1. Introduction
Meteorological data are a necessary input for most atmospheric transport and dispersion models. Instead of using observations directly, it is common to rely upon meteorological model results to provide variables such as wind, temperature, and precipitation for plume transport and dispersion calculations because the meteorological model interpolates these variables in space and time consistent with the atmospheric equations of motion (Bowman et al. 2013). The meteorological model output is typically archived hourly or for regional and global models at even less frequent time intervals (Kalnay et al. 1996; Janjic 2003; Janjic et al. 2005; Kanamitsu 1989). Although the meteorological fields vary continuously in time, typically they are output as instantaneous values at specific time intervals. Alternative approaches using time-averaged output fields (Brioude et al. 2012) are not yet in common use as a result of the additional computational and memory cost in computing time averages during the meteorological simulations. Most dispersion models will either read these data fields directly or, through a preprocessing step, extract relevant meteorological variables as well as convert or diagnose other parameters required for the dispersion model. In offline dispersion models, the hourly meteorological data are interpolated, usually linearly, to the integration time step of the dispersion simulation, which may be on the order of minutes or even seconds.
As the demand for a more rapid response to local-scale emission incidents increases, the higher model spatial resolution required to resolve important local flow features also requires a concomitant increase in the temporal resolution of the meteorological data needed for dispersion modeling. However, the temporal interpolation of hourly meteorological data will introduce aliasing errors (flow features move multiple grid points between the output time intervals) and such interpolation does not provide an accurate representation of the state of the atmosphere on the time scales required for high-resolution dispersion applications. Alternatively, it is possible to output the meteorological fields more frequently, however at the cost of greater data storage requirements. The inline approach to dispersion modeling provides for meteorological data at a finer temporal resolution and avoids the data storage cost, but it does require running the complete meteorological prediction with each dispersion simulation. However, a one-way nesting-down approach can be used so that only the finest-resolution calculation is repeated if multiple dispersion simulations are required. The greatest strength of the inline dispersion modeling is not just with the improved fidelity of the representation, but through the potential of the interaction between the pollutant being modeled and the different development of meteorological features (Grell et al. 2005) that may be sensitive to the pollutant concentrations. Here, we focus on the blending of a dispersion model and a meteorological forecast model for a more integrated approach to pollutant simulations and we test these simulations using a nonreactive pollutant. The integration of a Lagrangian dispersion model into the Weather Research and Forecasting (WRF) Model is fundamentally different than just computing pollutant dispersion directly in WRF because the near-source Lagrangian transport and dispersion is not affected by artificial diffusion associated with Eulerian solutions.
The Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT; Draxler and Hess 1997) had originally been designed to run offline utilizing meteorological data such as are provided by the Advanced Research WRF (WRF-ARW; Skamarock et al. 2008). Such a modeling approach was applied to different kinds of studies like Chen et al. (2012), Srinivas et al. (2012), and Hegarty et al. (2013). A new model framework has been developed such that the revised HYSPLIT model is coupled inline with WRF-ARW to take advantage of the higher temporal frequency of the meteorological variables produced by the meteorological model, as well as using WRF-ARW’s vertical coordinate scheme. The result is a more consistent depiction of the state of the atmosphere available to the dispersion model through the elimination of the need to interpolate the meteorological fields temporally and vertically. In this study, the Cross-Appalachian Tracer Experiment (CAPTEX; Ferber et al. 1986) and Atmospheric Studies in Complex Terrain (ASCOT; Gudiksen et al. 1984) are simulated using both the inline and offline versions of HYSPLIT. The inline results are compared with the offline outputs and evaluated against measurements obtained from the tracer experiments. Although our focus for the coupling is on the use of a dispersion model that can be applied to fine-resolution plume simulations, where the size of the plume may be much smaller than the grid spacing of the meteorological data, we also need to test the inline approach at larger scales, hence the application of the model to both the ASCOT and CAPTEX experiments. The coupling of dispersion and meteorological models is not unique from the Eulerian perspective (Grell et al. 2005; Zhang 2008), where the pollutant species is carried as another diagnostic variable within the meteorological model prediction. In these situations, the minimum plume size is limited to the grid resolution of the meteorological prediction.
An overview of the models and the inline coupling technique is described in the next section. The model configuration and the experiment designs are described in section 3. The results and discussions comparing the inline and offline results with observations for both CAPTEX and ASCOT are presented section 4. Section 5 summarizes the results and discusses the future research direction.
2. Model description
a. Meteorological model: WRF-ARW
The WRF-ARW (Skamarock et al. 2008) is a widely used mesoscale meteorological model applied to weather prediction and atmospheric research and used for various offline simulations such as chemical-transport, air quality, and plume dispersion modeling (Wong et al. 2012; Hegarty et al. 2013). The model is built in a fully compressible Euler nonhydrostatic atmosphere and the governing set of equations is cast in conservative (flux) form for conserved variables; nonconserved variables such as pressure and temperature are diagnosed from the prognostic equations. The Arakawa C grid is used as the horizontal coordinate system while the vertical grid is defined on time-dependent terrain-following hydrostatic pressure coordinates.
WRF-ARW has been applied to diverse research topics in a broad range of spatial scales from thousands of kilometers down to scales of a few meters, including global scale, the mesoscale, and even the microscale (Skamarock et al. 2008). Similar to other numerical weather prediction models, in WRF-ARW the vertical mixing for the planetary boundary layer (PBL) is parameterized using the surface fluxes estimated from a surface layer scheme and a land surface model. The turbulent kinetic energy (TKE) can be generated as a prognostic variable if a high-order closure scheme (i.e., TKE-based scheme) is selected to parameterize the vertical mixing. At scales of 10 m or less, WRF-ARW is also capable of resolving turbulent eddies by its governing equations with surface-layer and subgrid turbulence schemes for large-eddy simulations.
b. Dispersion model: HYSPLIT
The HYSPLIT (Draxler and Hess 1997) model is designed to compute both simple air parcel trajectories and complex dispersion and deposition simulations. It has been used to identify the source–receptor relationship of air pollutants using trajectory analysis and is also run for plume dispersion predictions for a variety of events such as nuclear incidents, volcanic eruptions, wildfire smoke transport, and dust storm episodes (http://ready.arl.noaa.gov/index.php). HYSPLIT is a Lagrangian model, which means that the dispersion calculation follows the transport vector and only the meteorological fields about the computational point are required. The horizontal coordinate and map projections used by HYSPLIT are identical to the meteorological input. For the vertical grid, the meteorological profiles are linearly interpolated onto an internal model terrain-following coordinate. In a particle dispersion simulation, particles released from a source are advected following the mean wind field and dispersed according to a random component caused by the atmospheric turbulence. For each meteorological input data source, a customized preprocessing program is required to convert the meteorological data fields from the meteorological model formats [Gridded Binary (GRIB1 and GRIB2), netCDF, MM5, etc.] into a common format required for running HYSPLIT. Typically, 1- or 3-hourly meteorological data are the most commonly provided for the transport and dispersion calculations, and then they are temporally linearly interpolated between the input data times to the integration time required by HYSPLIT.
c. WRF–HYSPLIT inline coupling
The HYSPLIT model (version 4) is coupled into WRF-ARW (version 3.5.1) by modifying the process of getting the meteorological input for the dispersion computation and its vertical coordinate system. The inline coupling occurs at the time integration level of WRF-ARW where HYSPLIT obtains the required meteorological fields (Table 1) directly from the meteorological model and then advances the dispersion computation to the current time of WRF-ARW’s clock. This means the dispersion model is called as a subroutine like all the other physical packages within the computational loop of WRF-ARW. This becomes the frequency at which updated meteorological data are supplied to HYSPLIT in time intervals of minutes or even seconds depending on the user-defined time step in WRF-ARW. The use of the high-temporal-frequency meteorological data for running inline HYSPLIT avoids dealing with large volumes of disk output that may slow down the WRF simulation and the subsequent offline HYSPLIT simulation that would need to read all those data.
List of WRF-ARW internal variables used in the coupled HYSPLIT simulations.
To use different meteorological data sources for input for the same dispersion calculation, standard offline HYSPLIT manages its own internal vertical coordinate system. It is a terrain-following system (σ) expressing heights relative to mean sea level and the top of the dispersion model with a quadratic relationship between height and model level. The constants in the quadratic relationship can be altered to get the same or better vertical resolution than the meteorological input. The input data for HYSPLIT need to be interpolated vertically from the data coordinates to match HYSPLIT’s internal layers. For the coupled HYSPLIT, the WRF-ARW vertical coordinate, a terrain-following hydrostatic pressure coordinate (η), is used (Fig. 1). The layer height that is used to locate particle positions is expressed as the geopotential height (divided by gravitational constant) and is predefined by WRF-ARW. The transport and dispersion of particles are given in terms of the η layers configured for the meteorological simulations.
Schematic of the vertical coordinate used in the inline coupled HYPSLIT model. ZNW represents the eta values on full (vertical velocity w) levels in WRF-ARW, and ZNU is for values on half (mass) levels.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
Schematic of the vertical coordinate used in the inline coupled HYPSLIT model. ZNW represents the eta values on full (vertical velocity w) levels in WRF-ARW, and ZNU is for values on half (mass) levels.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
Schematic of the vertical coordinate used in the inline coupled HYPSLIT model. ZNW represents the eta values on full (vertical velocity w) levels in WRF-ARW, and ZNU is for values on half (mass) levels.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
The computational cost for running HYSPLIT within the WRF-ARW framework depends on the number of Lagrangian particles that are used for the simulations. For instance, running HYSPLIT inline for CAPTEX using 50 000 particles required an additional ~7% of computational time when compared with the stand-alone WRF run. For ASCOT, which used 250 000 particles, the inline run took approximately 25% more time than running the WRF for meteorology only (Table 4). The data storage requirement for the WRF meteorological files to run HYSPLIT offline is 6.3 gigabytes (GB; hourly output) and 21 GB (5-min output) for each episode of CAPTEX and ASCOT, respectively. For the inline approach, it is not necessary to save the WRF output files since the inline HYSPLIT uses the meteorological variables during the WRF computation.
3. Evaluation protocol
a. Experimental data: CAPTEX
CAPTEX (Ferber et al. 1986) was conducted from mid-September through the end of October 1983 using an inert perfluorocarbon (PMCH; C7F14) tracer to simulate the long-range transport and diffusion of pollutants. There were a total of six 3-h releases, four episodes (releases 1–4) from Dayton, Ohio (DAY), and two episodes (releases 5 and 7) from Sudbury, Ontario, Canada (SUD) (see Table 2). Measurements were taken at a ground-level air sampling network (~80 sites; dark gray dots in Fig. 2) distributed 300–800 km from the source, collecting 3- and 6-h-average concentrations, in eastern Ohio, Pennsylvania, New Jersey, New York, Vermont, New Hampshire, and southeastern Canada. One additional short 30-min tracer release from Dayton (release 6) was not evident in the sampling data, so it is not included in this study. This experiment has been widely used in evaluating dispersion models, and studies were published by Stohl et al. (1998), Peltier et al. (2010), Lei et al. (2012), and Hegarty et al. (2013).
CAPTEX tracer releases and simulation period of each release. Note that a single 3-h tracer release started at the HYSPLIT start time.
(a) The CAPTEX study showing the two nested domains. Outer box is 27-km grid spacing for the downscaling to the inner box (9 km). (b) The inner domain with the large black dots showing the DAY and SUD release locations. Dark gray dots are the tracer sampling stations.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
(a) The CAPTEX study showing the two nested domains. Outer box is 27-km grid spacing for the downscaling to the inner box (9 km). (b) The inner domain with the large black dots showing the DAY and SUD release locations. Dark gray dots are the tracer sampling stations.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
(a) The CAPTEX study showing the two nested domains. Outer box is 27-km grid spacing for the downscaling to the inner box (9 km). (b) The inner domain with the large black dots showing the DAY and SUD release locations. Dark gray dots are the tracer sampling stations.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
Similar synoptic conditions were observed for other releases at the DAY site that took place in the afternoon when tracers were well mixed throughout the boundary layer. The experiment area was behind a high pressure system centered to the south of the release such that the associated flow featured moderate southwesterly winds for releases 1, 3, and 4. During release 2, a high pressure system centered over the northeastern United States featured clear skies and southerly winds advecting the tracer to Canada at the early stages of the episode (Fig. 3a). The synoptic flow shifted to southwesterly, bringing the plume back to New York and Pennsylvania as a weak cold front from the Midwest pushed the high pressure toward the southeast. The other two releases (5 and 7) from the SUD site occurred after midnight behind cold fronts such that the northwesterly winds would carry the tracers throughout the sampling network. Figure 3b shows the synoptic weather chart at 850 hPa for CAPTEX release 5.
The 850-hPa temperature (°C; shaded), geopotential height (m; blue lines), and wind barbs [kt (1 kt = 0.514 m s−1)] at (a) 0000 UTC 25 Sep and (b) 0000 UTC 25 Oct 1983.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
The 850-hPa temperature (°C; shaded), geopotential height (m; blue lines), and wind barbs [kt (1 kt = 0.514 m s−1)] at (a) 0000 UTC 25 Sep and (b) 0000 UTC 25 Oct 1983.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
The 850-hPa temperature (°C; shaded), geopotential height (m; blue lines), and wind barbs [kt (1 kt = 0.514 m s−1)] at (a) 0000 UTC 25 Sep and (b) 0000 UTC 25 Oct 1983.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
b. Experimental data: ASCOT
ASCOT (Gudiksen et al. 1984) was a series of field tracer experiments designed to study transport and dispersion of pollutant associated with nocturnal drainage flows. This study focuses on one of these experiments conducted in mid-September of 1980, which included five releases (see Table 3), each on a different night. The perfluorocarbon tracer was released for an hour duration at the Anderson Creek valley in northern California and more than 50 sampling sites were placed in an area 10 km × 10 km away from the release site to the southeast (Fig. 4). Collection of samples took place over 10-, 15-, 20-, 60-, 120-, and 180-min durations at different samplers and lasted for about 7 h from the start of the tracer release.
ASCOT tracer releases and simulation period of each release. Note that an hour tracer release started at the HYSPLIT start time at Anderson Creek (38.785°N, 122.725°W).
(a) The ASCOT study consisting of five nested domains. Outer boxes show 27-, 9-, 3-, and 1-km grid spacings for the downscaling to the most inner box (333 m). (b) The inner domain with the large black dot representing the release location at Anderson Creek (38.785°N, 122.725°W); the dark gray dots surrounding the black dot are tracer sampling stations; background colors represents terrain height (m).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
(a) The ASCOT study consisting of five nested domains. Outer boxes show 27-, 9-, 3-, and 1-km grid spacings for the downscaling to the most inner box (333 m). (b) The inner domain with the large black dot representing the release location at Anderson Creek (38.785°N, 122.725°W); the dark gray dots surrounding the black dot are tracer sampling stations; background colors represents terrain height (m).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
(a) The ASCOT study consisting of five nested domains. Outer boxes show 27-, 9-, 3-, and 1-km grid spacings for the downscaling to the most inner box (333 m). (b) The inner domain with the large black dot representing the release location at Anderson Creek (38.785°N, 122.725°W); the dark gray dots surrounding the black dot are tracer sampling stations; background colors represents terrain height (m).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
The Anderson Creek valley has the geographical features for drainage flow development when radiation cooling from the surface happens on clear nights with weak ambient winds. Orgill and Schreck (1985) summarized the conditions for drainage winds during ASCOT by analyzing the tethered-balloon and tower wind and temperature data taken during each experimental night. According to their analysis, three of five release nights (1, 4, and 5) were classified as favorable for the development of drainage flow. Release 2 was on a fair or marginal drainage night while on the night of release 3, the drainage flow was poorly developed because of the strong westerly winds accompanied by low-level cloudiness and high humidity over the experiment area. Figure 5 shows the 700-hPa weather chart on the day of ASCOT 2 and 4.
As in Fig. 3, but for 700 hPa at 0000 UTC (a) 16 Sep and (b) 20 Sep 1980.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
As in Fig. 3, but for 700 hPa at 0000 UTC (a) 16 Sep and (b) 20 Sep 1980.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
As in Fig. 3, but for 700 hPa at 0000 UTC (a) 16 Sep and (b) 20 Sep 1980.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
c. Model configuration and experiment designs
The WRF model was initialized by the North American Regional Reanalysis (NARR; Mesinger et al. 2006) with 32-km grid spacing and available every 3 h. Two sets of nested domains were configured for CAPTEX (Fig. 2) and ASCOT (Fig. 4). The NARR data were downscaled to provide the initial and lateral boundary conditions (IC/LBC) to the outermost WRF domain. Then, WRF results of coarse domains were nested down to the innermost domain, which was used for the inline and offline HYSPLIT dispersion modeling. The output frequency of WRF for the offline HYSPLIT calculations and the time step for meteorological computations for the inline HYSPLIT simulations were set differently in accordance with the spatial resolution. The model configuration and physics options used in the simulations are summarized in Table 4.
Summary of WRF and HYSPLIT model setup for CAPTEX and ASCOT.
CAPTEX uses two nested domains, as shown in Fig. 2. The inner domain at 9-km grid spacing is used for the dispersion calculation. Both inline and offline HYSPLIT simulations were run for each of the six CAPTEX episodes listed in Table 2. Each episode consisted of a single 3-h tracer release at the hour when the HYSPLIT simulation was initiated. Calculated concentrations were averaged over 3 h for comparison with the 3- and 6-h-average samples taken during the CAPTEX experiment.
For ASCOT, five nested domains were configured (Fig. 4). For the finest grid domain (333.3-m grid spacing) used for the dispersion calculation. The IC/LBC are obtained from the nest-down of its mother domain. There were five tracer releases at Anderson Creek in northern California during the ASCOT tracer experiment (Table 3). The emission of each release lasted 1 h starting around midnight local time. Model concentrations were output every 5 min to be compared with the various time duration samples taken during the experiment.
The subgrid cloud scheme and the analysis nudging were applied to CAPTEX simulations but were not used in ASCOT cases because of the high spatial resolution of its configuration. Other physics options were set up identically for both CAPTEX and ASCOT runs (Table 4). For such spatial grid resolutions in ASCOT, two approaches, the large-eddy simulation (LES) and a PBL parameterization, should be taken into consideration. The LES is required for grid size on the order of 10 m or less. The use of the LES option is recommended in the WRF model to resolve eddies for a domain with less than a 100-m grid size (Dudhia and Wang 2014). On the other hand, a PBL parameterization should be used for a horizontal grid spacing of 500 m or larger because eddies are unresolved at that scale. Therefore, for a grid size of 100–500 m, like the finest domain in ASCOT, an LES is not required but a PBL parameterization is still needed (Bryan 2014; Dudhia and Wang 2014). The dispersion simulation configurations for both experiments are included in Table 4. Note that they are identical for the inline and offline versions of HYSPLIT.
d. Statistical metrics for evaluation
4. Results and discussion
a. Comparison with CAPTEX experiment
The controlled tracer release data from the CAPTEX experiment were used to evaluate the results of the inline coupling. The statistical parameters computed using the 3- and 6-h-averaged experimental data for the six CAPTEX episodes are presented in Table 5. The configuration of HYSPLIT for the plume calculation was identical for the offline and inline simulations. The offline version used hourly WRF-ARW fields, which were linearly interpolated to the plume calculation time step while the inline version used WRF-ARW data available at the model’s integration time step (1 min). The statistical scores of all five CAPTEX releases were very close between the coupled and standard HYSPLIT runs. The newly developed model framework of HYSPLIT embedded in WRF-ARW generates similar results to the standard HYSPLIT model driven by offline meteorological data for this regional-scale experiment. Note that about 300 samples were available for the statistical analysis for each CAPTEX release except release 7, which consisted of about 200 samples. Because the sampler spacing was designed to capture the location of the tracer plume rather than its internal structure and the smoothing effects of the time-averaged samples, the statistical evaluation with tracer measurements may not be able to fully reveal the differences between the inline and offline plumes.
Statistical summary of inline and offline HYSPLIT results for six CAPTEX tracer releases.
A spatial comparison of the inline and offline plumes for CAPTEX episodes 2 (top panel) and 5 (bottom panel) is shown in Fig. 6 as difference plots between the two simulations. In episode 2, the tracer was released at the DAY site, which was influenced by a high pressure system centered over the mid-Atlantic region and a weak cold front approaching the sampling area on 26 September. The surface wind flow near the release location (i.e., backside of the high) was southwesterly, which veered with height to westerly and northwesterly (Brown et al. 1984). The inline plume accumulated more tracer particles near the surface, resulting in higher concentrations than the offline plume. The concentration differences between the two approaches became less evident the next day as the plume moved away from the release site. For episode 5, the plume started at the SUD site (in Canada) and was transported to the south by a northwesterly wind associated with a cold front developed ahead of the leading edge of a high pressure system passing through the western section of the sampling array on 26 October (Brown et al. 1984). The tracer concentration corresponding to the inline plume was again higher than offline one at the trailing edge of the plume (Fig. 6, bottom). The differences were smaller the next day, but there is a consistent pattern evident in both cases in that the trailing edge shows higher concentrations in the inline version than the offline version. Similar patterns can be seen in other CAPTEX episodes despite the varying transport patterns. That means more tracers accumulated near the surface in the inline case than the offline simulation. The largest concentration differences between the two approaches occurred in a region of sparse sampling just downwind of the SUD release location.
Spatial differences in tracer concentration (pg m−3) for inline minus offline plumes for CAPTEX episodes (a),(b) 2 and (c),(d) 5. The black dot shows the tracer release site.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
Spatial differences in tracer concentration (pg m−3) for inline minus offline plumes for CAPTEX episodes (a),(b) 2 and (c),(d) 5. The black dot shows the tracer release site.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
Spatial differences in tracer concentration (pg m−3) for inline minus offline plumes for CAPTEX episodes (a),(b) 2 and (c),(d) 5. The black dot shows the tracer release site.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
Tracer concentrations were also collected by aircraft to understand the vertical distribution at various times after release. There were almost 400 samples available for CAPTEX release 2 at five different altitudes. The expectation was that the more instantaneous nature of the aircraft sampling at 6-min intervals would be more sensitive to the inline versus offline model differences. As shown by the statistics in Table 6, both inline and offline HYSPLIT again generated similar results when using the aircraft measurements. The CAPTEX tracer experiment was originally designed to investigate the transport and dispersion of pollutants in the hundreds to thousands of kilometers scale with synoptic features dominated by a high pressure system. The inline coupling system does not show any major differences over the offline approach for this type of application, probably because the regional flows changed gradually in all episodes and the tracer was generally well mixed throughout the boundary layer. However, the consistently higher concentrations shown by the inline version at the trailing edge of the plume, especially just downwind of the source, suggests a closer examination of more local-scale dispersion is required. The comparable results at least confirm that the restructuring of the code from the offline to the inline version was successful.
Statistical summary of inline and offline HYSPLIT results compared with aircraft measurements taken at multiple levels during CAPTEX release 2 on 26 Sep 1983.
b. Comparison with ASCOT experiment
Simulations using the inline coupled WRF–HYSPLIT model were conducted for ASCOT using fine spatial resolution: 333.3 m for the horizontal grid and ~14 m thickness for the first model layer. High temporal frequency (5 min) meteorological data were provided for the offline dispersion calculation. The inline HYSPLIT approach used the meteorological data available at each WRF integration time step at 1-s intervals. All five releases during ASCOT took place at night for the purpose of studying drainage flows. The release site was located in the outflow region of the drainage flow in an open but sheltered area along Anderson Creek, toward the southeast of the ridge (Fig. 4). The statistical results are summarized in Fig. 7, which shows the total rank and its four components including correlation coefficient, fractional bias, figure of merit in space, and the Kolmogorov–Smirnov parameter. For four of five releases, the inline results were better or slightly better than the offline results. The improvement found in the inline simulations was especially evident for releases 2, 4, and 5, primarily because of an increase in the correlation coefficient, implying that the inline HYSPLIT simulations predicted the variations of plume concentration at the sampling sites more accurately than did offline run. For release 1, the improvement was smaller. The inline simulation showed slightly less bias (lower normalized FB) and its correlation went up to 0.16 from almost zero for offline cases.
The statistical evaluation for the inline and offline results from five ASCOT episodes. Different colors represent the four components of rank shown in Eq. (5).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
The statistical evaluation for the inline and offline results from five ASCOT episodes. Different colors represent the four components of rank shown in Eq. (5).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
The statistical evaluation for the inline and offline results from five ASCOT episodes. Different colors represent the four components of rank shown in Eq. (5).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
Release 3 was the only episode where inline HYSPLIT did not outperform the offline run because of the larger fractional bias of the inline results. Among all five release nights, the atmospheric conditions during release 3 were not favorable for drainage flow development. Although the inversion formation associated with nocturnal cooling occurred earlier that night, the drainage conditions did not develop because of the presence of strong westerly winds aloft (Orgill and Schreck 1985). Despite the fact that the inline version for release 3 showed a lower overall rank, the other statistical metrics present a mixed result. The FMS is the percentage of overlap area between the measured and simulated areas while the KSP indicates the maximum difference between cumulative distributions of observed and predicted concentrations. The inline version showed better spatial coverage of the plumes (higher FMS) but a larger difference between the cumulative concentration distributions (lower KSP) than did the offline approach.
Figure 8 shows the difference plot of tracer concentrations and plume overlap comparisons for ASCOT releases 2 and 4, the two releases that showed the best performance for inline HYSPLIT. The illustration shows a comparable number of offline (blue) and inline (pink) grid cells extending beyond the overlap region. Consistent with the CAPTEX results, the inline calculation tends to keep the plume more intact closer to the release site and therefore maintains higher concentrations at the center of the plume that better matched the observations, which showed a narrow tracer plume moving toward the southeast. Because the inline and offline simulations were identically configured (tracer concentration grid, mixing parameters, and PBL settings), the improvement seen in the inline results can be attributed to 1) the higher temporal frequency of the meteorological data and 2) using more layers in the WRF vertical coordinate system for the dispersion calculation. These factors are the primary differences between the two calculations and are possibly the cause of the greater rate of plume spreading in the offline version.
Spatial comparison between inline and offline plumes for ASCOT episodes (a),(b) 2 and (c),(d) 4. (left) Difference plot showing tracer concentration (pg m−3) for inline minus offline plumes. (right) The inline plume (pink), offline plume (light blue), and their overlap (gray). The black dot in each panel shows the tracer release site, and the gray lines are contours (m) of the terrain height.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
Spatial comparison between inline and offline plumes for ASCOT episodes (a),(b) 2 and (c),(d) 4. (left) Difference plot showing tracer concentration (pg m−3) for inline minus offline plumes. (right) The inline plume (pink), offline plume (light blue), and their overlap (gray). The black dot in each panel shows the tracer release site, and the gray lines are contours (m) of the terrain height.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
Spatial comparison between inline and offline plumes for ASCOT episodes (a),(b) 2 and (c),(d) 4. (left) Difference plot showing tracer concentration (pg m−3) for inline minus offline plumes. (right) The inline plume (pink), offline plume (light blue), and their overlap (gray). The black dot in each panel shows the tracer release site, and the gray lines are contours (m) of the terrain height.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0247.1
The offline run used the WRF data within 5-min intervals and interpolated it linearly to the time step needed for dispersion calculation. However, changes in the wind vectors and mixing parameters may not be varying linearly within this 5-min time window and the wind interpolation may be missing shorter time period changes that are occurring in WRF-ARW. The inline system provided meteorological data at WRF’s computational time step (every 1 s) and no temporal interpolation was involved in the subsequent dispersion calculation. In the offline HYSPLIT simulations, the default internal vertical coordinate was used and its resolution was coarser than the WRF’s vertical structure. Another set of offline simulations was conducted with more layers near the surface by altering the constants in the equation of HYSPLIT’s vertical coordinate. The results showed some improvements compared to the original offline results but the inline runs still showed better performance statistics than did the increased-layer offline simulations. In addition, the vertical resolution of the inline version, although comparable to the revised offline version, requires no vertical interpolation, and therefore the complex vertical wind structure near the ground remains unaltered.
The inline coupling of the dispersion model with the meteorological model would, in principle, better depict weather situations at finer spatial scales, which are associated with finer temporal resolution and rapid nonlinear changes in the wind or mixing. Using subhourly interpolated meteorological data provided at hourly intervals for plume calculations may not be good enough to describe rapidly changing meteorological conditions. The simulation results for ASCOT demonstrated that the inline approach is beneficial for the application of dispersion calculations at the fine scales required in complex terrain. However, in the case of regional-scale events dominated by anticyclonic systems (like CAPTEX), the results are not sensitive to the coupling approach and offline dispersion simulations are comparable to the inline methods. Indeed, even using high temporal frequency meteorological data (5-min offline) for CAPTEX, the differences between inline and offline simulations were very small (Ngan et al. 2013).
5. Summary and future work
In this study, the inline linkage of the dispersion model, HYSPLIT, with a mesoscale meteorological model, WRF-ARW, has been presented. HYSPLIT is run during the WRF-ARW calculation, taking advantage of the higher temporal frequency of the meteorological variables. The movement of particles is computed in the WRF-ARW domain along its terrain-following hydrostatic pressure coordinate. When compared to the traditional offline approach using meteorological data generated by a WRF-ARW simulation interpolated to the times required between the data output hours, the inline coupling is expected to more accurately depict the state of the atmosphere and improve the fidelity of the dispersion simulation.
The newly developed inline system was evaluated by comparing inline and offline model simulation results with measured tracer concentrations from controlled tracer experiments: CAPTEX and ASCOT. Simulations for CAPTEX were configured over a 9-km grid spacing domain across the northeastern United States while for ASCOT the domain was at 333.3-m grid spacing with more vertical layers near the surface over the complex terrain area in northern California than for the CAPTEX configuration. The temporal resolution was set up differently for two experiments because of the different grid spacings. For CAPTEX, the meteorological model was advanced in time every 60 s and output was written every hour. The offline HYSPLIT scheme took the hourly meteorological data and interpolated it to the HYSPLIT integration time step. The inline run used the meteorological fields at the WRF-ARW time step every 60 s. However, for all six CAPTEX releases the inline and offline simulations showed comparable performance statistics when compared to the tracer measurements. Even though differences in detail between the two HYSPLIT plumes were observed in areas of sparse sampling, the benefit of using the inline approach was not evident for the CAPTEX tracer experiment, which was originally designed to investigate the transport and dispersion of pollutants from synoptic to regional scales with most individual releases primarily influenced by a high pressure system.
For the ASCOT experiment, the offline dispersion was calculated by using meteorological data from WRF-ARW written every 5 min while the inline computation was performed at the model’s time step every second. The statistics of the model results compared with the tracer measurements indicated that the inline results clearly outperformed the offline simulation in three releases of five and were slightly better in another one. For those releases where the inline version had higher total rank (a cumulative statistical score) than the offline version, the correlation coefficient increased and the fractional bias decreased. Furthermore, the inline version showed better plume coverage of the measurements than did the offline HYSPLIT model. The inline coupling system is especially beneficial for this type of application because the drainage flow occurred over an area of complex terrain area, over a short time period (less than 10 h), and in fine spatial resolution, which required very detailed and rapid changes in the wind vectors and mixing parameters to accurately model the movement of tracer particles. In addition, the ASCOT subhourly sampling intervals provided a detailed view of the plume structure for comparison with the simulation results that was not possible for the 3- or 6-hourly time-averaged tracer data from CAPTEX.
In the next development stage, the inline HYSPLIT model will be further modified to use WRF’s domain decomposition for parallelization and computational efficiency. For most local-scale simulations where the inline version provides the most benefit, domain parallelization is not expected to provide much computational time reduction because the tracer plume will be limited in its spatial extent. The inline HYSPLIT approach will be also tested using others parameters from WRF-ARW that are most relevant to plume mixing, stability, and convection, as well as the time-averaged wind fields available in the model. In both the inline and offline versions of HYSPLIT, many of these parameters are currently diagnosed from the wind and temperature profiles rather than using the comparable parameters directly from WRF-ARW. Thus, any inconsistencies or errors due to the rediagnosis of these parameters can be minimized. The terrain data used for the ASCOT case are the 30-s elevation data (~900 m), which is the finest resolution commonly used in WRF-ARW. However, it is still much coarser than the grid spacing used to model ASCOT (333.3 m). Future work will include the processing of higher-resolution elevation data from satellites and their implementation into WRF-ARW. We will also compare tracer results simulated by inline HYSPLIT with other models such as WRF-CHEM to characterize the difference in the plumes generated by Lagrangian and Eulerian approaches. Furthermore, the inline HYSPLIT model will be further tested using other local-scale tracer experiments such as the Metropolitan Tracer Experiment (METREX; Draxler 1987) for the urban environment and other complex terrain studies.
Acknowledgments
The authors thank Nick Heffter for digitizing the ASCOT data from the report.
REFERENCES
Bowman, K. P., J. C. Lin, A. Stohl, R. Draxler, P. Konopka, A. Andrews, and D. Brunner, 2013: Input data requirements for Lagrangian trajectory models. Bull. Amer. Meteor. Soc., 94, 1051–1058, doi:10.1175/BAMS-D-12-00076.1.
Brioude, J., W. M. Angevine, S. A. McKeen, and E. Y. Hsie, 2012: Numerical uncertainty at mesoscale in a Lagrangian model in complex terrain. Geosci. Model Dev. Discuss., 5, 967–991, doi:10.5194/gmdd-5-967-2012.
Brown, R. M., M. J. Leach, G. S. Raynor, and P. A. Michael, 1984: A summary and index of the weather documentation for the 1983 Cross-Appalachian Tracer Experiments. Informal Rep. BNL36879, Brookhaven National Laboratory, Upton, NY, 41 pp.
Bryan, G., 2014: Recommendations for WRF simulations with horizontal grid spacing between 100 m and 1 km. 15th WRF Users Workshop, Boulder, CO, NCAR, 4.1. [Available online at http://www2.mmm.ucar.edu/wrf/users/workshops/WS2014/ppts/4.1.pdf.]
Chen, B., A. F. Stein, N. Castell, J. D. de la Rosa, A. M. de la Campa, Y. Gonzalez-Castanedo, and R. R. Draxler, 2012: Modeling and surface observations of arsenic dispersion from a large Cu-smelter in southwestern Europe. Atmos. Environ., 49, 114–122, doi:10.1016/j.atmosenv.2011.12.014.
Chen, F., and J. Dudhia, 2001: Coupling and advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585, doi:10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.
Deng, A., and Coauthors, 2009: Update on WRF-ARW end-to-end multi-scale FDDA system. 10th WRF Users Workshop, Boulder, CO, NCAR, 1.9. [Available online at http://www2.mmm.ucar.edu/wrf/users/workshops/WS2009/presentations/1-09.pdf.]
Draxler, R. R., 1987: One year of tracer dispersion measurements over Washington, D.C. Atmos. Environ., 21, 69–77, doi:10.1016/0004-6981(87)90272-1.
Draxler, R. R., 2006: The use of global and mesoscale meteorological model data to predict the transport and dispersion of tracer plumes over Washington, D.C. Wea. Forecasting, 21, 383–394, doi:10.1175/WAF926.1.
Draxler, R. R., and G. D. Hess, 1997: Description of the HYSPLIT_4 modeling system. NOAA Tech. Memo. ERL ARL-224, NOAA/Air Resources Laboratory, Silver Spring, MD, 24 pp. [Available online at http://www.arl.noaa.gov/documents/reports/arl-224.pdf.]
Dudhia, J., 1989: Numerical study of convection observed during the Winter Monsoon Experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46, 3077–3107, doi:10.1175/1520-0469(1989)046<3077:NSOCOD>2.0.CO;2.
Dudhia, J., and W. Wang, 2014: WRF advanced usage and best practices. 15th WRF Users Workshop, Boulder, CO, National Center for Atmospheric Research, 35 pp. [Available online at http://www2.mmm.ucar.edu/wrf/users/workshops/WS2014/ppts/best_prac_wrf.pdf.]
Ferber, G. J., J. L. Heffter, R. R. Draxler, R. J. Lagomarsino, F. L. Thomas, and R. N. Dietz, 1986: Cross-Appalachian Tracer Experiment (CAPTEX -83) final report. NOAA Tech. Memo. ERL ARL-142, NOAA/Air Resources Laboratory, Silver Spring, MD, 60 pp. [Available online at http://www.arl.noaa.gov/documents/reports/arl-142.pdf.]
Grell, G. A., and D. Devenyi, 2002: A generalized approach to parameterizing convection combining ensemble and data assimilation techniques. Geophys. Res. Lett., 29, 1693, doi:10.1029/2002GL015311.
Grell, G. A., J. Dudhia, and D. Stauffer, 1994: A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note NCAR/TN-398+STR, 121 pp. [Available online at http://nldr.library.ucar.edu/repository/assets/technotes/TECH-NOTE-000-000-000-214.pdf.]
Grell, G. A., S. E. Peckham, R. Schmitz, S. A. McKeen, G. Frost, W. C. Skamarock, and B. Eder, 2005: Fully coupled “online” chemistry within the WRF model. Atmos. Environ., 39, 6957–6975, doi:10.1016/j.atmosenv.2005.04.027.
Gudiksen, P. H., G. J. Ferber, M. M. Fowler, W. L. Eberhard, M. A. Fosberg, and W. R. Knuth, 1984: Field studies of transport and dispersion of atmospheric tracers in nocturnal drainage flows. Atmos. Environ., 18, 713–731, doi:10.1016/0004-6981(84)90257-9.
Hegarty, J., and Coauthors, 2013: Validation of Lagrangian particle dispersion models with measurements from controlled tracer releases. J. Appl. Meteor. Climatol., 52, 2623–2637, doi:10.1175/JAMC-D-13-0125.1.
Hong, S.-Y., J. Dudhia, and S.-H. Chen, 2004: A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation. Mon. Wea. Rev., 132, 103–120, doi:10.1175/1520-0493(2004)132<0103:ARATIM>2.0.CO;2.
Hong, S.-Y., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341, doi:10.1175/MWR3199.1.
Janjic, Z. I., 2003: A nonhydrostatic model based on a new approach. Meteor. Atmos. Phys., 82, 271–285, doi:10.1007/s00703-001-0587-6.
Janjic, Z. I., T. Black, M. Pyle, E. Rogers, H.-Y. Chuang, and G. DiMego, 2005: High resolution applications of the WRF NMM. Preprints, 21st Conf. on Weather Analysis and Forecasting/17th Conf. on Numerical Weather Prediction, Washington, DC, Amer. Meteor. Soc., 16A.4. [Available online at https://ams.confex.com/ams/pdfpapers/93724.pdf.]
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471, doi:10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.
Kanamitsu, M., 1989: Description of the NMC Global Data Assimilation and Forecast System. Wea. Forecasting, 4, 335–342, doi:10.1175/1520-0434(1989)004<0335:DOTNGD>2.0.CO;2.
Lei, L., D. R. Stauffer, and A. Deng, 2012: A hybrid nudging-ensemble Kalman filter approach to data assimilation in WRF/DART. Quart. J. Roy. Meteor. Soc., 138, 2066–2078, doi:10.1002/qj.1939.
Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343–340, doi:10.1175/BAMS-87-3-343.
Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and S. A. Clough, 1997: Radiative transfer for inhomogeneous atmosphere: RTTM, a validated correlated-k model for the longwave. J. Geophys. Res., 102, 16 663–16 682, doi:10.1029/97JD00237.
Ngan, F., A. F. Stein, and R. R. Draxler, 2013: Inline coupling of WRF-HYSPLIT: Development and results for CAPTEX-83. Traversing New Terrain in Meteorological Modeling, Air Quality and Disperison, Davis, CA, Air Quality Research Center, University of California, Davis. [Available online at http://aqrc.ucdavis.edu/event/conference-traversing-new-terrain-in-meteorological-modeling-air-quality-and-dispersion/.]
Orgill, M. M., and R. I. Schreck, 1985: An overview of the ASCOT multi-laboratory field experiments in relation to drainage winds and ambient flow. Bull. Amer. Meteor. Soc., 66, 1263–1277, doi:10.1175/1520-0477(1985)066<1263:AOOTAM>2.0.CO;2.
Peltier, L. J., S. E. Haupt, J. C. Wyngaard, D. R. Stauffer, A. Deng, J. A. Lee, K. J. Long, and A. J. Annunzio, 2010: Parameterizing mesoscale wind uncertainty for dispersion modeling. J. Appl. Meteor. Climatol., 49, 1604–1614, doi:10.1175/2010JAMC2396.1.
Skamarock, W. C., and Coauthors, 2008: A description of the Advanced Research WRF verson 3. NCAR Tech Note NCAR/TN-475+STR, 125 pp. [Available online at http://www2.mmm.ucar.edu/wrf/users/docs/arw_v3.pdf.]
Srinivas, C. V., R. Venkatesan, R. Baskaran, V. Rajagopal, and B. Venkatraman, 2012: Regional scale atmospheric dispersion simulation of accidental releases of radionuclides from Fukushima Dai-ichi reactor. Atmos. Environ., 61, 66–84, doi:10.1016/j.atmosenv.2012.06.082.
Stohl, A., M. Hittenberger, and G. Wotawa, 1998: Validation of the Lagrangian particle dispersion model FLEXPART against large-scale tracer experiment data. Atmos. Environ., 32, 4245–4264, doi:10.1016/S1352-2310(98)00184-8.
Wong, D. C., and Coauthors, 2012: WRF-CMAQ two-way coupled system with aerosol feedback: Software development and preliminary results. Geosci. Model Dev., 5, 299–312, doi:10.5194/gmd-5-299-2012.
Zhang, Y., 2008: Online-coupled meteorology and chemistry models: History, current status, and outlook. Atmos. Chem. Phys., 8, 2895–2932, doi:10.5194/acp-8-2895-2008.
