Data quality: we have error bars for each individual data point available (estimated from photon counting statistics) which can be used for data quality classification (e.g., automatic editing) or weighting (e.g., weighted nocturnal means). However, these error bars are not shown in the figures to avoid clutter.

"Int." are integrated O2 (or OH) band intensities expressed in relative units (photon counts with scale factors consistent with our older data, divided by the long-term average).
Intensities for November 2002 to January 2003 are uncorrected for an accidental loss of instrument sensitivity, by approximately 50%. These data are preliminary.
Since 2006, intensities (including background) are adjusted to compensate for the lower sensitivity of the new photomultiplier to maintain a consistent intensity scale.

"Temp." are rotational temperatures derived for the O2 (or OH) bands, corresponding to neutral temperature at 95 (87) km. Although both are in kelvins, they certainly contain unknown systematic errors (uncorrelated but probably constant, for each band) and cannot be directly compared with each other.

"Bckgrd" is a sample of the spectral background used to background-correct both band intensities, and temperatures. It is essentially stellar, during moonless nights, with characteristic galactic signatures, in winter. When moonlit, it also shows scattered moonlight and, eventually, clouds (good for editing). Units are photon counts (mean dark count corrected). Long-term stability verified by clear night background comparison at same celestial coordinates (see example).

Unlike in some of our earlier papers, none of these parameters is low-pass filtered, but reflects the total data variance (including instrumental noise).

"Local Standard Time" is Universal Time - 3:00;  note that times for data before April 1998 are corrected for clock drift and therefore are reliable within a few seconds; the others may still be in error by about ±5 min. Since February 2006, a clock drift correction derived from a new daylight sensor again reduces timing uncertainty.

Note that the number of data per night given in the monthly overview table are the arithmetic mean for the four data "channels" (OH/O2 intensities and temperatures). This point is normally irrelevant, except if the differences are appreciable (for instance, no reliable intensities available because of absorption, but temperatures still good).

Figures marked "update 2" show an improved data version avoiding spurious discontinuities that occasionally affected O2 temperatures. The other parameters are practically unaffected by this change. "update 3" corrects a bug in background correction for data with reduced detector sensitivity.