Prediction of hydrogeological risks in Italy: data taxonomy, database creation and preliminary regressor analysis

In an increasingly globalized environment, vulnerabilities emerge from an increasingly interdependent and interconnected world, and risks are no exception: climate change is an example of how disasters can express themselves on a global scale and transfer rapidly from one sphere to another. Arisk offers itself as a solution by developing software with underlying algorithms that can measure any type of risk objectively and comparably over time and space. The final objective of the work is to analyze the correlation between hydrogeological risk and the financial performances of small and medium-sized enterprises, with a focus on wineries. This thesis covers only the first part of the overall work and aims to create accurate models for forecasting total annual rainfall. Specifically, it seeks to understand what correlation (and if any) exists between inputs such as temperature and humidity and annual rainfall. An input is a variable defined as a characteristic surveyed or measured on statistical units and can be: quantitative variable (modes are real numbers) or qualitative variable (modes are non-numeric attributes). To perform predictive modeling (the problem of developing a model using historical data to make a prediction on new data for which the answer is unknown) an initial database containing data from 31 municipalities in the Langhe (Piedmont) for six different years was initially created, and 3 different approaches were used on this: Linear Regression, Neural Network, and Random Forest. Later with the aim of finding the best model, a second database (municipalities in the Prosecco area in Veneto) was created, because by expanding the database the predictive modeling is more effective. The results showed that random forest is the best estimator for this type of analysis and that the total precipitation range can be predicted with 78% accuracy. These results suggest that by further expanding the database and covering diversified areas of Italy, the accuracy of prediction can be improved. On this basis, we can say that it is absolutely worth continuing with the next phases of the study.