Data is from Munich Re, the large German reinsurer, and the United Nations, and is discussed in this peer reviewed paper by Neumayer and Barthel.
One thing I wish is that economists (etc.) would put error bars on their data, so when it's aggregated like this there's some idea of the quality of the result. This graph suggests that we know global losses due to weather disasters to the nearest 0.01% of global GDP, or less -- or about $7 billion. Given the wide range in the quality of governments around the world, and the secrecy of many (especially China), I'm dubious their data provides this kind of resolution.
We're all trained to ask about error bars on results from the physical sciences, but never ask for them from the softer sciences that are trying to be hard. Here they may well be important, or at least helpful.