Monitoring and Modeling Gentrification Using Remote Sensing and Geospatial Technologies




Journal Title

Journal ISSN

Volume Title



Understanding gentrification is crucial for urban poverty reduction and sustainable development. Traditional studies mainly focus on conceptualizing the causes and impacts of urban gentrification. However, delineating spatial and temporal distributions of gentrification is often missing in these studies. Recent advancements in technology, including Remote Sensing, Geographic Information Systems, and machine learning, have provided new opportunities to better understand the spatial and temporal variations of urban gentrification. This dissertation presents a novel approach for monitoring and modeling gentrification by taking advantage of these technological advancements. In this study, two machine learning-based models have been developed to identify gentrification in two cities, Washington, DC, and New York City, using remote-sensing time series and building reconstruction data. The New York City building reconstruction dataset is used to represent the crucial mid-phase for gentrification projects, and provides a large number of training samples in modeling gentrification. The gentrification pattern identification model and early detection model are validated with various evaluation metrics. Results indicate that the models developed in this study can accurately distinguish gentrified buildings from regular urban structures through measuring and generalizing the remote-sensing time series. The validating experiments suggest that the overall accuracy reaches 0.87 for the early detection model. Additional experiments show that the precision and recall for identifying gentrified buildings reach at least 0.80 and 0.74, respectively. Moreover, the performance metrics for identifying ungentrified buildings are relatively higher than gentrified buildings, indicating that the developed models are not overtrained. The study also demonstrates the possibility of transferring the early detection model to border regions when construction data are unavailable. An alternative data source is needed for cities that do not have building construction data to map large-scale gentrification. The experiment compares the GEDI L2 canopy heights and nDSM measures in Washington, DC. The result shows that the canopy heights from GEDI and nDSM data are highly correlated (R2 = 0.79) in certain areas. The experiment suggests that it is possible to extract building height information from the waveforms captured by the GEDI sensor, providing a potential alternative to construction datasets for training machine learning models and mapping gentrification globally. Future research will be devoted to modeling gentrification with the GEDI data and examining more cities outside the two study areas. This study utilizes new technology to identify gentrification in cities, which can help scientists better understand the spatial-temporal distribution of gentrification. The developed machine learning-based models can easily be transferred into different regions, especially cities experiencing rapid urban expansion in developing nations. Mapping gentrification timely and accurately will help policymakers and urban planners make better decisions. The research strengthens the understanding of gentrification in cities for scientists and experts, enhances social and environmental justice for urban poor and displaced populations, and contributes to the 2030 United Nations Sustainable Development Goals.



Gentrification, GIS, Land use change, Machine learning, Remote Sensing, Urban poverty