Case study
Turning river forecasts into lifesaving flood intelligence

Earth Sciences New Zealand researchers predict river flows in near real time

Floods are New Zealand's most frequent and costly natural disaster, but until recently, the country lacked a consistent way to forecast them nationwide. Now, by combining climate data with high-resolution weather and hydrological models using a supercomputer, the Aotearoa Flow Awareness System (AFAS) delivers hourly river forecasts with a 48-hour lead time across 60,000 river reaches. This capability gives regional councils and emergency services vital time to prepare, protect infrastructure, and warn communities at risk.

A red fire truck is parked on a flooded street while firefighters work in heavy rain.
Table of Contents
    Earth Sciences New Zealand logo.

    Summary

    Customer: Earth Sciences New Zealand

    Industry: Emergency management

    Country: New Zealand

    Vision

    Provide at-risk communities and responders with fast and accurate flood forecasts on a national scale

    Strategy

    Develop a system that combines weather, satellite, and climate data, and links hydrological and weather models, using supercomputing to generate reliable flood forecasts

    Outcomes

    • Delivers hourly river flow forecasts with a 48-hour lead time across 60,000 river reaches
    • Predicts when and where rivers are likely to overflow to give communities time to evacuate
    • Improves the accuracy of forecasts by running ensemble simulations to show a range of flood scenarios
    • Fills gaps where local maps don't exist and allows for long-term planning and flood risk avoidance
    • Supports protective measures such as stopbanks, reducing damage and keeping communities safer

    Generating lifesaving flood intelligence

    Flooding is New Zealand's most frequent and costly natural disaster, impacting communities and insurers at a scale of billions of dollars each decade. "As climate change progresses, flooding has become a bigger issue in New Zealand, and we know it's going to become even more frequent and severe in the future," says Dr. Emily Lane, principal scientist, Natural Hazards and Hydrodynamics at Earth Sciences New Zealand. "Fast and accurate forecasting of flood impacts is crucial for reducing the risk to life, property, and infrastructure." Earth Sciences New Zealand is the country's leading provider of natural and environmental science services and is internationally recognized for its climate research. To help inform flood disaster preparedness and response efforts, its researchers collaborated with regional councils, government agencies, insurers, and infrastructure providers to codesign a new national river flow forecasting system. Known as the Aotearoa Flow Awareness System (AFAS), the platform combines high-resolution weather, satellite, and climate data with a national hydrological model to generate precise forecasts of when and where rivers are likely to rise. "By generating a reliable picture of river flow increases during extreme rainfall events, our forecasting system is helping councils and emergency services strengthen emergency readiness, reduce blind spots, and better aid public safety," says Dr. Lane.

    Replacing fragmented flood warning systems

    As an island nation, New Zealand is especially vulnerable to atmospheric rivers, which are narrow bands of water vapor that can unleash heavy rain when they encounter land, especially coastal mountains. While not all atmospheric rivers produce extreme events, slow-moving and moisture-rich weather patterns can trigger severe rainfall, flooding, and landslides. Yet until recently, the country lacked a consistent nationwide system for predicting flood risks created by these events.

    "Previously, flood forecasting and mapping were handled by regional councils with varying levels of capacity, creating a postcode lottery where some communities received strong flood warnings and others little to none," says Dr. Lane. With AFAS, councils across the country can access a consistent nationwide baseline of information that complements existing regional models." The consistency our system provides can also help guide land-use planning in the longer term, preventing homes and infrastructure from being built in flood-prone areas," Dr. Lane explains.

    By generating a reliable picture of river flow increases during extreme rainfall events, our forecasting system is helping councils and emergency services strengthen emergency readiness, reduce blind spots, and better aid public safety.

    Dr. Emily Lane
    Principal Scientist, Natural Hazards and Hydrodynamics, Earth Sciences New Zealand

    An adult and a child stand at the edge of a flooded roadway, holding umbrellas in the rain.

    Mapping flood risk in near real time

    To build AFAS, Earth Sciences New Zealand integrated 40 years of climate and hydrological data from its Virtual Climate Network Station with detailed terrain maps, rainfall records, and information on river infrastructure.

    Researchers relied on the organization's high performance computing system, an HPE Cray supercomputer, to process these massive datasets. By combining computing power with sophisticated hydrological and hydrodynamic models, they can produce high-resolution simulations for rivers in both upland areas and lowland floodplains.

    "Some New Zealand floodplains receive water from over 100 rivers, so we had to come up with innovative methods to efficiently incorporate the vast amounts of data we had on all contributing waterways into the forecasts," recalls Dr. Lane.

    Using high performance computing proved vital to overcoming these challenges. "The scale of the computations—high-resolution simulations across thousands of catchments—simply could not have been achieved otherwise," says Dr. Lane.

    The team also developed Cylc, a custom workflow platform that orchestrates thousands of computational steps across catchment areas to feed data efficiently into AFAS's models. This has enabled hourly forecasts with a 48-hour lead time across 60,000 river reaches nationwide.

    In addition to the forecasting work, the Cylc workflow and methodology are also being used to produce nationally consistent flood inundation maps that accurately identify the areas likely to be submerged during floods and estimate the depth of the water.

    "We've just launched a public viewing tool for these maps so residents can see whether their homes are at risk," shares Dr. Lane.

    The consistency our system provides can also help guide land-use planning in the longer term, preventing homes and infrastructure from being built in flood-prone areas.

    Dr. Emily Lane
    Principal Scientist, Natural Hazards and Hydrodynamics, Earth Sciences New Zealand

    Strengthening public readiness

    The new system has changed the way New Zealand communities prepare for floods. With longer range and more accurate forecasts, people can move stock or belongings to higher ground, sandbag vulnerable areas, and protect important infrastructure. Longer lead times also make it more practical to use temporary measures such as portable stopbanks, which are barriers that can be quickly set up to redirect floodwaters.

    Forecast ensembles, which run multiple simulations to show a range of possible flood scenarios, give councils and emergency services an early heads-up. They can then stock generators, ready evacuation plans, and take other protective actions before issuing public warnings, reducing damage and keeping communities safer.

    Dr. Lane points to Westport, a flood-prone town on New Zealand's South Island, as an example of the system's strengths in action. In July 2021, floods damaged or destroyed nearly 500 homes. "By February 2022, our system correctly predicted the impact of two rainfall events: one heavy storm that caused only moderate flooding, and a smaller event that triggered worse flooding due to saturated soils," she recalls. "Councils used AFAS guidance to respond swiftly and appropriately".

    Scaling with supercomputing and AI

    To expand its research and forecasting capabilities, Earth Sciences New Zealand has now added Cascade, a new HPE Cray XD2000 powered by 4th Gen AMD EPYC processors. The new supercomputer triples available processing capacity and is connected with NVIDIA® InfiniBand for rapid data movement while using GreenLake for File Storage to handle the massive datasets required for hydrological modeling. Together, these technologies will support multiple AI powered simulations, higher-resolution runs, and extreme scenarios such as probable maximum floods.

    "With Cascade, we can run multiple AI-powered simulations simultaneously for more accurate weather predictions. We'll also be able to support higher resolution runs and model more extreme events like probable maximum floods while also potentially shortening lead times," notes Dr. Lane.

    Creating a foundation for next-gen forecasting

    Looking ahead, the research team also plans to integrate AI more fully into its approach to make its forecasts more precise and actionable. 

    By combining AFAS's models with AI and the cutting-edge RiskScape platform—a risk-modeling tool that the team co-developed with environmental research institute GNS Science—forecasts can evolve into even more detailed guidance for councils, emergency services, and the public.

    As Dr. Lane points out, harnessing AI can support a shift from relative river flow forecasting categories, such as exceptionally high, to absolute measurements in cubic meters per second to make flood data fully actionable. Councils can then better understand not just when rivers may rise but also exactly when and where they are likely to overflow.

    Ultimately, the goal is to move to forecasting inundation, to predict precisely where and when flooding is likely to occur. This next phase will be built on various work programs that are already underway.

    "Without that level of insight, we can't reliably know if water will spill out of a river channel or which communities will be cut off," Dr. Lane explains. "AI-powered impact-based forecasts will further strengthen our ability to turn data into decisions that save lives."

    Dark storm clouds loom over a distant mountain range while dramatic beams of sunlight break through the sky.

    We've just launched a public viewing tool for these maps so residents can see whether their homes are at risk.

    Dr. Emily Lane
    Principal Scientist, Natural Hazards and Hydrodynamics, Earth Sciences New Zealand

    Solution

    Hardware

    Existing system

    • HPE Cray Supercomputer

    Hardware

    New system

    • HPE Cray XD2000
    • 4th Gen AMD EPYC processors
    • GreenLake for File Storage
    • NVIDIA InfiniBand
    • Direct liquid cooling
    • HPE Performance Cluster Manager

    Explore more

    HPE supercomputers

    Discover more Digital Game Changers stories

    Document ID: a50014156enw