Assessing risks at the land-sea interface: the case study of the Veneto coastal area

Extreme weather events are causing severe threats all over the world, posing increasing environmental and socio-economic risks, which are amplified by the effect of climate change. Coastal areas are particularly vulnerable to extreme marine and weather events (e.g., storm surges, extreme rainfall) given the high exposure of population, settlements, and economic activities at the land-sea interface.
Therefore, understanding the main risk factors of these extreme events is necessary to implement suitable disaster risk management measures, which could guide coastal authorities and policy-makers in improving the resilience of coastal communities to natural hazards and climate change.
Nevertheless, identifying the triggering factors of such risks has always been challenging, since the complex dynamics driving the coastal systems. In this regard, in order to unveil relations between hazards and their cascading effects, in recent years, Machine Learning (ML) algorithms have gained popularity due to their ability to extract information from a huge quantity of data, by overcoming the limits of traditional physical-mathematical models. However, the outcomes of these advanced methods, to be reliable, must be corroborated through traditional statistical analysis and scientific reasoning.
Based on these needs, my Thesis was aimed at investigating the factors that played a key role in the occurrence of damages (e.g., damages to people, buildings and infrastructures, agriculture, tertiary sector) generated by extreme weather events in the coastal municipalities of the Veneto region, focusing on the 2009-2019 timeframe. Accordingly, the aim was achieved by reviewing the scientific literature concerning the state-of-the-art ML algorithms implemented for assessing risks and impacts caused by natural hazards in coastal areas, as well as by applying traditional and ML-driven techniques (i.e., a Random Forest Model) of data science to find relations between the analyzed factors and the damage occurrences.
Overall, the developed methodology pointed out some interesting relationships between the triggering factors and the occurred damages in the case study area within the analyzed timeframe. These findings can pave the way for guiding decision-makers and local stakeholders in the development of suitable disaster risk reduction and climate adaptation measures, aimed at increasing the resilience of coastal communities to extreme weather events.
Finally, although the research encountered some limits due to the type and the resolution of the data, the results and the criticalities evidenced by this study could be useful for the implementation of advanced ML algorithms (e.g., Graph Neural Networks, Artificial Neural Networks) intended to predict damage occurrences in coastal areas.