From Heuristic Models to GeoAI: A Scientometric and Thematic Evolution of Landslide Susceptibility Mapping Approaches

Landslides are among the most destructive forces on Earth, capable of swallowing highways, burying villages, and reshaping landscapes in moments. For decades, scientists have tried to predict where these events are most likely to happen, creating maps that highlight areas of high risk. These maps, known as landslide susceptibility maps, rely on understanding the relationship between the ground itself and the factors that trigger movement, such as steep slopes, heavy rainfall, and the type of rock or soil present. In the early days of this work, experts relied heavily on their own experience and judgment to weigh these factors, drawing conclusions based on what they knew about the terrain. Over time, however, the tools available to these researchers have changed dramatically. The rise of powerful computers, high-resolution satellite images, and advanced software has allowed scientists to move beyond simple observation and into the realm of data-driven prediction, where algorithms can process vast amounts of information to find patterns invisible to the human eye.

A new study by researchers at Morgan State University and the Federal School of Surveying in Nigeria offers a comprehensive look at how this field has transformed over the last twenty years. The team did not conduct a new experiment in the field; instead, they performed a massive review of existing scientific literature to trace the evolution of landslide prediction. They gathered 795 peer-reviewed research papers published between 2006 and 2026 from major scientific databases. Using specialized software designed to analyze how research themes connect and change over time, they mapped the intellectual journey of the field. Their goal was to understand not just which methods were used, but how the entire approach to the problem shifted from human intuition to artificial intelligence.

The analysis reveals a clear and steady transition in how scientists approach landslide mapping. In the earliest period, from 2006 to 2012, the research was dominated by what the authors call heuristic models. These are methods where experts assign weights to different factors based on their knowledge, often using simple statistical tools to check their work. During these years, the primary focus was on gathering better data through remote sensing, which involves using satellites and aircraft to observe the Earth from above. As the timeline moved into the period between 2013 and 2018, the focus shifted toward more quantitative approaches. Researchers began to rely heavily on machine learning, a type of computer science where algorithms learn from data to make predictions without being explicitly programmed for every specific rule. This era saw a surge in the use of digital elevation models, which are detailed 3D representations of the terrain, allowing for a much more precise understanding of slope and stability.

By the time the researchers examined the period from 2019 to 2022, the field had entered a new phase characterized by the rapid adoption of deep learning. This is a more advanced form of artificial intelligence that mimics the way the human brain processes information through layers of interconnected nodes. The study found that during these years, the number of published papers exploded, with nearly 90 percent of all the research in their dataset appearing after 2018. The themes in the literature expanded to include complex ensemble models, which combine the predictions of multiple different algorithms to create a more robust and accurate result. The researchers also noted a growing integration of technologies like LiDAR, which uses laser pulses to create incredibly detailed maps of the ground surface, and InSAR, a technique that uses radar to detect tiny movements in the earth's surface before a landslide occurs.

In the most recent period, from 2023 to 2026, the study identifies a mature landscape where the focus has shifted toward making these powerful artificial intelligence systems understandable and reliable. While the core goal of predicting landslides remains the same, the methods have become highly specialized. The researchers found that scientists are now working on explainable artificial intelligence, a field dedicated to opening the "black box" of complex algorithms so that humans can understand why a model made a specific prediction. This is crucial for decision-makers who need to trust the maps before they can use them to plan roads or evacuate communities. The study also highlights a growing emphasis on infrastructure, with more research dedicated to protecting highways and urban areas, reflecting the increasing value of these systems in a world where climate change and urbanization are making landslides more frequent and dangerous.

The review also uncovered the human side of this scientific progress. The researchers mapped the global network of collaboration and found that the work is heavily concentrated in Asia, particularly in China, which produced the highest number of studies. India, Vietnam, and the United States also emerged as major hubs of activity. The analysis showed that these researchers do not work in isolation; instead, they form dense networks of collaboration, with teams in different countries working together to share data and refine their models. This international effort has been essential in driving the field forward, allowing for the testing of new methods across diverse landscapes and geological conditions.

Despite these significant advances, the study points out that challenges remain. The researchers note that while the models are becoming more accurate, they still struggle with uncertainty. It is often difficult to know exactly how a model will perform in a new location that it has never seen before, a problem known as transferability. There are also concerns about the quality of the data used to train these systems; if the records of past landslides are incomplete or inaccurate, the predictions will suffer. Furthermore, the complexity of the newest artificial intelligence models can make them difficult to interpret, raising questions about how to ensure they are making decisions based on sound geological principles rather than just statistical coincidences.

Ultimately, this review serves as a roadmap for the future of landslide science. It confirms that the field has moved far beyond the simple, expert-based maps of the past, evolving into a sophisticated discipline driven by big data and artificial intelligence. The authors suggest that the next frontier lies in creating systems that are not only powerful but also transparent and capable of operating in real-time. By combining the latest in geospatial technology with advanced computing, scientists aim to build a future where communities can be warned of impending landslides with greater precision and reliability, turning the tide against one of nature's most unpredictable hazards.