peter haplof
UCLA researchers collected data on a variety of landslides, including one on this mountain in the eastern Himalayas.
A new technology developed by UCLA geologists that uses artificial intelligence to more accurately predict where and why landslides will occur could help save lives and property in some of the world’s most disaster-prone regions. may strengthen.
The new method, described in a paper published in Communications Earth & Environmental, improves the accuracy and interpretability of AI-based machine learning techniques, requires far less computational power, and is more efficient than traditional predictive models. is also widely applicable.
The approach, the researchers say, would make droughts, wildfires, and earthquakes the perfect source of landslides, and climate change could produce more powerful and wetter storms, worsening conditions like California. The location is especially valuable.
Many factors influence where landslides occur, including terrain shape, slope and drainage area, soil and rock material properties, environmental conditions such as climate, rainfall, hydrology, and seismic ground motions. With so many variables, predicting when and where a chunk of Earth might lose its grip is as much an art as it is a science.
Geologists have traditionally estimated local landslide hazards by incorporating these factors into physical and statistical models. Given enough data, such models can achieve fairly accurate predictions, but physical models are time and resource intensive and are not widely applicable. Statistical models, on the other hand, offer little insight into how to assess different risk factors to arrive at a prediction. .
Predicting landslides using artificial intelligence
In recent years, researchers have trained AI machine learning models known as deep neural networks (DNNs) to predict landslides. Given a set of landslide-related variables and historical landslide information, these large, interconnected networks of algorithms can very quickly process and “learn” this data to produce highly accurate predictions. can be done.
However, like statistical models, DNNs, despite their advantages in processing time and learning ability, cannot “show results”, so researchers need to interpret their predictions and use them to prevent potential landslides. It is becoming difficult to know which causal factors to target. future.
Kevin Hsiao, a doctoral student in Earth, Planetary and Space Sciences, said, “DNNs display the probability of a landslide as a percentage, and it may be accurate, but what caused the landslide and why were certain variables most important?” cannot be elucidated,” he said. He is also the co-first author of a journal article.
According to co-first author Khalid Youssef, a former biomedical engineering student and postdoctoral fellow at UCLA, the problem is that the various network layers of a DNN constantly interact with each other during the learning process, and their analysis It is said that it is impossible to solve A new technique from UCLA researchers was intended to address this.
“We tried to ensure a clear separation of results from different data inputs, which makes our analysis more useful in determining which factors are the most important contributors to natural disasters.” will be,” he said.
Youssef and Shao collaborated with co-corresponding authors, Seulgi Moon, UCLA Associate Professor of Earth, Planetary, and Space Sciences, and Louis Bouchard, UCLA Professor of Chemistry and Bioengineering, to bring the analytical power of the DNN to the database. developed an approach to decouple from It has complex adaptability to give more practical results.
Their method uses a type of AI called superposition neural networks (SNNs). In this network, the various layers of the network run in parallel with each other, maintaining the ability to evaluate complex relationships between data inputs and output results, but converging only on the most important parts. . Finally, we derive a prediction.
Researchers input SNN data for 15 geospatial and climatic variables associated with the eastern Himalayas. This region was chosen because the majority of human losses from landslides occur in Asia, with a significant portion occurring in the Himalayas. SNN models were able to predict landslide hazards in the Himalayan region with accuracy comparable to DNNs, but most importantly, researchers wanted to decompose variables to determine which variables played a greater role in producing the results. I was able to confirm that there was
“Just as an autopsy is required to determine cause of death, identifying the exact trigger of a landslide always requires on-site measurements and a history of soil, hydrology, and climatic conditions such as rainfall and intensity. We need records, but these are hard to come by, in remote places like the Himalayas,” Moon said. “Nevertheless, our new AI prediction model is able to identify key variables and quantify their contribution to landslide hazard.”
Researchers’ new AI programs require far fewer computer resources than DNNs and can run effectively with relatively little computing power.
“In contrast to DNNs, which require powerful computer servers to train, SNNs are so small that they can run on the Apple Watch,” said Bouchard.
The team plans to extend the study to other landslide-prone regions of the world. For example, in California and similar areas where frequent wildfires and earthquakes exacerbate the risk of landslides, the new system could serve as an early warning system that considers numerous signals to predict various surface hazards, including flooding. May be useful for development.
/ Open to the public. This material from the original organization/author may be of the nature of its time and has been edited for clarity, style and length. Mirage.News does not take any organizational positions or positions and all views, positions and conclusions expressed herein are those of the authors only. Read the full article here.
