Confusion with results produced when changing sampling time interval and type

Topics about the HYSPLIT dispersion model.
GarySutlieffUoB
Posts: 10
Joined: March 22nd, 2021, 11:21 am
Registered HYSPLIT User: No

Confusion with results produced when changing sampling time interval and type

Post by GarySutlieffUoB »

I have been using HYSPLIT (the desktop client, to be specific) to retrieve a vertical profile of concentration of Sulphur Mustard gas for use in a radiative transfer model, the objective being to assess the capability of detecting this gas with a specific instrument.

To collect test results I had initially modelled a 100kg release over a 36 second time period and sampled within a 1x1 km grid, with vertical levels every meter up to around 300 m. The sampling interval was every 15 minutes within a 3 hour period, which produced one set of viable results (at the first 15 minute mark after the release was completed, with all subsequent results showing zero concentration). This, once processed and used with the radiative transfer model, led to the sort of result I was expecting/aiming for, but my next task was to establish how the concentration (and therefore the detectability) of the gas varied over time. Since the gas was gone from the sampling area by the 30-minute mark, I thought it logical to switch to a 30 minute sampling duration and to sample every minute, collecting 30 sets of data to then produce a graph of concentration over time from. However, I found that in the results of this new data set, there was only a non-zero value for the concentration at the 1-minute mark, with the value changing to zero from minute two onwards! This was unexpected and confusing, and led to me realising that I had overlooked the sampling type setting in the sampling grid set up.

However, I was more confused when I tested to see what results looked like with both the snapshot and time-averaged sampling types for both the 1-minute and the 15-minute sampling intervals, and found that my original configuration (time-averaged sampling type, 15-minute interval) was the only configuration that produced a concentration at any point beyond 1 minute. While it makes sense that there would be a value for the concentration at 15 minutes in a time-averaged configuration since the average would include the reading from minute 1 alongside all of the readings of zero for subsequent minutes, what's strange is that the values obtained in my original dataset at the 15-minute point, despite being time averaged over the fifteen minute period which presumably includes several readings of zero concentration, are of the same order as the concentrations recorded in the data sets with a one-minute sampling interval.

My questions are these:
- Why would these values be of the same order if the time periods over which they are averaged are drastically different?
- Why does changing one or both of the sampling time and sampling type have an unexpectedly drastic impact on the results? I understand from the user manual what each setting does in theory, but parts of the results (such as the similarity in order from results with drastically different inputs) have me confused.
- What configuration should I be using to try and collect the data that I need?

I'm sure I'm just missing something simple but I'm still not super experienced with the model, so if anyone could enlighten or guide me with this I'd be greatly appreciative. I'm happy to provide screenshots of my settings if necessary, along with whatever extra information I can give.
alicec
Posts: 339
Joined: February 8th, 2016, 12:56 pm
Registered HYSPLIT User: Yes

Re: Confusion with results produced when changing sampling time interval and type

Post by alicec »

Can you attach.

1. The MESSAGE files produced by the different runs. In particular the 15 time averaged as well as the 1 minute snapshots.
2. CONTROL and SETUP.CFG files used.

Thanks.
GarySutlieffUoB
Posts: 10
Joined: March 22nd, 2021, 11:21 am
Registered HYSPLIT User: No

Re: Confusion with results produced when changing sampling time interval and type

Post by GarySutlieffUoB »

My apologies for the late response! I was not notified by email that a reply had been posted, like I am used to, so I was not aware you had replied. Thank you for replying to my original post so swiftly. I don't seem to have SETUP.CFG files in my working directory, but I can supply the CONTROL and MESSAGE files for each configuration. However, I've had to change them to .txt files since the .cfg extension wouldn't allow me to upload them, and also I can only upload the 8 files three at a time, so they will be spread over the next few replies.
Attachments
control-15minutesampling-averaged.txt
CONTROL File for the time-averaged results with a 15-minute sampling interval
(395 Bytes) Downloaded 31 times
control-1minutesampling-snapshot.txt
CONTROL File for the snapshot results with a 1-minute sampling interval
(396 Bytes) Downloaded 76 times
control-1minutesamping-averaged.txt
CONTROL File for the time-averaged results with a 1-minute sampling interval
(394 Bytes) Downloaded 28 times
GarySutlieffUoB
Posts: 10
Joined: March 22nd, 2021, 11:21 am
Registered HYSPLIT User: No

Re: Confusion with results produced when changing sampling time interval and type

Post by GarySutlieffUoB »

Posting another three of the MESSAGE and CONTROL Files here:
Attachments
MESSAGE-1minutesampling-snapshot.txt
MESSAGE File for the snapshot results with a 1-minute sampling time.
(19.22 KiB) Downloaded 30 times
MESSAGE-1minutesampling-averaged.txt
MESSAGE File for the time-averaged results with a 1-minute sampling time.
(19.22 KiB) Downloaded 27 times
control-15minutesampling-snapshot.txt
CONTROL File for the snapshot results with a 15-minute sampling time.
(397 Bytes) Downloaded 27 times
GarySutlieffUoB
Posts: 10
Joined: March 22nd, 2021, 11:21 am
Registered HYSPLIT User: No

Re: Confusion with results produced when changing sampling time interval and type

Post by GarySutlieffUoB »

Attaching the final two CONTROL and MESSAGE files:
Attachments
MESSAGE-15minutesampling-snapshot.txt
MESSAGE File for the snapshot results with a 15-minute sampling time.
(18.38 KiB) Downloaded 28 times
MESSAGE-15minutesampling-averaged.txt
MESSAGE File for the time-averaged results with a 15-minute sampling time.
(18.38 KiB) Downloaded 28 times
alicec
Posts: 339
Joined: February 8th, 2016, 12:56 pm
Registered HYSPLIT User: Yes

Re: Confusion with results produced when changing sampling time interval and type

Post by alicec »

Can you make sure you are not comparing values from the deposition layer? These values will be the same since they are accumulated rather than averaged.

From the message files it looks like the 1 minute time step is being used in all the simulations so that your 15 minute average concentration should be about 15 times smaller than the 1 minute averaged/snapshot concentrations. Note that since minimum time step is a minute there is no difference between a 1 minute average and a 1 minute snapshot. However, you should make sure you are comparing concentrations (you have a deposition layer defined. concentrations would be in levels about the first one).

Other thoughts:
HYSPLIT minimum time step is 1 minute. In short, HYSPLIT is not designed for this high spatial resolution. While, HYSPLIT will let you define as fine a concentration grid as you want, you will start getting results that are not particularly meaningful if you start going below a few km in the horizontal resolution and tens of meters in the vertical.

Your vertical resolution is defined as 1m.

Your horizontal resolution is about 125 m. Any wind speed greater than 2 m/s will result in computational particles skipping grid boxes in the horizontal. Your entire concentration grid domain is only 500m in horizontal and 30m in vertical, so it is not surprising that you are only seeing concentrations in the first minute.

You should evaluate whether you need this fine of spatial resolution. If you do, then you will need an urban scale model or modify HYPSLIT to have smaller time step. You should also consider the temporal and spatial resolution of your input meteorology. For this scale, you might need something like an LES.
GarySutlieffUoB
Posts: 10
Joined: March 22nd, 2021, 11:21 am
Registered HYSPLIT User: No

Re: Confusion with results produced when changing sampling time interval and type

Post by GarySutlieffUoB »

I may have been. The ground level (which I assume is the extent of the deposition layer, I don't see why it wouldn't be?) concentrations do match, and the results above that altitude are about 15x smaller for the 15 minute average compared to the 1 minute, so those two points of confusion make a lot more sense now, thanks.

As for the other thoughts - would expanding the horizontal extent of the sampling area be a good way to verify that this is what's happening, before I look at changing resolutions?

My requirements for resolution might be fixed, since I've been processing HYSPLIT's outputs to create atmospheric profiles of concentration that have a specific structure. Unusually, when I researched different models I couldn't find any indication that HYSPLIT was unsuitable, so I may need to re-evaluate there, but thank you for the warning.

On the topic of meteorology files, I don't know if this is discernible from the MESSAGE or CONTROL files, but I've been using the GDAS 0.25 degree global dataset, which is as I understand it one of the finer resolution data sets available for this kind of modelling. Am I wrong about that or am I missing something else? Lastly, I'm not familiar with LES as an acronym, can you elaborate on that please?
alicec
Posts: 339
Joined: February 8th, 2016, 12:56 pm
Registered HYSPLIT User: Yes

Re: Confusion with results produced when changing sampling time interval and type

Post by alicec »

GFS 0.25 degree meteorology data is one of the finer scales global meteorological datasets available.
Resolution of the meteorological fields for this dataset is about 27km in the horizontal, 50 m in the vertical (near the surface), and 3 h temporal resolution.

There are higher resolution datasets available for more limited area (e.g. CONUS, Hawaii etc).
https://www.ready.noaa.gov/READYmetdata.php

Also if people need a high resolution dataset for a particular area, they may run their own WRF run.
LES is large eddy simulation.

If you are running HYSPLIT with horizontal resolution of 125 m, vertical resolution of 1m and temporal resolution of 1 minute there are a few
things to be aware of.

HYSPLIT minimum time step is 1 minute. At that resolution, computational particles may very well 'skip' over concentration grid boxes. Thus using an average may be problematic. Using snapshots may be ok.

You need to make sure you have enough computational particles. A good rule of thumb is that the concentrations you need to resolve
should be represented by at least 10 computational particles in a grid box per time step. See section 2.4 for a more in depth explanation.
https://www.mdpi.com/2073-4433/11/12/1369/htm

If your meteorological data that you're using has a much coarser spatial and temporal resolution (say the GFS 0.25 degree) then HYSPLIT results will probably be similar to what you would get with a Gaussian plume model (assuming you are using enough computational particles) which assume constant meteorological conditions (wind speed, direction, stability class).
GarySutlieffUoB
Posts: 10
Joined: March 22nd, 2021, 11:21 am
Registered HYSPLIT User: No

Re: Confusion with results produced when changing sampling time interval and type

Post by GarySutlieffUoB »

Thank you very much for all the information. Looking into acquiring higher-resolution meteorology seems like a wise choice, as does switching to snapshots instead of averages.

Having read your paper that you linked, I think I understand conceptually what you mean by computational particles, and that they're linked the minimum simulate-able concentration by the equation you provided, but could you kindly elaborate a bit on what the actual process would be for calculating how many particles are present? I'm not sure I understand how to proceed - do I simply need to take one of my concentration results and try to feed that back through the equation?
GarySutlieffUoB
Posts: 10
Joined: March 22nd, 2021, 11:21 am
Registered HYSPLIT User: No

Re: Confusion with results produced when changing sampling time interval and type

Post by GarySutlieffUoB »

Having thought about this for a few days and also discussed with my supervisor, I have a few more questions I'd like to ask in addition to my previous one about computational particles:

1. Do you have any specific recommendations about which alternative models would be best to use if HYSPLIT turns out to be unsuitable/if we can't find a solution to make HYSPLIT produce the results I require? I've looked at NAME, CALPUFF, and AERMOD so far but my research suggested that each of them had their own reasons not to be suitable so I was left not totally sure as to what the best choice was. I didn't find anything of that kind for HYSPLIT until you first answered my questions.

2. Are there any zero-wind meteorology datasets in the HYSPLIT archive that I can download, or does the archive only contain actual recorded/forecasted datasets? How would one go about producing a zero-wind dataset compatible with HYSPLIT if need be?
Post Reply

Return to “Dispersion Model”