Also, I am not sure of the choice of the Poisson distribution suggested above. Roofers are experienced at walking on rooftops, and die due to, well, accidents. That and roofing work is fairly consistent. This number would follow some sort of normal or Poisson distribution, but homeowners are typically only used to a ladder to clean rain gutters out, not roofing or rooftop walking.
Death of roofers/installers during initial installation were the numbers I was quoting from Australia. I think homeowner fatalities will top the roofers in areas with increasing solar panel installations. I would think a Weibul distribution with a Beta of 0.5 would better suit this scenario to model all of the inexperienced homeowners cleaning their solar panels monthly, and then dropping off (no pun intended) after reports and education on working at heights diffuses into the populace.
Deaths from fusion and nuclear on the other hand are very unpredictable. It is easy to imagine that year after year, no people die. And then suddenly millions of people die. That means that equating actual danger with observed-deaths-so-far is just not possible.