Centaur VGI: An Evaluation of Engagement, Speed, and Quality in Hybrid Humanitarian Mapping

Abstract

Volunteered geographic information (VGI) is often cited as a potential solution to persistent global inequalities in map data, particularly in areas undergoing humanitarian crises. Poor volunteer engagement, slow data production, and low-quality outputs have limited progress, however, and can unintentionally exaggerate inequalities. Hybrid machine learning–VGI (ML–VGI) frameworks can help to overcome these challenges through a combination of workflow automation and purposive human input, but the use of these workflows is rare in practice. Here, we implement an ML–VGI framework (Centaur VGI) and undertake a detailed comparative usability assessment against an existing, widely used VGI mapping platform to demonstrate its potential to improve volunteer engagement, mapping speed, and data quality. Our results suggest that through automated building, searching, and labeling, the Centaur VGI platform provides greater usability, quicker data production, and improved data quality for most users. Consequently, we provide the first evidence that hybrid ML–VGI approaches can be used to facilitate increased public participation in humanitarian building mapping efforts and thus help reduce global inequalities in map data.