Date: Tue, 01 Oct 2002 16:33:03 MDT To: cases99-list@lists.Colorado.EDU cc: sean@ucar.edu, mahrt@oce.orst.edu, vickers@oce.orst.edu, reina@oce.orst.edu From: "Sean Burns" Subject: ATD 5-min fluxes vs OSU "gap" fluxes... Hi, As mentioned at the meeting last week I compared the OSU "gap" fluxes to the ATD 5-min fluxes. These comparisons are on the web at: http://www.mmm.ucar.edu/science/abl/cases/meeting_storrs02.html Also on this webpage are the haar wavelet plots that Jielun showed in her presentation. A brief explanation of the flux comparison plots will probably be helpful...if you look at the "Example from one night" pdf the top panel shows the the ATD 5-min fluxes along with the 1-hr avg from these 5-min fluxes...the next panel plots both the 1-hr avg of the 5-min-ATD fluxes and the OSU fluxes...the 3rd panel plots the difference between the hr-averaged-5-min fluxes and the OSU fluxes. note the y-scale of the 3rd panel is a factor of 10 smaller than the one above it. the bottom 2 panels are wind speed and wind direction. This particular night is one which showed the largest differences between the OSU and ATD fluxes. the "Overall Summary" pdf is a 9-page pdf file. each page is for a different sonic level. There are the frequency distributions of the difference between the hourly OSU and ATD fluxes for all the "good" data time periods. The difference is expressed as a percent difference betweeen OSU and ATD...ie: percent diff = 100 * (wTv_OSU - wTv_ATD) ./ abs(wTv_OSU) the normalization is done by the absolute value of the OSU flux so the sign doesn't change. As an example page1 of the pdf file is for the sonic at 0.5m. . .the blue line (solid dots) is for the daytime data and the green line (open circles) is for the night time data. Note that i didn't include the transition time periods (ie, 6-7 and 18-19 CST). The numbers in the upper left corner indicate the number of valid hourly pts used for that particular sonic...e.g., for the sonic at 0.5m during the daytime the number of valid pts consists of 59 hrs of data out of a possible 341 hrs. For the sonic at 0.5m the number of "valid" data is rather small since the sonic was moved from 1.5m to 0.5m partway through the experiment. How to intrepret this plot (using 0.5m sonic page as an example): The peak of the distribution for the daytime data (blue line) is around 67% at an error of zero. . .since the bin size for each point is 5%, this means that 65% of the hourly differences had an error of between -2.5 to +2.5%. if you look at the next point to the right it is at about 8% which means about 8% of the data had an error between 2.5% to 7.5%...and so on. The daytime data are fairly symetrically distributed around zero, but the nighttime data show a tendency to errors which are greater than zero (this means that ATD fluxes are MORE NEGATIVE than the OSU fluxes at night). The points at the extreme ends indicate the percentage of hourly points that had an error of more or less than 40%. Here are my conclusions based on these plots: 1. During the day the OSU fluxes are slightly greater than the ATD 5-min fluxes (due to 5-minutes not being long enough to sample all the flux-carrying eddies). The bias is about 5% and is more pronounced for the higher level sonics. 2. At night the ATD 5-min fluxes are typically slightly larger (ie more negative) than the OSU fluxes. The bias is also about 5% and is more consistent at all tower levels. . .I am not sure why the OSU fluxes would be SMALLER than the ATD fluxes at night?? 3. At higher levels on the tower there are more large errors (ie, greater than 40% differences). The percentage of pts with large errors is greater for the night-time data than the daytime data. For example the sonic at 50m shows about 12% of the nightime data have errors greater than 40%, and 12% have errors less than -40% (these are cases where even the sign of the flux is wrong). My interpretation of this is that the 5-min flux window is too long in these cases and causes the flux to be poorly estimated. 4. At night some of the fluxes are quite small (order of a few W/m2) so a 10% error means the difference between ATD and OSU is only about 0.1 to 0.2 W/m2. There are probably other (more insightful) ways to do these comparisons but this gives a rough estimate on the magnitude of errors in the ATD 5-min flux calculations. If you have any questions or comments let me know... Sorry for the length of this e-mail! SpB. ------------------------------------------------- Date: Tue, 01 Oct 2002 16:10:15 PDT To: Sean Burns cc: cases99-list@lists.Colorado.EDU, mahrt@coas.oregonstate.edu, vickers@coas.oregonstate.edu, reina@coas.oregonstate.edu From: Larry Mahrt Subject: Re: ATD 5-min fluxes vs OSU "gap" fluxes... Sean, I think your interpretation is correct. Apparently at night, the 5 min flux for the most stable cases tends to include a little bit of flux in the same direction as the turbulent flux with a larger random component. The random part of the flux becomes much more dominant if one uses a traditional 30 min. averaging time, which can lead to many cases with bad flux estimates. Fortunately, much of the this error, but not all it, was eliminated with the use of 5 min. averages. As a reminder to those who are not flux people, these averaging times refer to time used to compute the perturbations. Then one can average the fluxes over any arbitrarily longer averaing time to reduce the random flux error. Traditional is 30 min., we use one hour. Larry