A Data-Driven Approach to Assess Coastal Vulnerability: Machine Learning from Hurricane Sandy

Abstract:

As climate changes and population living along the coastlines continues to increase, an understanding of coastal risk and vulnerability to extreme events becomes increasingly important. With as many as 700,000 people living less than 3 m above the high tide line, New York City (NYC) represents one of the most threatened among major world cities. Recent events, most notably Hurricane Sandy, have put a tremendous pressure on the mosaic of economic, environmental, and social activities occurring in NYC at the interface between land and water. Using information on property damages collected by the Civil Air Patrol (CAP) after Hurricane Sandy, we developed a machine-learning based model able to identify the primary factors determining the occurrence and the severity of damages and intended to both assess and predict coastal vulnerability. The available dataset consists of categorical classifications of damages, ranging from 0 (not damaged) to 5 (damaged and flooded), and available for a sample of buildings in the NYC area. A set of algorithms, such as Logistic Regression, Gradient Boosting and Random Forest, were trained on 75% of the available dataset and tested on the remaining 25%, both training and test sets being picked at random. A combination of factors, including elevation, distance from shore, surge depth, soil type and proximity to key topographic features, such as wetlands and parks, were used as predictors. Trained algorithms were able to achieve over 85% prediction accuracy on both the training set and, most notably, the test set, with as few as six predictors, allowing a realistic depiction of the field of damage. Given their accuracy and robustness, we believe that these algorithms can be successfully applied to provide fields of coastal vulnerability for future extreme events, as well as to assess the consequences of changes, whether intended (e.g. land use change) or contingent (e.g. sea level rise), in the physical layout of NYC.