Elena Guk — Antarctic meteorology & forecast verification

Weather forecasts, checked against reality.

I'm an independent researcher in Antarctic meteorology and forecast verification. Now I work outside an academic institution and looking for research collaborations.

The poles warm roughly twice as fast as the rest of the planet, and Antarctica holds the largest ice mass on Earth. Its weather feeds into sea-level estimates, iceberg formation, and the AI forecast models now used there — where there is very little independent data to check them against.

Write to me →

For a collaboration, a PhD opening, or a research-role enquiry.

Three-way wind-speed verification across Antarctic stations

The finding. I compared GraphCast (via WeatherBench2) against ERA5 reanalysis and SCAR READER station observations, for near-surface wind at 25 stations. Checked against reanalysis, the forecast looks accurate. Checked against the stations, real errors show up. Verifying AI forecast against reanalysis hides them.
Why it matters. It's a reason to be careful with AI models like AIFS where there are few observations. The check that matters is against the stations, not against another model.
Current status. Accepted at EMS2026 and presented at NL Polar Day 2026.
Detailed project portfolio

Why I check against station data, not only against models illustrative

How to read it: each line is near-surface wind over a stretch of time. The forecast (amber) stays close to the reanalysis (teal), so checked against reanalysis alone it looks accurate. The station record (dark) shows a real rise in wind that both underestimate. That gap is what my work surfaces, and it matters most where, as in Antarctica, there's very little to check against. The lines here are illustrative.

How I work without an institution

I set the direction

Without a supervisor, I follow recent conference programmes and current papers to find a question that's genuinely open and that I can answer on my own.

I build on open data

Station, reanalysis and model data, brought together in Python. Most of what I need is openly available; some needs API access.

I check it twice

I check each step against published methods. There's no supervisor to catch a mistake, so I do that part myself.

I put it in the open

I present at conferences, and prepare papers and publishable reproducible Python notebooks.

How I built my relevant skills over time

EMS2026 — poster accepted at a major European meteorology conference

NL Polar Day 2026 — poster presented

Peer-reviewed paper — Arctic recreation, Geography and Natural Resources (2015).

NSF Arctic-FROST Fellow (2017)

ECMWF course Machine Learning for Weather System Modelling — (2026, in progress)

KNMI Open Data API Workshop (2026)

MSc Human Geography (2014), magna cum laude — thesis "Peculiarites of recreation in Norilsk Region", spatial analysis skills

Since 2004 — training and practice in statistics and machine learning, originally within economics studies

RYA sailing skipper (2021) — applied marine meteorology

Fieldwork — Taymyr expedition, work in the Nordic countries and Greenland

AI-assisted coding: Claude, github, Python, HTML (I've built this website myself)

Not just Antarctica: check my Montenegro heat comfort window analysis and other weather visualisations!

How I got here

I came to meteorology indirectly. My background is in geography, economics and statistics — about twenty years working with regressions — and since 2021 I've been teaching myself meteorology at university level, because the standard study route didn't fit how I work.

Working on my own, no supervisor signs off my questions, so I check them against the literature myself, twice. I choose what to work on, and in exchange the work has to stand on results rather than a job title.

I have a second line of researchHow autistic migrants are read at work, and where competence gets lost.

Autism & employment