Mapping Informal Settlements Using Machine Learning Techniques, Object-Based Image Analysis and local Knowledge

Abstract

The existence of informal settlements in Riyadh City, the Kingdom of Saudi Arabia (KSA), has given rise to some urban planning issues. To provide improvements to mapping and planning processes, the current study aims to evaluate and characterize informal settlements within the city using object-based machine learning (ML) techniques (specifically, Random Forest (RF) and Support Vector Machine (SVM)), expert knowledge (EK) and satellite data. An examination of four defined locales has produced a comprehensive, local, informal settlement ontology. Four main categories (shape, geometry, texture, and pattern) were used to build the ontological framework. Expert local knowledge was employed to produce a ruleset to accurately identify and map these areas. Specific indicators identified by the specialists were used in a combined object-based ML and image analysis (OBIA) approach, with high-resolution worldview-3 imagery used as input data. Results demonstrated that combining EK and ML with remotely sensed data can efficiently, effectively and accurately distinguish informal settlement areas. This work has shown that an object-based ML technique (RF), in combination with EK about important local environment indicators, provides a useful method for mapping informal settlements.