If the cause is electrical, there are two possibilities that occur to me. Both result in localized heating, which when it is intense enough can raise the local temperature high enough to start a fire.
The first possibility can occur anywhere in any electrical system. Namely, a high resistance connection. If the resistance is higher than it should be, but not so high as to limit the overall circuit current, then the power dissipated at the connection causes localized heating at the rate of I^2*R. I.e. 10 amps flowing through a 1 ohm connection will produce 100W in a very small location.
The second possibility is unique to PV but requires a double failure. If a PV cell fails it can turn from a power source into a load. The voltage created by all the other functioning cells in series with it is enough to overcome the breakdown voltage of the failed cell to make it conduct while reverse biased. E.g. if a 400V string has a 10A current, and one cell has a breakdown voltage of 30V, then you end up with a 300W heater in a very small location.
The safety function of bypass diodes is to prevent this scenario by being activated. Each PV cell on a module has a Voc of under 1V and a Vmp of around 0.5V. So if you put a diode with a breakdown voltage of 30V across a substring of say 24 cells, under normal operation the 24 cells produce a voltage of 12V, less than the breakdown voltage of the diode (as is the open circuit voltage). While if one cell fails and the substring would otherwise end up with a voltage drop across it of (30 - 0.5*23) = 18.5V the other way, the bypass diode instead starts conducting with a voltage drop of only around 0.5V. So with 10A on the string you get 5W of heating at the bypass diode instead of 300W of heating at the failed cell.
However, if a cell and its bypass diode both fail, you are back to have 300W of heating at the failed cell.
[Numbers are approximate but should be plausible to within a factor of two.]
Cheers, Wayne