TIL that retired chemist, Khonghao Shou, predicted the 2003 earthquake in Bam, Iran by observing cloud formations.
He not only predicted the approximate date, but also the magnitude and epicenter.
His method (cloud formations) is not widely accepted by the scientific community. Perhaps because his geological and meteorological skills are self-taught. He claims a prediction success rate >60%, and that it would be higher if he had better access to satellite imagery.
"Shou’s unconventional theory is as follows: intense heat and pressure at faults, due to friction between sections of crust, enables underground water vapour to collect in cracks around larger faults. These can be released to the surface, where the amount of vapour can indicate the strength, and thus magnitude, of the impending earthquake. Once the vapour is released from the fault, which signals roughly where the epicentre will be, it heats up the surrounding air. This enables it to rise, and, once at a suitable altitude (as is the standard procedure for cloud formation), cool and condense - forming clouds roughly along the fault line, fuelled from the point of vapour release. This release is accompanied by a drop in strength of the surrounding rock, hence making an earthquake likely."
Or, for a quasi-animated version, skip to 0:48 in the provided youtube link.