Xenon-133 and caesium-137 releases into the atmosphere from the Fukushima Dai-ichi nuclear power plantdetermination of the source term, atmospheric dispersion, and deposition

The hardest problem after a nuclear accident isn't the explosion. It's knowing what you released, how much, and where it went. On March 11, 2011, no one at Fukushima Dai-ichi knew. The only way to find out, after the fact, was to run the atmosphere backward. The earthquake that triggered the disaster measured a magnitude of 9.0, striking about 130 kilometers off Japan's east coast. The tsunami that followed flooded the emergency diesel generators at Fukushima Dai-ichi roughly 50 minutes after the automatic shutdown of reactor units 1 through 3. Cooling failed. Water levels fell. Cores degraded and partially melted. Hydrogen explosions damaged the upper structures of units 1 and 3. The spent-fuel pond in unit 4, reported at 84 degrees Celsius on March 13, was later involved in its own explosion. What escaped into the atmosphere during all of this—how much, through which pathways, at what height, and when—was entirely unmetered. Stohl and colleagues set out to reconstruct that release for two isotopes with fundamentally different personalities. Xenon-133 is a noble gas: chemically inert, it doesn't stick to surfaces or wash out in rain. It travels with the air mass and decays with a half-life of 5.25 days. Caesium-137, by contrast, binds to aerosol particles. It deposits on land and sea, enters food chains, and lingers—its half-life is 30 years. Because they behave so differently, these two isotopes become complementary detectives. Xenon tells you when and where the plant vented gases. Caesium tells you where the contamination landed. The method Stohl and colleagues used is called inverse modeling, and the core idea is elegant. Instead of starting at the source and asking where the plume went, you start at the measurements and ask what source could have produced them. Their tool was FLEXPART, a Lagrangian particle dispersion model—Lagrangian meaning it tracks parcels of air as they move through the atmosphere rather than solving equations on a fixed grid. They built a first guess of the emissions from fuel inventories and a careful chronology of documented events at the plant: venting operations, explosions, pressure, and temperature changes. That first guess corresponded to roughly 12.4 exabecquerels of xenon-133 and about 21.5 petabecquerels of caesium-137, but with deliberately large uncertainties so the measurements could reshape the solution. The inversion divided time into 324 three-hour intervals, from March 10 through April 20, and height into three layers: zero to 50 meters, 50 to 300 meters, and 300 to 1,000 meters. That gave 972 unknown emission elements. For each one, FLEXPART was run forward with a unit release, building a source-receptor matrix that encodes how sensitive each downwind measurement is to each possible release. The inversion then finds the emissions that minimize a cost function balancing three things: misfit with the observations, departure from the first guess, and rapid fluctuations in time. In plain terms, the goal is to find the release history that fits the data, stays reasonably close to what we expect physically, and doesn't oscillate wildly. Measurement data came from dozens of stations across Japan, North America, and beyond—atmospheric concentration measurements for both isotopes, and deposition measurements for caesium. Now for what that inversion actually found—and here the xenon result is the one that stops you cold. The total a posteriori release of xenon-133 is 15.3 exabecquerels, with an uncertainty range of 12.2 to 18.3. That is 23 percent higher than the first guess and more than twice the estimated xenon-133 release from Chernobyl, which was 6.5 exabecquerels. Stohl and colleagues conclude this is likely the largest radioactive noble gas release in history. The entire noble gas inventory of reactor units 1 through 3 escaped between March 11 and 15, with no significant emissions detected after the 15th. There's a puzzle embedded in that number. The retrieved release is actually higher than the total estimated xenon-133 inventory of the plant at the time of the earthquake. The explanation lies in iodine-133, which has a half-life of 20.8 hours. As fuel damage progressed, iodine-133 decayed continuously into xenon-133, generating additional activity. ORIGEN calculations showed the iodine-133 inventory at the time of the accident was almost identical to the xenon-133 inventory. Accounting for that daughter-product contribution brings the effective xenon-133 source into alignment with the inversion result—the numbers work out. What doesn't quite work out, at least not with the official accident narrative, is the timing. The inversion's emission time series starts roughly six hours earlier than the first reported successful venting of unit 1. Stohl and colleagues tested this feature under alternative inversion settings and found it robust. The early onset predates active venting—which means noble gases were escaping through some other pathway. The paper discusses several candidates: structural damage from the earthquake itself, leaks caused by overpressure, thermal stress from cold-water injection into hot fuel, and a documented incident in which workers opened an air lock and observed a white cloud escaping. The inversion also retrieves large emissions beginning around 12:00 Coordinated Universal Time on March 11, which the authors associate with a suspected primary containment failure. The forensic implication is clear: the plant's barriers failed earlier and more extensively than planned venting operations alone would suggest. Caesium-137 tells a different kind of story—one about where the contamination landed and why it landed there rather than somewhere else. The total a posteriori release is 36.6 petabecquerels, with an uncertainty range of 20.1 to 53.1 petabecquerels. That is about 43 percent of the estimated Chernobyl caesium-137 emission. Of that 36.6 petabecquerels, only 18 percent of total modeled deposition by April 20 fell on Japanese land—6.4 petabecquerels. Just under 2 percent landed on other land areas. The remaining roughly 80 percent went into the oceans. The reason most of it ended up in the ocean comes down to meteorology during a two-week window. For most of the accident, westerly winds carried the plume eastward over the North Pacific, and in-cloud scavenging removed caesium from the air over the sea. However, during the strongest emission period—March 14 and 15, when modeled release rates peaked at around 400 gigabecquerels per second—a developing cyclone over southern Japan pulled contaminated air back over eastern Honshu. Rain from that cyclone produced intense wet scavenging over land, driving simulated local deposition near the plant up to nearly 1,000 kilobecquerels per square meter. A second episode followed on March 20 through 22, when a frontal system again brought contaminated air over Honshu, producing additional deposition that reached Tokyo. The authors are explicit: had the winds blown offshore instead of turning back over Japan during those emission peaks, the land contamination would have been far smaller. Two weeks of weather decided the map. The caesium results also exposed a methodological sensitivity worth noting. Stohl and colleagues ran the model with two different meteorological inputs—National Centers for Environmental Prediction Global Forecast System and European Centre for Medium-Range Weather Forecasts—and found Global Forecast System agreed better with measurements. European Centre for Medium-Range Weather Forecasts produced stronger wet scavenging: in direct comparison for March 13 and 14, European Centre for Medium-Range Weather Forecasts deposition was 22 percent larger than Global Forecast System. When the inversion was rerun using European Centre for Medium-Range Weather Forecasts meteorology, the total caesium emission estimate dropped by 32 percent. Which weather model you use materially changes what you conclude about how much was released. That's a significant source of uncertainty, and it's honest of the authors to say so. Stohl and colleagues close with a call that is worth amplifying. Their entire analysis depended on measurement data from the Comprehensive Nuclear-Test-Ban Treaty Organization radionuclide network and supplementary datasets from European, Taiwanese, and American stations, along with Japanese deposition measurements. Almost none of those datasets are publicly available. The paper explicitly recommends creating a central data repository—quality-assured, standardized, and open to the public—stating that such sharing would allow considerable improvements to source-term estimates in the future. Given that the uncertainty range on the caesium estimate spans from 20 to 53 petabecquerels—a factor of more than two—the practical stakes of that recommendation are real. What this study ultimately demonstrates is that the atmosphere keeps records. Even when releases go unmetered, even when the accident unfolds in chaos, the fingerprints of what escaped are preserved in air samples from Sacramento to Stockholm. Inverse modeling, given a good transport model, enough measurement data, and a sensible first guess, can read those fingerprints. It's forensic atmospheric science—and Fukushima is now its most consequential case study. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.