1. Topic

2. Introduction

Air quality forecasting is a natural extension of the knowledge and experience built up from air quality (AQ) assessment and modelling. In its most general form it entails the prediction of air quality on any time scale, from hours to years, and on any spatial scale, from street to global. When considering short-term air quality forecasts, one is limited to forecasts of a maximum of 2 days, with the most common time frame being 12 to 24 hours. Forecasts on this time scale can be local, regional or even global though the most common use of short-term air quality forecasts is on the urban scale. AQ forecasts can be used for several applications:

• As a warning system to reduce the risk of exposure of high-risk categories, e.g. asthma patients.
• As input to planning short-term abatement strategies e.g. redirection of traffic, reduction of high traffic speeds, limitations on emissions etc.
• As public information to encourage a self-regulating abatement system, e.g. recommendations for citizens to reduce emissions by using public instead of private transport.
• As emergency warning systems for accidental release of pollutants, e.g. nuclear power plant accidents, oil refinery flares etc.

Forecast systems are of essentially 2 different types, statistical forecasts or prognostic (dynamic) forecasts. Combinations of the two are also possible. Statistical forecasts are based on observed relationships between air quality and some meteorological predictors, e.g. wind speed and direction, temperature or stability. Prognostic forecasts, on the other hand, use chemical transport models (CTMs), or other types of mathematical/parameterised (simplified) models, to make prognosis of the concentration of pollutants based on forecasted meteorology.

Short-term forecasts can be made on both regional and local scales. Often, when long range transport of pollutants such as ozone or dust are important, there is a need to predict on both scales, the regional scale forecast being used as input for the local scale forecast. Most forecasting systems, however, do focus on the urban scale. When statistical models are used then the urban region is generally treated as a whole, the forecast being given one value for the entire city. Prognostic models, on the other hand, can give predictions down to street scale. Gridded models can produce air quality forecast maps down to resolutions of around 1 km.

Both statistical and prognostic forecasting systems can be implemented as short-term forecasts on both regional and urban scales. The choice of which system to use is dependent on local conditions, the availability of historic air quality data and the knowledge and experience to run prognostic CTMs.

3. Discussion

Statistical model implementation

Statistical models require a well established relationship between measured air quality and meteorological predictors, which can be used to build up the statistical model. This type of forecasting is most effective when there are clear relationships between local meteorological predictors and air quality. Statistical models are most effective when long range transport is not a factor, i.e. local emissions dominate.

There are several types of statistical models. The simplest forms are persistence and climatology. Persistence is the most relevant and simplest method for short-term forecasts and requires little or no effort. More advance statistical methods, such as criteria selection, decision trees, regression functions and neural networks can also be used. Statistical models do not generally provide concentration values, but pool air quality into indexes, usually 3 or 4 levels of air quality.

The following steps are required for the implementation of statistical forecasts:

Obtain an archive of monitoring data for the desired forecast components. Minimum of 2 years.

• Have access to measured or modelled meteorological data for the corresponding archive period. Models used are generally synoptic scale models, e.g. HIRLAM and ECMWF where the data can be downloaded via web or ftp portals.
• A statistical model can be established using the relationships between air quality and the meteorological model. The relationships should be established between the observations and the meteorological forecast model to be used as this gives the most direct link.
• The model must be tested, validated and improved over at least a 1 year period. Continuous improvements can be carried out, depending on the statistical model used.

Prognostic model implementation

Prognostic models can come in many forms, dependent on the application. Accidental releases are often modelled with parameterized plume models while urban scale forecasts generally use parameterised line source and/or gridded Eulerian models, coupled with meteorology to produce forecasts of air quality.

Prognostic models give a spatial and time dependent distribution of pollutants and require predictions for emissions, as well as meteorology, to determine concentrations through a prognostic CTM. They do not require large amounts of observational data and can be used, but not validated, where no observational data is available. In principle the same guidelines laid out for air pollution modelling (see the topic descriptions on modelling) can also be applied in regard to prognostic forecasting.

The following steps are required for the implementation of prognostic forecasts:

• Access to and expertise in meteorological forecast models. This includes synoptic scale models, e.g. HIRLAM and ECMWF (see links in the web section below) where the data can be downloaded via web or ftp portals, and meso-scale models, e.g. MM5 and RAMS, which would usually require in-house capabilities to run. The meteorological side of air quality forecasting is often undertaken by the regional meteorological authorities.
• Access to and expertise in CTMs
• An extensive emissions database is required
• The model must be tested, validated and if required improved over at least a 1 year period.
• If important aspects of air quality are regional, e.g. ozone, then input from a regional CTM may be required

Abatement strategies

Abatement strategies used in conjunction with short-term forecasts are often related to traffic, if this is the major cause of poor air quality. These include for instance redirection of traffic, reducing speed limits at high-speed roads, and limiting the number of cars on the road. The general concept is that forecasts are required at least 1 day in advance in order to organize and inform the public of any changes. Experience with testing the effectiveness of such strategies, or even implementing them, is very limited. Such measures have an associated economic cost and so trust in the forecast system and a clear understanding of the benefits is also required. A specific topic description is dedicated to the possible short term measures: “Short Term Planning and Actions”.

Short-term measures can also be applied to industrial sources, when these contribute significantly to the air pollution concentrations. These may include reducing or shutting down production facilities. Such measures often have a more demonstrable effect.

Forecast quality

There are currently no official guidelines available for measuring the quality of air pollution forecasts. However several scoring systems such as skill measurement, which judges the predictive ability of the forecast in reference to persistence, and indicators such as the number of ‘false alarms’ can be used. It is clearly important from a public trust point of view that forecasts are perceived to be accurate. If this is the case then the implementation of abatement strategies is more likely to succeed.

4. Recommendation / Conclusion

· Though statistical models are simpler to implement and can often function well once tuned to specific sites, conditions and emissions, they cannot be extended beyond their already defined scope. So, a statistical model cannot be used at another site, under a different climate or address changes in emissions. As such they are limited in their application, e.g. they cannot be used for scenario calculations.

· In contrast, prognostic models that contain the physical description of the processes can be used at any location or for any emission scenario. Their disadvantage is their need for accurate input in the form of meteorology and emissions. They also require more effort and expertise to implement but their use as a forecast tool is recommended above statistical models if the models and expertise are available for implementation locally. It is important to note that such models can also be directly applied to other aspects of air quality assessment.

· An important aspect of air quality forecasts is their perceived and actual reliability. The use of short-term abatement strategies, which often have associated economic or personal disadvantages, to improve air quality should only be applied once the trust in the forecast system is high. The use of forecast systems as warnings for public health is less sensitive and more wide-spread.

· Short term air quality forecasting implies the prediction of short term emission sources behaviour and the use of appropriate air quality models. Details on such groups of models are given in the “modelling” sections of the INTEGAIRE Best Practice database.

5. Examples / Further Reading

Short term air quality forecasting in Bristol

Short term AQ forecast methods in Seville

PM10 AND O3 Forecast bulletins for the Veneto Region (I)

Short term air quality forecasting in Oslo

A proposal for a short term AP forecasting system for individual planning of urban travel routes

6. Additional Documents / Web Links

· Guidelines for Developing an Air Quality (Ozone and PM_2.5) Forecasting Program, from EPA in U.S.A. (English) http://www.epa.gov/airnow/aq_forecasting_guidance-1016.pdf

· French based European regional forecast using the Prev’air system (French): http://prevair.ineris.fr/

· Better City Air. Air quality forecasts for Norwegian cities (Norwegian): http://www.luftkvalitet.info/

· The Australian Air Quality Forecasting System (English): http://www.dar.csiro.au/information/aaqfs.html

· Danish based multi-scale air quality forecast system Thor (English/Danish): http://www2.dmu.dk/1_viden/2_Miljoe-tilstand/3_luft/4_Spredningsmodeller/5_thor/default_en.asp

· European Centre for Medium-Range Weather Forecast (http://www.ecmwf.int/) provides 3-6 day meteorological forecasts for Europe. These can be ordered for an area at this link: http://cobranett.no-ip.info/meteo/ecmwfk1.htm

· The HIRLAM synoptic scale model for regional weather forecast: http://met.no/english/r_and_d_activities/method/num_mod/hirlam.html

Last Updated

25th January 2005