Geologists at UCLA have developed a groundbreaking technique that utilizes artificial intelligence (AI) to improve the prediction of landslides. This innovative method, described in a paper published in Communications Earth & Environment, offers enhanced accuracy and interpretability compared to traditional predictive models. By harnessing the power of AI, geologists aim to protect lives and property in regions prone to natural disasters, such as California, where the convergence of droughts, wildfires, and earthquakes increases the risk of devastating landslides.
Predicting the occurrence of landslides is a complex task influenced by multiple factors, including terrain shape, slope, drainage areas, soil and bedrock properties, as well as environmental conditions like climate, rainfall, hydrology, and ground motion resulting from earthquakes. Geologists traditionally incorporate these factors into physical and statistical models to estimate landslide risks. While these models can provide reasonably accurate predictions, physical models require significant time and resources and are not suitable for large-scale application. On the other hand, statistical models lack transparency regarding the assessment of risk factors.
Harnessing AI for landslide prediction
For years, researchers have been training AI machine-learning models, specifically deep neural networks (DNNs), to predict landslides. These interconnected networks of algorithms excel in processing vast amounts of data and quickly learning from it, resulting in highly accurate predictions. However, DNNs have limitations when it comes to explaining their predictions and identifying the specific variables that contribute to landslide occurrences.
To address the limitations of DNNs, UCLA researchers developed a method utilizing a type of AI called a superposable neural network (SNN). Unlike DNNs, SNNs separate the results from different data inputs, allowing for more useful analysis of the most important contributing factors to natural disasters. The research team fed the SNN data on various geospatial and climatic variables relevant to the eastern Himalaya mountains, an area known for its high frequency of landslides. The SNN model demonstrated accuracy comparable to DNNs in predicting landslide susceptibility, while also providing insights into the variables that play significant roles in the results.
Unraveling variables and reducing computing resources
The ability of the SNN model to identify key variables and quantify their contributions to landslide susceptibility offers valuable insights to geologists. In addition, this approach requires fewer computer resources compared to DNNs, making it efficient and adaptable even with limited computing power. The researchers emphasized that their SNN model could run on devices as small as an Apple Watch, eliminating the need for powerful computer servers.
Future applications and early warning systems
The UCLA research team plans to expand their work to other landslide-prone regions worldwide. Particularly in areas like California, where the combination of frequent wildfires and earthquakes exacerbates landslide risks, the new AI prediction model could aid in the development of early warning systems. These systems would incorporate multiple signals and predict a range of surface hazards, including floods, to enhance preparedness and response efforts.
Geologists at UCLA have successfully employed AI technology to revolutionize landslide prediction. Their method, utilizing superposable neural networks, offers improved accuracy and interpretability compared to traditional models. By identifying key variables and quantifying their contributions, this AI approach enables geologists to better understand the factors influencing landslides. With the ability to run on devices with limited computing power, this model has the potential to enhance early warning systems and protect lives and property in landslide-prone regions globally.
