5.1 Sensitivity of models to meteorological data
Meteorological data are one of the most important inputs into any air dispersion model. Ground-level concentrations of contaminants are primarily controlled by two meteorological elements: wind direction and speed (for transport), and turbulence and mixing height of the lower boundary layer (for dispersion).
The meteorological data requirements for steady-state Gaussian-plume models and advanced dispersion models vary considerably. Steady-state Gaussian-plume models require meteorology data from a single surface station. They assume that the single station data are applicable to the whole modelling domain up to the top of the boundary layer and that conditions do not vary with height.
Advanced dispersion models - including puff, particle and grid models - allow meteorological conditions to vary across the modelling domain and up through the atmosphere. This is a much more complex situation than for steady-state modelling and thus requires much more complex meteorological data. Because meteorological sites do not provide the relevant data at every point in the modelling domain, a meteorological model is used to predict and provide the meteorological variables at sites where information is not available. The advanced dispersion model then uses this pre-processed meteorological data for analysis.
Because the meteorological data requirements vary greatly between these two model types, the choice of which dispersion model to use can depend on questions regarding the expected meteorological conditions. The question, Will the meteorological conditions be uniform across the modelling domain? (or can they be approximated this way?) needs to be answered. You therefore need to consider the:
- boundary layer structure
- atmospheric turbulence
- modelling domain topography
- mesoscale meteorology (air-pollution meteorology).
There is a range of options for collecting and processing land-based meteorological data, including surface meteorological stations, tethered balloons, radiosonde upper air balloons, manual observations, remote sensing systems (SODAR/RASS, Radar, Lidar) and satellites. Various meteorological processors are also available to process raw data into formats required by air dispersion models.
Meteorological data must be treated as a critical input for any modelling study.
Steady-state Gaussian-plume models require meteorological data from a single surface station.
Advanced dispersion models allow meteorological conditions to vary across the modelling domain.
5.2 Meteorological data for steady-state Gaussian-plume models
Steady-state Gaussian-plume models require meteorological data from a single site. These data requirements can be met by three approaches, which are discussed in order below. The use of each approach will strongly depend on the:
- meteorological data available
- purpose for which the model is being used
- scale and significance of the potential effects of the discharge
- accuracy of information and level of detail required by the regulatory authority.
The approach taken should match the scale and significance of the discharge being assessed, while making use of the best available meteorological data.
5.2.1 Screening meteorological data
As a first step and when worst-case events are of primary concern, it is generally recommended to use a standard screening meteorological data set as an initial air dispersion modelling assessment. Most commercially available models such as AUSPLUME or ISCST3 supply screening data sets with the model.
Screening meteorological data sets have been developed using standard combinations of wind speed, stability class and mixing heights, which should mimic the range of atmospheric conditions that are likely to occur in any given location. They provide a simple option to run the air dispersion model and can be applied in most locations. The maximum ground level concentration predicted using a screening data set is normally regarded as conservative. This means that it is likely the model over-predicts concentrations expected to occur in reality, assuming that other input data are of good quality. The results from a screening model are often termed 'worst-case scenario' impacts.
There are several limitations to these data. They can only model one-hour averages, not longer time-averaging periods such as eight hours, 24 hours or annual averages. This means that certain contaminants that have ambient guidelines for longer periods - such as PM10 (24 hours) - cannot be directly assessed using a screening data set. However, the model CTSCREEN, which comes with a screening meteorological data set, can provide estimates of 24-hour averages.
Another limitation is that these data cannot provide an indication of how frequently an event might occur, what the spatial distribution of the impact is, nor average concentrations.
Screening meteorological data sets should therefore not be used for:
- averaging periods longer than one hour
- PM10 sources that are likely to produce significant downwind concentrations
- frequency assessment of pollution events
- airshed sources.
Screening data sets should only be used to gain a 'first cut' estimate of the magnitude of the maximum ground-level concentration for a particular source.
When a screening data set is used, the modeller must ensure it contains mixing heights and stability classes which realistically represent the location being modelled.
To estimate the 'worst-case' scenario, all other model inputs, such as emission rates, must be selected and shown to produce 'conservative' results.
5.2.2 Ready-made, site-specific data sets
When screening meteorological data cannot be used, it may be appropriate to use ready-made, site-specific meteorological data sets for modelling. These situations include those when a screening data set does not:
- provide sufficient accuracy
- meet the criteria of the ambient air quality requirements of the local council
- suit the source or type of pollutant being modelled (e.g. PM10 from a large coal-fired boiler).
Some urban and regional ready-made meteorological data sets have been produced for some regional councils and are also available for a number of the larger cities in New Zealand (see Appendix B). These regional councils should be able to provide advice on the availability and appropriateness of any ready-made site-specific data set for a specific modelling project.
Some private consultants have also produced site-specific data sets. Normally this data is carefully guarded intellectual property or owned by their clients. But depending on the task for which it is intended and the amount you are prepared to pay, this source of data may be worth exploring. Again the relevant regional council should be able to provide advice on what, if any, data has been produced by consultants for a specific area. The data sets developed by consultants normally cost between $1000 and $5000.
5.2.3 Developing a site-specific data set
If a suitable ready-made meteorological data set is not available or is not applicable to the site in question, one needs to be developed. Provided it is of good quality, on-site data are often the preferred source of meteorological input data even if other nearby sets are available. A distinct advantage of having on-site data is that they can also be used for dispersion model valuation studies.
The predecessor to this Guide is the document Guidelines for the use of Dispersion Models (NIWA, 1998). Part 2 of this document provides an overview on the use of meteorological data as input for Gaussian-plume dispersion models. It also contains detailed descriptions of some of the methods used to calculate derived meteorological parameters. While Part 2 of the original modelling guideline provides a useful introduction to the use of meteorological data in modelling, recent developments in meteorological modelling have rendered some of the detail out of date. The following sections provide current recommended practices.
Developing a meteorological data set can be expensive and time-consuming. Depending on the complexity of the site, a degree of meteorological expertise may be required to make sure the data are accurately representing the conditions experienced at the site. It is recommended that if the data are to be used as part of an AEE, they are put through a thorough quality assurance process and/or peer reviewed before use.
The collection of site-specific meteorological data has been fully covered in the documents On-site Meteorological Program Guidance for Regulatory Modelling Applications (US EPA, 1987) and Part 51, Guideline on Air Quality Models (US EPA, 1999). The former provides details on site location, recording mechanisms, data communication, sampling rates, system accuracies, data handling, quality control and treatment of missing data. It is recommended that this guidance be adopted as best practice for the collection and processing of meteorological data for use in dispersion modelling applications. This is consistent with the approach taken in the Good Practice Guide for Air Quality Monitoring and Data Management (Ministry for the Environment, 2000b).
When producing a site-specific data set there are generally two sources of data that can be used: data collected on site, or data collected from an existing nearby source.
a Data collected on site
A meteorological station should be located away from the influences of obstructions such as buildings and trees to ensure that the general state of the environment (wind direction and temperature) is best represented. It is recommended that you use a 10 m high mast for measuring wind direction and speed and temperature differentials. However, where the mast is located in good free-flow conditions and there are height restrictions from local council bylaws, a 6m high mast can be used.
For major industrial sources with tall stacks, or a site within a complex terrain environment, higher monitoring masts (30 m and higher) are recommended to adequately monitor lower boundary-layer wind and temperature profiles. It may be necessary for these situations to supplement or even replace a tall mast with monitoring via remote sensing instruments such as SODAR/RASS or tethered-sonde systems.
On-site data should be reduced to hourly averages for all parameters. To develop a meteorological data set for air dispersion modelling the following parameters need to be monitored from the site:
- temperature difference (between 1.5 m and 10 m or higher)
- relative humidity
- wind speed
- wind direction
- solar radiation.
While all the above variables provide valuable information for modelling, the most important variables are wind speed and direction, and temperature. Setting up a station to record and log these three parameters costs approximately $12,000 (in 2004). There will also be relatively small additional costs associated with site maintenance and data management.
Depending on what instrumentation is employed on site, the data collected may need to be supplemented with the following off-site data from the National Institute of Water and Atmospheric (NIWA) Climate Database (CLIDB) system:
- hourly cloud cover and height for the region
- twice-daily upper air temperature, relative humidity, and wind speed and direction from the closest upper air radiosonde station.
When developing a meteorological data set, the representativeness of the data set must be assessed, and demonstrated, in terms of climatic means and extremes. This can essentially be established in two ways: by undertaking long-term (three to five years) monitoring of on-site data collection, or by establishing correlations between on-site data, climatic averages and regional extremes. Average climatic conditions for the region can be obtained from NIWA's CLIimate DataBase (CLIDB) http://www.niwa.co.nz/services/clidb.
b Data from locations removed from but close to the site
As a rule, site-specific data are always preferred when developing a meteorological data profile for a specific source. However, sometimes this is not possible or other suitable surface meteorological data from other local sources may be available. For simple single-station plume modelling, off-site data should only be used if the nearby site has similar topographic characteristics which are likely to result in similar meteorological conditions for the site concerned. For example, when both sites are located in the same valley system, or in close proximity along a coastline. The representativeness of off-site data must be established before being used in any dispersion study.
c Where to get raw data from
The three principal sources of meteorological data are:
- Climate Database (CLIDB)
- New Zealand Meteorological Service
- regional councils, which operate ambient air quality monitors.
The Climate Database (CLIDB) (http://www.niwa.co.nz/services/clidb) is administered by NIWA in Wellington. Raw data can be downloaded if you are a registered user and familiar with structured query language (SQL). Otherwise NIWA CLIDB staff can download data for an administrative fee. Data are available through a subscription-based web service (CliFlo). Ad hoc or complex data requirements can be requested via the website's 'climate-enquiries' link.
5.2.4 Limitations associated with developing meteorological data sets
Limitations associated with developing meteorological data sets include the treatment of missing data and calm or stagnant conditions. These require careful consideration.
Gaussian-plume models assume that concentrations of pollutants are inversely proportional to wind speed, therefore concentrations become unrealistically large as wind speeds approach calm conditions. Two of the commonly used Gaussian-plume models deal with calms in the following manner.
- AUSPLUME calculates pollutant concentrations for a minimum wind speed of 0.5 m/s. Wind speeds in the model's meteorological data input file that are less than 0.5 m/s are substituted with a wind speed of 0.5 m/s.
- ISCST3 calculates pollutant concentrations for a minimum wind speed of 1 m/s. However the criterion of l m/s wind speed is referenced to the point of release (i.e. stack height). Wind speeds generally increase with height above the ground and ISCST3 recognises this. Consequently, depending upon the height of the stack and stability conditions, the wind speeds at point of release may be higher than those recorded in the meteorological data input file (which are generally taken at a reference height of 10 m). ISCST3 does not calculate pollutant concentrations for wind speeds of less than 1 m/s at release height and assigns the concentration for a wind speed of 1 m/s to any hours in the data set where the wind speed is between 0.5 and 1.0 m/s. Any wind speed less than 0.5 m/s is treated as invalid data.
Neither model treats low wind speeds in a realistic manner and effectively throws away worst-case dispersion conditions for many types of sources. If all hours of wind speed less than 0.5 m/s or 1.0 m/s were treated as invalid/missing data and removed from the data set, this may distort the frequency distribution of predicted concentrations. For example, if worst-case conditions are F stability with wind speeds of 0.5 m/s or less and 1% of the data is treated as invalid/missing, then the 99.9 or 99.5 percentile concentration may be very much lower than it would be in reality. For this reason it is recommended that when using steady-state models, all wind speeds less than 0.5 m/s contained in the meteorological data be set to 0.5 m/s. The amount of adjusted wind speed data must be quantified when presenting the modelling results and the potential implications of the data adjustment must be addressed in the assessment.
The potential effect of low wind speeds on assessments undertaken using Gaussian-plume models depends quite strongly on the nature of local wind flow, and the accuracy (or otherwise) of the hourly average wind direction. In some situations, the wind direction may be steady at low wind speeds (e.g. cold drainage flow down-slope or a land breeze), while in other situations the wind direction may be highly variable over a short time scale. In the former situation, the hourly average wind direction may be quite accurate (i.e. the wind direction is quite steady), and the low wind speed prediction from the Gaussian-plume model may be reasonable. In the latter situation, the hourly average assumption results in an over-estimation as the wind direction meanders over a wide range. At any particular site either situation can probably arise at different times. This situation emphasises (again) the value of local meteorological data.
If calm conditions are recognised as a potential issue for a specific site, an advanced model may be used as these still operate no matter how low the wind speed. CALPUFF assumes that hourly average winds below 0.5 m/s are calms and uses its specific algorithms to deal with them as such. Particle dispersion models, such as that included in TAPM, may give a better picture of dispersion in calm conditions, as they can account for sub-hour fluctuations in the wind and particle distributions are not restricted to being Gaussian in shape. However, there is still some debate on these issues, which remain unresolved.
When modelling with steady-state models, all wind speeds less than 0.5 m/s contained in the meteorological data set must be:
a) quantified and reported when presenting modelling results
b) set to 0.5 m/s before modelling.
The implications of not being able to model calm conditions must be addressed in the assessment.
Where maximum concentrations are predicted for low wind speeds, local meteorological monitoring is highly desirable. Use this data to resolve questions about the variability of wind direction and the accuracy (or otherwise) of the hourly average wind direction.
b Missing data
Most meteorological processing programs and air dispersion models require a full data set of all parameters for all hours. Missing data must be replaced with synthesised data to ensure that the air dispersion models can function. Where there are only one or two hours of missing data, linear interpolation of the data is acceptable.
In New Zealand, with many remote automated surface weather stations, longer periods of missing data (in the order of weeks) may occur. For periods of up to seven days, synthesised averages from a longer-term record of the station may be substituted into the data set. For continuous periods of longer than seven days, the data should be considered to be missing and the length of the data set reduced by the length of the missing data. For example, with three weeks of continuous missing data, the total length of the data would cover 49 weeks instead of a standard 52 weeks. It is important, however, to ensure that an adequate coverage of all seasons is obtained within the data.
All missing or synthesised meteorological data should be clearly documented and discussed in the method.
Periods of missing data that are less than seven days in length may be replaced with synthesised data produced from long-term seasonally adjusted records.
Periods of missing data that are longer than seven days in length must be recorded as missing data.
5.2.5 Derived meteorological parameters
For steady-state plume modelling there are two key meteorological parameters that are not likely to be directly measured and are required for single-station meteorological files only: stability and mixing height.
a Stability classification schemes
Atmospheric stability is a measure of the propensity for vertical motion and hence is an important indicator of the likely magnitude of pollutant dispersion.
A simplified measure of stability was developed by Pasquill (1961) and later modified by Gifford. This is called the Pasquill-Gifford (PG) Stability Classification and is based on a fairly restricted set of measurements of an unspecified averaging time. The measurements were made in the 1950s. This classification consists of six classes, which include A (extremely unstable), B (moderately unstable), C (slightly unstable), D (neutral), E (slightly stable) and F (moderately stable) (see Appendix A).
In 1967 Turner developed a classification scheme based on the original Pasquill-Gifford scheme. This consists of seven classes including 1 (extremely unstable), 2 (unstable), 3 (slightly unstable), 4 (neutral), 5 (slightly stable), 6 (stable) and 7 (extremely stable).
These classification schemes assume that stability in the layers near the ground are governed by convective fluxes from solar radiation (day), cloud cover (night) and mechanical fluxes from wind speed. Although there are more superior dispersion coefficient schemes, most models offer the P-G dispersion coefficient scheme due to its long and relatively successful history of use.
The Pasquill-Gifford stability classification scheme can be used unless an alternative method can be shown to produce more accurate results.
Full details of methods used to assign stability classes using routinely monitored meteorological data are given in US EPA, 1987.
b Mixing height
The mixing height or mixing depth is the height to which the atmosphere is uniformly mixed. Mixing height is determined by either upper atmosphere temperature inversions or wind shear (changes in wind speed with height). Mixing heights have a diurnal variation and rapidly change after sunrise and at sunset (Figure 5.1). Research shows an inverse relationship between pollutant concentrations and mixing height, so mixing height is often used as, and is a critical guide of, the pollution potential in an area (Oke, 1987). Dispersion model predictions can be highly sensitive to changes in mixing height.
If a plume penetrates up through, or is released above, the mixing height, the pollutants will be trapped aloft and their effect will not be observed at ground level. If a plume is trapped within a shallow mixed layer the vertical dispersion will be limited and high ground-level concentrations are likely to occur.
Four methods that are commonly used to determine mixing height are:
- derivation from upper air data (e.g. radiosonde measurements)
- ground-based remote sensing (e.g. Doppler SODAR)
- derivation from routinely measured surface meteorological data (e.g. using a US EPA meteorological pre-processor model such as RAMMET)
- using a prognostic meteorological model (e.g. TAPM, see section 5.3.2).
Determining mixing height is usually an expensive and complex task requiring considerable expertise and should therefore not be undertaken lightly. The uncertainty of mixing heights determined by the methods referred to above increases in the lowest level of the atmosphere. It is generally accepted that mixing heights determined to be less than 50 m contain a significant degree of uncertainty.
When mixing height data are required but not available, determine if the model results are sensitive to changes in mixing height by undertaking a sensitivity analysis of the model results to this parameter.
When it can be demonstrated that mixing height data are not a critical parameter, use data that are likely to be representative of the patterns expected in the area of interest.
When mixing height are a critical parameter, derive these data set using the following hierarchy of methods:
a) ground-based remote sensing
b) derivation from upper air data
c) derivation from routinely measured surface meteorological data
d) estimation using a validated meteorological model.
Set the minimum mixing height in a meteorological data set to 50 metres unless there is evidence to show that mixing heights of less than 50 metres do actually occur.
5.2.6 Meteorological conditions that Gaussian-plume models cannot account for
In situations of complex terrain or near coastal boundaries, meteorological conditions such as calms, coastal fumigation, sea/land breeze re-circulation, and mountain and valley winds can significantly affect the dispersion of pollutants. These meteorological conditions are highly complex in a spatial (vary quickly from place to place) and temporal (vary within periods of minutes rather than hours) sense.
Gaussian-plume models cannot account for these meteorological conditions adequately because of the steady-state formulation (which assumes uniform meteorological conditions) and their inability to retain a memory of the preceding hour's emissions. The following examples highlight meteorological conditions that Gaussian-plume models cannot adequately simulate and for which an advanced dispersion model should be used instead.
a Calm and low wind speed conditions
Under stable, high-pressure synoptic (large-scale) weather conditions, calm conditions often occur near the ground, especially at night and early morning. These stable conditions can often result in elevated pollution episodes as vertical and horizontal mixing of the lower boundary layer is inhibited. Calms are of particular concern when dealing with sources that release contaminants close to the ground or when looking at airshed systems. Gaussian-plume models break down during low wind speed or calm conditions due to the inverse wind speed dependence of the steady-state plume equation, and this limits their application. How to deal with calm conditions when using Gaussian-plume models is discussed in section 5.2.5(a).
Temperature inversions are caused by a number of different mechanisms. Surface or ground-based inversions often occur on clear, cold nights when there are low wind speeds. Under these conditions the ground cools more quickly than the air immediately above it, causing a pool of cooler, more dense air to accumulate at ground level. These ground-based inversions occur frequently throughout New Zealand, especially in hilly terrain. If temperature inversions develop in a valley, pollutants can often be trapped under the inversion layer and result in high pollution episodes. The break-up of a surface inversion is shown in Figure 5.2, where the layer of air near the ground (at 1000 hPa) heats up during the day by radiative heating while the upper air temperature (above 900 hPa) remains relatively constant.
An advection inversion often occurs when warm air passes over a cooler surface, which can result in the development of a low-level inversion and the formation of ground-level fog. This type of inversion occurs less frequently in New Zealand.
A subsidence inversion or upper air inversion develops within a high-pressure system when the subsiding air is compressed and the upper air becomes warmer than the air below.
Inversion conditions are difficult to simulate with Gaussian-plume models, due to associated low wind speeds (see section 5.2.7a), the appearance of multiple layers of pollution, and the difficulty of defining the mixing height.
c Fumigation during inversion break-up
Nocturnal ground-based and upper air inversions start to break up during the first few hours of sunlight as the net heat flux (heating of the surface by the sun) becomes positive. As the ground heats from below, convective mixing (heat transfer) takes place, effectively breaking up the ground-based inversion from below. The growing vertical eddies (caused by heating) mix the air above the surface inversion down to ground level, a process called inversion break-up.
Inversion break-up fumigation is the process whereby pollutants emitted above the inversion layer during the night are fumigated down to the ground during this break-up process. Inversion break-up fumigation is often associated with very high pollutant concentrations at some distance from the source. This process is fairly transient, taking place over tens of minutes and typically during mid-morning (Kerman et al., 1982).
The transient nature of these events, and the difficulties that Gaussian-plume models have identifying different layers aloft and the interaction between layers, makes fumigation during inversion break-up an issue to be wary of when using these models. However, one Gaussian-plume model, SCREEN 3, the US EPA's screening version of ISC3, does incorporate code specifically written to enable it to provide estimates of maximum concentrations during inversion break-up.
d Sea- and land-breeze circulations
Because land surfaces heat and cool quicker than the sea or other water bodies, temperature gradients develop that can result in the generation of localised wind flows (Figure 5.3). A sea breeze develops during the day as the air over the land warms more quickly than the air over the sea. It rises, bringing in an onshore breeze, with a return flow aloft. At night the opposite occurs and a land breeze develops, flowing towards the sea under an area of subsidence.
Sea breezes are generally strongest during the day in summer and land breezes strongest during winter nights. They can both have significant effects on air quality over urban areas, as they are recirculating air currents that can return pollutants (instead of remove them) to an area from which they were released earlier in the day.
e Mountain-valley winds
Mountain and valley winds are generated due to similar heating and cooling mechanisms to sea-land breezes. During the day the air above a slope is heated and becomes warmer than neighbouring air at the same height above sea level, but further above the ground. It rises due to convection, and upslope mountain winds occur (Figure 5.4). At night the mountain slopes cool more quickly than the surrounding air, and the cool air drains down the slope, generating valley winds. This heating and cooling often results in closed circulation patterns, which can trap and/or recirculate air pollution in the mountain-valley system.
Where there are meteorological conditions that Gaussian-plume models cannot account for, an advanced model (which is more capable of handling these conditions) should be used.
5.3 Meteorological data for advanced dispersion models
Advanced dispersion models require more complex meteorological data than steady-state models. This includes inputs from surface networks (land and sea) and upper air stations. Because there will not be meteorological sites at every point on the ground in the modelling domain, and monitoring in the upper air (anything above the height of a tower) is normally very sparse, meteorological models must be used to provide this 'missing data'.
There are two different types of meteorological model that can be used to provide a three-dimensional grid of meteorological data:
- diagnostic wind models (DWM), which interpolate and/or extrapolate meteorological observations
- prognostic models, also known as a 'mesoscale' models.
The meteorological model outputs are then used to drive a dispersion model.
Meteorological models can either form part of an air dispersion modelling system (e.g. CALMET provides meteorological fields for CALPUFF, RAMS for HYPACT, and TAPM calculates both the meteorology and dispersion), or they can be stand-alone. Prognostic models (from mesoscale to global scales) are used to provide national weather forecasts. RAMS is used worldwide for this purpose; LAPS - whose analyses are used to drive TAPM - is used in Australia. The NZ Meteorological Service uses MM5 for mesoscale weather forecasts.
Meteorological models of the type described in this section - and their associated dispersion models - have rarely been used in New Zealand for regulatory impact assessments, largely because:
- the models have not been user friendly and needed large computing resources to run them
- the network of meteorological stations for input data to a diagnostic model (especially upper air) is relatively sparse
- the format for data storage is sometimes not compatible with that required by the model.
More recently a number of advanced dispersion models have been released that are much more user friendly. Efforts are always being made by developers to enable models to run faster, and with increased computing power available it is becoming feasible for all users to run these models on a modern personal computer. It is the rapid increase in computing power over recent years that has resulted in an increase in the number of people using these tools.
Examples of diagnostic and prognostic meteorological models are provided in the following two sections. Recommendations outlining which meteorological models to use in the New Zealand situation are given in section 5.3.5.
5.3.1 Diagnostic meteorological models
Diagnostic meteorological models use data from available locations and assign values to the meteorological variables throughout a three-dimensional grid by interpolation and extrapolation. The conservation of mass principle is applied throughout the process. The term 'diagnostic' is used because the input data and model results are for the same time period. Diagnostic models are not predictive, and their calculated fields for each time interval do not depend on fields at previous times. The model's output is a data file in a format required by a particular air dispersion model.
An example is CALMET, the pre-processor to CALPUFF. In recent years CALMET has been increasingly used in the USA and Australasia, and is used here to illustrate the features of a diagnostic meteorological model.
The CALMET meteorological model (Scire et al., 2000) is a diagnostic meteorological model developed as a component of the CALPUFF modelling system for use in air quality applications. CALMET in its basic form is designed to produce hourly fields of three-dimensional winds and various micro-meteorological variables based on the input of routinely available surface and upper air meteorological observations only. CALMET consists of a diagnostic wind field module and micro-meteorological modules for over-water and over-land boundary layers.
The diagnostic wind field module uses a two-step approach to the computation of the wind fields (Douglas and Kessler, 1998). In the first step, the initial-guess wind field is adjusted for terrain effects to produce a step 1 wind field. The second step consists of an objective analysis procedure to introduce observational data into the step 1 wind field to produce a final wind field, the step 2 wind field. Some of the advantages and disadvantages of this model are detailed below.
Advantages of CALMET
- Observations can be incorporated into the model, to produce realistic meteorological fields.
- CALMET can reproduce fine-scale effects (down to a couple of hundred metres' resolution) and still maintain efficient model run times on a personal computer.
- Output from the prognostic meteorological models such as MM5 and TAPM can be incorporated into the CALMET run, providing information in data-sparse regions. This combined approach is the preferred way of operating CALMET.
Disadvantages of CALMET
- The CALMET/CALPUFF system is technically more advanced than a plume model and is perceived as being difficult to regulate and complex to use.
- Routine meteorological data are sparse in New Zealand.
- There are potentially extra costs of running CALMET.
With regard to ease of use, CALPUFF can be run in a steady-state mode using the same meteorological data that are required to run AUSPLUME or ISCST3. The minimum requirements for CALMET are similar to those for the steady-state models. An ISCST3 or AUSPLUME meteorological file can be used to drive CALPUFF, but again there is the option for a more refined treatment when it is necessary and the data are available.
Costs to industry may be higher for a full CALMET/CALPUFF analysis than a simple steady-state analysis. In many cases, though, the differential is very small compared to other fixed costs of a project, and the differential tends to decrease with increasing modeller experience. Also, a more accurate answer can mean large savings in a project, and in some cases can make the difference between obtaining approval for a project or being rejected. Industry will also save costs from the model's ability to handle multiple effects within one model framework; i.e. once set up the modeller can model anything from long-distance visibility to aqueous phase chemistry to plume visibility applications, without requiring the services and set-up costs of another model.
5.3.2 Prognostic meteorological models
Prognostic models are driven by large-scale synoptic analyses and numerically solve the equations of atmospheric dynamics to determine local meteorological conditions. They do not require local meteorological data to run, although if data are available they should be compared with model results to validate the model. Prognostic models are able to represent all scales, from global down to features on scales in the range 1-10 km. Most are run in a nested format with the outer domain covering distances in the order of 500-1000 km - the regional scale.
All domains are initialised using analyses from global or limited-area models, usually run by national weather services. These are provided by many forecasting agencies or similar institutions, such as the US National Meteorological Center, the European Centre for Medium-Range Weather Forecasts, the UK Meteorological Office, or the Australian Bureau of Meteorology. The outer domain is also driven at its boundaries by the global or limited-area models as the run progresses - this feeds in the effects of weather systems to the domain of interest. The prognostic models describe the three-dimensional fields of temperature, wind speed and direction, and moisture through the region at high spatial resolution.
Prognostic models all contain realistic dynamical and physical formulations, and potentially produce the most realistic meteorological simulations for regions where data are sparse or non-existent. The high resolution needed for regulatory assessments means that these models have historically been seldom used as regulatory models. The computing costs of long-term simulations have been prohibitive, although more recently this is less true. And, if local meteorological data are absent, the use of a prognostic modelling system could be a sensible option as part of a regulatory assessment.
RAMS is the most commonly used prognostic meteorological model in New Zealand (Wratt et al., 2001), followed by MM5, ARPS, and (more recently), TAPM.
a RAMS and MM5
RAMS and MM5 are three-dimensional, non-hydrostatic prognostic mesoscale models. MM5 is the fifth-generation NCAR/Penn State Mesoscale model. The model includes a multiple-nesting capability, non-hydrostatic dynamics and four-dimensional data assimilation (Dudhia et al., 1999). MM5 is free to users, while RAMS is subjected to licensing costs. Both models enjoy widespread use throughout the world, are well supported, continually under development, have been used in many studies, and appear regularly in the scientific literature. The main advantages and disadvantages of these models are detailed below.
Advantages of RAMS and MM5
RAMS and MM5:
- have the ability to assimilate local meteorological data
- have realistic dynamical and physical formulations, suitable for simulations in New Zealand's complex environment
- can produce realistic meteorological fields in data-sparse regions
- are flexible enough to couple output meteorological fields to dispersion model runs at any resolution (e.g. RAMS coupled to HYPACT).
Disadvantages of RAMS and MM5
RAMS and MM5:
- have relatively high computational demands
- require a large amount of user knowledge and expertise to produce reliable and convincing results
- do not themselves include dispersion models, and the associated dispersion models do not necessarily comprise all of the features required for regulatory assessments (e.g. building effects).
At present, most prognostic models require significant computer resources to run. They also describe a comprehensive collection of meteorological phenomena and are widely used in meteorological research. However, some features that contribute significantly to the computational cost of mesoscale modelling are not important for air quality simulations, such as gravity waves and complicated microphysical processes. Careful formulation of the model dynamics so as to omit or filter out these features can increase the run speed, enabling longer runs to be contemplated for regulatory applications. This has been done with the CSIRO's TAPM.
TAPM is a PC-based three-dimensional prognostic meteorological modelling system, including various dispersion modules, as described in section 2.2.2. TAPM has a GUI that allows the user to set up and run the model under the Windows operating system. It connects to databases of terrain, vegetation, soil type, sea surface temperature and synoptic-scale meteorological analyses for Australia and New Zealand, as well as most regions throughout the world. TAPM is driven by six-hourly synoptic analyses at approximately 75 km resolution. This database is derived from LAPS analysis data from the Bureau of Meteorology.
Advantages of TAPM
- It is easy to use and completely self-contained, with good visualisation of model results.
- The model output is easy to convert for input into other models, such as CALMET, AUSPLUME, DISPMOD and ISCST3.
- As for any prognostic model, it requires no local data to run, although it has the ability to assimilate local surface meteorological data.
- It is designed to run on a modern personal computer.
- Describes the effects of point, line and volume sources, simulates the effects of buildings on dispersion, and simulates chemical reactions between pollutants.
- Resolution of the pollution dispersion models can be higher than that of the meteorological model - and will usually need to be for regulatory assessments.
Disadvantages of TAPM
- Although easy to use, a high level of understanding of boundary-layer meteorology and pollution dispersion is needed, as with all prognostic model systems, to produce meaningful results.
- The maximum horizontal resolution of the meteorological model component of TAPM is of the order of a 1 km grid-size. If meteorological features are expected, or geographical forcing is present at smaller scales, then the user should take care. Although assimilation of meteorological data is possible, care must be taken to ensure that the meteorological data are representative of the scales modelled by the meteorological model.
5.3.3 Prognostic model output as inputs to Gaussian-plume models
Some prognostic meteorological models produce output data in a format that can be used by plume models. Prognostic model results may be extracted at a single location (the site of pollution emissions) in a format compatible with, say, AUSPLUME or ISCST3, and treated as pseudo-observations for input to the dispersion models. This is a possible alternative if there are no site-specific observations, and has the advantage that there would be no missing data. The pseudo-data would also be compatible with CALPUFF running in a single-site mode. TAPM has options to produce meteorological output compatible with most commonly used Gaussian-plume models.
However, extracting results from a single point to run a plume model ignores many of the advantages of undertaking sophisticated - and more realistic - meteorological modelling, as most information in the prognostic model results would never be used. If a model such as TAPM is being run to produce the meteorological information, then it can be run as a dispersion model at little extra cost.
Care should also be taken when extracting only the mixing height from a single point. Although the model mixing height should be consistent with other model parameters at that location, it may not be consistent with observed parameters (e.g. wind and temperature), which are being used as inputs to the plume model. It would be more realistic to derive the mixing height from meteorological observations.
The practical advantage of extracting single-point meteorological data for a plume model is that the meteorological model need only be run once, no matter how many dispersion model runs are required. As TAPM is a self-contained meteorological and dispersion model, with the two processes running at the same time, the meteorology has to be re-run for each dispersion model case, and this is relatively computer resource intensive. However, other meteorological and dispersion models (e.g. RAMS/HYPACT or CALMET/CALPUFF) carry out the two processes separately. The meteorological data need only be calculated once and this consideration does not apply.
Prognostic model output should only be used as meteorological input data for Gaussian-plume models when:
a) it is appropriate to use a Gaussian-plume model for dispersion
b) there is no other source of meteorological data available.
However, using the TAPM meteorological output in a simple (quick to run) dispersion model can be attractive when you want to quickly test a wide range of options.
5.3.4 A combined prognostic/diagnostic approach
Both the prognostic and diagnostic approaches to meteorological modelling have advantages for the production of realistic meteorological fields for input to dispersion models, as follows.
- Prognostic models do not need local meteorological observations to run, so can simulate the meteorology of regions where few data are available.
- Diagnostic models can incorporate available measurements, and - provided the measurements are interpolated realistically - can potentially produce meteorology close to that observed (indeed, at the monitoring sites the modelled meteorology should be exactly the same as that observed).
Two variations on these approaches may be identified in which each model type incorporates the beneficial features of the other.
a Data-assimilating prognostic models
Prognostic models can take advantage of local meteorological data by the process of meteorological data assimilation. There are several ways of accomplishing this. One common method is known as 'nudging'. Essentially, the prognostic model solution is forced towards the observations during the model run. At best, the model solution is already close, so the forcing is small - hence the term 'nudging'. Nudging can have beneficial effects on the model solution, but must be used carefully. For example, the local meteorological data input to the prognostic model should be representative of the observed meteorology on scales resolved by the prognostic model. If the monitoring site is in complex terrain which is not resolved by the prognostic model grid, then its data should not be assimilated.
b Prognostic model output as input to a diagnostic model
Diagnostic models may be run in data-sparse areas through the incorporation of output from a prognostic model. The prognostic model provides a 'first-guess field', which is then modified by the diagnostic model to take account of terrain or land-use features that are at a smaller spatial scale than the terrain used by the prognostic model. The main purpose of this approach is to increase the horizontal resolution of the meteorological fields, which is necessary if there are important terrain or land-use features at the higher resolution. The procedure is far less computationally demanding than running a prognostic meteorological model (with or without data assimilation) at sub-km resolutions.
It is worth noting that:
- approaches (a) and (b) are not new, but they are discussed here as practical approaches to combining results from prognostic meteorological models (which simulate the atmospheric dynamics according to physical laws) with available meteorological observations
- the combined approaches work in both data-sparse and data-abundant regions
- most, if not all, prognostic models have data assimilation routines, so they may be used in approach (a)
- diagnostic models such as CALMET are set up to combine prognostic model output with meteorological measurements in approach (b)
- meteorological observations may be used twice, being both assimilated into the prognostic model run and used in the objective analysis stage of the diagnostic model run.
The choice between approaches (a) and (b) involves considering the best resolution that may be practically attained in a long-term prognostic model simulation. This should be high enough to resolve the important meteorological features which can only be simulated by a prognostic model, such as land and sea breezes, and developing cyclones and fronts. If this is sufficient to resolve terrain and land-use effects on the local meteorology, then approach (a) is appropriate. If there are, say, terrain-forcing effects, such as blocking, channelling or slope flows which are not resolved in the prognostic simulation, then these may be incorporated using a diagnostic model; that is, following approach (b).
It is important to note that the choice between (a) and (b) and the choice of model resolution (for both prognostic and diagnostic) depends on meteorological considerations only. The resolution of the dispersion model is independent of the resolution of the meteorological model(s), and is generally equal to or (much) higher than the resolution of the input meteorology.
These approaches, despite their potential to produce realistic mesoscale meteorological features (which have important consequences for pollution dispersion), have not been widely adopted in New Zealand. They are becoming more common overseas for regulatory impact assessments, and have been used worldwide for many years for scientific research.
Approach (b) involves combining the three-dimensional prognostic model output with meteorological observations (from the surface and from vertical profiles) in a diagnostic model. A variation on this uses a set of key, user-selected vertical profiles, extracted from the prognostic model results, and used as if they were observations in the diagnostic model. The diagnostic model extrapolates to provide three-dimensional fields. The extracted profiles are used in place of the full three-dimensional prognostic model fields. However, three-dimensional model output can occupy an extremely large amount of disk space - files need to be as text rather than binary format so they can be read by a different model. For example, TAPM allows this variation of approach (b) in the extraction of profiles for input to CALMET as pseudo-data.
However, the latest version of TAPM (v 2.0) allows output of the full three-dimensional meteorological fields (as text), which may be converted by the user and read by CALMET.
c Hazards associated with combining model results with observations
Careful checking needs to be carried out with the approach described in section 5.3.4a. Data assimilation works well if the model prediction at the data point is already close to the observation at that location. If this is not the case, the model solution at surrounding and downwind grid points can become nonsensical.
It must be assumed that over a 12-month period the prognostic model will not predict some days well in (probably) all regions. If the intention is to run a dispersion model for 12 months and examine annual statistics, it may be safely assumed that the meteorological model will predict the right types of weather and at the right annual frequency, even if not on the correct day all the time. It is perhaps safer to use the observations to validate the modelled meteorology, rather than assimilating them and potentially generating unrealistic model results. Extra care must be taken if the dispersion modeller wishes to use the meteorological model to simulate a particular day. In that case, the meteorology has to be correct and must be validated against suitable observed data.
Similar considerations apply when adopting the approach in section 5.3.4b. If the prognostic model output used in the initial-guess phase of CALMET's wind field calculation differs from the observations used in the subsequent objective analysis, the resulting wind field will be unrealistic. This can occur particularly if the prognostic model does not resolve terrain effects which are resolved by CALMET. If observations are plentiful, a more realistic wind field may be obtained without the prognostic fields as an 'initial guess'. If scarce, it could be safer to run CALMET in a 'no-observations' mode, where the wind field is a perturbation of the prognostic model output.
5.3.5 The future use of non-steady-state meteorological data in New Zealand
As indicated earlier in the document, an important difference between Gaussian-plume dispersion models and more advanced dispersion models is in their requirements for meteorological data. Advanced dispersion models require fully three-dimensional, time-dependent meteorological data (i.e. 'non-steady state'), which are provided by advanced meteorological models such as TAPM, MM5 and CALMET.
In New Zealand, the creation of non-steady-state meteorological data sets has been mainly carried out by scientists as part of research programmes, rather than consultants with more limited time and resources. This situation is gradually changing, as advanced models are steadily becoming more widely used. This is encouraging, as many dispersion-modelling exercises ought to be carried out using advanced models rather than Gaussian-plume models. Criteria for deciding the kind of model to use have already been discussed.
Once it has been decided that your project requires an advanced dispersion model - and therefore requires three-dimensional meteorological fields - there are several factors to consider when deciding on the most appropriate meteorological model. In other words, there are still some New Zealand-specific issues to address.
For one thing, many areas of New Zealand have very few surface meteorological data sites, and there are only three routine radio-sounding sites providing vertical profiles (Whenuapai, Paraparaumu and Invercargill). This poses a challenge when running a diagnostic wind model using observational data only.
New Zealand's complex terrain poses a different challenge for prognostic meteorological models. As already discussed, attempts at the combined prognostic/diagnostic approach may lead to problems when the prognostic model results are not consistent with observations.
The dispersion modeller, though acknowledging it is necessary, may still be daunted by the task of meteorological modelling. However, in time, as the modelling community in New Zealand becomes more experienced, the consequences of these issues will become better understood, and will be accounted for at the reporting stage. Also, meteorological models are continually improving, and in future they will be able to better handle meteorological conditions in complex terrain and coastal areas.
Finally, it is unlikely there will be an increase in the number of routine meteorological sites around New Zealand.
When carrying out non-steady-state meteorological modelling:
a) assess the availability of meteorological data in the region to be modelled
b) consult a topographic map of the region to gauge its geographical complexity
c) determine what spatial resolution is likely to be required
d) decide whether the required resolution is feasible in a prognostic model
e) consider the prognostic model approach if feasible, or else the combined prognostic/ diagnostic approach
f) consider the diagnostic approach alone if meteorological data are abundant
g) take care when assimilating observations into a prognostic model in regions of complex terrain
h) take care when incorporating observations into the diagnostic stage of the combined approach in regions of complex terrain.
These recommendations give only a general indication, and modellers should be guided by their own experience and expertise.
If advanced modelling is necessary, but considered too onerous by the modeller, then the meteorological component should be contracted out, rather than avoided through the use of a Gaussian-plume model.