Detecting, Classifying and Locating Informal Settlements in Megacities such as Cape Town Using Sentinel-1 data

Mgaga, Mfanafuthi

Detecting, Classifying and Locating Informal Settlements in Megacities such as Cape Town Using Sentinel-1 data

Thesis / Dissertation

2026

Publisher

University of Cape Town

Department

Department of Electrical Engineering

Faculty

Faculty of Engineering and the Built Environment

Abstract

The use of remote sensing is a method for generating land cover maps. This research utilizes radar remote sensing, specifically the Sentinel-1 sensor. A major advantage of radar remote sensing is its ability to penetrate clouds and vegetation, enabling data collection in all weather conditions. It also provides valuable information on surface roughness and moisture content, improving the accuracy and detail of land cover maps. The sensor was chosen for its ability to acquire data in any weather conditions, as well as during both day and night. Images were downloaded and processed in SNAP software. The objective of this study was to detect, classify, and map land use and cover, with a focus on informal settlements, using radar remote sensing and GIS techniques. The research had two main objectives: firstly, is to classify land use and land cover using supervised classification, and secondly to conduct an accuracy assessment using QGIS with relevant plugins. The methodology is outlined in Chapter 3, and implementation is detailed in Chapter 4. Results are discussed in Chapter 5, and the conclusion is presented in Chapter 6. Although there are multiple algorithms for supervised classification, random forest was chosen due to its ability to reduce overfitting in decision trees, improve accuracy, and handle both regression and classification problems. It works well with continuous and categorical values and automates missing data. Classified land cover categories included agricultural land, grassland, bare soil, forested areas, formal settlements, informal settlements, industrial areas and water bodies. While SNAP lacks built-in functionality for accuracy assessment, it can provide classification accuracy based on the classifier's threshold and replication of classes from training data. The overall classification accuracy was found to be 72.83% with a kappa coefficient of 0.65 when considering no-data pixels. Excluding no-data pixels improves accuracy to 83.24% with a kappa coefficient of 0.78. Informal settlements, the main class of interest, achieved a consumer accuracy of 91.8% and a producer accuracy of 90.7% Based on the results in Chapter 5, the research met its objectives. The results are valuable for city planners and decision-makers, particularly for rapid response and rescue during natural disasters like floods. It also aids in planning services for informal communities that may not otherwise receive attention due to their exclusion from city or municipal maps.