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

3.1.2. Statistical Study of rock falls on the RN1 Many different factors are known to cause rock falls. The two main ones are the pressure caused by water infiltrating and freeze-thaw cycles. A study made by Delonca [Delonca, 2014] analysed the correlation between the different meteorological factors and rock falls trying to determine which factors are linked to rock falls and which one is the most important. The general difficulty of studying rock falls is the scarcity of data, making it difficult to find important correlations between the event and the meteorological condition. A new method is proposed in order to find correlation when events are rare. The proposed methodology weights the number of rockfalls by the probability of occurrence of the studied triggering factor and this is done by grouping the days based on the descriptive feature, calculating the probability of having a day with an event inside each grouping. The final step is a linear regression analysis which is performed to compare the magnitude of the triggering factor and the probability of occurrence for each value. The grouping is arbitrary but two rules should be followed, the number of days in each grouping should be superior to 5 and each grouping should contain at least one event. This analysis was made for the rain, the accumulated rain, the maximum temperature, the minimum temperature and the difference between both. For each factor studied it would also be compared with the value of the previous days as the triggering factor may take some time to detach the block. It is important to note that the correlation between all of those factors and the presence of an event was calculated, but only the correlation between the rain and the events was seen as sufficient. Figure 7 shows the values of the cross-correlation coefficient between the daily number of rockfalls and the daily rainfall at RN1 (La Réunion) for several delays and how the most significant coefficient is obtained for the rain from the previous day. Figure 7: Cross-correlation coefficient between the daily number of rockfalls and the daily rainfall at RN1 (La Réunion) for several delays. The significance threshold, equal to 0.031, is represented by the dashed lines. Source: Delonca et al. (2014) 14