be used in combining forecasts and making predictions about future events (Clemen, 1989), for example by using prediction markets to forecast election results, stock prices, or the outcomes of sport events (Wolfers and Zitzewitz, 2004, Tziralis and Tatsiopoulos, 2012). While the ability of collectives to outperform individual experts in certain situations is by now well recognized (Surowiecki, 2005), our understanding of the factors contributing to collective intelligence is far from complete. To date, there is little research regarding the use of collective intelligence for prediction of weather forecasting. Collective intelligence has been recently used in the context of climate, for example in employing citizens to collect and report meteorological data 1 (Nov et al., 2014) and in the collaborative development of proposals for tackling issues related to climate change 2 (Introne et al., 2013). Notwithstanding the critical importance of meteorological phenomena to individuals, economy, and the society at large, and despite the centuries-old human interest in weather forecasting (Cox, 2002), collective intelligence systems have been rarely employed in the prediction of weather events 3 . The objective of this study is to investigate the extent to which collective intelligence could be utilized to accurately predict weather events, and in particular rainfall. The literature points to the importance of several key factors in determining the accuracy of human prediction (Mellers et al., 2014) and the determinants of collective intelligence (Larrick et al., 2012). Building on these prior studies, we explore the effect of three factors on the accuracy of the group’s collective prediction, namely: (a) design mechanisms (specifically, transparency: the extent to which information of other group members’ predictions are made available); (b) group members’ knowledge (both knowledge of climate phenomena and familiarity with the study’s geographical site); and (c) groups’ diversity (in terms of members’ place of residence). Our analyses employ metrics of group intelligence (i.e. Collective Intelligence Quality, Win Ratio), as well as compare the accuracy of groups’ predictions against the predictions of the standard model used by the National Meteorological Services (European Centre for Medium-Range Weather Forecasts; ECMWF). We chose Israel as the site for this investigation, given the country’s irregular 1 For example, the Citizen Weather Observer Program (CWOP; http://wxqa.com/) or the Community Collaborative Rain, Hail and Snow Network (CoCoRaHS; http://www.cocorahs.org/). 2 For example, see the works of the MIT Climate CoLab (http://climatecolab.org/). 3 For an exception, see Hueffer et al. (2013).