The rows_per_key for idx_obs_daily_1 and idx_obs_daily_2 are nowhere near the same. The rows_per_key for the first element on ibx_obs_daily_2 on the other 5 database slave servers of identical content was been ~2K and ~8K.
Its highly suspicious that exactly the same number got calculated here for rows_per_field when the first element of the each of these indexes is a different field.
ANALYZE table was run on all tables 2 weeks ago when the first instance of an incorrect query plan was observed. Though data changes this index stats also got corrupted again.