A 2026 study mapped areas susceptible to landslides across eThekwini, South Africa, by combining satellite imagery, geographic information systems and machine learning. The study is cited as work by Cher Petersen and colleagues in Scientific Reports. An account provided by Wits University says the researchers aimed to identify vulnerable areas and the factors associated with landslide susceptibility. The resulting maps point to a role for both the natural landscape and the way land is developed.
Land use, elevation, vegetation cover and geological conditions were among the factors the study identified as important. The account describes one especially strong pattern: built-up and cultivated areas were more likely to experience landslides than less disturbed environments. That is an association reported by the study; it does not by itself show that development caused a particular landslide. It does, however, suggest that mapping where people build and cultivate may be relevant to understanding where susceptibility is higher.
The researchers used two machine learning approaches, random forest and support vector machines. The account reports that both models achieved accuracy levels above 98%. That number describes model accuracy as reported; it is not a statement that any individual slope has a 98% chance of sliding, or that a future event can be predicted with that certainty. The account does not explain how the accuracy was calculated or tested, so the result needs that context to assess how well the maps might generalize.
The maps identify the coastal belt of eThekwini as particularly susceptible, while mountainous inland areas generally show lower susceptibility. These are regional susceptibility patterns, not a forecast of when a landslide will happen. The account gives no event timing or neighborhood-level probabilities. Its value, as described, is in helping planners see where conditions associated with landslides are concentrated.
The report relates the work to the devastating floods and landslides of April 2022 in KwaZulu-Natal. It notes that both informal and formal settlements were affected, and discusses damage in places such as Umdloti alongside questions about development, vegetation removal and building on geologically sensitive land. The reported study does not attribute the 2022 disaster to any single cause. Instead, it presents landslide susceptibility as shaped by interactions among environmental conditions and human decisions about where and how development occurs.
The researchers say removing vegetation can reduce slope stability, while construction may change drainage patterns or add pressure to vulnerable terrain. Those explanations help show why land-use information matters, though the account does not establish that a specific building project caused a particular slide. Dr. Laven Naidoo, identified in the report as a senior researcher at the Gauteng City-Region Observatory, said the work sought to identify factors driving landslides and areas that could be vulnerable to future events.
The maps could give municipalities and planners evidence to consider when managing development and prioritizing preparedness. The study account presents machine learning and satellite observation as decision-support tools that may make regional assessments more efficient. It does not report that the maps have been adopted in planning or tested as a way to reduce losses. Their contribution is a clearer picture of where susceptibility is higher, which can inform decisions alongside local knowledge and other risk assessments.