Search

The Emissions Database for Global Atmospheric Research (EDGAR) estimates the human-induced emission rates on Earth. EDGAR collaborates with atmospheric modelling activities and aids policy in the design of mitigation strategies and in evaluating their effectiveness. In these applications, the uncertainty estimate is an essential component, as it quantifies the accuracy and qualifies the level of confidence in the emission. This study complements the EDGAR emissions inventory by providing an estimation of the structural uncertainty stemming from its base components (activity data, AD, statistics and emission factors, EFs) by (i) associating uncertainty to each AD and EF characterizing the emissions of the three main greenhouse gases (GHGs), namely carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O); (ii) combining them; and (iii) making assumptions regarding the cross-country uncertainty aggregation of source categories. It was deemed a natural choice to obtain the uncertainties in EFs and AD statistics from the Intergovernmental Panel on Climate Change (IPCC) guidelines issued in 2006 (with a few exceptions), as the EF and AD sources and methodological aspects used by EDGAR have been built over the years based on the IPCC recommendations, which assured consistency in time and comparability across countries. On the one hand, the homogeneity of the method is one of the key strengths of EDGAR, on the other hand, it facilitates the propagation of uncertainties when similar emission sources are aggregated. For this reason, this study aims primarily at addressing the aggregation of uncertainties' sectorial emissions across GHGs and countries. Globally, we find that the anthropogenic emissions covered by EDGAR for the combined three main GHGs for the year 2015 are accurate within an interval of −15 % to +20 % (defining the 95 % confidence of a log-normal distribution). The most uncertain emissions are those related to N2O from waste and agriculture, while CO2 emissions, although responsible for 74 % of the total GHG emissions, account for approximately 11 % of global uncertainty share. The sensitivity to methodological choices is also discussed.

An accurate simulation of the absorption properties is key for assessing the radiative effects of aerosol on meteorology and climate. The representation of how chemical species are mixed inside the particles (the mixing state) is one of the major uncertainty factors in the assessment of these effects. Here we compare aerosol optical properties simulations over Europe and North America, coordinated in the framework of the third phase of the Air Quality Model Evaluation International Initiative (AQMEII), to 1 year of AERONET sunphotometer retrievals, in an attempt to identify a mixing state representation that better reproduces the observed single scattering albedo and its spectral variation. We use a single post-processing tool (FlexAOD) to derive aerosol optical properties from simulated aerosol speciation profiles, and focus on the absorption enhancement of black carbon when it is internally mixed with more scattering material, discarding from the analysis scenes dominated by dust. We found that the single scattering albedo at 440 nm (ω0,440) is on average overestimated (underestimated) by 3–5 % when external (core-shell internal) mixing of particles is assumed, a bias comparable in magnitude with the typical variability of the quantity. The (unphysical) homogeneous internal mixing assumption underestimates ω0,440 by ∼14 %. The combination of external and core-shell configurations (partial internal mixing), parameterized using a simplified function of air mass aging, reduces the ω0,440 bias to -1/-3 %. The black carbon absorption enhancement (Eabs) in core-shell with respect to the externally mixed state is in the range 1.8–2.5, which is above the currently most accepted upper limit of ∼1.5. The partial internal mixing reduces Eabs to values more consistent with this limit. However, the spectral dependence of the absorption is not well reproduced, and the absorption Ångström exponent AAE675440 is overestimated by 70–120 %. Further testing against more comprehensive campaign data, including a full characterization of the aerosol profile in terms of chemical speciation, mixing state, and related optical properties, would help in putting a better constraint on these calculations.

This paper describes, documents, and validates the TM5-FAst Scenario Screening Tool (TM5-FASST), a global reduced-form air quality source–receptor model that has been designed to compute ambient pollutant concentrations as well as a broad range of pollutant-related impacts on human health, agricultural crop production, and short-lived pollutant climate metrics, taking as input annual pollutant emission data aggregated at the national or regional level. The TM5-FASST tool, providing a trade-off between accuracy and applicability, is based on linearized emission-concentration sensitivities derived with the full chemistry-transport model TM5. The tool has been extensively applied in various recent critical studies. Although informal and fragmented validation has already been performed in various publications, this paper provides a comprehensive documentation of all components of the model and a validation against the full TM5 model. We find that the simplifications introduced in order to generate immediate results from emission scenarios do not compromise the validity of the output and as such TM5-FASST is proven to be a useful tool in science-policy analysis. Furthermore, it constitutes a suitable architecture for implementing the ensemble of source–receptor relations obtained in the frame of the HTAP modelling exercises, thus creating a link between the scientific community and policy-oriented users.

This study evaluates simulated vertical ozone profiles produced in the framework of the third phase of the Air Quality Model Evaluation International Initiative (AQMEII3) against ozonesonde observations in North America for the year 2010. Four research groups from the United States (US) and Europe have provided modeled ozone vertical profiles to conduct this analysis. Because some of the modeling systems differ in their meteorological drivers, wind speed and temperature are also included in the analysis. In addition to the seasonal ozone profile evaluation for 2010, we also analyze chemically inert tracers designed to track the influence of lateral boundary conditions on simulated ozone profiles within the modeling domain. Finally, cases of stratospheric ozone intrusions during May–June 2010 are investigated by analyzing ozonesonde measurements and the corresponding model simulations at Intercontinental Chemical Transport Experiment Ozonesonde Network Study (IONS) experiment sites in the western United States. The evaluation of the seasonal ozone profiles reveals that, at a majority of the stations, ozone mixing ratios are underestimated in the 1–6 km range. The seasonal change noted in the errors follows the one seen in the variance of ozone mixing ratios, with the majority of the models exhibiting less variability than the observations. The analysis of chemically inert tracers highlights the importance of lateral boundary conditions up to 250 hPa for the lower-tropospheric ozone mixing ratios (0–2 km). Finally, for the stratospheric intrusions, the models are generally able to reproduce the location and timing of most intrusions but underestimate the magnitude of the maximum mixing ratios in the 2–6 km range and overestimate ozone up to the first kilometer possibly due to marine air influences that are not accurately described by the models. The choice of meteorological driver appears to be a greater predictor of model skill in this altitude range than the choice of air quality model.

The evaluation and intercomparison of air quality models is key to reducing model errors and uncertainty. The projects AQMEII3 and EURODELTA-Trends, in the framework of the Task Force on Hemispheric Transport of Air Pollutants and the Task Force on Measurements and Modelling, respectively (both task forces under the UNECE Convention on the Long Range Transport of Air Pollution, LTRAP), have brought together various regional air quality models to analyze their performance in terms of air concentrations and wet deposition, as well as to address other specific objectives.This paper jointly examines the results from both project communities by intercomparing and evaluating the deposition estimates of reduced and oxidized nitrogen (N) and sulfur (S) in Europe simulated by 14 air quality model systems for the year 2010. An accurate estimate of deposition is key to an accurate simulation of atmospheric concentrations. In addition, deposition fluxes are increasingly being used to estimate ecological impacts. It is therefore important to know by how much model results differ and how well they agree with observed values, at least when comparison with observations is possible, such as in the case of wet deposition.This study reveals a large variability between the wet deposition estimates of the models, with some performing acceptably (according to previously defined criteria) and others underestimating wet deposition rates. For dry deposition, there are also considerable differences between the model estimates. An ensemble of the models with the best performance for N wet deposition was made and used to explore the implications of N deposition in the conservation of protected European habitats. Exceedances of empirical critical loads were calculated for the most common habitats at a resolution of 100  ×  100 m2 within the Natura 2000 network, and the habitats with the largest areas showing exceedances are determined.Moreover, simulations with reduced emissions in selected source areas indicated a fairly linear relationship between reductions in emissions and changes in the deposition rates of N and S. An approximate 20 % reduction in N and S deposition in Europe is found when emissions at a global scale are reduced by the same amount. European emissions are by far the main contributor to deposition in Europe, whereas the reduction in deposition due to a decrease in emissions in North America is very small and confined to the western part of the domain. Reductions in European emissions led to substantial decreases in the protected habitat areas with critical load exceedances (halving the exceeded area for certain habitats), whereas no change was found, on average, when reducing North American emissions in terms of average values per habitat.

In the framework of the third phase of the Air Quality Model Evaluation International Initiative (AQMEII3), and as contribution to the second phase of the Hemispheric Transport of Air Pollution (HTAP2) activities for Europe and North America, the impacts of a 20 % decrease of global and regional anthropogenic emissions on surface air pollutant levels in 2010 are simulated by an international community of regional-scale air quality modeling groups, using different state-of-the-art chemistry and transport models (CTMs). The emission perturbations at the global level, as well as over the HTAP2-defined regions of Europe, North America and East Asia, are first simulated by the global Composition Integrated Forecasting System (C-IFS) model from European Centre for Medium-Range Weather Forecasts (ECMWF), which provides boundary conditions to the various regional CTMs participating in AQMEII3. On top of the perturbed boundary conditions, the regional CTMs used the same set of perturbed emissions within the regional domain for the different perturbation scenarios that introduce a 20 % reduction of anthropogenic emissions globally as well as over the HTAP2-defined regions of Europe, North America and East Asia. Results show that the largest impacts over both domains are simulated in response to the global emission perturbation, mainly due to the impact of domestic emission reductions. The responses of NO2, SO2 and PM concentrations to a 20 % anthropogenic emission reduction are almost linear (∼ 20 % decrease) within the global perturbation scenario with, however, large differences in the geographical distribution of the effect. NO2, CO and SO2 levels are strongly affected over the emission hot spots. O3 levels generally decrease in all scenarios by up to ∼ 1 % over Europe, with increases over the hot spot regions, in particular in the Benelux region, by an increase up to ∼ 6 % due to the reduced effect of NOx titration. O3 daily maximum of 8 h running average decreases in all scenarios over Europe, by up to ∼ 1 %. Over the North American domain, the central-to-eastern part and the western coast of the US experience the largest response to emission perturbations. Similar but slightly smaller responses are found when domestic emissions are reduced. The impact of intercontinental transport is relatively small over both domains, however, still noticeable particularly close to the boundaries. The impact is noticeable up to a few percent, for the western parts of the North American domain in response to the emission reductions over East Asia. O3 daily maximum of 8 h running average decreases in all scenarios over north Europe by up to ∼ 5 %. Much larger reductions are calculated over North America compared to Europe. In addition, values of the Response to Extra-Regional Emission Reductions (RERER) metric have been calculated in order to quantify the differences in the strengths of non-local source contributions to different species among the different models. We found large RERER values for O3 (∼ 0.8) over both Europe and North America, indicating a large contribution from non-local sources, while for other pollutants including particles, low RERER values reflect a predominant control by local sources. A distinct seasonal variation in the local vs. non-local contributions has been found for both O3 and PM2.5, particularly reflecting the springtime long-range transport to both continents.

In this study we introduce a hybrid ensemble consisting of air quality models operating at both the global and regional scale. The work is motivated by the fact that these different types of models treat specific portions of the atmospheric spectrum with different levels of detail, and it is hypothesized that their combination can generate an ensemble that performs better than mono-scale ensembles. A detailed analysis of the hybrid ensemble is carried out in the attempt to investigate this hypothesis and determine the real benefit it produces compared to ensembles constructed from only global-scale or only regional-scale models. The study utilizes 13 regional and 7 global models participating in the Hemispheric Transport of Air Pollutants phase 2 (HTAP2)–Air Quality Model Evaluation International Initiative phase 3 (AQMEII3) activity and focuses on surface ozone concentrations over Europe for the year 2010. Observations from 405 monitoring rural stations are used for the evaluation of the ensemble performance. The analysis first compares the modelled and measured power spectra of all models and then assesses the properties of the mono-scale ensembles, particularly their level of redundancy, in order to inform the process of constructing the hybrid ensemble. This study has been conducted in the attempt to identify that the improvements obtained by the hybrid ensemble relative to the mono-scale ensembles can be attributed to its hybrid nature. The improvements are visible in a slight increase of the diversity (4 % for the hourly time series, 10 % for the daily maximum time series) and a smaller improvement of the accuracy compared to diversity. Root mean square error (RMSE) improved by 13–16 % compared to G and by 2–3 % compared to R. Probability of detection (POD) and false-alarm rate (FAR) show a remarkable improvement, with a steep increase in the largest POD values and smallest values of FAR across the concentration ranges. The results show that the optimal set is constructed from an equal number of global and regional models at only 15 % of the stations. This implies that for the majority of the cases the regional-scale set of models governs the ensemble. However given the high degree of redundancy that characterizes the regional-scale models, no further improvement could be expected in the ensemble performance by adding yet more regional models to it. Therefore the improvement obtained with the hybrid set can confidently be attributed to the different nature of the global models. The study strongly reaffirms the importance of an in-depth inspection of any ensemble of opportunity in order to extract the maximum amount of information and to have full control over the data used in the construction of the ensemble.

The impact of air pollution on human health and the associated external costs in Europe and the United States (US) for the year 2010 are modeled by a multi-model ensemble of regional models in the frame of the third phase of the Air Quality Modelling Evaluation International Initiative (AQMEII3). The modeled surface concentrations of O3, CO, SO2 and PM2.5 are used as input to the Economic Valuation of Air Pollution (EVA) system to calculate the resulting health impacts and the associated external costs from each individual model. Along with a base case simulation, additional runs were performed introducing 20 % anthropogenic emission reductions both globally and regionally in Europe, North America and east Asia, as defined by the second phase of the Task Force on Hemispheric Transport of Air Pollution (TF-HTAP2). Health impacts estimated by using concentration inputs from different chemistry–transport models (CTMs) to the EVA system can vary up to a factor of 3 in Europe (12 models) and the United States (3 models). In Europe, the multi-model mean total number of premature deaths (acute and chronic) is calculated to be 414 000, while in the US, it is estimated to be 160 000, in agreement with previous global and regional studies. The economic valuation of these health impacts is calculated to be EUR 300 billion and 145 billion in Europe and the US, respectively. A subset of models that produce the smallest error compared to the surface observations at each time step against an all-model mean ensemble results in increase of health impacts by up to 30 % in Europe, while in the US, the optimal ensemble mean led to a decrease in the calculated health impacts by ∼ 11 %. A total of 54 000 and 27 500 premature deaths can be avoided by a 20 % reduction of global anthropogenic emissions in Europe and the US, respectively. A 20 % reduction of North American anthropogenic emissions avoids a total of ∼ 1000 premature deaths in Europe and 25 000 total premature deaths in the US. A 20 % decrease of anthropogenic emissions within the European source region avoids a total of 47 000 premature deaths in Europe. Reducing the east Asian anthropogenic emissions by 20 % avoids ∼ 2000 total premature deaths in the US. These results show that the domestic anthropogenic emissions make the largest impacts on premature deaths on a continental scale, while foreign sources make a minor contribution to adverse impacts of air pollution.

The work here complements the overview analysis of the modelling systems participating in the third phase of the Air Quality Model Evaluation International Initiative (AQMEII3) by focusing on the performance for hourly surface ozone by two modelling systems, Chimere for Europe and CMAQ for North America. The evaluation strategy outlined in the course of the three phases of the AQMEII activity, aimed to build up a diagnostic methodology for model evaluation, is pursued here and novel diagnostic methods are proposed. In addition to evaluating the base case simulation in which all model components are configured in their standard mode, the analysis also makes use of sensitivity simulations in which the models have been applied by altering and/or zeroing lateral boundary conditions, emissions of anthropogenic precursors, and ozone dry deposition. To help understand of the causes of model deficiencies, the error components (bias, variance, and covariance) of the base case and of the sensitivity runs are analysed in conjunction with timescale considerations and error modelling using the available error fields of temperature, wind speed, and NOx concentration. The results reveal the effectiveness and diagnostic power of the methods devised (which remains the main scope of this study), allowing the detection of the timescale and the fields that the two models are most sensitive to. The representation of planetary boundary layer (PBL) dynamics is pivotal to both models. In particular, (i) the fluctuations slower than ∼ 1.5 days account for 70–85 % of the mean square error of the full (undecomposed) ozone time series; (ii) a recursive, systematic error with daily periodicity is detected, responsible for 10–20 % of the quadratic total error; (iii) errors in representing the timing of the daily transition between stability regimes in the PBL are responsible for a covariance error as large as 9 ppb (as much as the standard deviation of the network-average ozone observations in summer in both Europe and North America); (iv) the CMAQ ozone error has a weak/negligible dependence on the errors in NO2, while the error in NO2 significantly impacts the ozone error produced by Chimere; (v) the response of the models to variations of anthropogenic emissions and boundary conditions show a pronounced spatial heterogeneity, while the seasonal variability of the response is found to be less marked. Only during the winter season does the zeroing of boundary values for North America produce a spatially uniform deterioration of the model accuracy across the majority of the continent.

Through the comparison of several regional-scale chemistry transport modeling systems that simulate meteorology and air quality over the European and North American continents, this study aims at (i) apportioning error to the responsible processes using timescale analysis, (ii) helping to detect causes of model error, and (iii) identifying the processes and temporal scales most urgently requiring dedicated investigations. The analysis is conducted within the framework of the third phase of the Air Quality Model Evaluation International Initiative (AQMEII) and tackles model performance gauging through measurement-to-model comparison, error decomposition, and time series analysis of the models biases for several fields (ozone, CO, SO2, NO, NO2, PM10, PM2. 5, wind speed, and temperature). The operational metrics (magnitude of the error, sign of the bias, associativity) provide an overall sense of model strengths and deficiencies, while apportioning the error to its constituent parts (bias, variance, and covariance) can help assess the nature and quality of the error. Each of the error components is analyzed independently and apportioned to specific processes based on the corresponding timescale (long scale, synoptic, diurnal, and intraday) using the error apportionment technique devised in the former phases of AQMEII. The application of the error apportionment method to the AQMEII Phase 3 simulations provides several key insights. In addition to reaffirming the strong impact of model inputs (emission and boundary conditions) and poor representation of the stable boundary layer on model bias, results also highlighted the high interdependencies among meteorological and chemical variables, as well as among their errors. This indicates that the evaluation of air quality model performance for individual pollutants needs to be supported by complementary analysis of meteorological fields and chemical precursors to provide results that are more insightful from a model development perspective. This will require evaluation methods that are able to frame the impact on error of processes, conditions, and fluxes at the surface. For example, error due to emission and boundary conditions is dominant for primary species (CO, particulate matter (PM)), while errors due to meteorology and chemistry are most relevant to secondary species, such as ozone. Some further aspects emerged whose interpretation requires additional consideration, such as the uniformity of the synoptic error being region- and model-independent, observed for several pollutants; the source of unexplained variance for the diurnal component; and the type of error caused by deposition and at which scale.

We present an overview of the coordinated global numerical modelling experiments performed during 2012–2016 by the Task Force on Hemispheric Transport of Air Pollution (TF HTAP), the regional experiments by the Air Quality Model Evaluation International Initiative (AQMEII) over Europe and North America, and the Model Intercomparison Study for Asia (MICS-Asia). To improve model estimates of the impacts of intercontinental transport of air pollution on climate, ecosystems, and human health and to answer a set of policy-relevant questions, these three initiatives performed emission perturbation modelling experiments consistent across the global, hemispheric, and continental/regional scales. In all three initiatives, model results are extensively compared against monitoring data for a range of variables (meteorological, trace gas concentrations, and aerosol mass and composition) from different measurement platforms (ground measurements, vertical profiles, airborne measurements) collected from a number of sources. Approximately 10 to 25 modelling groups have contributed to each initiative, and model results have been managed centrally through three data hubs maintained by each initiative. Given the organizational complexity of bringing together these three initiatives to address a common set of policy-relevant questions, this publication provides the motivation for the modelling activity, the rationale for specific choices made in the model experiments, and an overview of the organizational structures for both the modelling and the measurements used and analysed in a number of modelling studies in this special issue.

Simulations from chemical weather models are subject to uncertainties in the input data (e.g. emission inventory, initial and boundary conditions) as well as those intrinsic to the model (e.g. physical parameterization, chemical mechanism). Multi-model ensembles can improve the forecast skill, provided that certain mathematical conditions are fulfilled. In this work, four ensemble methods were applied to two different datasets, and their performance was compared for ozone (O3), nitrogen dioxide (NO2) and particulate matter (PM10). Apart from the unconditional ensemble average, the approach behind the other three methods relies on adding optimum weights to members or constraining the ensemble to those members that meet certain conditions in time or frequency domain. The two different datasets were created for the first and second phase of the Air Quality Model Evaluation International Initiative (AQMEII). The methods are evaluated against ground level observations collected from the EMEP (European Monitoring and Evaluation Programme) and AirBase databases. The goal of the study is to quantify to what extent we can extract predictable signals from an ensemble with superior skill over the single models and the ensemble mean. Verification statistics show that the deterministic models simulate better O3 than NO2 and PM10, linked to different levels of complexity in the represented processes. The unconditional ensemble mean achieves higher skill compared to each station's best deterministic model at no more than 60 % of the sites, indicating a combination of members with unbalanced skill difference and error dependence for the rest. The promotion of the right amount of accuracy and diversity within the ensemble results in an average additional skill of up to 31 % compared to using the full ensemble in an unconditional way. The skill improvements were higher for O3 and lower for PM10, associated with the extent of potential changes in the joint distribution of accuracy and diversity in the ensembles. The skill enhancement was superior using the weighting scheme, but the training period required to acquire representative weights was longer compared to the sub-selecting schemes. Further development of the method is discussed in the conclusion.

In this study, methods are proposed to diagnose the causes of errors in air quality (AQ) modelling systems. We investigate the deviation between modelled and observed time series of surface ozone through a revised formulation for breaking down the mean square error (MSE) into bias, variance and the minimum achievable MSE (mMSE). The bias measures the accuracy and implies the existence of systematic errors and poor representation of data complexity, the variance measures the precision and provides an estimate of the variability of the modelling results in relation to the observed data, and the mMSE reflects unsystematic errors and provides a measure of the associativity between the modelled and the observed fields through the correlation coefficient. Each of the error components is analysed independently and apportioned to resolved processes based on the corresponding timescale (long scale, synoptic, diurnal, and intra-day) and as a function of model complexity.The apportionment of the error is applied to the AQMEII (Air Quality Model Evaluation International Initiative) group of models, which embrace the majority of regional AQ modelling systems currently used in Europe and North America.The proposed technique has proven to be a compact estimator of the operational metrics commonly used for model evaluation (bias, variance, and correlation coefficient), and has the further benefit of apportioning the error to the originating timescale, thus allowing for a clearer diagnosis of the processes that caused the error.

The multimodel ensemble exercise performed within the HTAP project context (Fiore et al., 2009) is used here as an example of how a pre-inspection, diagnosis and selection of an ensemble, can produce more reliable results. The procedure is contrasted with the often-used practice of simply averaging model simulations, assuming different models produce independent results, and using the diversity of simulation as an illusory estimate of model uncertainty. It is further and more importantly demonstrated how conclusions can drastically change when future emission scenarios are analysed using an un-inspected ensemble. The HTAP multimodel ensemble analysis is only taken as an example of a widespread and common practice in air quality modelling.

Online coupled mesoscale meteorology atmospheric chemistry models have undergone a rapid evolution in recent years. Although mainly developed by the air quality modelling community, these models are also of interest for numerical weather prediction and regional climate modelling as they can consider not only the effects of meteorology on air quality, but also the potentially important effects of atmospheric composition on weather. Two ways of online coupling can be distinguished: online integrated and online access coupling. Online integrated models simulate meteorology and chemistry over the same grid in one model using one main time step for integration. Online access models use independent meteorology and chemistry modules that might even have different grids, but exchange meteorology and chemistry data on a regular and frequent basis. This article offers a comprehensive review of the current research status of online coupled meteorology and atmospheric chemistry modelling within Europe. Eighteen regional online coupled models developed or being used in Europe are described and compared. Topics discussed include a survey of processes relevant to the interactions between atmospheric physics, dynamics and composition; a brief overview of existing online mesoscale models and European model developments; an analysis on how feedback processes are treated in these models; numerical issues associated with coupled models; and several case studies and model performance evaluation methods. Finally, this article highlights selected scientific issues and emerging challenges that require proper consideration to improve the reliability and usability of these models for the three scientific communities: air quality, numerical meteorology modelling (including weather prediction) and climate modelling. This review will be of particular interest to model developers and users in all three fields as it presents a synthesis of scientific progress and provides recommendations for future research directions and priorities in the development, application and evaluation of online coupled models.

We explicitly address the fundamental issue of member diversity in multi-model ensembles. To date, no attempts in this direction have been documented within the air quality (AQ) community despite the extensive use of ensembles in this field. Common biases and redundancy are the two issues directly deriving from lack of independence, undermining the significance of a multi-model ensemble, and are the subject of this study. Shared, dependant biases among models do not cancel out but will instead determine a biased ensemble. Redundancy derives from having too large a portion of common variance among the members of the ensemble, producing overconfidence in the predictions and underestimation of the uncertainty. The two issues of common biases and redundancy are analysed in detail using the AQMEII ensemble of AQ model results for four air pollutants in two European regions. We show that models share large portions of bias and variance, extending well beyond those induced by common inputs. We make use of several techniques to further show that subsets of models can explain the same amount of variance as the full ensemble with the advantage of being poorly correlated. Selecting the members for generating skilful, non-redundant ensembles from such subsets proved, however, non-trivial. We propose and discuss various methods of member selection and rate the ensemble performance they produce. In most cases, the full ensemble is outscored by the reduced ones. We conclude that, although independence of outputs may not always guarantee enhancement of scores (but this depends upon the skill being investigated), we discourage selecting the members of the ensemble simply on the basis of scores; that is, independence and skills need to be considered disjointly.

In this study we present a novel approach for improving the air quality predictions using an ensemble of air quality models generated in the context of AQMEII (Air Quality Model Evaluation International Initiative). The development of the forecasting method makes use of modelled and observed time series (either spatially aggregated or relative to single monitoring stations) of ozone concentrations over different areas of Europe and North America. The technique considers the underlying forcing mechanisms on ozone by means of spectrally decomposed previsions. With the use of diverse applications, we demonstrate how the approach screens the ensemble members, extracts the best components and generates bias-free forecasts with improved accuracy over the candidate models. Compared to more traditional forecasting methods such as the ensemble median, the approach reduces the forecast error and at the same time it clearly improves the modelled variance. Furthermore, the result is not a mere statistical outcome depended on the quality of the selected members. The few individual cases with degraded performance are also identified and analysed. Finally, we show the extensions of the approach to other pollutants, specifically particulate matter and nitrogen dioxide, and provide a framework for its operational implementation. *One out of many