November 9, 2012 at 5:58 am #1408
I've been using EddyPro 4.0 advanced processing for our open path eddy covariance data. Currently, I am doing a flux comparison between Ameriflux processing and Eddypro using a 10 days data set. I am getting very poor correlations specially in FC; LE and H are pretty much spot on.
Could you indicate me which output files can I find the WPL terms applied to FC?.
Or could you please give me any suggestions as for were can I trace this differences within EDDYPRO processing? . Please see attached picture for a daily cycle comparison. Although, in only one day the difference is small, it makes a bigger impact in the cumulative values.
[attachment=966,98]November 9, 2012 at 6:40 am #1650
the reason why you cannot find the details of the WPL terms on EddyPro’s output is that, mostly depending on the instrument type (open or closed path), the processing follows a different path and in some cases the WPL correction is inextricably interwoven to the spectral corrections and to the humidity correction of sensible heat flux (please refer to the software documentation for the details). For this reason, it isn’t obvious how to output unambiguous “WPL contributions”. However, we are getting this request from time to time, so we will be thinking about a convenient way to provide this information on output.
That said, the graph certainly shows a major discrepancy that seems to be related to WPL and/or spectral corrections. I would be available (and interested) to analyze this data in details. I would thus suggest you contact me by email (firstname.lastname@example.org) to exchange raw files and results. Once we get things clear, we can post the outcome here on the forum.
GerardoNovember 12, 2012 at 10:58 pm #1651
Sounds great. I just sent you the info to your email.
ThanksNovember 16, 2012 at 9:32 pm #1654
I tracked down the origin of the difference in Fc, completely explained by the difference in the spectral correction factors, which is due to different correction methods (Moncrieff et al. 1997 in EddyPro, Massmann 2000, in Ameriflux). The difference in final Fc appears "amplified", with respect to the actual difference in the spectral correction factors, due to the fact that here final Fc are small numbers resulting from the difference between two relatively large numbers, the spectrally corrected raw CO2 flux (where most of the difference shows up) and the WPL term. Attached a summary plot showing this.
Note also that raw CO2 fluxes before spectral corrections are identical (not shown in the plot).
[attachment=989,103]April 9, 2013 at 9:47 am #1753
I also have compared EddyPro results with other software, in this case with my Alteddy package. I compared for two sites, a forest site in the Netherlands (Loobos) and a grassland site in Spain. Both sites have LI7500, the grassland site has a low mast (instruments at 2.35m agl) and lower fluxes.
For both sites raw covariances and final H are virtually the same but final LE and Fc are consistently higher for EddyPro. For the forest site the differences are small, +4% for LE and +2% for Fc. For the grassland site LE results are +8% and Fc is +14%. See attached graphs. I suspect this is mainly a result of the frequency response corrections.
My program uses Kaimal 1972, with EddyPro I used Moncrieff 2004/1997.
Is there a way to extract the spectral correction factors or raw spectrally corrected fluxes from EddyPro? How do we determine which flux comes closest to the "truth"?
[attachment=1318,173] [attachment=1318,174]April 9, 2013 at 1:14 pm #1754
Is there a way to extract the spectral correction factors or raw spectrally corrected fluxes from EddyPro?
In the full_output file from EddyPro there is a section towards the end of the columns labeled as ‘uncorrected_fluxes_and_spectral_correction_factors_(scf)’. That section will have the correction factors and the uncorrected flux.
How do we determine which flux comes closest to the “truth”?
I will let the others discuss this one.April 9, 2013 at 3:40 pm #1755
I suggest we perform a systematic sw intercomparison. Lately I have been performing a comparison with Matthias Mauder between TK3 and EddyPro, that we are presenting in these days at the EGU (Matthias will be at the poster) and that you can find attached to this post. As you can read, we encountered a similar issue, although in that case it gave smaller differences.
As you said, it is likely that the difference is due to the difference in the spectral correction methods, which is then possibly “amplified” in the final fluxes as discussed in the first part of this post. The larger the WPL term with respect to the raw flux, the larger the apparent amplification of the spectral correction difference into the final fluxes.
As Israel mentioned, the spectral correction factors can be found in the last part of the full output file, but the only way to reconstruct all flux contributions is to start from the covariances and recalculate WPL after the spectral correction.
I think it would be more insightful and instructive to perform a comparison EddyPro vs. Alteddy. We may use the same dataset (open and closed path) used in the comparison with TK3, and dig a bit into the details.
How about it?
GerardoApril 9, 2013 at 3:45 pm #1756
In my full_output file I find the following headers:
file_info, corrected_fluxes_and_quality_flags, storage_fluxes, vertical_advection_fluxes, gas_densities_concentrations_and_timelags, air_properties, unrotated_wind, rotated_wind, rotation_angles_for_tilt_correction, turbulence, footprint, uncorrected_fluxes, statistical_flags, spikes, variances, covariances
So no spectral correction factors. I use version 4.1.0. Am I missing something obvious?April 9, 2013 at 3:54 pm #1757
Jan, look in the second line of the header. Along with each uncorrected flux (e.g. un_LE) you’ll find something like LE_scf which stands for Spectral Correction Factor.
GerardoApril 9, 2013 at 4:18 pm #1758
Ah, found it!
Will get back on your proposal tomorrowApril 10, 2013 at 3:46 pm #1760
The deeper I dig into this the harder it becomes to come to conclusions. I now found that EddyPro’s u* values are a bit higher than mine and found a difference in the angle of attack calibration results for w. Which routine do you currently use, aoa_cal_nakai_20212.f90 or aoa_calibration_nakai.f90 ?
Maybe we could switch to a different dataset although that might result in some more work for you…April 10, 2013 at 5:44 pm #1762
if you are using EddyPro 4.1 in Express mode, then the angle of attack correction is after Nakai and Shimoyama (2012). If you use it in advanced mode, of course it depends on your selection. I recently successfully crosschecked (for the second time) the implementation of the 2012 version with Taro Nakai, so I assume it is correctly implemented in EddyPro.
It is ok for me to switch to a different dataset: more chances for new findings. Please think of a way to share your dataset with me (dropbox or similar?), and make sure you send me also the “.metadata” and the “.eddypro” file that you created with EddyPro.
Also, I guess we can continue this conversation by email (email@example.com) and post here the results of our comparison? 🙂
GerardoApril 11, 2013 at 8:25 am #1763
I use advanced mode and thought I was using Nakai2006 (because this is implemented in Alteddy). However I found out that this setting is not saved in the metadata file.
In other words, if I select Nakai2006, save the metadata file, close EddyPro, open EddyPro, open the metadata file, then the setting reverts to Nakai2012.
So the only way to use Nakai2006 is to select it and then run the data without closing EddyPro.
Hope this explanation is clear.
Let’s continue by email, will send you a message today.
JanApril 17, 2013 at 12:39 pm #1764
An update on EddyPro 4.1 vs Alteddy 3.74 for an open path gas analyzer on a crop site. It turns out there was a bug in the frequency response correction routine of Alteddy 3.74. The bug has been traced back to version 3.70.
After correction the comparison of final fluxes looks much better. See graphs. LE is still about 3% larger for EddyPro, I will look into this in due time.
[attachment=1355,177] [attachment=1355,178] [attachment=1355,179]
You must be logged in to reply to this topic.