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

Figure 18:Variation of the TWI along the road. This spatial study aimed to better understand the characteristics of the cliff which were related to the rock falls. Different values were plotted in function of the pr. The clearest correlation is that in zones with protection there are less rock falls. Observing Figure 9 and 11 it is possible to observe the difference of the amount of rock falls before and after the addition of protective elements along the road. Analysing the data before the constructions of 2008 the data already showed that the initial section (1+700 - 3+500) the number of events was visibly lower. The lithological structures do not seem to be strongly related with the distribution of events, but a deeper study could show some correlation. On The other hand, there are some correlations between the topography and the number of events. For values of the standardized area (Figure 17) closer to 0 (example 4+000 - 6+000) more rock falls tend to fall (Figure9). The region with a higher standardized area close to 0,5 ( example 6+200 - 7+500) has fewer rock falls. This spatial analysis shows how the number of rock falls are related to a number of different characteristics of the bedrock. A spatial analysis could be done using those features as the input in order to predict the location a rock will fall. 4. Methodology used for prediction models AI was used to create models that will predict the probability of rockfall that reach road. if rock falls will reach the road based on the data. There are various ways to create prediction models and various criterias to measure their capacities. The objective is, given a certain amount of information, what will happen in another case. All the data could be given to the program as an input, but then it would only be possible to evaluate the model based on the information it used to produce the model. One solution is separating the data into two parts, training data and 25