Data-mining for multi-variable flood damage modelling with limited data

Abstract

Flood damage assessment is usually done with damage curves only dependent on the water depth. Recent studies have shown that data-mining techniques applied to a multi-dimensional dataset can produce significantly better flood damage estimates. However, creating and applying a multi-variable flood damage model requires an extensive dataset, which 10 is rarely available and this can limit the application of these new techniques. In this paper we enrich a dataset of residential building and content damages from the Meuse flood of 1993 in the Netherlands, to make it suitable for multi-variable flood damage assessment. Results from 2D flood simulations are used to add information on flow velocity, flood duration and the return period to the dataset, and cadastre data is used to add information on building characteristics. Next, several statistical approaches are used to create multi-variable flood damage models, including regression trees, bagging regression trees, 15 random forest, and a Bayesian network. Validation on data points from a test set shows that the enriched dataset in combination with the data-mining techniques delivers a significant improvement over a simple model only based on the water depth. We find that with our dataset, the trees based methods perform better than the Bayesian Network. IntroductionBecause flood risk management becomes increasingly risk-based, flood damage estimation is increasingly important in flood 20 risk assessment. Flood risk assessment supports policy makers to decide which flood risk management measures are most efficient in reducing flood risks and how much investment is cost-efficient. With the European Union Floods Directive (EC, 2007) now fully in place, national flood risk assessment are being developed with the final aim to support flood risk management plans. In the Netherlands, such flood damage assessment has been used to derive the optimal protection standard for flood protection (Kind, 2013; van der Most, 2014), using the current Dutch standard method for damage 25 modelling (Kok et al., 2005). Also for insurance applications, more precise estimates of flood damages are required.Flood risk assessments require flood damage models. These models typically predict the fraction of damage based on the water depth, and average building repair and replacement costs for different types of buildings (Messner et al., 2007;Jonkman et al., 2008). When validated, such simple flood damage models often don't perform well (e.g. Jongman et al., 2012). This is because water depth alone cannot explain the full complexity of the flood damaging processes and several 30 studies have only found low correlation coefficients (typically below 0.5) between the water depth and the flood damage (e.g. Merz et al., 2013, Pistrika&Jonkman, 2009). Furthermore, often no local data is available on flood damage and Nat. Hazards Earth Syst. Sci. Discuss.,

Testing empirical and synthetic flood damage models: the case of Italy

“…8). This confirms previous findings (Wagenaar et al, 2017b) and justifies the use of depth-damage curves for post-disaster assessment. Flow velocity and geometric characteristics of buildings (area and perimeters) are also important (factor 2.7 to 2.3), followed by other predictors such as building value, flood duration, number of storeys, finishing level (quality) and type of structure (factor 1 or less).…”

“…The vast majority of univariable flood damage models account for water depth as the only explanatory variable to explain the often complex relationship between the magnitude of a flood event and the resulting damages; however, other parameters may influence the flood damage process, such as flow velocity (Kreibich et al, 2009), flood duration and water contamination (Molinari et al, 2014;Thieken et al, 2005), just to name a few. In addition, non-hazard factors can be significantly different from one place to another, such as the type and quality of buildings, presence of basements, density of dwellings, early warning systems and precautionary measures (Cammerer et al, 2013;Carisi et al, 2018;Kreibich et al, 2005;Merz et al, 2013;Penning-Rowsell et al, 2005;Pistrika and Jonkman, 2010;Schröter et al, 2014;Smith, 1994;Wagenaar et al, 2017b). Multivariable models (MVMs) can account for such additional factors and thus are able to adapt to the characteristics of a specific event and location.…”

“…artificial neural networks and random forests) (Merz et al, 2013;Spekkers et al, 2014;Kreibich et al, 2017;Carisi et al, 2018), Bayesian networks (Vogel et al, 2013) and Tobit estimation (Van Ootegem et al, 2015). Moreover, some MVMs support probabilistic analysis of damage (Dottori et al, 2016;Essenfelder, 2017;Wagenaar et al, 2017a). MVMs need to be validated against empirical records from the region where they are applied in order to produce reliable estimates (Hasanzadeh Nafari et al, 2017;Molinari et al, 2014Molinari et al, , 2019Scorzini and Frank, 2017;Zhou et al, 2013).…”

Analysing flood fatalities in Vietnam using national disaster database and tree-based methods

“…Recently they have been applied in hydrology studies (Ali et al, 2010;Carlisle et al, 2010;Loos and Elsenbeer, 2011) and flood risk studies (Merz et al, 2013;Spekkers et al, 2014;Chinh et al, 2015;Hasanzadeh Nafari et al, 2016;Wagenaar et al, 2017). A review of recent literature on the topic reveals that applications in flood risk studies are very recent.…”

“…A review of recent literature on the topic reveals that applications in flood risk studies are very recent. Four in five articles used tree-based methods to select the substantial flood damage influencing parameters for different case studies 20 using MATLAB software (Merz et al, 2013;Chinh et al, 2015;Hasanzadeh Nafari et al, 2016;Wagenaar et al, 2017). Spekkers et al (2014) explored damage-influencing factors on insurance claims regarding water-related damage using decision-tree analysis and variable importance with statistical software R. There has not been any empirical study on the application of tree-based methods approach to analyse the damage-influencing parameters on flood fatalities.…”