The various extreme weather events that occurred globally in 2021, from Europe to China to North America, served as yet another reminder that robust strategies for climate adaptation are crucial at a time of rapid global warming. Building resilient communities and lessening the impact that natural disasters have on vulnerable infrastructure can be aided by automated systems driven by machine learning algorithms trained on Earth observation data. When deployed, computer vision models can analyze satellite imagery in real time and inform decision makers and nongovernmental organizations about the timely and targeted allocation of resources and humanitarian aid personnel to affected areas. Here, we overview several specific 2021 extreme events and the factors that caused the loss of life, damage to infrastructure, and economic loss. The events surveyed include flooding in Germany, wildfires in Greece, and Hurricane Ida in the Eastern United States. Taking this information into account, we further discuss barriers to the large-scale deployment of current machine learning technologies, especially models trained on Earth observation data. We examine the limitations of satellite imagery and big data applications in detecting damage and building collapse and how Interferometric Synthetic Aperture Radar (InSAR) can be a tool to resolve existing issues. The aim of this work is to understand why many state-of-the-art models being developed have not yet been successfully and extensively deployed in the real world and to foster discussion about optimizing the use of deep learning technology to save lives and lead effective disaster management efforts.
Natural disasters ravage the world's cities, valleys, and shores on a monthly basis. Having precise and efficient mechanisms for assessing infrastructure damage is essential to channel resources and minimize the loss of life. Using a dataset that includes labeled pre- and post- disaster satellite imagery, the xBD dataset, we train multiple convolutional neural networks to assess building damage on a per-building basis. In order to investigate how to best classify building damage, we present a highly interpretable deep-learning methodology that seeks to explicitly convey the most useful information required to train an accurate classification model. We also delve into which loss functions best optimize these models. Our findings include that ordinal-cross entropy loss is the most optimal loss function to use and that including the type of disaster that caused the damage in combination with a pre- and post-disaster image best predicts the level of damage caused. We also make progress in the realm of qualitative representations of which parts of the images that the model is using to predict damage levels, through gradient class-activation maps. Our research seeks to computationally contribute to aiding in this ongoing and growing humanitarian crisis, heightened by climate change. Specifically, it advances the study of more interpretable machine learning models, which were lacking in previous literature and are important for the understanding of not only research scientists but also operators of such technologies in underserved regions.