What is a trajectory analysis?
Trajectory or backtrajectory analyses use interpolated measured or modeled
meteorological fields to estimate the most likely central path over geographical
areas that provided air to a receptor at a given time. The method essentially
reverses the wind field and moves a parcel of air backward in time.
Backtrajectories are an oversimplification of the atmosphere in that dispersion
is not accounted for and the potential source areas contributing to a receptor
are underestimated for any given trajectory. Two of the most
commonly used atmospheric Lagrangian trajectory models include
HYSPLIT (Draxler and
Hess, 1997) and
FLEXPART (Stohl and Siebert, 2001). The commonly applied HYSPLIT model
uses archived 3-dimensional meteorological fields generated from observations
and short-term forecasts. HYSPLIT has five options for vertical transport; the
default uses the average interpolated vertical velocity. he trajectory analysis
in this project used the HYSPLIT model maintained by NOAA Air Resources
Investigators often run HYSPLIT backtrajectories from multiple heights to
capture the effects of vertical variation of horizontal winds within the mixed
layer depth. The models produce a series of “endpoints” representing longitude,
latitude, and elevation of the parcel at one-hour intervals.
For individual days, plots of the individual trajectories are informative. For
periods of many days or years, or for certain conditions, such as best 20%
visibility days, a graphical method of summarizing the data is best. Residence
time analysis computes the amount of time (e.g. hours) or percent of time the
parcel is in a horizontal grid cell. In the figures presented below, residence
time is shown as percent of total hours in each grid cell. The domain of
interest is divided into areas such as one-degree latitude by one-degree
longitude cells. This data can then be contoured and plotted on a map.
Geographic areas associated with bad and good visibility days can be determined
by running trajectories and plotting residence times separately for these
conditions. Conditional probability maps show the fraction of times a certain
condition is met when a trajectory passes over a grid cell enroute to the
receptor. A conditional probability map for 20% worst visibility days shows the
likelihood of having 20% worst visibility occurring when air passed over each
grid cell (The average would be 20%). Areas with high conditional probability
indicate that when air passes over those grid cells, it is very likely to be
associated with poor visibility at the receptor. It says nothing about the
frequency of airflow from the grid cells - the residence time plots give that
When interpreting the results of backtrajectory analysis, one must keep in mind
the density of emissions along and near the backtrajectory. Because geographic
areas may be high emitters of some haze causing compounds and low for others,
trajectory calculations should be made and plotted for conditions in addition to
high or low reconstructed haze, such as high or low sulfate, nitrate, and
organic carbon. It should also be recognized that the meteorological input
fields typically used represent large-scale flows and cannot accurately
represent local to mesoscale flows such as topographically influenced flow,
nocturnal jets, and seabreeze/landbreeze. It some cases systematic biases may
occur that could lead to invalid conclusions regarding source-receptor
relationships. The Nested Grid Model (NGM) meteorological fields used in some
analysis for the Grand Canyon Visibility Transport Commission was shown by Green
et. al. (2000) to have a systematic bias in wind directions in the southwestern
US that resulted in significant biases regarding source receptor relationships
in this region. Evaluation of the meteorological fields used in the analysis
here is beyond the scope of this scoping study, but is recommended for future
work. We will for the time being assume that the errors are random and that the
regional to large-scale transport patterns are reasonably accurate averaged over
a significant number of trajectories.
How did we compute the trajectories?
For the continental US, the NOAA ARL
Assimilation System (EDAS) meteorological data was used in the model.
For Alaska and Hawaii, the
FNL data set
was used. EDAS assimilates observed data into short-term Eta model
calculations to obtain meteorological fields. The EDAS fields are archived
by the National Oceanic and Atmospheric Administration’s (NOAA) Air Resources
Laboratory (ARL) at 80 km horizontal resolution. ARL archives the FNL
meteorological field at 190 km resolution. In general, higher resolution
fields are desirable in that flow features on smaller scale may be captured.
However, all the meteorological fields mentioned above cannot capture local or
Backtrajectories spanning the years 2000 through 2002 were computed
for each IMPROVE and Class I area. So far we have calculated
backtrajectories for 186 sites in the US. Some of the model parameters are as
||192 hours (8 days) backward in time
|Top of model domain
|Vertical motion option
||used model data
||10, 500 and 1500 meters
HYSPLIT calculated backtrajectories from each site every 3 hours. The
model was run in a batch mode with over a million separate trajectories.
How do we use the model output?
HYSPLIT produced a trajectory for each 3-hour ...more to come
What tools did we use?
HYSPLIT model batch input tool: This is a program that was written in Visual
Basic to generate the HYSPLIT model input files.
HYSPLIT model output combine tool
Trajectory tool to convert endpoints to shapefile
Trajectory gridding tool
Draxler,R.R.; and Hess,G.D. (1997). Description of the Hysplit_4 modeling
system. Report No. NOAA Tech Memo ERL ARL-224, December 1997. Prepared by Air
Resources Laboratory, NOAA, Silver Spring, MD.
Green et al. (2000).
Stohl,A.; and Seibert,P. (2001). The FLEXPART particle dispersion model version
4.0 user guide. June 4, 2001.