The Escalating Wildfire Challenge in New Zealand

New Zealand's landscapes, from the rugged South Island mountains to the expansive forests of the North, are increasingly vulnerable to wildfires. Climate change is intensifying this threat, with warmer temperatures, drier conditions, and longer fire seasons becoming the norm. Fire and Emergency New Zealand (FENZ) reports that extreme fire weather days are on the rise, and projections indicate urban fires could increase by over 40% by 2100 due to hotter air temperatures. Recent events underscore the urgency: in January 2026, a heatwave prompted extreme wildfire danger warnings, with fires like the 22-hectare blaze near Pan Pac forest highlighting rapid spread risks. Human activity ignites 97% of these fires, threatening homes, biodiversity, agriculture, and forestry—a sector vital to the economy.

As urban sprawl encroaches on fire-prone rural areas, the need for precise prediction tools grows. Enter Lincoln University's cutting-edge research, where scientists are developing smarter fire modelling to bolster national resilience.

Lincoln University Leads with Machine Learning for Fire Refugia Prediction

Te Whare Wānaka o Aoraki | Lincoln University is at the forefront of this effort through Dr. Helen M. de Klerk's co-authored study, Predicting Persistent Forest Fire Refugia Using Machine Learning Models with Topographic, Microclimate, and Surface Wind Variables, published December 5, 2025, in the International Journal of Geo-Information. Dr. de Klerk, affiliated with Lincoln's Centre for Geospatial and Computing Technologies, adapted overseas techniques to map unburnt areas—or refugia—in fire-prone forests. Though tested in South Africa's Southern Cape and Tsitsikamma regions, the model's variables mirror New Zealand's South Island topography: steep coastal mountains, variable winds, and foehn effects.

"When landscapes burn, it's important to understand which areas may remain unburnt... they protect flora that can regenerate ecosystems and provide safe havens for animals and insects," Dr. de Klerk explains. This work positions Lincoln as a hub for geospatial fire science, training future experts in environmental modelling.

Decoding Forest Fire Refugia: Nature's Safe Havens

Forest fire refugia (Latin for "refuge") are patches within burnscapes that escape flames over decades or centuries, preserving seed banks and wildlife corridors. They are pivotal for post-fire recovery, as resilient plants like podocarps or beech reseed devastated areas. In New Zealand, where indigenous forests cover 31% of land but face invasion by flammable gorse and pines, identifying refugia prevents biodiversity loss.

Traditional mapping relied on historical fire perimeters, but machine learning (ML) enhances precision by integrating real-time variables. Lincoln's approach uses ensemble models—combining Random Forest (RF), Extreme Gradient Boosting (XGBoost), and K-Nearest Neighbours (KNN)—achieving 82% average accuracy and up to 92% AUC-ROC scores. For Kiwi landscapes, this means proactive protection of South Island valleys shielded from nor'west winds.

Inside the Machine Learning Magic: Step-by-Step Breakdown

The model's power lies in its data fusion. Step 1: Gather topographic data from SRTM Digital Elevation Models (aspect, slope, elevation). Step 2: Layer microclimate metrics like solar irradiation (Global Solar Atlas) and topographic wetness index. Step 3: Incorporate surface winds from Computational Fluid Dynamics (CFD) simulations in OpenFOAM. Step 4: Apply Principal Component Analysis (PCA) to reduce dimensions, ADASYN oversampling for class balance, and grid search for hyperparameter tuning. Step 5: Train on 70/30 spatial blocks, iterating 10 times.

• Aspect: Top predictor (27-51% importance), as sun-facing slopes dry faster, fueling flames.
• Solar Radiation & Wind Direction: Secondary drivers, creating moisture deficits and channeling blazes.
• Ensemble Superiority: Outperforms singles, ideal for variable NZ terrains.

Omission tests confirmed robustness: skipping elevation barely dented accuracy, proving focus on micro-scale winds.Read the full study.

South African Success, Kiwi Potential: Landscape Parallels

South Africa's fynbos shares New Zealand's steep, coastal gradients and foehn winds, yielding refugia on shaded, leeward slopes. Dr. de Klerk notes: "Both regions have mountains with steep slopes... exposed to changing wind directions." In NZ's Fiordland or Kaikōura ranges, similar "wind shadows" could shield podocarp forests. Adapting this via Lincoln's geospatial tools could map refugia nationwide, aiding FENZ operations.

For students eyeing research jobs in ecology, Lincoln offers hands-on ML training amid rising demand.

Photo by Олег Мороз on Unsplash

Prof. Tim Curran's Plant Flammability Lab: Complementing the Models

Associate Prof. Tim Curran, Lincoln's fire ecology lead, pioneered flammability testing with the "Plant BBQ"—torching samples to quantify ignite time, flame height, and burn rate. Over 12 years, his team cataloged 470 low-flammability natives like tōtara and kōwhai, launching a directory with FENZ in 2025.

Findings: Dense, moist foliage slows spread, creating green firebreaks. Curran's work integrates with de Klerk's models, recommending plantings around predicted refugia.Explore the directory.

Beech Forests Under the Microscope: Latest Bioeconomy Insights

Recent PhD work by Georgia Stevenson, under Bioeconomy Science Institute (BSI) and Lincoln collaboration, tests beech litter flammability—once thought fireproof. Field burns reveal dry litter ignites under extreme conditions, challenging assumptions as climate dries forests. BSI's extreme wildfire programme, involving Lincoln, Scion, and FENZ, models these risks for better preparedness.

From Prediction to Action: Transforming Fire Management

Smarter modelling identifies defensible spaces, natural breaks, and priority zones, optimizing FENZ resources during fast fires. Landowners can plant low-flamm species; councils zone developments away from refugia. Policy-wise, integrate into district plans, echoing FENZ's CheckIt'sAlright.nz tools.

• Predict wind channels for backburning.
• Protect biodiversity hotspots.
• Reduce response times, saving lives/property.

Bolstering Biodiversity and Ecosystem Recovery

Refugia host fire-sensitive species, enabling rapid regeneration. In NZ, this safeguards kauri, rimu, and endemic insects amid gorse invasions. Lincoln's holistic approach—ML plus flammability data—ensures ecosystems rebound faster, vital as fires intensify.Career advice for ecologists.

Economic Safeguards for Primary Industries

Forestry (3% GDP) and farming face billions in losses; better models minimize disruptions. "Our food and fibre industries... would benefit greatly," says de Klerk. Insurers and exporters gain predictability, fostering resilience.

Photo by Wolfgang Hasselmann on Unsplash

Looking Ahead: Lincoln's Vision and Career Pathways

Ongoing BSI projects and international ties (US Forest Service) promise NZ-tailored models. Lincoln trains geospatial experts via degrees in environmental science. Aspiring researchers, explore higher ed jobs, university jobs, or rate your professors. Check higher ed career advice for wildfire science paths. Lincoln exemplifies how NZ universities drive climate adaptation.