LOS ANGELES – Geologists at UCLA have unveiled a groundbreaking technique that utilizes artificial intelligence (AI) to enhance the accuracy of landslide predictions, potentially saving lives and safeguarding property in regions prone to natural disasters. This innovative method offers improved interpretability and precision compared to traditional predictive models, requiring less computational power.

The researchers highlight the significance of this approach, particularly in areas like California, where the convergence of droughts, wildfires, and earthquakes increases the risk of devastating landslides. Predicting landslides is a complex task influenced by numerous factors, including terrain shape, slope, drainage areas, soil properties, and environmental conditions such as climate, rainfall, hydrology, and ground motion caused by earthquakes. 

In recent years, researchers have turned to AI-based deep neural networks (DNNs) to predict landslides. These complex networks can process vast amounts of data and learn from historical landslide information, enabling highly accurate predictions. However, DNNs lack transparency, making it challenging for researchers to understand the reasons behind their predictions and identify specific variables contributing to landslides.

To address this limitation, the UCLA team developed a new method utilizing a type of AI called a superposable neural network (SNN). By employing the SNN, the researchers successfully predicted landslide susceptibility in the eastern Himalayan mountains using 15 geospatial and climatic variables. The model achieved accuracy comparable to DNNs but, crucially, enabled the identification of key variables and their respective contributions to landslide susceptibility.