Navigating a changing ice world

Navigating Arctic sea ice is a critical challenge for wildlife, researchers and maritime operators alike. But IceNet, a cutting-edge forecasting system, predicts sea ice changes with remarkable accuracy. As BRYN NOEL UBALD and JONATHAN SMITH write, from ship routing to wildlife monitoring, IceNet supports safer, more sustainable polar operations and offers a powerful tool in the face of a changing climate.

Every day, billions of people check the weather to plan everything from what to wear to how they would escape a potential climate emergency, such as a hurricane or wildfire. But in the Arctic, it’s not just the weather that dictates the routines of daily life—it’s the sea ice.

Shifting ice conditions can have a profound impact on the day-to-day lives of Arctic wildlife, maritime crews and Indigenous communities. As climate change thins sea ice, fuelling a step-change in historical patterns, the frozen ocean is becoming harder than ever to read. Traditional physics-based models are falling short when it comes to capturing the complex interactions between ice, ocean and atmosphere, leaving a critical shortfall in forecasting ability.

Training on decades of data

Fortunately, the IceNet project may help close the gap. IceNet is a cutting-edge initiative led by the British Antarctic Survey (BAS) in collaboration with the Alan Turing Institute. It uses an artificial intelligence (AI)-based system to predict future sea ice by learning from past observations to interpret the impact of climate change on the Arctic. IceNet combines satellite data and weather observations to forecast both pan-Arctic and pan-Antarctic sea ice concentrations on a daily timescale up to several months ahead. These forecasts can help Arctic communities, researchers, conservationists and shipping operators prepare for increasingly unpredictable conditions.

To train the system, IceNet combined decades of satellite observations with detailed historical weather data, such as wind speeds, temperatures and other atmospheric variables. Over time, the system has evolved from producing monthly forecasts to generating daily sea ice concentration maps at 25 km resolution, with experimental work underway to train it on 6 km resolution data (which would offer greater detail and precision). The model learns how sea ice responds to different weather and sea ice patterns over time—information that enables it to forecast sea ice concentrations up to three months ahead of time.

This approach has proven to be remarkably effective: an earlier version of IceNet outperformed the leading physics-based models in forecasting summer sea ice, which is particularly difficult due to rapid melting and variability.

Diverse use cases

To bring these forecasts into operation, research software engineers developed a fast, flexible digital ecosystem that enables the generation of real-time predictions and integrates with other tools, leading to a wide range of downstream applications. Workflows have been created to process the large amounts of data needed to train the model and generate daily sea ice forecasts. Researchers are using this digital ecosystem to develop more advanced models.

In addition, a digital ecosystem has been created that enables users to visualize the sea ice forecasts. For example, researchers with the Government of Nunavut (Canada) trialled use of the ecosystem during their survey of the Foxe Basin polar bear population last year. The IceNet forecasts highlighted areas where sea ice was likely to linger—meaning polar bears might gather. These places might have been missed by a survey that focused solely on coastal regions. Because the team was often based at remote Arctic field camps, low-bandwidth delivery systems were developed to ensure they could access the forecasts during the survey.

Sea ice also has direct impacts on the movement of ships in the Arctic. In addition to IceNet, the BAS AI Lab is developing a suite of state-of-the-art AI systems to support ship navigation and operations in changing sea ice conditions. As you might imagine, navigating around or through sea ice can be tricky. It can burn more significant amounts of fuel, or you might be unable to traverse certain ice-packed areas at all. The AI systems will use IceNet forecasts as inputs to map out ideal ship routes to enable sustainable navigation. The systems will include ecological information—such as areas where animals are likely to be present—to help ships steer away from fragile ecosystems.

In our work to close a critical gap in sea ice forecasts for a region on the frontline of global warming, we hope to make IceNet as powerful and everyday as the weather forecasts available in our pockets—and in an openly accessible manner, both in terms of the code it is built on and the forecasts it generates.