Hello,
I am a research assistant in a social science department working on a project which requires a mapping between emissions sources and their destinations in the context of vehicle exhaust CO, HC, and NOX. Avoiding specifics for the sake of brevity: the end goal is a mapping between source locations and the resulting spatial distribution of pollution damages. I have two questions (among many others, which I might follow up on):
1. Generally, is HYSPLIT an appropriate model to be using for this purpose? I am aware that there are other chemical transport models out there that have been used by other research groups to create similar mappings before (see AP3, EASIUR, InMap, etc.).
2. Where should I look to find appropriate parameter values for vehicle exhaust, especially regarding the deposition setup?
I am happy to provide further details as needed. Thanks in advance!
Andrew
Vehicle emissions dispersion
Re: Vehicle emissions dispersion
Hi, thanks for your questions and consideration of HYSPLIT.
Part of the answer really depends on what length scale you are looking at. If you are trying to consider impacts close to roadway, e.g., adjacent to the roadway and up to, say, 500 m or so away, then HYSPLIT is not going to be able to provide very useful results. The minimum time step of 1 minute means that the minimum resolution is on the order of 200-300 m. Since the concentrations will drop off dramatically as you go away from a roadway, I don't think HYSPLIT would be a good tool.
If the situation was, say, an large road that was 1-10 km or more away from any relevant "receptor" -- i.e., a person being exposed -- then you could potentially use HYSPLIT, subject to the caveats below.
As with any transport and dispersion model, the simulation is only as good as the meteorological data driving the simulation. HYSPLIT is no exception. If you are doing your analysis in an urban area, and are concerned with near-roadway meteorological phenomena, then most if not all of the available met data that you could use to drive HYSPLIT -- or any other model -- will likely not be all that accurate.
My suggestion would be to look at the literature to see what others have done for near-roadway exposure modeling. The impact of vehicles on the nearby atmospheric flow is likely significant, as would general complications of urban meteorology. You might need some sort of computational fluid dynamics approach. I suspect there have been numerous studies that have attempted to quantify this, and I suggest you see what you can find. As noted above, HYSPLIT is not going to be able to give you very accurate information for the regions of highest exposure near the roadway.
Another caveat concerns chemical transformations. CO and some HC have relatively long half-lives, but NOx and some HC do not. You can try to treat some simple chemistry in HYSPLIT, e.g., with a linear decay rate, if you can estimate that. But, for complex, nonlinear chemical simulations you may need a different model. I'd try to consider the time and length scale of the atmospheric transformations for the pollutants you are considering. If they don't transform too much over the time and length scales you are interested in, then this phenomena can be "safely" ignored in your analysis. But, this will not be true for some pollutants over some time and length scales.
Regarding parameter values for vehicle exhaust, e.g. for deposition, there are some suggestions in the HYSPLIT users guide at
https://www.ready.noaa.gov/hysplitusersguide/S314.htm
Hope this helps a little.
Part of the answer really depends on what length scale you are looking at. If you are trying to consider impacts close to roadway, e.g., adjacent to the roadway and up to, say, 500 m or so away, then HYSPLIT is not going to be able to provide very useful results. The minimum time step of 1 minute means that the minimum resolution is on the order of 200-300 m. Since the concentrations will drop off dramatically as you go away from a roadway, I don't think HYSPLIT would be a good tool.
If the situation was, say, an large road that was 1-10 km or more away from any relevant "receptor" -- i.e., a person being exposed -- then you could potentially use HYSPLIT, subject to the caveats below.
As with any transport and dispersion model, the simulation is only as good as the meteorological data driving the simulation. HYSPLIT is no exception. If you are doing your analysis in an urban area, and are concerned with near-roadway meteorological phenomena, then most if not all of the available met data that you could use to drive HYSPLIT -- or any other model -- will likely not be all that accurate.
My suggestion would be to look at the literature to see what others have done for near-roadway exposure modeling. The impact of vehicles on the nearby atmospheric flow is likely significant, as would general complications of urban meteorology. You might need some sort of computational fluid dynamics approach. I suspect there have been numerous studies that have attempted to quantify this, and I suggest you see what you can find. As noted above, HYSPLIT is not going to be able to give you very accurate information for the regions of highest exposure near the roadway.
Another caveat concerns chemical transformations. CO and some HC have relatively long half-lives, but NOx and some HC do not. You can try to treat some simple chemistry in HYSPLIT, e.g., with a linear decay rate, if you can estimate that. But, for complex, nonlinear chemical simulations you may need a different model. I'd try to consider the time and length scale of the atmospheric transformations for the pollutants you are considering. If they don't transform too much over the time and length scales you are interested in, then this phenomena can be "safely" ignored in your analysis. But, this will not be true for some pollutants over some time and length scales.
Regarding parameter values for vehicle exhaust, e.g. for deposition, there are some suggestions in the HYSPLIT users guide at
https://www.ready.noaa.gov/hysplitusersguide/S314.htm
Hope this helps a little.
Re: Vehicle emissions dispersion
Hi Mark,
Thanks for the reply, this is very helpful! To give some additional context, the spatial resolution I am interested in is also somewhat coarse, so I have a little trouble fully calibrating what you mean by accuracy within a specific range. The question I am looking to answer is "on average, if cars emit X pollutants in area A, what proportion of that pollution ends up in areas A, B, C, etc". The spatial resolution of "areas" I am considering is on the order of census block/block group/tract, which tend to be sized on the order of 10^4-10^5, 10^5-10^6, and 10^6+ square meters respectively. Based on your numbers, using HYSPLIT would probably only be correct (not factoring in your other caveats) at the census tract (and maybe census block group?) level (and the overwhelming majority of the distribution would probably end up within the source tract) - is this accurate?
In case it helps, the methodology I was considering to create the mapping I described was, for a given census tract (or other level of spatial aggregation), create a number of test points within it (e.g. along roadways), and use HYSPLIT to calculate the distribution of pollutants at a number of different test times. I would then average those distributions across sample runs, and spatially collapse the averaged distribution to the census tract level. In this sense, I only care that, on average, HYSPLIT will get the proportion of the resulting pollutant spread that lies within each census tract about right.
Your other caveats are appreciated - I'll look into those!
Thanks for the reply, this is very helpful! To give some additional context, the spatial resolution I am interested in is also somewhat coarse, so I have a little trouble fully calibrating what you mean by accuracy within a specific range. The question I am looking to answer is "on average, if cars emit X pollutants in area A, what proportion of that pollution ends up in areas A, B, C, etc". The spatial resolution of "areas" I am considering is on the order of census block/block group/tract, which tend to be sized on the order of 10^4-10^5, 10^5-10^6, and 10^6+ square meters respectively. Based on your numbers, using HYSPLIT would probably only be correct (not factoring in your other caveats) at the census tract (and maybe census block group?) level (and the overwhelming majority of the distribution would probably end up within the source tract) - is this accurate?
In case it helps, the methodology I was considering to create the mapping I described was, for a given census tract (or other level of spatial aggregation), create a number of test points within it (e.g. along roadways), and use HYSPLIT to calculate the distribution of pollutants at a number of different test times. I would then average those distributions across sample runs, and spatially collapse the averaged distribution to the census tract level. In this sense, I only care that, on average, HYSPLIT will get the proportion of the resulting pollutant spread that lies within each census tract about right.
Your other caveats are appreciated - I'll look into those!
Re: Vehicle emissions dispersion
Hi, glad that the earlier response was helpful. Let me try to address your additional questions here. To start, I am considering that 10^4, 10^5, and 10^6 square meters would have characteristic "downwind" lengths of ~100 m, ~300m, and ~1000m, respectively, in thinking about how HYSPLIT might be applied. Also, I am not sure, but I am thinking that you would be interested in estimating concentrations in areas adjacent to the roadway and also within the surrounding "region", e.g., say, out to maybe 10-20 km away from the roadway?
I think it sort of boils down to how accurate one can be -- and needs to be -- in the analysis, recognizing that whatever modeling tool you use will have limitations.
I know I mentioned this earlier, but I urge you to carefully study the literature on near-roadway exposures. I am not familiar with this literature, but I suspect that there is a rich history of published examples, and be careful not to overlook EPA and other agency reports, in addition to peer-reviewed literature. I am imagining that few if any of these studies have used HYSPLIT, but that some will have used other models, e.g., one or more different types of Gaussian Plume models. Other promising methodological approaches may already exist. In particular, if one or more of the studies contains any comparisons with actual measurements, to evaluate the accuracy of the modeling, this will be very useful.
However, you can certainly run HYSPLIT in the manner you suggest. The concentration grid that you define can be as "fine" as you like, but I'd say that a 300m horizontal grid (0.003 degrees) might be about as fine as you should reasonably go with HYSPLIT. However, you could do some numerical experiments with finer or courser grids to examine how reasonable the results look. The vertical spacing could be one layer with a top height of, say, 100 meters. One aspect to address is the number of computational point particles you use in the simulation. The finer the grid, the more computational point particles you need in the simulation. If you see results that look like "popcorn" -- discrete circles of concentration downwind, as opposed to a continuous plume -- the reason is likely that you don't have enough particles in the simulation. You can carry out numerical expts where you increase the number of particles and see if this has any influence on the simulation. If you increase the number and the results change significantly, then you generally don't have enough particles. Releasing 5000 or 10000 per hour may be sufficient but this will have to be tested. Once the results stabilize, then you shouldn't have to go any higher than that.
So, there are a lot of caveats with the use of HYSPLIT -- or any model -- in this application. However, probably a main one will be the wind direction and wind speed that drives the HYSPLIT model. You can use the 3 km HRRR model, and this should be decent, but, wind direction and wind speed in urban areas are not going to be that accurate. And if you are considering areas with tall buildings, where there would be a street canyon effect, then the wind direction and speed in the HRRR model will likely not be at all accurate. So, if a required goal of the project is to determine very accurate concentration levels for downwind areas, where the difference between adjacent areas is crucial, the wind direction errors of 10-15 degrees or more could play a major role. Of course, even if you used a Gaussian Plume model -- or any other model -- uncertainties in wind direction and wind speed (and in the mixing processes near the roadway) will be comparably uncertain.
If there are wind measurements in the simulation domain and you can compare them with the model you are using to drive HYSPLIT, then this can give you a sense of how uncertain things are. Maybe HRRR is very accurate in your domain -- maybe it is not...
Hope this helps a little more.
I think it sort of boils down to how accurate one can be -- and needs to be -- in the analysis, recognizing that whatever modeling tool you use will have limitations.
I know I mentioned this earlier, but I urge you to carefully study the literature on near-roadway exposures. I am not familiar with this literature, but I suspect that there is a rich history of published examples, and be careful not to overlook EPA and other agency reports, in addition to peer-reviewed literature. I am imagining that few if any of these studies have used HYSPLIT, but that some will have used other models, e.g., one or more different types of Gaussian Plume models. Other promising methodological approaches may already exist. In particular, if one or more of the studies contains any comparisons with actual measurements, to evaluate the accuracy of the modeling, this will be very useful.
However, you can certainly run HYSPLIT in the manner you suggest. The concentration grid that you define can be as "fine" as you like, but I'd say that a 300m horizontal grid (0.003 degrees) might be about as fine as you should reasonably go with HYSPLIT. However, you could do some numerical experiments with finer or courser grids to examine how reasonable the results look. The vertical spacing could be one layer with a top height of, say, 100 meters. One aspect to address is the number of computational point particles you use in the simulation. The finer the grid, the more computational point particles you need in the simulation. If you see results that look like "popcorn" -- discrete circles of concentration downwind, as opposed to a continuous plume -- the reason is likely that you don't have enough particles in the simulation. You can carry out numerical expts where you increase the number of particles and see if this has any influence on the simulation. If you increase the number and the results change significantly, then you generally don't have enough particles. Releasing 5000 or 10000 per hour may be sufficient but this will have to be tested. Once the results stabilize, then you shouldn't have to go any higher than that.
So, there are a lot of caveats with the use of HYSPLIT -- or any model -- in this application. However, probably a main one will be the wind direction and wind speed that drives the HYSPLIT model. You can use the 3 km HRRR model, and this should be decent, but, wind direction and wind speed in urban areas are not going to be that accurate. And if you are considering areas with tall buildings, where there would be a street canyon effect, then the wind direction and speed in the HRRR model will likely not be at all accurate. So, if a required goal of the project is to determine very accurate concentration levels for downwind areas, where the difference between adjacent areas is crucial, the wind direction errors of 10-15 degrees or more could play a major role. Of course, even if you used a Gaussian Plume model -- or any other model -- uncertainties in wind direction and wind speed (and in the mixing processes near the roadway) will be comparably uncertain.
If there are wind measurements in the simulation domain and you can compare them with the model you are using to drive HYSPLIT, then this can give you a sense of how uncertain things are. Maybe HRRR is very accurate in your domain -- maybe it is not...
Hope this helps a little more.