RINA - Démonstrateur d’utilisation de l’intelligence artificielle pour la gestion opérationnelle des risques naturels gravitaires géologiques | FEREC 2020

The proposed model also found the correlation with the rain and, as the classical method, indicated that the previous day was the most directly related to rock falls, but with the proposed method other correlations appear. The table ? shows the maximum correlation for each feature and it is interesting to see how the temperature, specially the minimum temperature, is related with the rock falls almost as much as the daily precipitation. In most cases this could be due to freeze thaw cycles, but it is not likely as temperatures in La Réunion do not descend below zero. An explanation could be the correlation between the temperature and rain as both are seasonal events. Table 1 Correlations between the chosen meteorological factors and the daily number of rockfalls; results obtained with the proposed method on the real databases. Only the maximum correlations are presented here. Source: Delonca et al. (2014) 3.2. First analysis of rockfall events Before creating models it is important to understand the behaviour of the cliff based on the data. Basic statistics and visual representations can help with understanding the relying dynamics of events and limitations of results. Observing the times series of daily number of events(Figure 8) the difference before and after 2008 is easy to see. It is also possible to observe a gap in 2004 as part of the data seems to be missing for this year. There are no remarks of any important changes during this year that would cause such a 15