7 National Environmental Research Institute, Roskilde, Denmark. 8.1 Introduction. The central purpose and scientific focus of GLOREAM, which stands for.
Chapter 8: Global and Regional Atmospheric Modelling Overview of Subproject GLOREAM Peter J.H. Builtjes1, Carlos Borrego2, Ana Cristina Carvalho2, Adolf Ebel3, Michael Memmesheimer3, Hans Feichter4, Annette Münzenberg5, Eberhard Schaller6 and Zahari Zlatev7 1
TNO-MEP, Department of Environmental Quality, Apeldoorn, The Netherlands University of Aveiro, Portugal 3 University of Köln, Germany 4 Max-Planck Institute of Meteorology, Hamburg, Germany 5 DLR, Bonn, Germany 6 Brandenburg Technical University, Cottbus, Germany 7 National Environmental Research Institute, Roskilde, Denmark 2
8.1 Introduction The central purpose and scientific focus of GLOREAM, which stands for GLObal and REgional Atmospheric Modelling, was the investigation - by means of advanced and integrated modelling - of the processes and phenomena which determine the chemical composition of the troposphere over Europe and on a global scale. An essential part of GLOREAM has been the development of stateof-the-art models and the determination of the model capabilities and model performance. The interaction between the different modelling groups and modellers in Europe to discuss problems, to exchange information and to share knowledge formed the heart of GLOREAM. The modelling subproject GLOREAM was the successor of the EUROTRAC, first phase, subprojects EUMAC - European Modelling of Atmospheric Constituents - and GLOMAC - GLObal Modelling of Atmospheric Chemistry. In EUMAC the focus had been on the development of an advanced model hierarchy for air pollution dispersion simulations. The central model was the three dimensional (3-D) long-range transport model EURAD covering Europe, which focused on episodic simulations. The aim of GLOMAC was the development and application of 3-D models of the global troposphere and the lower stratosphere focusing on ozone and acidification. The central models were MOGUNTIA, a relatively simple global model, and the on-line model ECHAM and the off-line model TM (Ebel et al., 1997). GLOREAM was built on the experience gained in both projects and continued working in further model development and application (Builtjes et al., 1999). GLOREAM was structured according to the following five, closely related working groups: ·
model investigation and improvement on a European scale,
·
model investigation and improvement on a global scale,
140
P.J.H. Builtjes et al.
·
computation aspects,
·
model evaluation and validation,
·
model application and assessment.
The EUROTRAC-2 Synthesis and Integration Project considered the research performed in the continuum Science-Tools-Application, which also shows the progress and sequence in time of research. Looked at in this way the first two working groups are more science oriented, the next two belong to tools and the final one is application. In GLOREAM a workshop was held every year, so that including the kick-off meeting a total of seven workshops have taken place. These workshops were in general attended by around 40 GLOREAM principal investigators and several guests. The number of peer-reviewed papers which have been written in the framework of GLOREAM was about 120 and about 15 theses. GLOREAM has also worked on capacity building. Quite a number of young scientists gave their first talks at GLOREAM workshops and many fruitful cooperations between institutes across Europe have been established. In the following sections, a summary is given of the results obtained according to the five GLOREAM tasks.
8.2
Model investigation and improvement on a European scale
Major progress has been made in GLOREAM concerning the development and improvement of 3-D regional chemistry-transport (CTM) models for the European scale. The modelling systems have been used to investigate the physical and chemical processes and phenomena which determine the chemical composition of the troposphere over Europe. New chemical and physical mechanisms have been implemented into the 3-D CTM models. The impact of the new parameterisations have been scientifically analysed to assess the performance of the models under different chemical and dynamical regimes for the European scale. All of the European scale 3-D Eulerian grid models can now do long-term calculations and some of them are used for daily air pollution-ozone forecasts available for the public on the Internet. More specifically, major work has been done with emphasis on the topics given below.
8.2.1 Input data: land use, emissions Improvement of input data such as land use data, topography, biogenic and anthropogenic emissions including the development of new tools and methods mainly based on Geographic Information Systems and Relational Databank Management Systems, has taken place. Considerable work has been done to use the harmonised anthropogenic emission data sets generated by the
GLOREAM
141
EUROTRAC-2 subproject GENEMIS for nested modelling applications. The impact of additional source categories has been investigated, e.g., ship emissions. Considerable effort has been undertaken to develop emission inventories for particulate matter.
8.2.2 Boundary values and treatment, coupling to hemispheric scale and stratosphere Improved treatment of lateral and vertical boundaries has been achieved by the coupling of regional limited area, European scale models to the hemispheric or global scale and their extension to the stratosphere to account for intercontinental transport as well as stratosphere – troposphere interaction. The model calculations contributed to a better understanding of the origin of air masses with enhanced ozone concentrations in the free troposphere over Europe. The origin of air masses can be traced back to the polluted planetary boundary layer over North America (lifting in a warm conveyor belt) or to the stratosphere (stratospheric intrusion). The model results have been compared to observations available from the TOR-2 measurement sites and to satellite data (CRISTA). Especially in the subproject EXPORT-E2, models and model experience from GLOREAM contributed to the analysis of the European ozone budget and the impact of hemispherical transport.
8.2.3 New chemical schemes, improved treatment of particulate matter Development, implementation and investigation of new chemical schemes, in particular for the heterogeneous phase to take account of the formation and transport of particulate matter in the troposphere and its interactions with the gas and aqueous phase. Major effort has been undertaken to include the chemical composition of the atmospheric aerosols within an ammonium-sulphate-nitratewater system and the formation of secondary organic particulate matter from gaseous precursors. Aerosol dynamics have been introduced into most of the 3-D CTMs normally using a modal approach.
8.2.4 Long-term simulations, model intercomparison study, daily air pollution forecast All models are able to do long-term calculations with an hourly output of atmospheric composition. Most of them include photochemistry and aerosols. A model intercomparison study focussing on the aerosol composition and including six 3-D CTMs of different complexity has been undertaken for the European scale for the growing season in the year 1995 (April – September). The results presented in Figure 8.1 show a considerable scatter in the modelled versus observed aerosol SO4. The same holds for total NO3, with a tendency to overprediction.
142
P.J.H. Builtjes et al.
GLOREAM AEROSOL INTERCOMPARISON STUDY 10,0 DEM EURAD/FFA EUROS
SO4 Model (ug/m3)
LOTOS MATCH REM3
1,0
Station Means 2.4. - 29.9.95 0,1 0,1
1,0 SO4 Observations (ug/m3)
10,0
10,0
TNO3 Model (ug/m3)
Station Means 2.4. - 29.9.95
1,0 DEM EURAD/FFA EUROS LOTOS MATCH REM3
0,1 0,1
1,0 TNO3 Observations (ug/m3)
10,0
Figure 8.1. Comparison of observations from EMEP stations and model results from the aerosol model intercomparison study for the growing season in 1995. The numerical simulations include photochemistry and treatment of particulate matter in all of the participating 3-D Chemistry Transport models.
These results also indicate that although clear progress has been made in aerosol modelling, there is also much more research and development needed before these models can be used to analyse abatement strategies.
GLOREAM
143
GLOREAM models have also been used in a model intercomparison study carried out within the TOR-2 subproject for the year 1997 aiming at the analysis of measured ozone trends. The response of the different models to emission scenarios for 1997 and 1987 has been calculated. Although all models show a similar behaviour, there is a difference in the magnitude of the response due to the emission changes. It could be shown that the ozone maxima are reduced considerably due to emission reduction whereas the average ozone concentrations show a reduction in the Mediterranean region but not in central and western Europe. More detailed information of this ozone model intercomparison study is described in the TOR-2 chapter of this book (Chapter 13). Some of the models are used for daily short term prediction of air pollution and most models are used as tools for abatement strategies, also with respect to the European directive 96/62 and its daughter directives on the improvement of air quality. Some of the models participate in the City Delta intercomparison study within the framework of the “Clean Air for Europe” programme (CAFE) recently established by the European Commission. Long-term applications, daily prediction and the comparison with observations in general lead to an improved scientific basis for the application of the models within the framework of the EU directives on air quality.
8.2.5 Nested modelling, horizontal and vertical grid refinement Nested modelling systems have been developed from the hemispheric scale down to the urban scale, even down to street canyon modelling. The impact of increased horizontal and vertical resolution of the models has been investigated. In particular, primary short-lived pollutants show a qualitative improvement of concentration patterns and maxima with refined vertical and horizontal resolution. Nested models have been applied for the planning and analysis of measurement campaigns (e.g., for the Milan and Berlin area).
8.2.6 Chemical and physical processes Models have been used to specify the physical and chemical processes determining the chemical composition of the troposphere for different meteorological and chemical regimes in different regions of Europe. It could be shown that central Europe/northern Italy show high net production of ozone and large vertical outflow, whereas the Mediterranean areas show relatively small net production of ozone and relatively small vertical and horizontal outflow. The role of the vertical transport caused by the Alps on the mass balance of ozone, PAN, other photo-oxidants and its precursors have been investigated. It could be shown that large amounts of precursors can be transported to elevated layers of the atmosphere due to the alpine circulation systems. This contributes considerably to the formation of ozone and other photo-oxidants at higher altitudes. Other studies aim at the investigation of the impact of future climatic changes (temperature, precipitation and humidity) on the air pollution levels.
144
P.J.H. Builtjes et al.
An example is shown in Figure 8.2. The Danish Eulerian Model DEM has been used to calculate the impact of a climate change scenario for the last 30 years of the 20th century on peak ozone levels. The results show an increase of up to 60% in the number of days in which the 60 ppb level is exceeded and an increase of up to 15% in the ozone daily maxima.
Figure 8.2. Impact of climate change on peak ozone levels.
8.2.7 Data assimilation Kalman Filter and four-dimensional variational data assimilation schemes (4-D) have been used as advanced data assimilation schemes. The aim of this research is to integrate model results and observations in an objective way and to calculate the chemical state evolution which gives the best description of observed data. Mesoscale modelling systems are used as dynamical constraints to perform the inverse modelling of air pollution concentration fields. Applications of the method could show a considerable improvement of short-term predictions of air pollution by initial value and emission rate optimisation. Data assimilation is also a powerful tool for further detailed scientific analysis of the modelling system and the parameterisations used (e.g., sensitivity). A data assimilation shell using the extended Kalman Filter has been developed around the chemistry transport model LOTOS. The system is able to assimilate hourly averaged measurements, daily averages and instantaneous data from satellite tracks simultaneously. For application of the Kalman Filter, the uncertainty in both the measurements and the model needs to be specified. Moreover, the system is able to produce estimates of the uncertain model parameters/noise factors, as specified by the modeller.
GLOREAM
145
0.00 0.05 0.10 0.15 0.20 0.30 0.40
to to to to to to to
0.05 0.10 0.15 0.20 0.30 0.40 0.50
Figure 8.3. Residues of AOD at 555nm, August 17 1997, 12:00 GMT. Not assimilated (left), assimilated (right).
In Figure 8.3 results are shown for a specific satellite track of the ATSRinstrument for which the Aerosol Optical Depth (AOD) values have been retrieved. The AOD is the column-integrated amount of fine aerosols. The figure shows the residues, i.e., the absolute difference between the observed values and the modelled values with and without assimilation along the track. The figure clearly shows the improvement when using data assimilation. In the assimilation ground level hourly ozone measurements and daily observations of SO4, NH4 and NO3 were also used. On average the differences between the observations and assimilated values for these components were reduced as well. Model uncertainties/noise factors taken into account were the various emission fields, the deposition velocities and photolysis rates. For the month of August 1997 the resulting emission estimates were within a margin of 20% of the a priori values. Although these emission estimates should be taken only as very indicative due to a number of remaining uncertainties, the results show that estimating uncertainties using the Kalman Filter is possible and a promising field to explore further.
8.3
Model investigation and improvement on a global scale
8.3.1 New model elements Studies with a focus on the global scale make use of CTMs and of GCMs (general circulation models). Within GLOREAM the following models were applied: Danish Eulerian Hemispheric Model, MOGUNTIA, TOMCAT, TM3 (all CTMs); ECHAM4 (GCM). The Danish Eulerian Hemispheric Model was further developed and applied to study air pollution in the Arctic. The model includes a photochemistry scheme predicting 55 chemical species. More recently, a scheme to calculate the atmospheric distribution and deposition fluxes of lead was introduced and successfully tested. A new mercury model is under development.
146
P.J.H. Builtjes et al.
A new field for application of CTMs is the “chemical weather” forecast. The goal is to provide the measurement community with predicted distributions of some chemical key species during campaigns to support the planning of flight tracks. The Max Planck Institute for Meteorology, Hamburg, developed a preliminary version of the new ECHAM5 GCM to provide forecasts of the chemical weather. This model version was equipped with several carbon monoxide (CO) tracer species using a technique called “tagging” to distinguish the influence of individual source regions and emission types on the simulated CO concentrations. The model was run in nudging mode which means relaxation of the model dynamics towards up-to-date forecast products from the European Centre for Medium Range Weather Forecast (ECMWF). The output is saved every two hours for up to five days in advance. The model version has been developed to support experimentalists in planning flight tracks. So far it has been applied in two field campaigns, GTE/TRACE-P and MINOS.
8.3.2 Use of satellite data Satellite data, and especially tropospheric data, are used more and more in the study of the chemical composition of the troposphere. A number of GLOREAM models, including TM-3, TOMCAT, LOTOS) have been used in the framework of the subproject TROPOSAT. An example is shown in Figure 8.3. More information is given in this book in the chapter on TROPOSAT (Chapter 15).
8.3.3 Effect of transcontinental transport on European pollution load A global to mesoscale model chain (ECHAM-REMO-GESIMA) focusing on Europe, Germany and Berlin-Brandenburg is applied for the investigation of the effect of long-range transport of pollution on surface air composition during a summer smog episode at the end of July 1994. Throughout this period, the global model simulation provides a first estimate of the photochemical composition of the troposphere twice a day in a relatively coarse horizontal resolution. Three nesting steps are performed so that the results of the respective lower resolution model simulation are used as initial and lateral boundary data for the respective higher resolution model simulation. The evaluation of model results with nearsurface observations of ozone reveals a more realistic reproduction of the variability of simulated ozone mixing ratios with increasing horizontal resolution. Ozone mixing ratios simulated by the mesoscale models in the planetary boundary layer (PBL) and the free troposphere (FT) are considerably closer to observed levels when initial and lateral boundary conditions are taken from the global model simulation. Ozone mixing ratios determined by the global model dominate the results of the higher resolution, limited area models in the free troposphere but also contribute significantly to near-surface ozone mixing ratios. Convective mixing induced by occasionally occurring thunderstorms couples the air masses of the free troposphere and the PBL over Europe. It is shown that ozone from the free troposphere is injected into the PBL contributing an amount of at least 5-10 ppbv
GLOREAM
147
to maximum near-surface ozone mixing ratios during the summer smog episode under investigation. It should be noted, however, that the contribution of longrange transported pollution to ozone concentrations in surface air calculated for that period might not be representative of typical situations because of especially high convective activity which contributes significantly to vertical transfer between the PBL and the free troposphere. Reducing the anthropogenic emissions by 25% (corresponding approximately to the emission reduction in Germany from 1994-2000) leads again to a modification of 5-10 ppbv in maximum near-surface ozone over Central Europe, a decrease in this case. From these results it can be concluded that intercontinental transport of pollution can obscure the results of local efforts to reduce critical exposure levels of ozone during summer smog conditions. European pollution might also be reduced by decreasing emissions elsewhere due to the decreasing contribution to the long-range transport of pollution. Besides local emissions and intercontinental transport, import of ozone-rich stratospheric air masses also affects the tropospheric ozone load. The TOMCAT model was used to explain the build-up of ozone over Europe in spring. The results suggest that 30-40% of ozone has stratospheric origin in winter and early spring but in summer less than 10% of ozone comes from the stratosphere and photochemistry dominates up to an altitude of 9 km.
8.3.4 Air pollution and climate Man-made emissions from fossil fuel combustion and biomass burning have considerably modified the chemical composition of the atmosphere. The effects of increasing levels of anthropogenic emissions were studied using a coupled atmospheric (AGCM) and oceanic (OGCM) general circulation model at the Max Planck Institute for Meteorology in Hamburg. The concentrations of the well-mixed greenhouse gases like CO2, CH4, N2O and several industrial gases like the CFCs are prescribed as observed (1860-1980) and according to scenario IPCC-IS92a thereafter. The space-time distribution of tropospheric ozone is prescribed, based on pre-calculated fields from simulations with an atmospheric chemistry model coupled to the same AGCM employed in this study. The tropospheric sulphur cycle is calculated within the coupled model using prescribed anthropogenic sulphur emissions of the past and projected until 2050 from IPCC-IS92a. The radiative impact of the aerosols is considered via both the direct and the indirect effect. The climate response is similar but weaker, if aerosol effects are included in addition to greenhouse gases. One notable difference to previous experiments is that the intensity of the global hydrological cycle becomes weaker in a warmer climate if both direct and indirect aerosol effects are included in addition to the greenhouse gases. Changes in the chemical composition affect climate but in turn climate change also affects the chemical composition of the atmosphere. Hence, applying the same source strength, we find a higher total aerosol load in a colder climate compared to a simulation reproducing higher temperatures. Obviously in a warmer climate the intensification of the hydrological cycle is associated with a
148
P.J.H. Builtjes et al.
faster turn-over of aerosol particles resulting in a lower aerosol load. An impact on climate is also expected to change the composition of gaseous species in the atmosphere. Other aspects that affect the long-range transport of pollution and its impact on European pollution levels are modifications in large scale dynamics. For example, a high correlation (r = 0.57) was found between the North Atlantic Oscillation (NAO) index and “North American ozone” at Mace Head, a remote site at the Atlantic coast of Ireland for the period 1993–1997. A decline in the NAO index would reduce transatlantic transport of North American pollution to Europe. Although a significant decline in the NAO index is predicted using a GCM with anthropogenic forcing from greenhouse gases and sulphate aerosols, a new study using a pattern-based measure of the NAO concludes the opposite – namely a weak increase in the NAO. This revised result concerning NAO could lead to an increase of transatlantic transport of North American pollution to Europe, emphasising that global restrictions of anthropogenic CO, VOC and NOx emissions are necessary to reduce and control the formation of photo-oxidants.
8.4 Computational aspects 8.4.1 Progress related to the numerical methods used in the models Numerical methods are important tools in the efforts to run comprehensive air pollution models efficiently and, what is even more important, in the efforts to run long sequences of scenarios by using one or more comprehensive air pollution models. While the accuracy required from the air pollution models is normally not very high, it is essential to achieve it. This is important when the relationship between model results and, say, emissions is studied by using scenarios with different values of the emissions. If the accuracy requirement is not satisfied, then numerical errors will influence the output results and it becomes impossible to ensure that the observed changes of the model results are only caused by the variation of the emissions. This is why efforts to develop numerical methods which are both sufficiently accurate and fast were carried out by many GLOREAM groups. The status at the end of 2002 can be summarised as follows: ·
Substantial improvements were achieved in the chemical algorithms.
·
It is still desirable to continue the search for more efficient numerical methods for the advection part.
·
The inclusion of new modules in the models (module for performing data assimilation, modules for treatment of particles, etc.) led to the development of essentially new numerical algorithms, which were not used in the models before the start of GLOREAM. The development of data assimilation modules, for example, required the use of optimisation algorithms as well as the solution of adjoint equations.
GLOREAM
149
8.4.2 Progress related to the efficient exploitation of computer power The development of new computers in the last decade is amazing. Although the computers stopped growing ever bigger, they are now much faster. Many big problems (including here many of the existing air pollution models) can at present be handled on workstations and PCs. However, scientists and engineers still do need faster computers if very large and time-demanding tasks are to be handled. This is also true for some tasks in the field of large-scale air pollution modelling. There are several reasons for this: ·
New and or more advanced modules are needed in the efforts to describe in a more adequate way the physical and chemical processes studied by the models. Such modules have to be incorporated in the models (such as, for example, modules for handling data assimilation, aerosols, cloud chemistry, etc.). This nearly always leads to an increase in the computational complexity of the model. In many cases the increase is very considerable.
·
The ultimate purpose when an air pollution model is used is to apply the model in a practical evaluation of possible damaging effects due to high pollution levels (such as, for example, losses of crops due to high ozone levels). This leads to a requirement to perform a long series of runs with different scenarios, which also increases the computational complexity.
·
Long-term computations are often needed in order to study the tendencies in the development of high pollution levels due to reductions of emissions. In the last 10-15 years the emissions in Europe as whole have been reduced; some of the reductions are rather considerable. The requirement for longterm runs is also contributing to an increase in the amount of computations. Long-term computations are highly appropriate when the influence of high pollution levels on climate changes is studied.
·
Sometimes more detailed information about pollution levels is needed. Such information can be achieved by using fine resolution models. The use of such models may lead to an increase in the number of computations by a factor of several hundreds.
·
Inverse problems have to be treated in the solution of certain tasks. The computational difficulties are enormous when inverse problems are formulated and handled on computers. These are very challenging tasks, both computationally and numerically. There are a lot of unresolved problems in this field.
This list can be continued, but this is not necessary because the five reasons given above show clearly enough that the air pollution models must continuously be improved in order to meet the requirements for achieving better (more accurate, more detailed and more reliable) results. The improvements imply increased computational complexity. Therefore, it is necessary to carry out the improvements together with attempts ·
to exploit in more efficient way the great potential power of the modern high-performance computers, and
150
·
P.J.H. Builtjes et al.
to visualise the results better in order to represent clearly the relationships between the investigated quantities which are normally hidden behind enormous amount of digital information (millions and millions of numbers stored in huge output files).
This short description of the importance of computational aspects in air pollution modelling explains why these issues were treated also in the Annual Reports of many participants in the GLOREAM subproject of EUROTRAC-2.
8.4.3 Numerical algorithms and the computational techniques The numerical algorithms and computational techniques must be permanently improved in the efforts to make the regional air pollution models to solve bigger tasks and more tasks. Most of the GLOREAM participants are making such improvements. The most important results are: ·
The further development of parallel computing.
·
The analyses of different kinds of splitting. The major purpose is to identify the splitting procedure, which minimises the error due to splitting.
·
The chemical part of a large-scale air pollution model is normally the most time-consuming part when the model is run on computers. Therefore, the task of finding fast and sufficiently accurate chemical mechanisms as well as fast and sufficiently accurate numerical algorithms for handling the chemical schemes on computers is very important. Progress has been made in this respect.
·
The treatment of some inverse problems is an important issue in air pollution modelling. Inverse problems lead to big computational tasks. These problems are normally very ill-conditioned (in the sense that small perturbations of the input data lead to big differences in the output results). Therefore, the search for efficient numerical algorithms is crucial in this field. The use of inverse dispersion modelling as a tool to derive emission data from measurements has been studied in GLOREAM.
8.4.4 Main benefits from successful resolution of the computational problems The computational issues are well represented in the individual Annual Reports of the participants in the GLOREAM subproject of EUROTRAC-2. In many of the reports it is documented that the major computational problems are successfully resolved. When this has been completed, then the following benefits have been (obtained or will be obtained in the near future): ·
it is possible to improve the description of the physical and chemical processes in the models,
·
it is possible to solve more tasks and bigger tasks,
GLOREAM
151
·
it is possible to carry out long simulations with different scenarios in order to study the response of the models to key parameters (anthropogenic and biogenic emissions, meteorological parameters, boundary conditions, etc.),
·
it is possible to start to run some of the models operationally, in an attempt to predict exceedance in the next two to three days, of critical levels (for example, ozone critical levels)
·
it is possible to start development of advanced control strategies for keeping the concentrations and/or the depositions of harmful pollutants under the prescribed critical levels.
8.5 Model evaluation and validation The expression ‘model evaluation’ summarises various assessment techniques for the quality of model simulation output in respect to the model’s intended use. Within GLOREAM, model evaluation has been done for two different applications: ·
the point-by-point reproduction of measured data in space and time needed for the interpretation and generalization of experiments,
·
the provision of near-surface concentration fields supporting both short-term and long-term air quality management aspects.
There is no doubt that the model evaluation procedures are different in details for each application, but some common features exist which are outlined in the next section.
8.5.1 General methodology Model evaluation includes three different elements, i.e., ·
it is based on a strategy protocol,
·
a number of model runs are carried out as the core activity, and
·
decision criteria for the success or failure of the model exist and have been defined prior to the model runs.
In the evaluation strategy protocol, the details of the performance tests are summarised including the time period and the model domain of the study. Agreement is needed in respect to the trace substance concentrations and/or the meteorological quantities that serve as target parameters. Furthermore, qualifying criteria for models, such as for example, the existence of a documentation, may be introduced. It is essential that model quality objectives (MQOs) with respect to accuracy, precision, representativeness and completeness are defined and accepted by the modellers before they do their simulations. These MQOs depend strongly on the measurement techniques and the data sets which are available for comparison with the model results. They will be more stringent if quality controlled and assured data with good four-dimensional coverage exist. On the other hand, larger differences have to be accepted between measurements
152
P.J.H. Builtjes et al.
originating from a routine network and modelled data. It is desirable to select a set of suitable data sets both showing a wide range of values for the target parameters and representing special atmospheric situations such as a summertime high ozone smog episode, for example. The criteria for the decision whether a model is able to reproduce successfully a significant portion of the measured data set or not are usually based upon predefined scores. One option is to use the percentage of simulated data lying within a certain span from the observations, which is determined by the preselected MQOs for accuracy and precision. This type of score serves as a quantitative measure. Finally, one should always keep in mind that a score may be sufficient for one application, whereas it is not acceptable in another context. The following procedure may be applied if more than one model is evaluated. The data used for comparison are open and as many test simulations as necessary are allowed for each model. The only constraint is that a fixed deadline exists for the delivery of the “best” simulation results to an “independent” group of scientists, who have no model of their own in the intercomparison. This group prepares the intercomparison along the lines that have been agreed upon in the strategy protocol. An element of model evaluation is the selection and use of appropriate statistical indices. These indices should be independent of the observational data set and should cover the full range of data, without giving a bias to lower or higher concentrations. This makes the analysis between different model evaluation exercises more transparent. A possible procedure, following studies performed by US EPA, is to focus, with ozone as an example, on the model’s ability to predict domain-wide peak ozone concentrations and the concentrations at all locations with observed ozone above 60 ppb. The indices which could be used are: ·
the normalised accuracy of domain-wide, maximum one-hour concentrations unpaired in space and time,
·
the mean normalised bias and mean normalised error of all simulated and observed concentration pairs with concentrations above 60 ppb.
8.5.2 Results for point-by-point reproduction of measured data Within GLOREAM, a joint effort of up to eight modelling groups teamed in the German Tropospheric Research Programme (TFS, 1996-2000) has been carried out to estimate quantitative performance measures for Eulerian CTMs under summer and autumn meteorological conditions. Potential temperature, specific humidity and the concentrations of both ozone and nitrogen dioxide have been selected as target parameters for the following reasons. The potential temperature provides information about the atmospheric stratification (with consequences for the atmospheric turbulence and mixing) and the daily variations in the reaction rate coefficients. The specific humidity on the one hand behaves like a passive tracer in a cloud-free or fog-free atmosphere and thus gives indications about the net effect of the various transport processes in the boundary layer. On the other hand it reveals whether the photochemistry in the model runs in a dry or wet
GLOREAM
153
environment. Ozone is the prime indicator species for photochemical smog and nitrogen dioxide has been chosen because its rather small atmospheric concentrations are an additional challenge for the models. Table 8.1 summarises the MQOs for the four target values. Table 8.1. Model quality objectives (MQOs) for the four target values used in the evaluation exercise carried out in the framework of the German Tropospheric Research Program (TFS, 1996-2000). The MQOs for specific humidity as well as ozone and nitrogen dioxide concentrations are defined as 10 percent of the median of the observations, respectively.
Model quality objective (MQO) for:
FluMoB, July 1994
BERLIOZ, July 1998
potential temperature
± 1.5
± 1.5
K
specific humidity
± 1.01
± 0.85
g/kg
ozone concentration
± 10
±7
ppb(v)
nitrogen dioxide concentration
± 1.25
± 0.75
ppb(v)
A set of four test cases was studied. Three data sets came from larger experimental efforts mainly designed to understand atmospheric photochemistry and are quality assured. For two of them, FluMoB and BERLIOZ, results from six different models are available and one model has additionally been run with a finer spatial resolution, making up a total of seven model runs available for the evaluation. Table 8.2 shows the horizontal resolution of the models, using an anonymous form for the models’ names. Table 8.2. Horizontal resolution of the seven models (names in anonymous form) participating in the evaluation exercise of the TFS.
Model name
horizontal resolution FluMoB, July 1994
BERLIOZ, July 1998
Model 1
18 km
18 km
Model 2
0.1666667 Grad
0.1666667 Grad
Model 3
6 km
6 km
Model 4
5 km
4 km
Model 5
4 km
4 km
Model 6
2 km
2 km
Model 7
2 km
2 km
Figures 8.4 and 8.5 show the scores, i.e., the percentage of differences between modelled and measured data that is less than or equal to the model quality
154
P.J.H. Builtjes et al.
objective (MQO) shown on the abscissa, for the four target parameters explained above. Figure 8.4 summarises the results from the first BERLIOZ intensive (20 and 21 July 1998) and Figure 8.5 those from two days (26 and 27 July 1994) of the FLUMOB experiment, respectively. Both experiments were carried out for the same purpose, i.e., to get more insight into the magnitude of and the reasons for the city plume of Berlin. Furthermore, the selected days had comparable meteorological fair weather conditions with high insolation, but also noticeable transport and thus advection of precursors for ozone formation. A more detailed description of the two experiments may be found elsewhere.
90 80
70
70
Score %
100
90 80
Score %
100
60 50 40 30 20 10 0
60 50 40 30 20 10 0
0.75 1.25 1.75 2.25 2.75 0.50 1.00 1.50 2.00 2.50 3.00
MQO for pot. temperature K
MQO for NO 2 concentration ppb(v) 90
80
80
70 60
70 60
Score %
100
90
Score %
100
50 40 30 20 10 0
1.00 1.50 2.00 2.50 3.00 0.75 1.25 1.75 2.25 2.75
50 40 30 20 10
0.75 1.25 1.75 2.25 2.75 0.50 1.00 1.50 2.00 2.50 3.00
MQO for spec. humidity g/kg
0
7.5 12.5 17.5 22.5 27.5 32.5 5.0 10.0 15.0 20.0 25.0 30.0 35.0
MQO for O 3 concentration ppb(v)
BERLIOZ, 20 and 21 July 1998 Model 1
Model 3
Model 5
Model 2
Model 4
Model 6
Model 7
Figure 8.4. Scores, i.e., percentage of differences between modelled and measured data in relation to the uncertainty range shown on the abscissa for potential temperature, specific humidity, nitrogen dioxide and ozone concentrations, respectively. one-minute-averaged aircraft data from the first BERLIOZ intensive are compared to the seven CTMs, whose horizontal resolution is given in Table 8.2. Both the scores for the MQOs listed in Table 8.1 and the changing scores resulting from other MQO choices can be easily obtained from this figure. From TFS (2002).
GLOREAM
155
The scores for the potential temperature are shown in the upper left graphs of Figures 8.4 and 8.5, respectively. The scores range from a few percent (below 10%) up to 40% for a MQO of ± 0.5 K and in both cases the highest scores are reached by the same models (no. 2, 4 and 6). In the FLUMOB case (see upper left graph in Figure 8.5) the range of the scores and thus the differences between the models becomes smaller and smaller for more relaxed MQOs of ± 1.5 K and more. All models have scores above 95% for a MQO of ± 2.5 K.
90
90
80
80
70
70
Score %
100
Score %
100
60 50 40 30
60 50 40 30
20
20
10
10
0
0
0.75 1.25 1.75 2.25 2.75 0.50 1.00 1.50 2.00 2.50 3.00
MQO for pot. temperature K
MQO for NO 2 concentration 90
80
80
70
70
Score %
100
90
Score %
100
60 50 40 30
40 30 20
10
10 0.75 1.25 1.75 2.25 2.75 0.50 1.00 1.50 2.00 2.50 3.00
MQO for spec. humidity g/kg
ppb(v)
60 50
20 0
1.25 1.75 2.25 2.75 1.00 1.50 2.00 2.50 3.00
0
7.5 12.5 17.5 22.5 27.5 32.5 37.5 5.0 10.0 15.0 20.0 25.0 30.0 35.0 40.0
MQO for O 3 concentration ppb(v)
FLUMOB, 26 and 27 July 1994 Model 1
Model 3
Model 5
Model 2
Model 4
Model 6
Model 7
Figure 8.5. Scores, i.e., percentage of differences between modelled and measured data in relation to the uncertainty range shown on the abscissa for potential temperature, specific humidity, nitrogen dioxide and ozone concentrations, respectively. One minute-averaged aircraft data from the FLUMOB experiment are compared to the seven CTMs, whose horizontal resolution is given in Table 8.2. Both the scores for the MQOs listed in Table 8.1 and the changing scores resulting from other MQO choices can be easily obtained from this figure. From TFS (2002).
156
P.J.H. Builtjes et al.
In contrast, in the BERLIOZ case (see upper left graph in Figure 8.4) only two models reach a score above 95% for a MQO of ± 2.5 K and one model (no. 7) shows less than 40% even for ± 3.0 K. Interestingly enough, a good score in the potential temperature comes with a bad score in the ozone concentration and vice versa. This is true for model no. 7, having the worst potential temperature score and by far the best score for the ozone concentration (see lower right graph in Figure 8.4). However, this is also true for models no. 2 and 4, which have the best potential temperature scores and at the same time comparably low scores for the ozone concentration. However, as an exception to this rule, model no. 5 shows low potential temperature scores and insufficient scores for the ozone concentration. Overall, the scores of model no. 5 are at the lower end in many of the eight model intercomparisons shown in Figures 8.4 and 8.5. This indicates that this model urgently needs improvement before it is applied for the interpretation and generalisation of experimental results. It can be seen from the upper right graphs in both figures that the simulated NO2 concentration values lie below 60% (BERLIOZ) and 80% (FLUMOB) even for MQOs that are larger than the median of the corresponding measurements. It is obvious that for this target parameter improvement in both the measurement capabilities and the models are necessary. The reader can find many more details in these figures.
8.5.3 Results for near-surface concentration fields Although the point-to-point reproduction exercise for the FLUMOB and BERLIOZ data has been less successful than expected beforehand, it is worthwhile to check the short-term prediction capabilities of mesoscale CTMs, i.e., to find out to what extent it is possible to do an ozone forecast (“chemical weather” forecast) over a few days using a procedure that is comparable to a routine weather prediction. For this application, approximate estimates for the location and the value of the maximum concentration are sufficient to support national or local environmental agencies, for example, in their execution of the environmental legislation. Thus less restrictive MQOs are needed. In addition, models (with noticeable differences in complexity) are for the first time accepted for use for the assessment of air quality in EU member countries under the European framework directive on ambient air quality assessment and management and the subsequent daughter directives. Therefore it is interesting to know how accurate and precise simulated atmospheric concentration fields are, i.e., whether the envisaged models are suitable for air quality management purposes or not. Hints about the performance of up to five different CTMs have been gained from an ozone forecast study performed in GLOREAM. In the upper part of Figure 8.6 the diurnal ozone concentration variation averaged over four months (May to September 1999) and spatially over Germany both from observations (hourly values from more than 300 stations) and from four models is shown. The ensemble of model data consists of daily ozone forecasts, i.e., up to 153 data sets for the period from 01 May to 30 September 1999. It is interesting to note that
GLOREAM
157
only one model is capable of reproducing the gross features (times of minimum and maximum concentrations, amplitude) of this “average German ozone day”.
Figure 8.6. Mean diurnal variation and bias of daily 24-hour ozone forecasts from four CTMs (DWD, FUB, NERI and SMHI) over a period of five months (May to September 1999). For two models the prognosis for the second day (DWD2 and FUB2) are included. The additional dashed curve in the upper figure represents the observations. The figure has been compiled from material provided by Stefan Tilmes through private communication. (Meaning of the acronyms: DWD German Weather Service and EURAD group, FUB Free University of Berlin, NERI National Environmental Research Institution, Roskilde, Denmark, SMHI Swedish Meteorological and Hydrological Institute).
158
P.J.H. Builtjes et al.
This can be seen more clearly in the lower part of Figure 8.6, where the bias for the four models is depicted as a function of time. The bias of the DWD model varies within a band of approximately ± 6 ppb(v) around the zero line. On the contrary, the other three models show the same functional form of the bias with higher values during the night time hours. The bias of the NERI and the SMHI model are positive with minimum values of a few ppb(v)s, i.e., the smaller biases around noon and in the afternoon are favourable for a forecast of the daily ozone maximum. However, the bias of the FUB model is negative throughout the complete day, i.e., the largest underestimation of the ozone concentration in this model occurs in the afternoon. Last but not least, it is worth mentioning, that the second day of the DWD forecast (DWD2) shows nearly no differences to the first day of the forecast (DWD), whereas the forecast quality for the second day in the model from the Free University of Berlin (FUB2) is reduced compared to the first day (FUB). Additional aspects of the forecast performance of the four models (plus a fifth model from NILU, for which only a reduced data set for July and August 1999 is available) can be found in the GLOREAM Final Report (Builtjes, 2003). It may be concluded from Figure 8.6 that it is justified to use the DWD, NERI and SMHI models for short-term ozone forecasts up to 48 hours on a routine basis. Some experiences from this application are described in the section 8.6. In respect to the FUB model it seems necessary to improve especially the day-time ozone formation before this model is ready for assessment studies and routine ozone prediction. Data assimilation, as described in the section 8.2.7 on model investigation and improvement, can also be used as part of model evaluation. The objective combination of model results and observations, especially using the Kalman filter approach, leads not only to a new model state, but also gives as output the noise factors of the selected parameters and input data, which are determined in the data assimilation process. These calculated noise factors give an indication of the sensitivity of the parameters and input data, and in this way of the model performance.
8.5.4 Summary Even at the end of EUROTRAC-2, after more than one decade of international research in the field of mesoscale experiments and model development, atmospheric chemistry and transport models are far from being complete. Both the evaluation activities described in brief in this summary and the various initiatives documented in the literature reveal that more research is needed in order to obtain reliable spatial and temporal distributions of atmospheric parameters suited to a specific purpose. This latter criterion is obligatory if one wants to forecast changes in air quality (and climate) on the longer time scales of a decade or even more (as is attempted in studies such as Auto Oil and CAFE).
GLOREAM
159
8.6 Model application and assessment An overview of all the models that have taken part in GLOREAM is given in the overview table at the end of this chapter. This table contains also a column with examples of application studies performed with these models. These application studies range from long-term model runs with scenario calculations, to model studies in the framework of the EU directives, to real-time ozone forecast runs. A number of examples are given below.
8.6.1 Models used for planning and forecasting Within GLOREAM, a major effort has been carried out in order to have an air quality forecast tool. In that respect, results from five air quality forecast systems are available on the Internet, namely: ·
http://artico.lma.fi.upm.es/, developed by the Environmental Software and Modelling Group, at the Technical University of Madrid, Spain;
·
http://www.eurad.uni-koeln.de, developed and implemented by EURAD, at the University of Cologne, Germany;
·
http://www.dmi.dk/vejr/index.html, developed and implemented by the Danish Meteorological Institute;
·
http://luft.dmu.dk, developed and implemented by NERI, Denmark;
·
http://trumf.fu-berlin.de, developed and implemented by the Free University of Berlin, Germany.
The first system uses the MM5 (PennState/NCAR Mesoscale Model Version 5) and the CMAQ (Community Multiscale Air Quality Modelling System US EPA). The MM5 is driven by the initial conditions available on the NOAA web site and assures the CMAQ proper meteorological fields, which produces concentration fields of air pollutants over the Iberian Peninsula. The EURAD forecast system consists of the mesoscale meteorological model MM5, the emission processor EEM (EURAD Emission Model) and the EURADCTM. The initial and boundary data for MM5 are obtained from the global AVN forecast (NCEP) at the start of the forecast cycle (00 UTC). The emission data are interpolated from the EMEP data base, in time and space, for three different regions of interest: Europe, Central Europe and the German state of North RhineWestphalia (Figure 8.7). In addition to the predicted gas phase concentrations, aerosol particles are also forecast during the cycle. For about 15 years the EURAD model has been developed and improved for applications within numerous case studies on the regional scale in Europe. The Danish Meteorological Institute (DMI) makes a prognosis of the surface ozone concentrations produced by the system of models composed by the meteorological model HIRLAM (HIgh Resolution Limited Area Model), the numerical weather prediction model from the DMI and the DACFOS model (Danish Atmospheric Chemistry FOrecasting System). DACFOS consists of two components: a 3-D Lagrangian chemical-transport, receptor-point model
160
P.J.H. Builtjes et al.
(DACFOS_L) and a statistical after-treatment of the ozone forecasts from the chemical transport model (DACFOS_S), when real-time ozone measurements are available. Different versions of the photochemical model in DACFOS_L have been developed during the GLOREAM project period.
Figure 8.7. Near surface concentrations of ozone (daily maximum) at June 18, 2002 for Europe, Central Europe and North Rhine-Westphalia (the concentrations are given in µg/m3) From Borrego et al., 2002.
The model system at NERI, the so-called Thor-system, uses a nested model hierarchy from the hemispherical scale down to, in principle, the street-level. The meteorological driver used is the Eta-model. At the Free University of Berlin an updated version of the REM3-Callgrid model is used, with a focus on the ozone forecasting for Germany.
8.6.2 Models capable of treating and/or modelling the EU directives A series of directives has been approved to control levels of some pollutants and to monitor their concentrations in the air. In 1996, the Environment Council adopted the Framework Directive (96/62/EC) on ambient air quality assessment and management. The first daughter directive (1999/30/EC) concerns limit values for sulphur dioxide, nitrogen dioxide and nitrogen oxides, particulate matter and lead in ambient air. The second daughter directive (2000/69/EC) deals with limit values for benzene and carbon monoxide in ambient air. More recently the third daughter directive (2002/3/EC), related to ozone, was launched. Besides the framework and daughter directives, the EU directive on National Emission Ceilings (2001/81/EC) is relevant for policy making. Tools for evaluating the impact of abatement strategies on ozone and other pollutants should provide fast assessments at low computational cost, in such a way that a large number of scenarios can be evaluated within a limited time frame. Concerning ozone effects, ozone exposure is to be assessed on a longterm basis, depending on the evaluation variable that is examined. Moreover, the formation and degradation of ozone can cover several days and the transport of ozone and its precursors is not limited to any national border.
GLOREAM
161
Several modelling tools have been designed for the assessment of abatement scenarios. One of these tools is the EUROS model, originally developed at RIVM (The Netherlands), an atmospheric model that simulates tropospheric ozone over Europe on a long-term basis. In the framework of the BelEUROS project, a new version of the EUROS model coupled with a state-of-the-art user interface has been installed at the Inter-regional Cell for the Environment (IRCEL/CELINE) in Brussels as a tool for policy support with respect to tropospheric ozone. Based on the results of extensive calculations by the EUROS and the LOTOS models, multiple linear regression techniques have been used to develop fast regression models. Although limited in accuracy, their results can provide a good insight to policy makers in the non-linear character of the response of ozone to its precursors. The EURAD modelling system has also been used to investigate the air quality in Europe, namely for strongly urbanised sub regions as required by the EU directives. The preparation of emission scenarios in relation to the CAFE programme is underway in close contact with the City-Delta initiative. One major problem in model applications is the quality and availability of appropriate emission data for particles. The above-mentioned LOTOS model (TNO, The Netherlands) is one of the participants in the discussion in CAFE about the model requirements in view of the EU directives. It is also suitable for UNECE calculations. In the past it has been applied to the evaluation of long-term values for ozone, sulphate and nitrate, and also to emission reduction studies and AOT 40 calculations. The Danish Eulerian Model has been applied in systematic studies concerning long time-period runs, economic evaluation of crop losses due to high ozone levels, development of biogenic emissions scenarios and the impact of climate change on air pollution levels. Results may be consulted on the following Internet site: http://www.dmu.dk/ AtmosphericEnvironment/DEM/.
8.6.3 Impact of vertical structuring on applied model results Simulated species in limited area models are sensitive to their vertical distribution at the lateral boundaries. The lifetime and distribution of ozone is also dependent on the larger scale conditions when investigating the processes of pollutant formation, as described in the section on global modelling above. Studies on the improvement of the accuracy of cloud parameters, for use in chemical transport modelling, are being carried out by the Free University of Berlin. Cloud parameters are important for boundary layer modelling, for the determination of photolysis frequencies and for chemical aerosol modules in CTMs. As stated above, the boundary and initial conditions are very important for improvements of CTM results. In this sense, studies about atmospheric dynamics, with emphasis on the vertical movement, are very important for the determination of concentration fluxes of atmospheric chemical species between the lower stratosphere and the free troposphere. Investigations of the tropopause region with the aim of exploring mesoscale features of chemistry and transport
162
P.J.H. Builtjes et al.
have already been carried out by the EURAD group during the first phase of EUROTRAC and have been continued in the framework of GLOREAM under EUROTRAC-2. They conclude that blocking synoptical situations have a pronounced impact on the exchange of air masses, including minor constituents, between the troposphere and stratosphere at middle latitudes. Their relative role in comparison with other known mesoscale processes has still to be quantified. Scientists at the University of Aveiro, Portugal, have identified vertical structures, namely thermal low systems, which are related to summer ozone episodes over Portugal. These kinds of situations are difficult to model by MM5, and consequently by the photochemical models, which requires a great effort on improving meteorological results.
8.7 Interaction with other EUROTRAC-2 subprojects The GLOREAM models have been applied in several EUROTRAC-2 subprojects to study specific phenomena. In TROPOSAT, the models have been used to analyse and assess tropospheric satellite data. In TOR-2, the models have been used to study ozone characteristics like the spring-time ozone maximum and the trends of ozone over the last decades. In EXPORT-E2 the models have been used to determine the impact of hemispherical transport on the ozone in the troposphere above Europe. The cooperation with SATURN has focussed on model evaluation and validation procedures and on the determination of boundary conditions for the urban scale models applied in SATURN. Most models in GLOREAM have used emission information of the subproject GENEMIS. Exchange of information also existed with the subprojects CMD and AEROSOL.
8.8 General conclusions and major achievements The general aims as formulated at the start of GLOREAM at the beginning of 1997 were: ·
to develop and improve three-dimensional regional and global scale atmospheric transport chemistry models,
·
to investigate, with the aid of models, processes and their interaction that control the chemical composition of the troposphere,
·
to apply complex and simplified models for specific environmental policy issues, and to assist other EUROTRAC-2 subprojects.
Looking to all the results obtained in GLOREAM, the double refereed papers, the theses and the reports describing the more policy-oriented results, it can be concluded that GLOREAM has done what it promised to do. Major achievements are the following:
GLOREAM
163
·
Continental/regional scale and global models have been improved considerable and have now reached a stage of maturity with respect to tropospheric ozone. Although progress has been made with respect to aerosols, this is however, much less.
·
Long-term (over years and more) model calculations are nowadays possible for most models on an hour-by-hour basis. Also ozone forecasting has become possible. This has been made possible by the increase of computer power and the improvements in the efficiency of numerical methods.
·
Models of different complexity, state-of-the-art and models of intermediate complexity are now available, enabling the calculations of many scenarios. Model intercomparison studies have been carried out as well as model evaluation and validation. In this way, model errors and flaws could be detected, and a beginning has been made to formulate criteria which models have to pass before they can be used in further studies.
·
New numerical methods for nesting/scale interaction and data assimilation have been developed which have greatly improved the capabilities of the models to address new science and policy issues.
8.9 Gaps in knowledge Currently the major gaps in knowledge are the following: ·
Clear progress has been made in aerosol modelling, but many unknowns still exist. This is to a large extent due to the lack of sufficient and detailed aerosol observations which makes model evaluation in part not possible.
·
Although model intercomparison studies and model evaluation have been carried out, a complete and generally accepted model validation system is not yet in place, nor have all the models undertaken a full model evaluation.
·
Numerical improvements are still necessary, especially to increase the computational speed/efficiency to enable the use of the full capabilities of data assimilation and (two-way) nesting.
·
One of the weakest points in modelling is the proper treatment of clouds (also in respect of aerosols) and the treatment of vertical exchange.
·
Air quality and climate are interconnected processes. This should be taken into account in future model improvements and model application and scenario studies.
More detailed information, including the full references to the studies and figures cited or used in this chapter, can be found in the separate Final Report of the GLOREAM subproject (Builtjes, 2003).
164
P.J.H. Builtjes et al.
References Borrego, C., P.J.H. Builtjes, A.I. Miranda, P. Santos, A.C. Carvalho ( 2002): Global and Regional Atmospheric Modelling, Proceedings of the 6th GLOREAM Workshop, Univ. of Aveiro, Portugal, ISBN 972-789-073-3. Builtjes, P.J.H, A. Ebel and J. Feichter. (1999): GLOREAM Subproject Description EUROTRAC-2, ISS, Munich. Builtjes, P.J.H., ed. (2003): Global and Regional Atmospheric Modelling, Final Report of Subproject GLOREAM, EUROTRAC-2, ISS, Munich, in preparation. Ebel, A., R. Friedrich and H. Rodhe. (1997): Tropospheric Modelling and Emission Estimation EUROTRAC Report, Volume 7, Springer Verlag, Berlin. ISBN 3-540-63169-0. TFS, (2002): Tropospheric Chemistry - Results of the German Tropospheric Chemistry Programme, Kluwer Academic Publishers, ISBN 1-4020-0694-2, and J. Atmos. Chem., volume 42.
