Arnoux, F.,; Abadie, S.; Bertin, X.; Kojadinovic, I. 2018. A database to study storm impact statistics along the Basque Coast. In: Shim, J.-S.; Chun, I., and Lim, H.S. (eds.), Proceedings from the International Coastal Symposium (ICS) 2018 (Busan, Republic of Korea). Journal of Coastal Research, Special Issue No. 85, pp. 806–810. Coconut Creek (Florida), ISSN 0749-0208. A database aiming at studying storm impact on the French Basque coast and covering the period 1949–2015 is presented in this paper Wave and water level data originate from synthetic or numerical hindcast complemented by buoys or tide gauge measurements. A statistical validation of each variable and, when needed, a distribution correction of biased variables was carried out. The constructed database will be used in confrontation with an historical review of past storm damages to try to find a statistical link between the two though a damage function.

A Database to Study Storm Impact Statistics along the Basque Coast

https://hal-univ-pau.archives-ouvertes.fr/hal-03200009/document

Coastal flooding event definition based on damages: Case study of Biarritz Grande Plage on the French Basque coast

Data about storm impacts are essential for the disaster risk reduction process, but unlike data about storm characteristics, they are not routinely collected. In this paper, we demonstrate the high potential of convolutional neural networks to automatically constitute storm impact database using timestacks images provided by coastal video monitoring stations. Several convolutional neural network architectures and methods to deal with class imbalance were tested on two sites (Biarritz and Zarautz) to find the best practices for this classification task. This study shows that convolutional neural networks are well adapted for the classification of timestacks images into storm impact regimes. Overall, the most complex and deepest architectures yield better results. Indeed, the best performances are obtained with the VGG16 architecture for both sites with F-scores of 0.866 for Biarritz and 0.858 for Zarautz. For the class imbalance problem, the method of oversampling shows best classification accuracy with F-scores on average 30% higher than the ones obtained with cost sensitive learning. The transferability of the learning method between sites is also investigated and shows conclusive results. This study highlights the high potential of convolutional neural networks to enhance the value of coastal video monitoring data that are routinely recorded on many coastal sites. Furthermore, it shows that this type of deep neural network can significantly contribute to the setting up of risk databases necessary for the determination of storm risk indicators and, more broadly, for the optimization of risk-mitigation measures.

Automatic creation of storm impact database based on video monitoring and convolutional neural networks

Bayesian networks are probabilistic graphical models that are increasingly used to translate hydraulic boundary conditions during storm events into onshore hazards. However, comprehensive databases that are representative of the extreme and episodic nature of storms are needed to train the Bayesian networks. Such databases do not exist for many sites and many Bayesian networks are trained on data generated by process-based models. To our knowledge, they have not been trained exclusively on observational data for storm impact modeling. This study aims to explore the performance in coastal flooding prediction of a Bayesian network exclusively based on observational data. To this end, we take the "Grande Plage" of Biarritz (South west of France) as a test case. The network is trained using data from several monitoring networks located near the study site. Because observational data about storm impact regime are limited, a second aim of this work is to propose a methodology based on statistical learning methods to complement the data about this variable. This methodology aims to select the statistical learning method with the best generalizing ability with a cross validation. Two Bayesian networks are trained, one exclusively on the observational data and one with both observational and predicted data. To compare the two networks, their performances are evaluated on the same events. We demonstrated that it is possible to predict coastal flooding risk in a qualitative manner with a Bayesian network based only on observational data with a $$F_1$$ -score, a measure combining precision and recall, of 0.628. However, the predictive skill of this network is questionable for the most intense storm impact regimes which are impact and overwash regimes. Storm impact data is extended with the random forest method which showed the best generalizing ability based on cross-validation. This extension of the database led to a better Bayesian network in terms of predictive skill, with precision, recall and $$F_1$$ -score 7% higher on average than for the network trained only on observational data. 

Bayesian networks to predict storm impact using data from both monitoring networks and statistical learning methods

A coastal storm damage database covering 64 years was built based on press and archives investgations. Flooding of Biarritz La Grande Plage sea front and related damages were characterized with a three level index along with binary reliability indicator. Damages to breakwaters protecting Saint Jean de Luz bay were also documented. Building such a database appeared to be extremely time consuming and also implies some limitations mainly due to the heterogeneity and incompleteness of the data sources. 30 flooding events were reported for Biarritz among which the five most destructive ones were December 28-29, 1951, November 9-10, 2010 and January 4-7, February 1-2, March 3-5, 2014. This clearly shows the extreme character of winter 2014. \textcopyright Coastal Education and Research Foundation, Inc. 2018.

Marion Beauvivre

Papers

A Database of Recent Historical Storm Impact on the French Basque Coast