Machine learning maps landslide risk zones in eThekwini to improve urban safety
Researchers in eThekwini use machine learning and satellite data to identify landslide-prone areas, aiding safer urban planning after 2022 floods.
Researchers in eThekwini have applied machine learning techniques combined with satellite data to identify the city's most landslide-prone areas. This follows the devastating April 2022 floods and landslides that caused hundreds of deaths, displaced over 40,000 people, and inflicted roughly R36 billion in damage.
The study, published in Scientific Reports, used geographic information systems, satellite imagery, and two machine learning models-Random Forest and Support Vector Machines-which achieved over 98% accuracy in mapping landslide susceptibility. The findings highlight that urban development activities such as road construction, property development, and vegetation loss significantly increase landslide risk.
Key factors influencing landslide occurrence include land use, elevation, vegetation cover, and geological conditions. Areas with built-up development and cultivated land showed higher susceptibility compared to less disturbed environments. The removal of vegetation reduces slope stability, while construction can alter drainage and stress vulnerable landscapes.
The research also found that landslide risk affects both informal and formal settlements, with areas along the coastal belt of eThekwini being particularly vulnerable. The disaster in 2022 impacted various communities, including formal residential areas like Umdloti, raising concerns about development on geologically sensitive terrain.
As extreme weather events linked to climate change become more frequent, the study suggests that machine learning and earth observation technologies can help municipalities plan safer urban growth and prioritise disaster preparedness. The approach is more efficient than traditional landslide assessments, which are often labour-intensive and costly.
Our take: We welcome the use of machine learning models achieving over 98% accuracy to map landslide risk in eThekwini, but we must ask how this data will be made accessible to local communities, especially those in informal settlements along the coastal belt. Will this technology translate into practical, affordable early warning systems that work on mid-range phones and in local languages, so it truly benefits those most vulnerable to such disasters?
Dr Laven Naidoo of the Gauteng City-Region Observatory said the research aims to support preventive planning and risk reduction by better understanding landslide drivers and vulnerable areas. This work provides a tool for evidence-based decision-making to protect infrastructure and communities in one of South Africa's fastest-growing metropolitan regions.
Sources
Source: Wits University
