Sólveig Haraldsdóttir

Reading the icebefore it goes.

Dr Sólveig Haraldsdóttir · climate AI scientist · Reykjavík

I build models that forecast how fast Iceland's glaciers melt, from satellite passes and weather data. Then I go up and check them on the ice.

Sólveig Haraldsdóttir in a navy shell jacket at the old harbour in Reykjavík, Mount Esja behind her across the bay

A melt forecast for every 100 metres of ice.

Leysing, Icelandic for the spring thaw, is the model I lead at Fönn Lab. Each morning it reads the newest radar and optical passes over Vatnajökull, Langjökull and Hofsjökull, adds the weather forecast, and predicts surface melt for the next ten days on a 100 metre grid. Hydropower and flood teams plan the water coming off the ice with it.

  • ReadsSentinel-1 radar, Sentinel-2 and Landsat optical passes, the weather forecast
  • PredictsDaily surface melt, ten days ahead, 100 m cells
  • ChangedTen-day runoff error 31 percent lower than the degree-day model it replaced, 2023 to 2025 seasons

Figures from the Leysing forecast archive.

Pixel 1 m64.045°N 16.898°WForecast cell 100 m

Move across the glacier to see the model's cells. Darker ice melts faster. Click to keep a reading, Escape clears them.

Three things the ice taught the model.

Close-up of layered blue glacier ice with trapped air bubbles and a thin dark band of volcanic ash
Pixel 1 m64.452°N 17.103°WCore FÖ-23-4, depth 11.2 m

2024

Ash on ice

A segmentation model that finds tephra and dust on the ice surface in 10 metre Sentinel-2 images. Dark ice absorbs more sun and melts faster, so the ash map now feeds Leysing every week.

What changedSummer melt error on the south outlets fell by a fifth once the ash map went in.

Sólveig at her desk at blue hour, a false-colour satellite image of a glacier on her monitor and snowy rooftops outside the window
Pixel 1 m64.156°N 21.951°WOn screen: Skaftá cauldrons

2025

Flood watch under the ice

When a lake beneath the glacier drains, radar interferometry shows the ice surface sagging by a few metres. My detector reads every pass and flags the sag on its own.

What changedIn 2025 it warned of two floods from the Skaftá cauldrons about 30 hours before the river gauges rose.

Sólveig in a lab coat and blue gloves examining a long ice core in a steel tray in a frosted cold room
Pixel 1 m64.156°N 21.951°WCold room, minus 20 °C

2023 to now

What the satellites cannot see

Each spring I drill shallow cores in the accumulation zone. Their layers tell the model how dense last winter's snow is and where old ash lies, which no satellite pass can measure.

What changedDensity error at depth roughly halved against the fixed-density guess the model used before.

Sólveig kneeling in the snow on the Vatnajökull ice cap beside a small GPS and weather mast with a solar panel
Pixel 1 m64.371°N 16.882°WMast V2, 1,340 m, Vatnajökull

Every forecast gets checked on the ice.

Fourteen ablation stakes and three weather masts on Vatnajökull report every hour. Each week I compare what Leysing said with what the stakes measured, and twice a summer I go up and read them myself.

Where it is still wrong

Leysing is least reliable in the first warm week of spring, when wet snow confuses the radar, and after a fresh ash fall that no optical pass has seen yet. In those weeks the map marks itself uncertain, and the people downstream decide with a person on the phone.

The path so far.

Posts and study

  • 2022 to nowLead scientist, climate AIFönn Lab, Reykjavík
  • 2019 to 2022Postdoctoral researcher, glacier remote sensingNordic Ice Observatory, Tromsø
  • 2018PhD, GlaciologyUniversity of Iceland
  • 2012BSc, PhysicsUniversity of Iceland

Talks and papers

  • 2025Leysing: ten-day surface melt forecasts for Icelandic ice capsPaper, open access
  • 2025What a 300 metre pixel knows about a glacierPublic lecture, Reykjavík, in Icelandic and English
  • 2024Tephra on ice: segmenting ash layers in Sentinel-2 imageryPaper
  • 2023Teaching a model what a stake knowsWorkshop talk on field validation

I supervise MSc and PhD projects in glacier remote sensing. Students can write through the form below.

Radio the station.

Ask about the forecasts, the fieldwork or the data. The station answers from Sólveig's field notes on this page.

A small grey research hut with a tall radio mast on the snow of the ice cap at dusk, one window lit
Pixel 1 m64.400°N 16.952°WFönn field hut, 1,520 m
FÖNN ONE · field radio log · channel 3 00:00:00 GMT

Field radio log. Answers from Sólveig's notes. Scripted, not Sólveig.

    Frequent traffic
    Answers from Sólveig's notes. Scripted, not Sólveig.

    Invite me to lecture, or ask for the data.

    Forecast archives and the ash maps are open to researchers and water managers. Lectures in Icelandic or English, for universities, schools and public audiences. I answer within five working days.

    What is this about?
    I read every request myself.

    Logged 00:00 GMT

    Thank you.

    Your request is in my queue.