Communities worldwide, including many throughout the United States, struggle to predict and manage significant landslide risk. Landslides prove difficult to predict because they are infrequent and their occurrence may depend strongly on the specific soil, rain, and wind conditions in each location. Effective warning proves hard to disseminate because community members have different risk perceptions and tolerances, and even the best scientific predictions of landslide risk are often imprecise. In this project, a team of geo, information, and social science research institutions, the Sitka Sound Science Center, and the Sitka Tribe of Alaska, will design a novel landslide risk warning system for Sitka, Alaska, a small, diverse coastal town of 9,000 pressed against the steep, landslide-prone slopes of the Tongass National Forest. Working with local students and other residents acting as citizen scientists, the project will deploy small, inexpensive, networked moisture sensors on the slopes above Sitka that, when combined with new methods for integrating diverse data streams, will improve landslide prediction. The project will map Sitka's social networks and residents' understanding of risk and will then use this information, along with new influence maximization methods, which identify well-connected 'key influencers' in each social network, to design effective dissemination channels for landslide warning. The project will use decision support tools to facilitate community deliberations and workshops with government officials on the appropriate design of the physical and social components of a warning system that will best balance timely warning with reduced incidence of disruptive false alarms. While focused on Sitka, this project's results should be widely applicable worldwide, especially in other small or remote towns or communities with landslide risk.

This project will advance geoscience, social science, information science, and risk management through innovative incorporation of multiple data streams from sources such as historical records and imagery, hydrologic sensors, and social networks. The project will advance information science by showing how diverse sources of data (of disparate time scales, dimensionalities, and levels of noise) can be integrated to improve decision-making and policy-making in highly uncertain environments. These diverse streams of data will allow us to utilize both existing machine learning methodologies, as well as novel influence maximization models for communicating natural hazard risk. The project will advance geoscience by improving predictive models through direct measurement of landslide triggering conditions and region-specific threshold calibration, and by testing how a vast increase in the number of in-situ sensors affects the design, implementation, and performance of landslide early warning systems. The project will advance social science through an improved understanding of risk perception and communication in social and cultural contexts. It will be among the first to study how network influence maximization can improve community education and natural hazard response. By linking an understanding of social networks and cultural frames of risk perception with a participatory, quantitative decision support system, this project will improve understanding of how data can be used to facilitate a fair, accountable, integrative, and transparent risk management process.