Methodology

---

There exist several definitions of slums, but in the context of this project we will refer to a ‘slum’ as being strictly a ‘morphological slum’, i.e., the morphological characteristics that can be observed on remote sensing images, such as the relative higher built-up densities, the lack of ‘formal vegetation’ (compared to regular built-up areas), the small building footprints and irregular layout patterns of areas, which commonly very well coincide with slum areas as defined by UN-Habitat using the five dimensions of slums (access to water, sanitation, overcrowding, housing durability and tenure).

Existing studies usually aim at providing information on the location and extension of slums within the city, or at characterizing the physical environment of slums (i.e., in terms of greenness, built-up density, lacunarity, etc.). In this study, we propose a general framework that allows for the integration of all this information at a reduced cost. First, the location and extent of slums are mapped citywide, using free or low-cost images (HR). This enables us to limit the acquisitions of more expensive imagery (VHR) only to slums for mapping and characterizing them in detail with SoA methods and FOSS solutions.

Our methods will be tested on a set of three sub-Saharan cities, each with slums presenting different characteristics, namely Nairobi (Kenya), Kisumu (Kenya) and Ouagadougou (Burkina Faso). The project will generate various deliverables under the form of GIS layers, thematic maps, databases and tools.

Expected outcomes:

• A better understanding of slum extension and intra-slum variations in sub-Saharan cities.
• Innovative methods for delineating slum areas in cities, using the lowest possible spatial and spectral resolutions.
• Going beyond slum delineation and providing tools and methods for deriving interpretable remotely-sensed indicators usable by authorities and stakeholders to characterize intra-slum variation.
• Exploiting novel and automated methods to extract satellite imagery variables for use as input into SoA image classifications.
• Implementing a novel open-source transferability approach for the detection and characterization of slums, both at the city- and the intra-slum scale.

Expected deliverables:

• Detailed land-cover maps of slums.
• Derived slum indicators useful for policymakers as a result of intra-slum analysis.
• Publication of our general framework as an open-access and open-source tool to perform identification and characterization of slums in sub-Saharan Africa, and to provide recommendations and good practices on the use of RS data for the task at hand.
• Valorization of our outputs in close coordination with stakeholders and local authorities in several SSA cities.
• Citywide probability maps of a location being a slum.

The project is structured according to several work packages:

Mapping Slums at the City Scale (WP2)

The potential of high-resolution imagery (ranging from 10m Sentinel-2 to 1.5m SPOT6-7) for scalable and transferable slum delineation will be assessed. We will carry out systematic exploration using FOSS, varying data sets and methods.

Characterizing Slums (WP3)

We will investigate the potential of very-high resolution imagery (ranging from 1.5m SPOT6-7 to 0.30m WorldView-3) for scalable and transferable slum characterization. State-of-the-art methods based on the use of FOSS will be used to extract a set of indicators from varying data sets, and a cost-benefit analysis will be performed.

Framework for mapping and characterizing slums (WP4)

This work package will link the technology with information needs at different levels and will ensure that research developments contribute to addressing information gaps for end users. So far, not much is known about what information is really required at the national, local and community levels, and how far remote sensing could go to reliably provide them. Ethics considerations on how to make spatial information on slums publicly available in support of pro-poor policies will be developed. The outcome of this circular feedback with end users will lead to the development of a general methodological framework.

Validation methods

A performance assessment will be conducted to compare the different methods and spatial/spectral resolutions. For each case study, the validation will help identify the best-performing approaches. The added value of the higher resolution against acquisition costs and computational resources will be quantified.