The reading list
Latest science.
Recent research and reports I'm reading, summarised in plain English.
Nature · Aug 2026
WeatherNext Cyclones: AI forecasts of cyclone track, intensity and wind structure
Alet F., Andersson T., Price I. et al. (Google DeepMind, Google Research, US National Hurricane Center, CIRA, UK Met Office)
Google DeepMind’s WeatherNext model reached state-of-the-art accuracy for cyclone track, intensity and wind structure. On 2023–2024 cyclones it gained more than 24 hours of lead time, so its three-day forecasts matched what earlier models gave for two days. A 1,000-member ensemble produces probability maps of tropical-storm to hurricane-force winds up to 15 days ahead. The code and model weights are open source, and live forecasts can be explored in Google’s Weather Lab.
Why it mattersAn extra day of warning is time for evacuation, pre-positioning and asset protection. The paper does not report results for the Australian region, so its value for Australian cyclones still needs local evaluation alongside Bureau of Meteorology warnings, which remain the official source.
AICycloneEarly warning
Read the source ↗ (opens in a new tab)CSIRO Northern Rivers Resilience Initiative · Jun 2026
Scenario testing and recommendations for future flood mitigation activities in the Richmond River catchment
Vaze J., Mateo C.M., Nguyen C. et al. (CSIRO), for NEMA
The first detailed hydrodynamic model of the whole Richmond River catchment tested community- and council-proposed mitigation bundles against the 2008, 2017 and 2022 floods. For the February 2022 peak at Lismore, one bundle lowered the modelled level by 2.07 m (14.87 m to 12.80 m), though the CBD levee was still overtopped. CSIRO notes a business case, cost-benefit analysis and detailed design are still needed.
Why it mattersA leading Australian example of catchment-scale, co-designed mitigation modelling, and an honest look at what mitigation can and cannot achieve.
FloodNSWMitigation
Read the source ↗ (opens in a new tab)Nature Communications · Jun 2026
Deep learning completes US flood hazard maps revealing millions exposed to previously unrecognized risk
Wu A.N., Zhang Y., Stouffs R.
Official US flood maps cover only about a third of river channels. An open-source deep-learning model learned from existing maps to complete a 30 m flood hazard map, finding 11 million more people and 4.1 million more buildings in flood zones than the official baseline, with vulnerable groups over-represented in unmapped areas.
Why it mattersMany councils have patchy or ageing flood studies; AI can help fill gaps for prioritisation, though it does not replace detailed studies.
AIFloodMapping
Read the source ↗ (opens in a new tab)Nature Climate Change · Jun 2026
Human-driven sea-level rise has quadrupled the frequency of coastal sea-level extremes since 1900
Dangendorf S., Sun Q., Maduwantha P. et al.
Combining tide-gauge records with climate model experiments, the study finds the median frequency of what were 1-in-100-year coastal water levels has risen about 12-fold since 1900. Human-driven forcing alone has quadrupled their likelihood, and natural variability is now secondary along most coastlines.
Why it mattersPlanning that relies on historical "1-in-100-year" storm tide levels is likely underestimating today’s coastal hazard.
CoastalSea levelAttribution
Read the source ↗ (opens in a new tab)Nature · May 2026
Enhanced response of extreme compound events to cumulative CO2 emissions
Li J., Zhang Y., Ciais P. et al.
A new metric tracks how compound extremes, such as hot-dry and hot-wet events, change per unit of cumulative CO2. Common compound events rise roughly in step with emissions, but rarer, more severe ones escalate disproportionately, and observation-constrained estimates are 37–75% higher than the model average.
Why it mattersRisk registers that assess heat, drought and flood separately will miss the fast-growing risk of hazards arriving together.
Compound hazardsExtremes
Read the source ↗ (opens in a new tab)Google Research blog · Mar 2026
Urban flash flood forecasts in Flood Hub, trained with Groundsource
Google Research
Google added urban flash flood forecasts to Flood Hub, predicting risk up to 24 hours ahead at 20 × 20 km resolution. The model was trained on Groundsource, a dataset of more than 2.6 million historical flood events across more than 150 countries that Gemini extracted from public news reports. Flash floods cause about 85% of flood-related deaths worldwide. Against US flash flood warnings, Google reports 22% recall and 44% precision.
Why it mattersIt shows how AI can turn public records into hazard data where gauges and studies are sparse. Google does not say whether Australian cities are covered, and the coarse resolution means it complements, not replaces, local flood studies and warnings.
AIFloodEarly warning
Read the source ↗ (opens in a new tab)WMO report · Mar 2026
State of the Global Climate 2025
World Meteorological Organization
2025 was about 1.43 ± 0.13 °C above the 1850–1900 average, and 2015–2025 are the 11 warmest years on record. Earth’s energy imbalance and ocean heat content both reached record highs, and global mean sea level is about 11 cm above the 1993 start of the satellite record.
Why it mattersRecord ocean heat feeds marine heatwaves, heavier rainfall and sea-level rise, all of which land directly on coastal councils.
Global climateOcean heat
Read the source ↗ (opens in a new tab)Australian Government (DCCEEW) · Sep 2025
National Climate Risk Assessment
Australian Climate Service
Australia’s first nationally comprehensive climate risk assessment identified 56 nationally significant risks and analysed 11 priority risks in depth, including cross-system risks for coastal communities and water security. It finds sea-level rise could put 1.5 million more people in high-risk coastal areas by 2050.
Why it mattersIt is the evidence base behind the National Adaptation Plan, and council adaptation and asset plans will increasingly be expected to align with it.
AustraliaClimate riskPolicy
Read the source ↗ (opens in a new tab)UN Office for Disaster Risk Reduction · May 2025
Global Assessment Report on Disaster Risk Reduction 2025: Resilience Pays
UNDRR
Counting cascading and ecosystem costs, disasters now cost more than US$2.3 trillion a year, compared with about US$202 billion in direct losses. The report argues that investing in risk reduction breaks the cycle of disaster, debt and uninsurability.
Why it mattersIt gives councils a global economic case for funding mitigation before disasters, not just recovery after them.
Disaster riskFinance
Read the source ↗ (opens in a new tab)