New Zealand Universities Develop AI Tech to Predict Climate-Driven Landslide Risks

New Zealand's universities are at the forefront of tackling one of the nation's most pressing natural hazards: landslides exacerbated by climate change. With steep terrain, frequent heavy rainfall, and vulnerable geology, the country has long grappled with these events, which claim more lives than earthquakes or volcanoes over centuries. Recent devastating storms like Cyclone Gabrielle in 2023, which unleashed around 800,000 landslides across the North Island, underscore the urgency. As global warming intensifies extreme rainfall, researchers from the University of Canterbury and collaborators are pioneering advanced prediction technologies to safeguard communities, infrastructure, and ecosystems.

The North Island's scarred landscapes from Gabrielle serve as a stark reminder. The cyclone's torrential rains saturated soils, triggering massive slope failures that buried homes, roads, and farmland, costing hundreds of millions and displacing thousands. Such rainfall-induced landslides (RILs), defined as shallow mass movements of soil and rock on slopes triggered by intense or prolonged precipitation, are New Zealand's deadliest hazard, responsible for NZ$250–300 million in annual damages.

University of Canterbury Leads Groundbreaking Landslide Models

At the University of Canterbury (UC), a team led by Livio Dreyer, Thomas R. Robinson, Marwan Katurji, and James H. Williams, in collaboration with GNS Science's Kerry Leith, has published pivotal research in Scientific Reports. Their study analyzes Gabrielle's landslide inventory of over 145,000 events, employing generalized additive models (GAMs)—statistical tools that capture non-linear relationships between variables—to forecast future risks.

Step-by-step, the process involves: (1) compiling high-resolution data from 1m LiDAR topography, land cover maps, lithology from the New Zealand Land Resource Inventory, and MetService's quantitative precipitation estimates (QPE) blending gauges, radar, and satellites; (2) delineating slope units using the r.slopeunits algorithm for geomorphological relevance; (3) training binary susceptibility and log-Gaussian intensity models via fivefold cross-validation; (4) simulating +2°C warmed storms using the Weather Research and Forecasting (WRF) model driven by CMIP5 data.

Results are alarming: a Gabrielle-like storm under +2°C warming could spawn up to 90,000 additional landslides, with extreme density areas (>86/km²) expanding 34%. Total numbers rise 7-14%, clustering near existing hotspots just 5m away on average. Read the full UC study here.

Machine Learning Ushers National-Scale Predictions

Complementing UC's work, Oliver Wigmore's preprint in EGUsphere deploys gradient boosted decision trees—a machine learning ensemble technique excelling at handling complex interactions—for a 25m-resolution national RIL susceptibility map. Trained on Gabrielle data from Hawke's Bay and Tairāwhiti, it integrates topographic, geologic, environmental factors, and rainfall triggers, achieving 0.94 ROC-AUC accuracy via SHAP explanations for interpretability.

Applied to NIWA's High-Intensity Rainfall Design System (HIRDS) under shared socioeconomic pathways (SSPs), it reveals disproportionate susceptibility hikes with warming, mitigated somewhat by forests. This first-of-its-kind dataset empowers climate-resilient planning. Antarctic Research Centre at Victoria University of Wellington affiliations highlight inter-university synergy. Access the preprint.

These models process vast datasets: satellite Copernicus DEM for elevation, LUCAS for land cover/forests, enabling rapid post-storm mapping and scenario testing.

GNS Science and University Partnerships Drive Innovation

GNS Science's Sliding Lands Hōretireti Whenua programme, partnered with University of Canterbury, Victoria University of Wellington, and Massey University, aims for national rapid landslide forecast models. ECLIPSE initiative maps post-storm damage using satellites, feeding AI tools for real-time hazard assessment.

The upgraded New Zealand Landslide Database (NZLD), launched October 2025, compiles hundreds of thousands of events, viewable interactively for planners. UC's geohazards research, including PhD projects on slow-moving landslides using satellite deformation and field data, enhances physics-based understanding.

iwi partnerships ensure culturally sensitive approaches, vital in Māori land contexts where whenua (land) holds ancestral significance.

Photo by Alexandre Lecocq on Unsplash

From Data to Action: How Prediction Tech Saves Lives

• Hazard Mapping: On-demand maps from rainfall forecasts pinpoint at-risk zones pre-storm.
• Climate Scenarios: SSP-tested projections guide infrastructure resilience, e.g., elevating roads in Hawke's Bay.
• Nature-Based Solutions: Models show forests reduce susceptibility; UC advocates expanding native cover on marginal slopes.
• Early Warning: Integrate with NIWA forecasts for alerts, as trialed post-Gabrielle.
• Land-Use Planning: Inform district plans, restricting development in expanding high-risk areas (projected 26-34% growth).

Stakeholders like local councils praise these tools for cost savings; e.g., Piha's Muriwai hybrid models assess susceptibility amid erosion.

Recent Case Studies Spotlight Urgent Need

Cyclone Gabrielle (Feb 2023): 800,000+ landslides, NZ's largest storm-triggered event, killed 11 indirectly, cost $14.5B. Tairāwhiti saw 1 in 10 properties hit.

January 2026 storms: Thousands in Tairāwhiti, 8 deaths Bay of Plenty, North Island evacuations—precursors to modeled futures.

West Coast: UC predicts landslide dams, temporary lakes breaching catastrophically; new zones flagged for monitoring.

Challenges and Multi-Perspective Views

Experts like UC's Robinson note non-linear risks: slopes near failure amplify small rainfall hikes. GNS emphasizes data gaps in remote areas, addressed by satellites.

Iwi perspectives: Whenua protection aligns with models favoring forests over intensification. Economists project billions in avoided losses via planning.

Govt response: Landslide Planning Guidance (GNS 2024) mandates susceptibility in consents; universities train future modelers.

Future Outlook: Resilient Aotearoa Through University Innovation

By 2050, intensified cyclones could double high-density zones. UC/Victoria ML frameworks scale nationally, integrating real-time data for apps like Landslide Watch Aotearoa.

Massey contributes hybrid models; Auckland explores urban risks. Training PhDs ensures continuity.

Optimism: Tech shifts from reaction to proaction, saving lives, NZ$ billions. Universities position NZ as landslide modeling leader.

Photo by NIR HIMI on Unsplash

Implications for Higher Education and Careers

UC's Geological Sciences program booms with demand for modellers; GNS-Victoria-Massey collaborations offer interdisciplinary PhDs in geohazards/climate.

Actionable: Aspiring researchers pursue Earth Sciences at Canterbury (world-ranked geohazards), leveraging tools like WRF, GAMs. Careers in hazard mitigation blend AI, geology, policy—vital amid warming.