Fun fact: In some statistics on the safety of different forms of electricity generation, the largest dam failure [1] is left out, which brings hydropower down to same level of safety as nuclear, wind and solar. Our World in Data is an honest source, and reports both values [2].
If you see hydro reported as safe as nuclear, solar and wind (for example in [3]), then they excluded the largest accident. If hydro is 2 orders of magnitude more deadly than nuclear, wind and solar, then they included everything.
solar 0.02 deaths/TWh
nuclear 0.03
wind 0.04
hydro 0.02 without, 1.5 with
The failure is of the Banqiao Dam in China, in 1975.
It was an extreme outlier amongst even major dam failures, and most of the contributing factors are at best distantly related to the technology itself, whilst common to any large and high-risk technical endeavour.
Repeating what I've written before:
[T]he failures largely accrued from institutional hubris, engineering insufficiency, lack of relevant domain knowledge (often deliberate ignorance or denial, see especially Vajont, also St. Francis), poor overall management, lack of disaster preparation, drilling, or readiness, limited resurces or capabilities (especially in developing countries), communications breakdown (see Banqiao's comms loss), and inadequate response in light of imminent or present threat.
None of these are domain-specific to hydraulic civil engineering or absent from nuclear engineering projects.
The specific failings at Banqiao were virtually all managerial and political, and not technical: poor engineering, inadequate safety provisions, underestimated environmental and operational risks, poor contingency planning, unforeseen perfect storm (literally), severed communications, insufficient warnings, no community disaster preparation, inadequate rescue and recovery. None of these failures are specific to hydro, all apply to nuclear power, and as non-engineering problems there is no technical fix that makes them go away.
In Banqiao, about 25,000 people died in the immediate inundation. Another 150,000 died in the following weeks of starvation and disease. There's no great mystery as to how such deaths are avoided: floodwaters are mitigated by high ground and evacuation centres; starvation and disease by food, water, and medical stocks; and rescue & recovery by trained teams and equipment. Reestablishment of communications, transport, and utilities is critical. These are all basic mitigations and are common to a wide range of foreseeable incidents.
China at the time was desperately poor, politically dysfunctional, institutionally corrupt and inept, and gambled hugely on risk and lost. Other major hydro disasters tend to share these traits.
The risks from Banqiao were short-term (acute), though profound. They could have been mitigated by planning, preparation, warnings, evacuation, and response. China at the time lacked the safety and civil defence maturity and mindsets, the political will, the engineering culture, and simply the capacity to respond appropriately given the magnitude of the disaster. The deaths were avoidable, and similar threats, say the 2017 Oroville Dam spillway failure. The engineering failure resulted in costs, but no loss of life, nor any significant damage to property other than the dam itself.
The U.S. still does see occasional dam failures, as with the Edenville and Sanford Dam failures of 19 May 2020. Here again poor engineering, construction, and operation created the risk, though again, preparedness and response prevented deaths, though in this case there was significant property damage.
Today, Chernobyl and Fukushima are exclusion zones, and they will be for centuries at a minimum. The reactor cores themselves, for tens of thousands of years. This is far longer than the companies, countries, cultures, and even languages extant at the time of the disasters --- our capacity to address risks and management at this scale is utterly nonexistent. Banqiao today is home to 17 million people who face no ongoing concerns from the incident.
Nuclear power also becomes a threat-multiplier: Chernobyl (and the neighbouring ZaporizhzhiaNuclear Power Station, which had been operating normally prior to Russia's 2022 invasion of Ukraine) have both proved militarily-significant points of leverage in the conflict, with potential impacts extending well beyond the immediate region and time. Yes, dams have also played that role, but the ultimate impacts there are localised in both space and time, as well as highly tangible in ways that radiation threats are not.
Dam failures are short-lasting disasters, occurring over hours, days, or perhaps weeks. Relative to total installations --- tens if not hundreds of thousands worldwide --- major incidents are relatively rare.
Nuclear failures are long-lasting disasters, occurring over decades, centuries, and perhaps millennia. With fewer than 400 nuclear power plants worldwide, we've seen numerous catastrophic failures as well as far more close calls.