The Escalating Landslide Threat in Aotearoa New Zealand
New Zealand's dramatic landscapes, with their steep hillsides and high rainfall, make the country particularly vulnerable to landslides. These events, often triggered by intense storms or earthquakes, pose significant risks to lives, infrastructure, and the economy. Recent studies highlight how climate change is amplifying this danger, with more frequent and severe rainfall events leading to a surge in landslide activity. Universities across the nation are at the forefront, harnessing artificial intelligence (AI) and machine learning (ML) to develop predictive tools that could save lives and guide land-use planning.
Landslides in New Zealand are not rare; the country experiences thousands annually. However, events like Cyclone Gabrielle in early 2023 demonstrated the scale of potential devastation, triggering around 800,000 landslides across the North Island alone. This catastrophe caused widespread damage, disrupted communities, and underscored the urgent need for advanced forecasting capabilities.
Cyclone Gabrielle: Catalyst for Innovation
Cyclone Gabrielle stands as one of the most extreme landslide-triggering storms in New Zealand's recorded history. The cyclone brought unprecedented rainfall, saturating soils and destabilizing slopes nationwide. Researchers mapped over 43,000 individual landslides from this event, providing a rich dataset for model development. This real-world data has been pivotal for New Zealand universities in refining their AI-driven prediction systems.
The aftermath revealed not just immediate impacts but long-term vulnerabilities. Roads, homes, and farmland were buried under debris, with economic losses running into billions. It also highlighted gaps in existing hazard models, spurring academic collaborations to integrate dynamic weather data with terrain analysis for more accurate forecasts.
University of Canterbury's Pioneering Research
The University of Canterbury (UC) in Christchurch is leading the charge in AI-enhanced landslide prediction. A landmark study published in Scientific Reports by UC researchers Livio Dreyer, Thomas R. Robinson, Marwan Katurji, and James H. Williams, alongside GNS Science's Kerry Leith, projects alarming increases under climate change scenarios. Under a +2°C warming world—aligned with current global trajectories—a Gabrielle-like storm could generate up to 90,000 additional landslides, with high-density areas expanding by 34%.Scientific Reports study
UC's approach combines high-resolution rainfall data from MetService with static factors like slope, geology, and land cover. Using generalized additive models (GAMs), they predict both susceptibility (likelihood of occurrence) and intensity (number per area). These models achieve high accuracy, validated through cross-validation techniques, offering a blueprint for national-scale applications.
Oliver Wigmore's work at UC further advances this with gradient boosted decision trees (GBDT), a machine learning algorithm excelling in handling complex interactions. Trained on Gabrielle data at 25m resolution, it forecasts rainfall-induced landslides (RIL) under future climates, emphasizing forest cover's mitigating role.EGU preprint
Machine Learning: The Core of Prediction Models
Machine learning algorithms are transforming landslide forecasting from static maps to dynamic, real-time tools. At UC, GBDT models process vast datasets—topography from LiDAR, rainfall from radars and satellites, antecedent moisture indices—to output probabilistic maps. These differ from traditional physics-based models by learning patterns from historical data, improving accuracy for rare extreme events.
Step-by-step, the process involves: (1) Data ingestion, blending real-time weather with geomorphic variables; (2) Feature engineering, like 24-hour max rainfall plus 7-35 day totals; (3) Model training on inventories like Gabrielle's 182 quality-controlled cells; (4) Validation via ROC-AUC scores nearing 0.94; (5) Forecasting under scenarios from Weather Research and Forecasting (WRF) simulations.
Victoria University of Wellington's Centre for Data Science and Artificial Intelligence (CDSAI) complements this by applying AI to broader environmental monitoring, tracking climate-driven changes that feed into landslide models.
National Collaborations Driving Progress
No single institution can tackle this alone. The Hōretireti Whenua – Sliding Lands programme, led by GNS Science, unites UC, University of Auckland (UoA), Massey University, and Victoria University of Wellington (VUW). Goals include probabilistic models for earthquake- and rainfall-triggered landslides, integrated with runout simulations via RiskScape and MERIT tools.GNS Sliding Lands
Massey contributes expertise in vegetation and soil science, UoA in seismic integration, and VUW in AI optimization. Case studies in Tairāwhiti (post-Gabrielle) and Auckland incorporate iwi knowledge and socio-economic vulnerability, ensuring culturally sensitive outputs.
CRISiSLab, funded by Natural Hazards Commission Toka Tū Ake, supports interns like Charutha Unni using AI and remote sensing for landslide detection, bridging academia and emergency response.
Climate Change: Quantifying the Risk Surge
Projections are stark: +7-14% more landslides per major storm under +2°C, clustering in high-susceptibility zones. North Island regions like Gisborne and Hawke's Bay face the brunt, with new risks emerging near existing hotspots. Forest cover emerges as a key mitigator, absorbing rain and stabilizing soils—prompting calls for reforestation in vulnerable areas.
These findings inform the National Adaptation Plan, urging updated building codes, evacuation protocols, and insurance models. Universities' open-source models democratize access, empowering local councils from Wellington to Dunedin.
From Models to Real-World Impact
AI tools enable nowcasting during storms, warning of 'danger windows' hours ahead. NEMA and MetService integrate them into apps like GeoNet, alerting communities via push notifications. In West Coast trials, early warnings reduced exposure during floods.
- Reduced fatalities through precise evacuations
- Infrastructure resilience via targeted reinforcements
- Economic savings: Billions from avoided Gabrielle-scale damages
- Equity: Prioritizing Māori land and remote areas
Challenges persist: Data scarcity in remote South Island, model uncertainty in earthquakes, and computational demands. Yet, hybrid physics-ML approaches are closing gaps.
Career Opportunities in NZ Geohazards and AI
NZ universities seek experts in geospatial AI, hydrologists, and data scientists. UC's School of Earth and Environment offers PhDs in ML hazard modeling; VUW's CDSAI recruits AI specialists. Roles blend research with policy, partnering GNS and iwi.
With funding from MBIE and Resilience to Nature's Challenges, programmes like Sliding Lands create pathways for postdocs and lecturers. Skills in Python, R, TensorFlow, and domain knowledge yield competitive salaries, around NZ$100k+ for mid-career.
Future Horizons: AI's Evolving Role
Looking ahead, integrating satellite constellations like Sentinel-1 with edge AI promises sub-hourly updates. Coupling with flood models addresses cascading risks. Universities advocate for national data platforms, enhancing Māori data sovereignty.
As climate pressures mount, NZ's academic ingenuity positions it as a global leader in resilient geohazards tech. These innovations not only protect Kiwis but offer exportable solutions for landslide-prone nations like Japan and Indonesia.