Air mass back trajectories

The periodically repeating critical situation in Kuala Lumpur and other Southeast Asian cities is a result of atmospheric long-range transport of air pollutants from Sumatra and Borneo where forests are being burnt off.

Atmospheric transport is one of the processes fundamental for the understanding of the environmental fate of pollutants. Some of these chemicals are subject to long-range transport. This section describes the context of this process and one of the approaches suitable for its investigation. As the RECETOX centre research activities are strongly oriented on the Stockholm Convention on persistent organic pollutants (POPs), this section is focused particularly on the long-range transport of these compounds. However, a high number of various studies deal with atmospheric transport of other organic and inorganic pollutants.

Background of the atmospheric long-range transport of POPs

Polychlorinated biphenyls (PCBs), substances with many useful technical properties, were produced and used mainly in Europe, North America and Japan. However, they have been detected also in Himalayan lakes, fat tissues of whale and seal blubber in Greenland and other cold areas. The well-known pesticide DDT is a similar case. Increased concentrations of this pesticide and its metabolite DDE have been found in breast milk of Inuit (Eskimo) women. DDT and DDE were also found in fish living in Arctic sea depths and measured in Arctic air. How is this possible considering none of these chemicals was produced or, in significant amounts, used in these clean areas?

One of the reasons is the half-life of POPs, i.e. the time-scale in which the concentration of a specific compound in a selected environmental compartment (water, soil, air etc.) decreases to half of its initial value. This quantity describes the resistance of the compound to decomposition or removal from a given environmental compartment. The half-life of POPs in the atmosphere is long enough such that these chemicals can be transported by air to polar regions, in which, bound to dust or snowflakes or transferred by atmospheric turbulence, they may reach the ground or water surface. Re-volatilisation and, hence, removal from the area is suppressed at low concentrations. Thus, polar regions are receptor areas for a number of pesticides and POPs transported in the atmosphere. However, POPs are not transported to these sensitive cold areas only, but are distributed globally. Atmospheric long-range transport of many, so-called semivolatile substances is enhanced by their ability to undergo more than one cycle of removal from atmosphere to water or soil and subsequent volatilization back to air (so called grasshopper effect). Long-range transport does not take place only in atmosphere, sea currents are another important agent.

Characteristics of POP sources

Due to the discovery of the harmful effects of POPs on human health and ecosystems, restrictions on the use and production of these chemicals started to be implemented over time. PCB production was being prohibited in various countries during 1970's and 80's. The use of DDT was prohibited in 1970 in Sweden and subsequently in other countries. However, until now DDT is being used for the control of vector born diseases (malaria). The applied amount of another pesticide from the POP group, lindane (γ-hexachlorocyklohexane), decreased significantly in European Union countries and its use is often fully prohibited. This trend leads to an increasing importance of emissions from so called secondary sources of these substances. A primary source is defined as the first entrance of a chemical into the environment, in the case of pesticides e.g. crop-spraying. Primary sources of PCBs can be e.g. cooling or dielectric liquids of devices containing these chemicals.

Agricultural and other soils that were subject to pesticide application function as a reservoir of these substances and their degradation products. They are also present in water streams and reservoirs, in which they accumulate in sediments. Secondary sources can be described as processes of retrograde release of the substances into the environment, e.g. evaporation from soils or leaching from sediments. Since POPs are characterized by long half-lives in these environmental compartments, soils and sediments function as sources of these substances even tens of years after their use has been stopped.

When studying atmospheric transport and other aspects of POP environmental fate, primary and secondary sources must be taken into account. Unfortunately, we encounter a problem at this point. While primary sources are usually at least partly described (e.g. in form of a so called emission inventory), we know very little about the location and particularly the intensity of secondary sources. However, there exist approaches able to cope with this problem.

How to describe the atmospheric transport of pollutants?

There are two approaches for the description of the atmospheric transport of POPs and a wide spectrum of other compounds. Both of them are based on a combined assessment of meteorological data and known concentrations of the studied pollutants. The first of them, dispersion modelling, studies the influence of well-described pollutant sources (e.g. industrial enterprises or complexes, self-heating buildings, main roads, dumping places, old burdens etc.) on their near surroundings and also farther regions. An appropriate section of this portal deals with this approach (link). On the contrary, receptor models describe the influence of known or unknown sources or source areas (e.g. agricultural lands, in which pesticides were applied, or urban and industrial areas) on pollutant concentrations recorded at the place of air quality measurement.

A significant part of the receptor and dispersion models makes use of so called air mass trajectories. A trajectory is the path of a hypothetic air parcel as it is acted on by winds. The hypothetic air parcel is a symbol for air masses which are a transport medium (carrier) of pollutants. One of the options when studying atmospheric transport is the determination of the parcel trajectory from its source to find out where the emitted substance may be transported. This is how we obtain so called forward trajectories, which are being used in some types of dispersion models.

For the interpretation of substance levels measured in air at a given locality and time, calculation of back trajectories is suitable for the determination of paths that the air mass followed before reaching the sampling place (receptor). A simplified presumption for this model type is that the air mass is subject to an uptake of pollutants emitted from source areas during its movement above the corresponding land surface. This is resulting in the concentration level measured at the receptor site (the influence of local sources is negligible or precisely quantified). The investigation of atmospheric transport based on this type of trajectories is discussed below.

An example of a back trajectory generated using the HYSPLIT model (Draxler and Rolph, 2003). The red line on the map shows the trajectory of a hypothetic air parcel within four days before arrival at the place of air quality measurement, in this case the regional observatory Kosetice, Czech Republic, on 5th May 2000 at 8am. The lower graph shows the changing height of the transported air parcel above ground.

Since the analysis of air mass back trajectories does not require information on sources and bears only on concentration measurements of the studied substances in air at the given locality, it is not dependent on the accuracy of emission inventories and, therefore, is able to reflect both known and unknown sources (often secondary sources in this case). This discriminates the analysis from other types of receptor models, which require very detailed (CMB models – Chemical Mass Balance) or at least partial knowledge of a previously localized source (multivariate statistical approaches such as factor analysis or principal component analysis). An overview of receptor models is given in the table below.

Types of receptor models, statistical evaluation of air mass back trajectories and related approaches: input requirements and output information

Receptor models based on air mass back trajectories may contribute to:

• determination of potential known and unknown source areas affecting substance concentrations at the site of air quality measurement (determination of potential source areas at longer or shorter distances)
• first description of potential sources, which may be inventorised after their verification, determination of appropriate emission factors and, therefore, contribute to the definition of effective measures of air pollution control
• explanation of fluctuations and concentration levels measured at background sites (i.e. sites with negligible effects of local sources) and their temporal trends
• description of changes in intensity and location of potential sources areas in time
• discussions on other questions associated with the environmental fate of POPs and other pollutants.

Investigation of air mass trajectories

A trajectory is a series of points in space and time which fulfill the trajectory equation (Stohl, 1998):

where t = time, X = position vector, and X' = wind velocity vector. If we know the initial position X₀ at time t₀ of the parcel, its path is completely determined through equation (4). It is possible to write:

It is also possible to find the inverse transformation:

which gives the initial coordinates of the parcel, which at time t is at position X. Thus, fictive air parcels may be followed in a wind field either forward (forward trajectories) or backward (back trajectories) in time.

4-day air mass back trajectories generated using the HYSPLIT model (Draxler and Rolph, 2003) for the regional observatory Kosetice in 2002. Only one trajectory per week was calculated, which is insufficient for a statistical analysis. However, the image allows for a simple look at the character of air flow affecting the selected receptor. Different velocities of air mass movement are evident, the highest can be observed in the western sector (distance of points defining the position of trajectory hourly segment endpoints describes the velocity of the moving air parcel – the closer the points are, the lower the velocity is).

The solutions of the trajectory equation can be kinematic using only wind information (Danielsen, 1961) or dynamic making use of velocity, mass field information and dynamic equations linking the two (Merrill et al., 1986). Since nowadays accurate wind fields with high space and time resolution are available, kinematic trajectories are more accurate (Stohl and Seibert, 1998).

Following input parameters must be selected before the trajectory generation:

• type of trajectory: isobaric, isentropic, or three-dimensional (these are currently considered as the most accurate; Stohl, 1998),
• trajectory length, which depends on the size of the studied region. Trajectories longer than 5 days are considered less accurate (Hopke et al., 1995),
• initial trajectory height, which depends on meteorological and orographical conditions in the receptor location and type of atmospheric transport under study,
• starting time of the trajectory (corresponding to the sampling date, it is recommended to generate multiple trajectories evenly covering the sampling time interval),
• geographical coordinates of the receptor.

Geographical coordinates determining the position of trajectory hourly segment endpoints (which define the path of the hypothetic air parcel) are the most important output from the trajectory generation model. Information on precipitation, temperature or atmospheric pressure for each endpoint may be chosen as supplementary parameters.

A frequently used model for air mass trajectory generation is HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory), which allows for the calculation of both forward and back trajectories and simulation of complex dispersion and deposition of pollutants (Draxler and Rolph, 2003). It is freely accessible at the READY website of the NOAA (Rolph, 2003). The FLEXTRA model is also widely used (Stohl et al., 1995). A convenient model for Europe and particularly the research of atmospheric transport in lower heights above ground is the TRAMPER model (Reimer and Scherer, 1991), which disposes of good resolution of orographic input data for this region.

Geographical information systems and other suitable software products may be employed for the statistical and spatial evaluation of trajectories. Two approaches of statistical evaluation of air mass back trajectories are introduced in the following text. The principle of spatial evaluation is common for both of them:

• the region of interest (e.g. Europe in case of Kosetice) is divided to individual cells by means of a grid, which can be imagined as square paper
• segment endpoints of all trajectories are drawn in the grid,
• cells with a number of trajectory segment endpoints lower than a selected criterion are excluded (no or very few air masses sampled in the given receptor passed over areas corresponding to these cells during the observed time period, thus, they had low or negligible effect on pollutant concentrations measured at the receptor site),
• the parameter value of the appropriate model is calculated for each grid cell. The values are subsequently used for creating maps depicting potential source areas.

Potential Source Contribution Function (PSCF)

This method was first used by Ashbaugh et al. (1985), who also described several variants of this model dependent on various levels of selected spatial resolution and inclusion of atmospheric transport in various heights above ground. Peng et al. (2007) summarized the PSCF method as follows: For a given grid cell ij, the PSCF value can be calculated by counting the trajectory segment endpoints that terminate within that cell. Assume the number of endpoints fall in the ij-th cell is n_ij, the number of endpoints for the same cell corresponding to pollutant concentrations higher than an arbitrary criterion value is defined as m_ij. The PSCF value for the ij-th cell is then defined as :

PSCFij = m_ij / n_ij

The PSCF value of the given cell may be understood as a conditional probability that the concentrations recorded at the receptor site that are higher than a selected criterion may be reached due to air masses travelling over a specific segment of land surface defined by this cell (Hsu et al., 2003). The higher PSCF value, the higher this probability is. Cells with a high PSCF value refer to areas with a potentially high contribution to the final pollutant concentration in the sample.

When a central value (e.g. median) is chosen as a criterion for discrimination between low and high pollutant concentrations in the corresponding air sample, the PSCF value in the given grid cell can result from considering samples with a pollutant concentration slightly or also considerably higher than the chosen criterion. Then, major sources can´t be discriminated from moderate ones. Therefore, e.g. 75% percentile may be used as the criterion to discriminate between sources of differing intensities (Hsu et al., 2003).

The PSCF model was applied to a long-term time series of POP concentration data from the regional observatory Kosetice by Dvorska et al. (2008). An example of the results calculation and interpretation is showed elsewhere at this portal.

Ground Mean Source Loading, L_g

Ground mean source loading is a continuous measure for the influence of ground surface emissions on the composition of passing air masses (Lammel et al., 2003). In the following equation, the L_g parameter is expressed in concentration units. We used a linear time weight, 1...0 within 96 hourly steps to account for the influence of dispersion on the accuracy of the location of the trajectory hourly points and of deposition on trace components abundances, both increasing back in time.

where I is the total number of trajectories; c_ij the total atmospheric concentration of pollutant at the receptor site upon arrival of the ith trajectory; n_g the total number of trajectory hourly points in grid cell g; h_ij the number of hours counted back from arrival time for the j_th hourly point, ith trajectory; 1-(h_ij/96) are time weights of hourly points and L₀ the unit loading of surface sources to air mass, set equal to 1 for trajectory hourly points which fall into grid cell _g and set equal to 0 for all other trajectory hourly points.

The calculation of ground mean source loadings was applied to a long-term time series of POP concentration data from the regional Kosetice observatory (Dvorska et al., 2009). An example of the calculation and interpretation of results is shown elsewhere at this portal (link).

Shortcomings in methods based on the statistical evaluation of air mass back trajectories

The mentioned (as well as other) methods for the statistical evaluation of air mass back trajectories may elicit some disadvantages and limitations:

• the trajectory itself is a model output and thus subject to inaccuracies,
• the spatial resolution of resulting maps is dependent on the spatial resolution of input data of models for trajectory generation, which is often coarse,
• it is difficult to determine the statistical significance of model outcomes (Hopke et al., 1995),
• the environment is often supposed to be ideal, i.e. the turbulent dispersion of pollutants in the atmospheric boundary layer and processes of their removal from atmosphere are neglected,
• it is likely that areas down and upwind from a potential source will be also determined as potentially sources due to the even distribution of weight over the whole trajectory. This effect may be eliminated by weighing and also using concentration data from multiple sampling sites (Hsu et al., 2003).

In spite of the stated disadvantages, statistical methods for the evaluation of air mass back trajectories are a suitable tool for obtaining first knowledge on potential source areas (Poirot and Wishinski, 1986).

References

Dvorská A., Lammel G., Klánová J., Holoubek I.(2008). Kosetice, Czech Republic - Ten years of air pollution monitoring and four years of evaluating the origin of persistent organic pollutants. Environmental Pollution 156, 403-408.

Dvorská A., Lammel G., Holoubek I., (2009). Recent trends of persistent organic pollutants in air in central Europe - Air monitoring in combination with air mass trajectory statistics as a tool to study the effectivity of regional chemical policy. Atmospheric Environment 43, 1280-1287.

Ashbaugh L.L., Malm W.C., Sadeh W.D. (1985). A residence time probability analysis of sulfur concentrations at Grand Canyon National Park, Amtospheric Environment 19, 1263-1270.

Draxler R.R., Rolph G.D. (2003).HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory) Model access via NOAA ARL READY Website (http://www.arl.noaa.gov/ready/hysplit4.html). NOAA Air Resources Laboratory, Silver Spring, MD, USA.

Hopke P.K., Li C.L., Landsberger S. (1995): The use of bootstrapping to estimate conditional probability fields for source locations of airborne pollutants. Chemometrics and Intelligent Laboratory Systems 30, 69-79.

Hsu Y.-K., Holsen T.M., Hopke P.K. (2003). Comparison of hybrid receptor models to locate PCB sources in Chicago, Atmospheric Environment 37, 545-562.

Lammel, G., Brüggemann, E., Gnauk, T., Müller, K., Neusüss, C., Röhrl, A. (2003).A new method to study aerosol source contributions along the tracts of air parcels and its application to the near-ground level aerosol chemical composition in central Europe. Journal of Aerosol Science 34, 1-25.

Peng, X.L., Choi, M.P.K., Wong, M.H. (2007). Receptor modeling for analyzing PCDD/F and dioxin-like PCB sources in Hong Kong. Environmental Modelling and Assessment 12, 229-237.

Poirot R.L., Wishinski P.R. (1986). Visibility, sulfate and air mass history associated with the summertime aerosol in Northern Vermont. Atmospheric Environment 20, 1457-1469.

Reimer E., Scherer B. (1991). An operational meteorological diagnostic system for regional air pollution analysis and long term modellin. In: Proceedings of the 19th ITM on air pollution modelling and its application. Crete 1991, Vol. II, 421-428.

Rolph G.D. (2003).Real-time Environmental Application and Display sYstem (READY) Website (http://www.arl.noaa.gov/ready/hysplit4.html). NOAA Air Resources Laboratory, Silver Spring, MD, USA.

Stohl A., Wotawa G., Seibert P., Kromp-Kolb H. (1995). Inerpolation errors in wind fields as a function of spatial and temporal resolution and their impact on different types of kinematic trajectories. Journal of Applied Meteorology 34, 2149-2165.

Stohl A. (1998). Computation, accuracy and applications of trajectories - a review and bibliography. Atmospheric Environment 32, 647-966.

Links:

Model HYSPLIT: http://www.arl.noaa.gov/HYSPLIT.php

Model TRAMPER: http://www.geo.fu-berlin.de/met/ag/trumf/Trajektorien/index.html

Model FLEXTRA: http://zardoz.nilu.no/~andreas/flextra+flexpart.html