Deprived area mapping using a scalable, transferable and open-source machine learning approach
Open AccessPopulation living in deprived areas continues to grow, highlighting the urgent need for accurate high-resolution maps and granular statistics to plan interventions and monitor changes. Satellite imagery provides a promising solution to offer consistent and accurate high-resolution maps globally. However, most studies have focused on using very high-resolution images, which often only cover small areas, and the transferability of models to other cities remains uncertain. Moreover, most researchers focused on the use of remote sensing data which provides a partial view of deprived areas missing equally important socioeconomic aspects. This study proposes an inexpensive, scalable and transferable approach to map deprived areas using freely available Sentinel-2 images and open geospatial data. The models were trained and tested on three individual cities: Lagos Nigeria, Accra Ghana, and Nairobi Kenya. The study first analyzed the morphological characteristics of deprived areas and grouped them into types based on aerial and ground-level information. This allows us to understand the inter- and intra-urban diversity of deprived areas. A multi-step modeling approach was carried out including an individual city modeling, city to city modeling and a generalized model using input features from contextual features, geospatial features, and a combination of both. Machine learning algorithms were evaluated for mapping deprived areas, including multi-layer perceptron (MLP), Random Forest, and XGBoost. The scalability of model performance was examined by using patches of the different deprived types identified. The study also tested the models' ability to map deprived areas in other cities, assessing their transferability and which deprived types can be easily mapped across cities. Results indicate that deprived areas have heterogeneous local characteristics that affect large area mapping, and the top features for each city show that models are sensitive to the spatial structures of deprived area types. Model trained in one city is able to predict deprived areas in other cities with an F1 score over 65%. The generalized model proves to be more beneficial for modeling multiple cities. This approach offers a promising solution for accurately mapping deprived areas globally, supporting stakeholders to plan geo-targeted interventions and monitor changes.
- 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.