A group of geologists out of UCLA have developed a new technique that uses artificial intelligence to better predict landslides. Not only that, but the AI will also better predict where and while landslides happen. If proven to be scaled, it could become a new tool to monitor areas that are prone to landslides. Saving both lives and property in the process.
The new method was first described in the journal Communications Earth & Environment, courtesy of Nature.com. According to the paper, the AI is able to improve the accuracy and interpretability of AI-based machine learning techniques. It also requires less computing power and can be layered with traditional predictive models.
Traditionally, geologists have estimated an area’s risk of landslides by measuring serval factors. These include climate, rainfall, ground motion from earthquakes, hydrology, and material properties such as soil and bedrock. All of these factors are incorporated into physical and statistical modeling and with enough data, the models are able to make reasonably accurate predictions.
The issue with these current models though, is that physical models are both resource and time-intensive. Moreover, they can’t be applied over broad areas. On another hand, statistical models give limited insight into how they are able to assess risk factors that bring them to their conclusions.
This is where deep neural networks currently aid geologists. Over the last several years, if these models are given massive information related to historical landslide information and risk-related variables, they’re able to process the information and “learn” from the data to create accurate predictions.
Co–first author of the journal and doctoral student in Earth, planetary and space sciences Kevin Shao, stated “DNNs will deliver a percentage likelihood of a landslide that may be accurate, but we are unable to figure out why and which specific variables were most important in causing the landslide.”
This problem was echoed by Khalid Youssef, co-first author and a former student of biomedical engineering and postdoctoral researcher at UCLA, “We sought to enable a clear separation of the results from the different data inputs, which would make the analysis far more useful in determining which factors are the most important contributors to natural disasters,“.
To contend with this issue. Both Youssef and Shao teamed with UCLA professors Seulgi Moon, associate professor of Earth, planetary and space sciences, and Louis Bouchard, professor of chemistry and bioengineering, to develop an approach that could decouple the analytic power of DNNs from their complex adaptive nature in order to deliver more actionable results.
Their method uses a superposable neural network, or SNN. It has different layers of the network that run with each other. This allows for the retention of relationships between data inputs and output results. Using the eastern Himalayan mountains as inspiration, the team fed the SNN data about 15 geospatial and climatic variables relevant to the location.
According to them, this was chosen due to the amount of lives lost in the area to landslides. The SNN model was able to predict landslide susceptibility with the accuracy of traditional DNN methods. But, with the SNN model, the researchers were able to peel back some of the variables and see which ones played a greater role in the results.
Co-author associate professor Seulgi Moon, “Similar to how autopsies are required to determine the cause of death, identifying the exact trigger for a landslide will always require field measurements and historical records of soil, hydrologic and climate conditions, such as rainfall amount and intensity, which can be hard to obtain in remote places like the Himalayas,“.
Professor Moon continued, “Nonetheless, our new AI prediction model can identify key variables and quantify their contributions to landslide susceptibility.” But that’s not the only advantage of this model. Unlike others, it uses so little computing power, “The SNN is so small it can run on an Apple Watch, as opposed to DNNs, which require powerful computer servers to train,”, according to Professor Louis Bouchard.
As for now, the team is hoping to extend their work to other landslide prone areas of the world. If successful, their method could help develop new early warning systems that take into account a great deal of variables with a greater range of prediction.