Chapter 15Dwelling Type Classification
—Md Nasir and Anshu Sharma
Executive Summary
Neighborhood vulnerability and risk assessment is essential for effective disaster preparedness. Due to their dependency on time-consuming and cost-intensive field surveying, existing systems of neighborhood vulnerability and risk assessment do not provide a scalable way to assess the precise extent of natural-hazard risk at a hyper-local level. Here, we used machine learning to automate the process of identifying dwellings and their type, allowing us to build a potentially more effective and expansive disaster vulnerability assessment system.
First, we used satellite images of low-income settlements and vulnerable areas in India to identify seven different dwelling types. Specifically, we formulated the dwelling type classification as a semantic segmentation task (identifying and classifying discrete dwellings and their types) and trained a U-Net–based neural network model with the data we collected. Then, we used a risk score assessment model that incorporated the determined dwelling type along with a flood inundation model of the regions. We used 2020 data obtained prior to natural hazards that occurred in India thereafter. We collected post-disaster ground-truth data from those regions to validate the efficacy of our model. We found that our model performed well. This work can drive preemptive action by providing household-level risk indicators that can inform decision-making by disaster ...