For pressure + temperature, the measurements were encoded in the first two consonant-vowel pairs. So, for example, in the word "diamond", the two measurements were "di" (= .26 inHg) and "mo" (= 58°). There are 50 possible pressure readings and 50 possible temperature readings, for a total of 2500 words; of these, a small fraction are encoded as "arbitrary" words when no English word could be found that matched the desired pattern. You can see a full list of all the words here: https://books.google.ca/books?id=i7YazQEACAAJ&pg=PA31#v=onep...

Keep in mind: the encoding process was meant to be more expensive to encode than to decode - each weather station only had to send out (encode) a few reports per day, but the receivers had to process (decode) hundreds.

I suspect the use of words, as opposed to just sequences of letters or numbers, was to improve robustness: it would be easier for operators to transmit recognizable words correctly (and error-correct them if need be) than if they were transmitting strings of random letters.