Experimental
Surface measurements
Measurements of CO in the surface layer were conducted at the University of Alaska/Geophysical Institute's Poker Flat Research Range (65.1 N, 147.5 W, 501 m asl) from March 15 to May 10, 1995. From May 24 to August 12, 1995 they were conducted at the 8th floor of the Geophysical Institute, University of Alaska, at the outskirts of the city of Fairbanks (64.9 N, 147.9 W, 205 m asl). A commercial API-300 (Advanced Pollution Instrumentation, Inc., San Diego, CA) gas filter correlation (GFC) instrument with automatic pressure and temperature compensation operated continuously from March 24 to May 11, 1995. Several modifications of the instrument have been made for this work, including addition of mass flow controllers, removal of water vapor using nafion tubing, and frequent checks of the zero level by removing CO from the airstream. These changes have lowered the detection limit to 14 ppb for a 7-minute average at a signal-to-noise ratio of 1. The instrument was installed in a trailer and controlled by a PC. Calibration was made by an addition of a NIST traceable CO standard once per day. During each hour, three 7-min zero cycles were alternated with three 7-min measurements (3 minutes of data were discarded after each mode change). Further details on the GFC methodology and intercomparisons of the UAF-GFC and NOAA measurements are reported by Jaffe et al. [1997]. Other measurements made during spring 1995 at this site include NOx, O3, PAN, JNO2 and hydrocarbons. These data are reported elsewhere [Beine et al., 1997; Herring et al., 1997].
At Barrow, located 750 km NW of Fairbanks, measurements were made at the NOAA Observatory (71.3 N, 156.6 W, 11 m asl), using a gas chromatograph and a mercuric oxide reduction detector with a frequency of 5 air samples/hour. A piecewise-linear calibration was referenced to the NOAA-CMDL gravimetric scale [Novelli et al., 1991]. Flask samples were also collected weekly as part of the NOAA-CMDL air sampling network [Novelli et al., 1992, 1994]. Once the sample was collected it was returned to the CMDL in Boulder, Colorado for analysis. The samples were analyzed by gas chromatography with mercuric oxide reduction detection and referenced to the NOAA-CMDL scale [Novelli et al., 1992, 1994] . In-situ and standard flask techniques agree within 1-2% [CMDL, 1992]. Mixing ratios are reported in parts per billion in mole fraction (ppb). Precision of the CO analysis is 1% (relative STD) and the overall accuracy is estimated to be ±2-3 ppb for mixing ratios between 100 and 300 ppb. Further details on the NOAA-CMDL sampling network and the analysis procedures can be found in Novelli et al. [1992, 1994].
Total Column Measurements
Spectra of solar radiation were recorded from the 8th floor of the Geophysical Institute, University of Alaska Fairbanks. An Ebert/Fastie-type spectrometer of 855 mm focal length with a grating of 300 groove mm-1 , equipped with a solar tracker, was used for measuring the total column CO and H2O contents of the atmosphere. The instrument was designed and constructed at the Institute of Atmospheric Physics, Moscow, Russia [Dianov-Klokov, 1984]. It provides a resolution nearly 0.2 cm-1 in the 2160 cm-1 spectral region. It has a Peltier cooled PbSe detector and a PC-based data acquisition system. A spectrum between 2153.0 cm-1 and 2160.0 cm-1 was recorded every 3 min (Fig.1).
Fig. 1. A spectrum,
observed in Fairbanks, 9 h 15 min Alaska Standard Time, May 3, 1995, solar
zenith angle z = 67.7, resolution 0.2 cm-1.
The retrieval algorithm, an improvement over our previous manual procedure [Dianov-Klokov et al., 1989], is a modification of the "curve-of-growth" technique [Goody, 1964]. For a comparison between the versions see Appendix A. A set of reference absorption spectra with triangle apodization and resolution of 0.0035 cm-1 for the entire atmosphere for varying temperature/humidity conditions were calculated. SFITM program (version 1.09d) [Rinsland et al., 1982] in "zero-iteration" mode (i.e., without fitting calculated spectra to observed ones) and the HITRAN 1992 database [Rothman et al., 1992] were used for this purpose. The entire atmosphere was assumed to be uniformly mixed for CO. These spectra were convolved with the instrument function of our grating spectrometer of 0.2 cm-1 in width. Integrated absorptions (or equivalent widths, EQW) of the CO R(3) line near 2158.300 cm-1 and the H2O line near 2156.564 cm-1 were determined from calculated and measured spectra the same way. Namely, the lines of zero absorption on the spectrum were assumed to pass through the intensity maxima at the sides of spectral lines ("microwindows")(Fig.1). This procedure resolves the problem of uncertainty in zero absorption line for a measured spectrum and makes it possible to make a direct comparison of the results to retrievals made by the SFIT code. First, the program determines the total column amount of H2O, comparing measured and calculated EQWs of the H2O line. This value is used for correcting the EQW of the CO line, since it is overlapped by a weak H2O line near 2158.105 cm-1. Absolute accuracy of the spectroscopic CO determination depends on the accuracy of spectral parameters from the HITRAN database; a comparison between different versions of this compilation (see Appendix B) shows that this does not exceed a few percent.
To compare the CO total column abundance with the in-situ measurements, the former is presented here as weighted mean mixing ratio in parts per billion in mole fraction. These values are close to the mean mixing ratios for the troposphere; for typical profiles they are 2-8% less than the tropospheric means ( Appendix A). One should multiply this by 2.124 E16 to convert into weighted total column amount (in mol/cm2 ). To get the tropospheric part of the total column abundance in mol/cm2 with the tropopause assumed at 265 mb, it should be multiplied by 1.54 E16.
The measurements were carried out between March 20 and August 11, 1995. More than 6,000 spectra were recorded when the sky was clear, and there were 77 days with measurements (including partially cloudy days). Between 20 - 250 spectra per day were measured. Relative standard deviation of individual measurements for 20-min intervals (as a rule, 6 spectra per interval ) is around +/- 3.2 %. This scatter is a result of instrumental, retrieval and atmospheric random errors.