Introduction

Last update: 29 October 2001

For the October 2001 exercise we have focused on the 1 degree & 1 hour matchups at /dom/files/ops/airs/tlscf/2000/12/airx2msy. There are 4 files for 15 December 2000 at 00, 06, 12, 18Z and one file for 16 Dec at 00Z. We have used the corresponding NCEP analysis files (not available on dom) to compare with the simulated observed radiances in the matchup files.

The first step was to read the matchup files and pull out the warmest single spot of the 3x3. For every matchup we calculate the radiance by feeding the NCEP profile thru our KLAYERS and SARTA programs. To speed things along, we only calculated the radiances for the handful of channels used by our simple nighttime cloud-free sea surface detection algorithm.

The cloud-free sea surface detection algorithm uses two window channels, one at ID 759 = 900.2 cm-1 and the other at ID 2333 = 2616.1 cm-1. We compute an effective sea surface temperature for each channel by using the NCEP profile calculated radiance to estimate the atmospheric effect. The estimated 2616 cm-1 surface temperature must agree to within 1 K of the NCEP sea temperature, and the 900 cm-1 estimated sea temperature must agree with the 2616 cm-1 temperature to within 0.5 K. We also check that we are over ocean (no land) and night (solar zenith > 100 degrees). We also check for ice using the NCEP ice flag as well as checking surface temperature is above 273 K.

A better description of this algorithm is available on pages 6 & 7 of Larrabee Strow's talk from the June 2001 Science Team meeting.

Once we've determined which locations are clear, we re-run the profile thru SARTA to compute the radiance for all channels. Total run time (start to finish) for any one matchup file is on the order of 3 minutes. The number of clear spots varies but is on the order of a couple hundred.

Results for the NCEP analysis profiles

Here are some plots of the mean difference in brightness temperature (calculated minus observed) for the five matchup files. (Blue=00Z, green=06Z, red=12Z, cyan=18Z, all 15 Dec, and magenta=00Z 16 Dec). The large differences at 15 um and 4um are probably due to variations in the upper atmosphere portion of the simulation profiles; the NCEP profiles have no info up high. The poor agreement in the ozone region is also likely due in part to differences up high. The most interesting result is the small bias visible in the two window regions. At 800-1000 cm-1 the bias is around 0.3 to 0.4 K and at at 2400-2700cm-1 it is 0.15 K. The cause is uncertain, but it may be due to a tiny amount of clouds in our "clear" observations.

Here is a plot showing the RMS difference and absolute value of the mean difference for the 06Z case alone.

Here is a plot comparing the mean difference for the "m16" (blue) and "nom" (red) channel sets for 06Z. In the windows regions and away from any local lines, the two differences are very similar, while the m16 shows the expected plus and minus difference on top of every line due to the frequency offset.

Here is scatter plot of the BT difference as a function of mean brightness temperature as well as channel frequency (color). The blue linear feature on the right angled down is Ozone. The red and dark blue features that slowly slope down until they bend sharply on the left are the fixed gases (mostly CO2) in the short and longwave. The cyan feature that slowly bends upward in the middle is water. The few cyan dots that follow the red and dark blue curve is methane.

scatter plot

Here is a plot showing the effect of scan angle on the mean BT difference. The colors in the plot are for scan angle: blue=0 to 10 degrees, green=10-20, red=20-30, cyan=30-40, and magenta=40-50. The data set is the combined 00, 06, 12, and 18Z NCEP matchups. Note the mean difference tends to be larger at the large angles.

Here is a plot showing the effect of latitude on the mean BT difference. Again the data set is the combined 00, 06, 12, and 18Z matchups, but this time grouped by absolute value of latitude. Blue=0 to 10 degrees, green=10-20, red=20-30, cyan=30-40, magenta=40-50, and yellow=50-90. The difference tends to be smaller at the equator (stable air masses) and at high latitudes (small water vapor).

Here's the same latitude plot except using the level 2 cloud cleared radiances for less than 15% mean cloudiness. The mean cloudiness comes from an average of the clouds in the 3x3 spots as reported in the level2 standard portion of the matchup. The mean cloudiness in the two cloud levels was summed. As this plot shows, the mean BT difference in the 800-1000 cm-1 window is larger than our level 1b results.

We also did a comparison of NCEP calculations to granule 84 observations. The mean brightness temperature difference is similar to the difference for the 1 hour 1 degree matchups.

Here is scatter plot showing the individual observations calc - obs for the 2616 cm-1 channel. Looks like part of the BT difference might be partially correlated with view angle. That may be due to errors in the profile which get magnified at larger angles. The NCEP profiles come from the 06Z analysis, while the mean time of the granule 84 observations is 08.4Z.

Results using ECMWF profiles

We did similar calculations using the ECMWF analysis profiles as we did for the NCEP profiles. The simulation data was based on NCEP forecast profiles with additional high altitude info, so the mean BT difference for the ECMWF profiles tends to be larger than for the NCEP profiles, especially for water vapor.

Clear detection comparison: our L1B test vs L2cc

We compared our (UMBC ASL) L1B clear detection to the cloud flag and cloud fraction from the L2cc. For the 00,06,12,18Z 15 December 2000 matchup data (version 85 files), our L1B clear detection routine found 1204 clear observations over sea at night. The L2cc had 2254 cloud-cleared observations, and contained 1135 of the 1204 spots found with our L1B test. Here is a histogram of the mean cloud fraction in these 1135 spots. The mean value is about 21 percent.

Here is aplot showing the cloud fraction for each of the 1135 spots. The data is shown using three colors/sysbols corresponding to the value of the L2cc "cloud_flag": not clear = cyan dots, clear = blue circles, unknown = red x. The cloud flag does not appear to be working.

Here is a plot that shows the mean BT difference between the observed radiances, L2cc - L1b (shown in blue). Also shown is the mean difference in calculated radiances (in red). The view angle of the two calculated radiances can differ by 1.1 degrees due to the L1b being any one of the 3x3 AIRS spots while the L2cc is always on the middle(?) spot. In the window regions the difference is negative, which might suggest the L1B is slightly more clear.