Small Area Population Modeling with Contextual Image Features and Random Forest Regression
Open AccessAccording to the United Nations, population growth is expected to occur primarily in urban areas, and urban areas in developing countries in Africa and Asia are likely to experience the most extreme urban population growth. To mitigate the environmental, economic, social, and human-health consequences associated with urbanization in such areas, access to spatially explicit information relating to population distribution is imperative. Decennial census data has often been used for this purpose, but when available, this data is regularly provided at a coarse scale or is out of date. To address these issues, modeling approaches that use ancillary data to disaggregate existing census data to a finer spatial resolution or directly predict population in areas where it is unknown have been developed. This research presents modeling techniques for population prediction and census data disaggregation that combine Random Forest regression and contextual features (i.e. filters capturing the texture and orientation of objects) derived from high-resolution satellite imagery. To assess their effectiveness in regions undergoing rapid population growth, the modeling approaches are tested in Colombo, Sri Lanka and Accra and Kumasi, Ghana. The disaggregation results demonstrated that the proposed technique could be used to accurately disaggregate census data onto an approximately 20-meter spatial resolution grid for each study area. While the results of the direct population prediction technique showed that a strong relationship exists between population density and contextual image features, it was unable to predict fine-scale population density with a high degree of accuracy. Overpredictions of population due to the presence of non-residential commercial buildings and underpredictions due to dense informal settlements were common for both the disaggregation and direct prediction techniques, but more severely impacted direct prediction. This indicates that additional information such as land use data may be required to accurately estimate populations at a fine scale.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
Copenhaver_gwu_0075M_15318.pdf | 2020-09-10 | Open Access |
|