Advancing Data for Street-level Flood Vulnerability: Extraction of Variables from Google Street View in Quito, Ecuador

TL;DR

This paper presents a novel methodology for remotely collecting flood vulnerability data using Google Street View and satellite images, enabling scalable and cost-effective mapping of flood risks in urban areas like Quito, Ecuador.

Contribution

It introduces a mixed methods approach combining crowdsourcing and machine learning to efficiently gather flood vulnerability variables from imagery.

Findings

Developed a scalable data collection methodology.

Created a labeled dataset for flood vulnerability analysis.

Demonstrated applicability in Quito, Ecuador.

Abstract

Data relevant to flood vulnerability is minimal and infrequently collected, if at all, for much of the world. This makes it difficult to highlight areas for humanitarian aid, monitor changes, and support communities in need. It would be time consuming and resource intensive to do an exhaustive study for multiple flood relevant vulnerability variables using a field survey. We use a mixed methods approach to develop a survey on variables of interest and utilize an open-source crowdsourcing technique to remotely collect data with a human-machine interface using high-resolution satellite images and Google Street View. This paper focuses on Quito, Ecuador as a case study, but the methodology can be quickly replicated to produce labelled training data in other areas. The overall project goal is to build training datasets that in the future will allow us to automate the mapping of flood vulnerability for urban areas in geographic regions.