search

davidcarslaw / openair

Tools for air quality data analysis
https://davidcarslaw.github.io/openair/
GNU General Public License v2.0
303 stars 113 forks source link

Where did I go wrong with SQTBA? #394

Open Emmanuel-Chevassus opened 3 weeks ago

Emmanuel-Chevassus commented 3 weeks ago

Question

Hi everyone, So I have been running SQTBA over the full 2023 year, with 120 hours long hysplit traj initialised every 3h, what I am looking at is methanesulphonic acid (MSA), an oxidation by-product of phytoplankton emitted DMS, so I find it very surprising that I am seeing this huge source region over Russia more than 1000km inland, as MSA is not emitted over continents as far as I know...Where do you think did I go wrong? I tried using 72h long traj and the issue persist, should I change the gaussian plume speed? maybe Introduce a specific min.bin value? I tried looping SQTBA over specific months instead of the full year, and the Russian area seems to appear in April, doesn´t really make sense...Any ideas why this happens? I would very much appreciate any advice on this,

trajLevel(mydatahourly, pollutant = "MSAOA", statistic = "SQTBA", col = "increment",origin=FALSE) sqtbanofilterfullyear

Cheers, Emmanuel

davidcarslaw commented 1 week ago

Hi Emmanuel

Thanks for your message - and apologies for the slow response; I have been on holiday. Where is the location of the site where these measurements are being made? The pattern of source origins looks rather patchy overall. Is it the case that MSA emissions are highly episodic, and is it also possible that very near-field sources could be important rather than regional sources? This type of analysis will only make sense for longer-range contributions e.g. where it takes many hours or days for a species to be formed. There will be more uncertainty at the edges of the plot due to a lack of trajectory points. What value did you assume for sigma - the default? It would be useful to see what your time series data looks like if it is possible you can share it...

All the best David

Emmanuel-Chevassus commented 1 week ago

Hi David, I hope you had a great time on your holidays, I have looked into this issue, my MSA time series weren´t reliable, we are having some important problems with the capture vaporizer on our aerosols mass spectrometer so that largely explains my issue. Measurements location is Mace Head, so we definitely have local tidal influence that I should have removed so as to not get biased SQTBA sources as you are recommending - I am thinking about using Stuart Grange´s random forest R package for this maybe https://github.com/skgrange/rmweather So I guess my questions now are: -How do we improve SQTBA once we are truly sure we are feeding it long range transport mass concentrations? I have only ever used standard sigma as recommended, did you ever notice improved results by estimating sigma through any models? Can we create an uncertainty map for SQTBA the way Zefir in Igor does uncertainty maps for PSCF/CWT? Is there anything you´d also specifically recommend? For example, some people sometimes introduce dispersion normalisation using the ventilation coefficient (VC) to better account for boundary layer dispersion, others weight far-end hysplit points for long trajectories, there are so many possibilities...Is there anything you personally like doing to improve sources identification? Sorry for the long message, Have a nice afternoon, Emmanuel

davidcarslaw commented 1 week ago

It will be well worth making sure the data are OK.

It's difficult I think differentiating between very close and distant sources. For near-field sources polar plots can be useful. They might also help isolate local sources so you can filter the data set to focus more on distant sources. The wind speed dependence will be a strong indicator of local sources i.e. ~ 1/x relationship. The standard sigma should work OK but if anything it might be useful to increase it.

There's no explicit quantification of uncertainties, although it can help running trajectory models at different starting heights to see how much dependence there is with those assumptions. Note that the SQTBA effectively does dispersion normalisation in that there is an increasing spread in source origins with distance from a source. Beyond that, I think you would need to look at modelling using a CTM - all the trajectory techniques have limitations, but should do a reasonable job when there is a lot of data available - like you have.

Emmanuel-Chevassus commented 1 week ago

Thank you for you advice! I´ll try all of this as soon as we have reliable data and I´ll keep you updated