Howdy!
I am trying to set up the "correct" parameters in HYSPLIT to retrieve impact maps of areas contributing to measured concentrations at a receptor site.
So far I have
NAM 12 km data
dispersion "back" setting at -12 h
emission set to generic "1" and unit to volume mixing ratio
STILT emulation setting in Advanced
The output looks like what I would expect it to (I have used flux footprint models in the past, and the retrieved map gives an integrated impact value of around 60%, aka 0.6, which appears a bit low but is not out of the range I expected; could be due to rounding?), but I am unsure about the above settings, such as whether 12 hour averages are meaningful, or whether the STILT setting should be combined with some other settings or better just ignored ....
Thanks for advice. - Gunnar
impact map for receptor, back dispersion setup
-
ariel.stein
- Posts: 660
- Joined: November 7th, 2012, 3:14 pm
- Registered HYSPLIT User: Yes
Re: impact map for receptor, back dispersion setup
The correct settings for your STILT emulation runs will depend on your application.
Setting the back dispersion to a short time will reflect only the nearby sources. In general, setting the duration of the run to 72 hours will give you a rough idea of the sinoptic scale influences to your receptor. You will have to increase the duration if dealing with continental scale source attribution.
On the other hand, the average time for footprints are set in the concentration grid setting. Do not confuse that parameter with the run duration as described above.
More info about the STILT emulation:
The STILT model is a variation of HYSPLIT developed by Lin et al.
(2003 - JGR, VOL. 108, NO. D16, 4493, doi:10.1029/2002JD003161) that can
be used to estimate upwind surface fluxes from atmospheric measurements.
Although HYSPLIT does not contain the mass conservation adjustments and coupled
vertical-mixing horizontal-transport features of STILT, setting this flag causes
the HYSPLIT concentration output field units to be similar to the STILT
configuration. Two changes are introduced; the mass summation is divided by air
density resulting in a mixing ratio output field (ICHEM=6) and the lowest
concentration summation layer (concentration layer top depth) is permitted to
vary with the mixed layer depth (ICHEM=9), so that only particles below 50% of the
mixed layer depth at each time step are summed to the lowest layer. The ICHEM=8
switch turns on both density and varying layer depth. A text file of particle
position information (PARTICLE.DAT) at each time step will also be created
unless the namelist parameter OUTDT defining the output
interval (min) is changed.
Some additional references:
http://journals.ametsoc.org/doi/pdf/10. ... -13-0125.1
Setting the back dispersion to a short time will reflect only the nearby sources. In general, setting the duration of the run to 72 hours will give you a rough idea of the sinoptic scale influences to your receptor. You will have to increase the duration if dealing with continental scale source attribution.
On the other hand, the average time for footprints are set in the concentration grid setting. Do not confuse that parameter with the run duration as described above.
More info about the STILT emulation:
The STILT model is a variation of HYSPLIT developed by Lin et al.
(2003 - JGR, VOL. 108, NO. D16, 4493, doi:10.1029/2002JD003161) that can
be used to estimate upwind surface fluxes from atmospheric measurements.
Although HYSPLIT does not contain the mass conservation adjustments and coupled
vertical-mixing horizontal-transport features of STILT, setting this flag causes
the HYSPLIT concentration output field units to be similar to the STILT
configuration. Two changes are introduced; the mass summation is divided by air
density resulting in a mixing ratio output field (ICHEM=6) and the lowest
concentration summation layer (concentration layer top depth) is permitted to
vary with the mixed layer depth (ICHEM=9), so that only particles below 50% of the
mixed layer depth at each time step are summed to the lowest layer. The ICHEM=8
switch turns on both density and varying layer depth. A text file of particle
position information (PARTICLE.DAT) at each time step will also be created
unless the namelist parameter OUTDT defining the output
interval (min) is changed.
Some additional references:
http://journals.ametsoc.org/doi/pdf/10. ... -13-0125.1
Re: impact map for receptor, back dispersion setup
ok, I increased the hours
the grid setting I had left at default because the description is a bit vague, mostly applies to forward dispersion; it only says that "Backward" calculations require that the stop time should come before the start time." But default stop time is a generic number, so that is confusing to me
When I instead set the interval (default was 00 12 00) to 01 01 00 I get maps for each hour but it appeared to re-assign the receptor each hour to the backward max location of the air parcel.
so I tried setting -1 04 00 for the interval, but it still moves the 2nd, 3rd , etc. 4-hour map out to the max location of the air parcel at that time. I presume this is intended since it reflects the equivalent to a forward model where the first number is the duration of the emissions, correct?
So if I want only the "short-term" influence, say the last four hours, which have had significant influences though from far away as well, what is the recommended setting? 00 04 00 ? Or should a stop time be given explicitly?
And how can prevent small numbers of the larger field to be dropped? Aka I get a lot of zeros in the output, but those are likely not zero but simply below a threshold, right?
Thanks,
Gunnar
the grid setting I had left at default because the description is a bit vague, mostly applies to forward dispersion; it only says that "Backward" calculations require that the stop time should come before the start time." But default stop time is a generic number, so that is confusing to me
When I instead set the interval (default was 00 12 00) to 01 01 00 I get maps for each hour but it appeared to re-assign the receptor each hour to the backward max location of the air parcel.
so I tried setting -1 04 00 for the interval, but it still moves the 2nd, 3rd , etc. 4-hour map out to the max location of the air parcel at that time. I presume this is intended since it reflects the equivalent to a forward model where the first number is the duration of the emissions, correct?
So if I want only the "short-term" influence, say the last four hours, which have had significant influences though from far away as well, what is the recommended setting? 00 04 00 ? Or should a stop time be given explicitly?
And how can prevent small numbers of the larger field to be dropped? Aka I get a lot of zeros in the output, but those are likely not zero but simply below a threshold, right?
Thanks,
Gunnar
Re: impact map for receptor, back dispersion setup
addendum to my last reply
I realized the emission settings were still at 1/hr and 1 hr ... the help file does not address what these should be, but I wonder if emission duration should be the same as total run hours for this application ...? If so, emissions per hour would need to be set to 1/(emission duration) to maintain fractional equivalents ... recommendation?
I realized the emission settings were still at 1/hr and 1 hr ... the help file does not address what these should be, but I wonder if emission duration should be the same as total run hours for this application ...? If so, emissions per hour would need to be set to 1/(emission duration) to maintain fractional equivalents ... recommendation?
-
ariel.stein
- Posts: 660
- Joined: November 7th, 2012, 3:14 pm
- Registered HYSPLIT User: Yes
Re: impact map for receptor, back dispersion setup
Here is the description of the way the sampling interval can be set in HYSPLIT's CONTROL file (lines 23 to 25:
Sampling start time: year month day hour minute
Default: [simulation start]
Each concentration grid may have a different starting, stopping, and output averaging time. Zero entry will result in setting the default values. "Backward" calculations require that the stop time should come before the start time.
Sampling stop time: year month day hour minute
Default: 12 31 24 60
After this time no more concentration records are written. Early termination on a high resolution grid (after the plume has moved away from the source) is an effective way of speeding up the computation for high resolution output near the source because once turned-off that particular grid resolution is no longer used for time-step computations.
Sampling interval: type hour minute
Default: 0 24 0
Each grid may have its own sampling or averaging interval. The interval can be of three different types: averaging (type=0), snapshot (type=1), or maximum (type=2). Averaging will produce output averaged over the specified interval. For instance, you may want to define a concentration grid that produces 24-hour average air concentrations for the duration of the simulation, which in the case of a 2-day simulation will result in 2 output maps, one for each day. Each defined grid can have a different output type and interval. Snapshot (or now) will give the instantaneous output at the output interval, and maximum will save the maximum concentration at each grid point over the duration of the output interval. Therefore, when a maximum concentration grid is defined, it is also required to define an identical snapshot or average grid over which the maximum will be computed. There is also the special case when the type value is less than zero. In that case the value represents the averaging time in hours and the output interval time represents the interval at which the average concentration is output. For instance, a setting of {-1 6 0} would output a one-hour average concentration every six hours.
Regarding the emission rate, the model uses a very small value for the emission rate (it overrides the value that you set in the CONTROL file) and a time step of 1 minute (it overrides the value set in SETUP.CFG). The footprint accumulation in the concentration grid assumes a constant mass value on each particle.
The equation that is used to calculate the footprint can be found in http://journals.ametsoc.org/doi/suppl/1 ... 0110.2.pdf Equation ES32.
Also, see Lin, J. C., C. Gerbig, S. C. Wofsy, A. E. Andrews, B. C. Daube, K. J. Davis, and C. A. Grainger, 2003: A near-field tool for simulating the upstream inf luence of atmospheric observations: The Stochastic Time-Inverted Lagrangian Transport (STILT) model. J. Geophys. Res., 108, 4493,
doi:10.1029/2002JD003161 for more details.
Sampling start time: year month day hour minute
Default: [simulation start]
Each concentration grid may have a different starting, stopping, and output averaging time. Zero entry will result in setting the default values. "Backward" calculations require that the stop time should come before the start time.
Sampling stop time: year month day hour minute
Default: 12 31 24 60
After this time no more concentration records are written. Early termination on a high resolution grid (after the plume has moved away from the source) is an effective way of speeding up the computation for high resolution output near the source because once turned-off that particular grid resolution is no longer used for time-step computations.
Sampling interval: type hour minute
Default: 0 24 0
Each grid may have its own sampling or averaging interval. The interval can be of three different types: averaging (type=0), snapshot (type=1), or maximum (type=2). Averaging will produce output averaged over the specified interval. For instance, you may want to define a concentration grid that produces 24-hour average air concentrations for the duration of the simulation, which in the case of a 2-day simulation will result in 2 output maps, one for each day. Each defined grid can have a different output type and interval. Snapshot (or now) will give the instantaneous output at the output interval, and maximum will save the maximum concentration at each grid point over the duration of the output interval. Therefore, when a maximum concentration grid is defined, it is also required to define an identical snapshot or average grid over which the maximum will be computed. There is also the special case when the type value is less than zero. In that case the value represents the averaging time in hours and the output interval time represents the interval at which the average concentration is output. For instance, a setting of {-1 6 0} would output a one-hour average concentration every six hours.
Regarding the emission rate, the model uses a very small value for the emission rate (it overrides the value that you set in the CONTROL file) and a time step of 1 minute (it overrides the value set in SETUP.CFG). The footprint accumulation in the concentration grid assumes a constant mass value on each particle.
The equation that is used to calculate the footprint can be found in http://journals.ametsoc.org/doi/suppl/1 ... 0110.2.pdf Equation ES32.
Also, see Lin, J. C., C. Gerbig, S. C. Wofsy, A. E. Andrews, B. C. Daube, K. J. Davis, and C. A. Grainger, 2003: A near-field tool for simulating the upstream inf luence of atmospheric observations: The Stochastic Time-Inverted Lagrangian Transport (STILT) model. J. Geophys. Res., 108, 4493,
doi:10.1029/2002JD003161 for more details.
Re: impact map for receptor, back dispersion setup
ok, I went about it a bit more systematically
I think the help texts are a bit confusing. The relevant part in your response is that the input values in the setup file are overwritten, aka the emission rate input is irrelevant. Equation ES32 is straightforward, but that was never my question.
The confusion, I my mind came from the help description of "grid averaging". Such "average" may make sense for the forward dispersion in terms of concentrations, but for backward, "averaging" meant something different to me than in the model. What actually happens is that particle numbers are "integrated" over time, not averaged. When I ran the model for 2-h, 4-h, and 6-h grid "averages", the first two 2-h values actually add to the first 4-h values, the first three 2-h values to the first 6-h value etc. ... that is not averaging in my view.
Only the plotting routine uses the wording "integrated", which makes more sense to me.
So, basically, the answer to my original question, I think should be that the "average time for footprints" in the grid settings should be set similar to the run time for numbers to make sense. Otherwise, short "averaging" times give way too low integral impacts (I find <1 % in some cases), especially in the near field. Since the longer times are integrated numbers, not averaged numbers of total particle number fractions, the longer the "average time" the closer the footprint integral gets to 100% impact area ... however, I also got values exceeding 100% when going over a certain number of hours, aka mass conservation was violated.
Lin et al. found that to be related to the met input data, so I wonder whether others here found a particular input data set working better than others.
I think the help texts are a bit confusing. The relevant part in your response is that the input values in the setup file are overwritten, aka the emission rate input is irrelevant. Equation ES32 is straightforward, but that was never my question.
The confusion, I my mind came from the help description of "grid averaging". Such "average" may make sense for the forward dispersion in terms of concentrations, but for backward, "averaging" meant something different to me than in the model. What actually happens is that particle numbers are "integrated" over time, not averaged. When I ran the model for 2-h, 4-h, and 6-h grid "averages", the first two 2-h values actually add to the first 4-h values, the first three 2-h values to the first 6-h value etc. ... that is not averaging in my view.
Only the plotting routine uses the wording "integrated", which makes more sense to me.
So, basically, the answer to my original question, I think should be that the "average time for footprints" in the grid settings should be set similar to the run time for numbers to make sense. Otherwise, short "averaging" times give way too low integral impacts (I find <1 % in some cases), especially in the near field. Since the longer times are integrated numbers, not averaged numbers of total particle number fractions, the longer the "average time" the closer the footprint integral gets to 100% impact area ... however, I also got values exceeding 100% when going over a certain number of hours, aka mass conservation was violated.
Lin et al. found that to be related to the met input data, so I wonder whether others here found a particular input data set working better than others.
-
ariel.stein
- Posts: 660
- Joined: November 7th, 2012, 3:14 pm
- Registered HYSPLIT User: Yes
Re: impact map for receptor, back dispersion setup
Regarding the use of different meteorological datasets to drive the inversion see the following from http://journals.ametsoc.org/doi/pdf/10. ... -13-0125.1
WRF can also produce additional outputs implemented specifically for the benefit of transport simulations (Nehrkorn et al. 2010). These include 1) time-averaged, mass-coupled horizontal and vertical velocities (using the WRF terrain-following vertical coordinate h) to improve mass conservation and the representation of wind variability and 2) time-averaged mass fluxes from the WRF moist convective parameterization scheme.
WRF can also produce additional outputs implemented specifically for the benefit of transport simulations (Nehrkorn et al. 2010). These include 1) time-averaged, mass-coupled horizontal and vertical velocities (using the WRF terrain-following vertical coordinate h) to improve mass conservation and the representation of wind variability and 2) time-averaged mass fluxes from the WRF moist convective parameterization scheme.
Re: impact map for receptor, back dispersion setup
Hello,
Following up on this thread, I have three questions about the STILT Emulation mode runs in Hysplit:
1) The output in the concentration display and in the cdump file are typically "per m3", but in STILT emulation mode they are expressed in "footprint" units from my understanding. However when i do backward simulations in STILT emulation mode (set in advanced settings) the display graphics still say "per m3" on them. Is this a mistake in the graphics? Or is it an indication that I have I not set something correctly?
2) From my understanding, the footprint units that HYSPLIT uses in STILT mode would be those of Equation ES32 in the BAMS supplement. Is this correct? If so, what are the exact units HYSPLIT uses to compute 'f' in ES32? If all is done in mks units, then f should be "1/(kmole/m2/time) where "time" is the units used to calculate the particle duration "delta_t" in ES32. Please confirm, and tell me also what the units of "time" are (hour? sec?) if this is correct understanding.
3) I am doing back calculations to assess footprints for an hourly average concentration at a receptor. I should then release particles over the hour corresponding to the time of the measurement, correct? The sampling time in the grid setting would then correspond to the time discretization of the footprints used to explain the hourly average concentration. However I wouldn't want to release particles continuously, but rather only over the hour of the measurement since this corresponds to the mass measured during the hour. Is this correct?
Please elaborate as necessary ... I want to make sure I get this correct. Thanks for your help. Frank
Following up on this thread, I have three questions about the STILT Emulation mode runs in Hysplit:
1) The output in the concentration display and in the cdump file are typically "per m3", but in STILT emulation mode they are expressed in "footprint" units from my understanding. However when i do backward simulations in STILT emulation mode (set in advanced settings) the display graphics still say "per m3" on them. Is this a mistake in the graphics? Or is it an indication that I have I not set something correctly?
2) From my understanding, the footprint units that HYSPLIT uses in STILT mode would be those of Equation ES32 in the BAMS supplement. Is this correct? If so, what are the exact units HYSPLIT uses to compute 'f' in ES32? If all is done in mks units, then f should be "1/(kmole/m2/time) where "time" is the units used to calculate the particle duration "delta_t" in ES32. Please confirm, and tell me also what the units of "time" are (hour? sec?) if this is correct understanding.
3) I am doing back calculations to assess footprints for an hourly average concentration at a receptor. I should then release particles over the hour corresponding to the time of the measurement, correct? The sampling time in the grid setting would then correspond to the time discretization of the footprints used to explain the hourly average concentration. However I wouldn't want to release particles continuously, but rather only over the hour of the measurement since this corresponds to the mass measured during the hour. Is this correct?
Please elaborate as necessary ... I want to make sure I get this correct. Thanks for your help. Frank
-
ariel.stein
- Posts: 660
- Joined: November 7th, 2012, 3:14 pm
- Registered HYSPLIT User: Yes
Re: impact map for receptor, back dispersion setup
Below are the answers to your questions:
1) The output in the concentration display and in the cdump file are typically "per m3", but in STILT emulation mode they are expressed in "footprint" units from my understanding. However when i do backward simulations in STILT emulation mode (set in advanced settings) the display graphics still say "per m3" on them. Is this a mistake in the graphics? Or is it an indication that I have I not set something correctly?
This seems to be a mistake in the graphics. If you set ICHEM=8 in the SETUP.CFG file then you should see a message in the screen when you run the model that tells you that you are running the in STILT mode.
2) From my understanding, the footprint units that HYSPLIT uses in STILT mode would be those of Equation ES32 in the BAMS supplement. Is this correct? If so, what are the exact units HYSPLIT uses to compute 'f' in ES32? If all is done in mks units, then f should be "1/(kmole/m2/time) where "time" is the units used to calculate the particle duration "delta_t" in ES32. Please confirm, and tell me also what the units of "time" are (hour? sec?) if this is correct understanding.
Yes, HYSPLIT uses that equation for calculating the footprint. However, the units are [ppm][m2][sec]/[micro-moles].
3) I am doing back calculations to assess footprints for an hourly average concentration at a receptor. I should then release particles over the hour corresponding to the time of the measurement, correct? The sampling time in the grid setting would then correspond to the time discretization of the footprints used to explain the hourly average concentration. However I wouldn't want to release particles continuously, but rather only over the hour of the measurement since this corresponds to the mass measured during the hour. Is this correct?
Yes, you should release your particles for the hour corresponding to the time of the measurement. You do not want to release the particles continously, only for the particular hour corresponding to you measurement.
1) The output in the concentration display and in the cdump file are typically "per m3", but in STILT emulation mode they are expressed in "footprint" units from my understanding. However when i do backward simulations in STILT emulation mode (set in advanced settings) the display graphics still say "per m3" on them. Is this a mistake in the graphics? Or is it an indication that I have I not set something correctly?
This seems to be a mistake in the graphics. If you set ICHEM=8 in the SETUP.CFG file then you should see a message in the screen when you run the model that tells you that you are running the in STILT mode.
2) From my understanding, the footprint units that HYSPLIT uses in STILT mode would be those of Equation ES32 in the BAMS supplement. Is this correct? If so, what are the exact units HYSPLIT uses to compute 'f' in ES32? If all is done in mks units, then f should be "1/(kmole/m2/time) where "time" is the units used to calculate the particle duration "delta_t" in ES32. Please confirm, and tell me also what the units of "time" are (hour? sec?) if this is correct understanding.
Yes, HYSPLIT uses that equation for calculating the footprint. However, the units are [ppm][m2][sec]/[micro-moles].
3) I am doing back calculations to assess footprints for an hourly average concentration at a receptor. I should then release particles over the hour corresponding to the time of the measurement, correct? The sampling time in the grid setting would then correspond to the time discretization of the footprints used to explain the hourly average concentration. However I wouldn't want to release particles continuously, but rather only over the hour of the measurement since this corresponds to the mass measured during the hour. Is this correct?
Yes, you should release your particles for the hour corresponding to the time of the measurement. You do not want to release the particles continously, only for the particular hour corresponding to you measurement.
Re: impact map for receptor, back dispersion setup
Great, thanks ...