Being remote sensing, algorithms for inverting the raw data to get physical quantities may be improving with time. Perhaps a slight adjustment in algorithm or calibration happened between July and Sept of this year. However you would expect this to introduce a systematic adjustment, not a scatter.
One might reasonably wonder why the data is steady before 1999. That may be because there were fewer (or different) remote sensor parameters measured then, or the data wasn't as precise.
Another possibility is that the calculation was simply done more precisely, and you are looking at the elimination of round off errors, or different round offs.