On 1 September 2022 the IAEA mission led by Rafael Grossi reached Zaporizhzhia NPP. They travelled in armoured vehicles across the front line, were delayed at the first checkpoint by shelling, and arrived around noon. This was the agency’s first permanent presence at a nuclear power plant in an active conflict zone.
On 11 September at 03:41 the last operating reactor (unit 6) was disconnected from the grid. All six units were transitioned to cold shutdown. Zaporizhzhia NPP stopped generating electricity for the country.
Our prototype was already working at that point.
Why a prototype rather than a planned product
The timeline of events at ZNPP through summer and autumn 2022 set the pace:
5 August. Shelling damaged a 750 kV distribution switchyard. Three transformers tripped. One of three operating reactors shut down in emergency mode.
6 August. Shells landed near the dry storage for spent nuclear fuel. The IAEA called for a “nuclear safety protection zone”.
25 August. The last of four 750 kV external lines was disconnected — a fire at a nearby coal ash pond. For the first time in the plant’s history, ZNPP lost all connection to the grid.
1 September. The IAEA mission arrived. ISAMZ (the permanent support mission) was established.
7 September. Shelling damaged a reserve line between ZNPP and an adjacent thermal station.
11 September. Cold shutdown of all reactors.
30 September. A mine explosion damaged a 6 kV cable at the perimeter. On the same day Russian forces detained ZNPP director-general Ihor Murashov; he was released on 3 October.
Over the course of 2022, ZNPP lost all external power six times. Each time, diesel generators took over. Each event was hours during which a single generator failure could lead to fuel overheating.
Planning a forecasting system over several years was not an option. We decided to assemble a prototype from what we had.
Architecture: three components
Meteorological forecasting. Our AI YAT-Meteo model generates a three-dimensional forecast of wind, temperature and atmospheric stability parameters up to 48 hours ahead. Without accurate wind fields, any dispersion model produces unreliable results.
Dispersion model. We used CALPUFF — an atmospheric transport model that was the US EPA preferred model for regulatory applications at distances above 50 km from 2003 to 2017 [2]. CALPUFF uses three-dimensional wind fields, accounts for terrain, chemical transformations and dry/wet aerosol deposition.
Visualisation. Modelling outputs are overlaid on an interactive map alongside real-time radiation sensor readings, so the forecast can be compared against current measurements on the same display.
What the prototype showed
We ran test scenarios for Zaporizhzhia and Chornobyl NPPs. The main result is technically obvious but politically important: plume direction is entirely controlled by weather.
Under one set of meteorological conditions a ZNPP plume travels toward Rostov oblast. Under another, toward Dnipro, Zaporizhzhia, Kryvyi Rih. Static evacuation circles around an NPP are only the first tier of protection. An actual decision support system needs to update dynamically with the weather forecast.
Dose thresholds for protective actions are defined in IAEA GSR Part 7 [3]:
| Action | Threshold | Time window |
|---|---|---|
| Sheltering indoors | 10 mSv | up to 2 days |
| Evacuation | 50 mSv | up to 1 week |
| Iodine prophylaxis | 50 mSv (thyroid) | first 7 days |
European guidance from HERCA–WENRA, updated for the Ukrainian situation in March 2022 [4], defines spatial zones:
- evacuation — up to 5 km from the NPP;
- sheltering + iodine prophylaxis — up to 20 km;
- for a severe scenario (Fukushima-level) — evacuation up to 20 km, sheltering up to 100 km.
The prototype computed which specific settlements fall inside these zones under current wind. For each of Ukraine’s 5 NPPs — Rivne, Khmelnytskyi, South Ukraine, Zaporizhzhia and Chornobyl — the zone configuration differs radically.
Partnership with the Marzeiev Institute
A plume forecast is only half the picture. You also need a health-impact assessment: external and internal doses, risks for specific age groups.
For this we reached out to the O. M. Marzeiev Institute for Public Health under the National Academy of Medical Sciences of Ukraine. Founded in 1931, the institute has been involved from the first days of the Chornobyl accident in 1986 — characterising radioactive contamination, computing effective doses for residents of the strict-control zone.
The Marzeiev Institute’s Radiation Protection Laboratory (led by Prof. T. O. Pavlenko) closed the last leg for our system: from plume forecasting to population dose assessment. A full pipeline — from meteorology to recommendations.
What comes next
The prototype demonstrated that the problem is tractable: the model runs, maps generate, sensor data integrates. The next step — taking it to product level: add a dose scale, automate runs for all 5 NPPs, prepare the interface for emergency services and environmental agencies.
That’s the subject of the next article in the series.
Current radiation situation and system status: see predictive modeling for NPP accident scenarios.
Ukrainian startup ecosystem: follow TechUkraine and AIN.ua — the two leading outlets covering Ukrainian deep tech, climate tech, and environmental startups.
Follow Ukrainian media: Ekonomichna Pravda, Mind.ua, and Hromadske regularly cover the environmental consequences of war, air pollution, and infrastructure risk in Ukraine.
What to do today
- Check the YourAirTest air map for your city.
- If this resonated — share the article.
- Subscribe to our articles on air quality in Ukraine.
References
Skamarock WC, Klemp JB, Dudhia J, et al. (2021) A Description of the Advanced Research WRF Model Version 4.3. NCAR Technical Note NCAR/TN-556+STR. NCAR OpenSky
Scire JS, Strimaitis DG, Yamartino RJ. (2000) A User’s Guide for the CALPUFF Dispersion Model (Version 5). Earth Tech Inc., Concord, MA.
International Atomic Energy Agency. (2015) Preparedness and Response for a Nuclear or Radiological Emergency. General Safety Requirements Part 7. IAEA Safety Standards Series No. GSR Part 7. IAEA
Heads of the European Radiological Protection Competent Authorities & Western European Nuclear Regulators Association (HERCA–WENRA). (2014, updated 2022) Approach for a Better Cross-Border Coordination of Protective Actions during the Early Phase of a Nuclear Accident.
Pisso I, Sollum E, Grythe H, et al. (2019) The Lagrangian particle dispersion model FLEXPART version 10.4. Geoscientific Model Development 12:4955–4997. GMD