Abstract Manager

In recent years, the integration of machine learning (ML) and artificial intelligence (AI) in transportation has revolutionized the way we manage and mitigate risks associated with natural disasters. This paper presents a novel approach to landslide susceptibility modeling at fine resolutions for linear asset management, specifically targeting Auckland's road network. The study was initiated in response to severe storms in early 2023, which resulted in approximately 2,000 landslides across New Zealand's North Island, causing significant disruption and financial burden on Auckland Transport.

The primary objective of this research was to develop a model that assesses landslide vulnerability on Auckland's 7,500 km road network. The model aims to transition from a reactive landslide repair strategy to a proactive prevention strategy, which has been suggested to provide a cost benefit to Auckland Transport. Traditional landslide susceptibility studies typically operate at resolutions of hundreds to thousands of meters, which are impractical for public transport operators due to funding limitations. Therefore, this study required a model resolution on the order of meters to tens of meters to be effectively implemented.

To achieve the necessary resolution and handle the large quantity of data inputs, Machine Learning (ML) and Novel Artificial Intelligent (AI) applications were utilized. The methodology involved determining input variables that influence landslide susceptibility, such as slope height and angle, geology type, proximity to geological faults, and rainfall depth. Data on these variables were acquired for each 12-meter road section using various open sources and Auckland Transport's internal records. The AI model was trained using data from past extreme weather events, including landslide location, type, extent, and activating rainfall depth.

The model employed a modern machine learning algorithm, which has been shown to outperform other algorithms in landslide susceptibility studies. The ML algorithm built multiple decision trees to find patterns in the input data that result in the presence or absence of a landslide. The trained AI model was then applied to future projected rainfall events to generate a risk rating for every 12-meter section of Auckland's road network under various climate scenarios. The output risk ratings were integrated into Auckland Transport's internal systems as a GIS overlay as well as FusionMap® realtime platform. 

The results of this study demonstrate the effectiveness of ML and AI in accurately assessing landslide susceptibility at fine resolutions. By incorporating high-level, publicly available data inputs and leveraging advanced machine learning techniques, the model provides a practical and cost-effective solution for managing landslide risks in transportation networks. This approach not only enhances the resilience of Auckland's road network but also serves as a valuable framework for other regions facing similar challenges.

In conclusion, the integration of ML and AI in transportation asset management offers significant potential for improving disaster preparedness and response. This research highlights the importance of adopting proactive strategies and leveraging advanced technologies to mitigate the impacts of natural disasters on critical infrastructure. The findings of this study contribute to the growing body of knowledge on the application of ML and AI in transportation and provide a foundation for future research and development in this field.