Uncertainties in future ozone and PM10 projections over Europe from a regional climate multiphysics ensemble
Abstract
[1] Due to the computational time required for modeling air quality climatologies, the characterization of processes introducing the largest uncertainty in air quality‐climate projections is a sound field of research. Here an air quality ensemble is assessed over Europe for present (1971–2000) and future (2071–2100, SRES A2) periods to characterize the sensitivity of regional air quality projections to the physics of the regional climate model driving the simulations. The ensemble comprises eight members resulting from combining two options of parameterization schemes for the planetary boundary layer, cumulus, and microphysics. The differences in the ensemble members (spread) for the concentration of tropospheric ozone and particulate matter (PM10) are strongly affected by the physics selected and could be considered as a matter of uncertainty in the change signals. Also, the leading processes causing the largest uncertainties in air quality projections have been identified and are mainly related to the election of the cumulus schemes.
1 Introduction
[2] The impact of regional climate changes on air pollution and their feedbacks over Europe is an important and innovative topic, increasingly receiving attention [e.g., Forkel et al., 2012; Meier et al., 2012; Tuccella et al., 2012; Hedegaard et al., 2013]. Changes in climate (i.e., temperature, wind speed, height of mixing layer, convection, fronts pass frequency) affect air quality by changing the dispersion, wet and dry deposition, photochemical processes, as well as natural emissions and background concentrations [Jacob and Winner, 2009]. The magnitude and extent of potential impacts of climate change on air quality in Europe urged major government agencies to undertake several important actions in this regard [Amann et al., 2004; European Commission, 2010]. The impact studies to be performed under this umbrella inevitably require downscaled scenarios of air quality under climate change scenarios, with as high spatial resolution as possible. To achieve this aim, the most common methodology is the online or off‐line coupling of chemistry transport models (CTMs) to regional climate models (RCMs). However, recent studies applying both modeling approaches [Manders et al., 2012; Juda‐Rezler et al., 2012, among many others] highlight a wide spread in the results related to the influence of climate change on regional air pollution at regional scales over Europe.
[3] In the development of downscaled air quality scenarios under future greenhouse gases (GHGs) and aerosol enhanced scenarios, there are several sources of uncertainty, including most notably the following: (1) the different possible evolutions of emissions (both of greenhouse gasses, trace gasses, and aerosols) [e.g., Brasseur and Roeckner, 2005], (2) those associated with global models themselves [Brands et al., 2013], (3) those introduced in the process of downscaling [e.g., Schmidli et al., 2007], and (4) those conditioned by the choice of the parameterization in the RCM [Yang et al., 2012; Vautard et al., 2013]. While the three first sources of uncertainty have been more widely studied, the latter has received less attention, not only when considering the problem from the air quality perspective but even just from the climate perspective. For example, while multimodel ensembles of regional climate simulations have been widely performed and investigated in an attempt to evaluate and overcome intermodel‐related uncertainties, few studies deal with similar multiphysics ensembles for elucidating intramodel uncertainties from a regional climate perspective [Yang and Arritt, 2002]. Moreover, to the best of our knowledge, there are no studies dealing with the uncertainty in air quality projections resulting from the physics employed in the RCM providing the driving regional climate simulations for the CTM. However, the use of different physical parameterizations alters the modeling of processes that may be intimately involved in the dispersion of air pollutants [Jerez et al., 2013a, 2013b].
[4] Under this umbrella, the main objective of this work is to conduct a comparative numerical modeling study of air quality projections from a climatic perspective using a multiphysics ensemble of simulations covering Europe. There are several aspects addressed in this work, namely the characterization of the mean change signals and the associated spread, and the identification of the processes determining the largest discrepancies among the various simulations. This latter goal is critical, since the large computational time needed for modeling air quality climatologies fosters the definition of the leading processes causing discrepancies when designing strategies for ensembles of air quality climate. In this sense, minimizing the number of simulations needed to characterize the uncertainty is demanding.
2 Simulations and Methods
2.1 Models and Domains
[5] This work assesses two sets of MM5‐CHIMERE (Fifth‐Generation Pennsylvania State University‐National Center for Atmospheric Research Mesoscale Model off‐line coupled to the CHIMERE chemistry transport model) simulations spanning the periods 1971–2000, as a control reference period (CTRL), and 2071–2100, as a future enhanced greenhouse gas and aerosol concentrations scenario (SCEN), respectively. The differences between the climatologies of these two runs will provide the change signal (CHNG).
[6] The regional climate simulations driving the CTM were performed using the regional climate model MM5 driven by the European Centre/Hamburg 5‐Run1 simulation forced by the SRES‐A2 scenario. No nudging of the regional model toward the global climate model was used. For further description of the regional climate simulations and the validation of the present‐day simulated climatologies, the reader is referred to Jerez et al.[2010, 2013a]. MM5 presents a wide spectrum of physics options, thus making it a suitable tool for assessing the role of the parametrization schemes on climate at regional scales. Each set of simulations consists of a multiphysics ensemble comprising eight members that result from varying the physical configuration of MM5 as follows. For the planetary boundary layer (PBL), we use either the Medium‐Range Forecast (MRF) [Hong and Pan, 1996] or the Eta [Janjic, 1994] model; for cumulus (CML), we use either the Grell (GR) [Grell, 1993] or the Kain‐Fritsch (KF) [Kain and Fritsch, 1990] scheme, and for the microphysics (MIC), we use either the Simple Ice (SI) [Dudhia, 1989] or the Mixed Phase (MP) [Reisner et al., 1998] model. Table 1 shows the eight distinct configuration schemes used in each ensemble member.
| Model | MM5 | CHIMERE (Common for all Members) | ||
|---|---|---|---|---|
| Member | PBL | CML | MIC | Chemical Mechanism: MELCHIOR2 |
| 1 | Eta | Grell | Simple Ice | Aerosol Chemistry: Inorganic |
| 2 | MRF | Grell | Simple Ice | (Thermodynamic Equilibrium With |
| 3 | Eta | Kain‐Fritsch | Simple Ice | ISORROPIA) and Organic (MEGAN SOA) |
| 4 | MRF | Kain‐Fritsch | Simple Ice | Scheme) Aerosol Chemistry |
| 5 | Eta | Grell | Mixed Phase | Natural Aerosols: Dust, Resuspension and |
| 6 | MRF | Grell | Mixed Phase | Inert Sea Salt |
| 7 | Eta | Kain‐Fritsch | Mixed Phase | BC: LMDz‐INCA+GOCART |
| 8 | MRF | Kain‐Fritsch | Mixed Phase | Emissions: EMEP |
[7] Although we are aware that other parametrized processes may play an important role, such as the land‐surface processes [Jerez et al., 2012], this ensemble was based on previous works that either identified a better performance of one scheme over the others (e.g., Jerez et al.[2010], regarding the Noah Land‐Surface Model) or concluded that none of the schemes outperforms systematically the others (e.g., Fernández et al.[2007] and Jerez et al.[2013a], regarding the PBL, CML, and MIC schemes). It is in this latter case where the characterization of the uncertainties is useful in the design of an ensemble.
[8] CHIMERE chemistry transport model [Rouil et al., 2009] was driven off‐line by the MM5 fields. CHIMERE's meteorological preprocessor was modified in order to prepare the MM5 variables to be read by the model core. Further detail on the full configuration on the runs can be found in Jiménez‐Guerrero et al.[2011, 2012] and Table 1. The chemical options have been kept constant in order to isolate the impact of changing the physical configuration of the driving RCM. Validation of present‐day climatologies runs can be found in Jerez et al. [2013c]. The spatial model configuration comprises a domain covering most of Europe with a resolution of 90 km for MM5‐RCM simulations (24 sigma levels are considered in the vertical, with the top one at 100 hPa) that are interpolated to a resolution of 25 km for the CTM.
[9] The role of the model physics in both tropospheric ozone (O3) and particulate matter (PM10) is analyzed here. Present and future periods are considered separately from projected changes (i.e., future‐minus‐present approach). Following the methodology proposed by Jerez et al.[2013a, 2013b], we have focused on the analysis on the Ensemble Mean (EM), mean value computed from the all values provided by every single member of the ensemble, the Ensemble Spread (ES), the maximum difference among the various ensemble members, and the Leading parametrized Process (LP), the scheme whose change provokes the largest differences among simulations.
3 Results
[10] Figure 1 (left) shows the present‐day climatologies depicted by the EM and the associated ES in the CTRL simulations in the winter (December‐January‐February, DJF) and summer (June‐July‐August, JJA) seasons for both O3 and PM10; the EM change (amplitude and signal) and the accompanying spread are shown in Figure 1 (right).
[11] In general, the EM from the CTRL simulations indicates that both seasons show a similar spatial distribution for ground‐level pollutants, although with higher concentrations of pollutants in summer. This is expected for O3 due to the enhanced photochemistry and for PM10 due to the abundant formation of secondary organic aerosols (mainly with a biogenic origin) especially over the Eastern Mediterranean Basin [Jiménez‐Guerrero et al., 2011], the more stagnant conditions and reduced precipitation. Looking at the JJA patterns for O3, Italy (specially the Alpine region), the Balkans, and the eastern coast of the Iberian Peninsula exceed the 100 μg m−3, with DJF values not exceeding 80 μg m−3. In the case of PM10, the most outstanding areas during JJA are the eastern Mediterranean and the southwestern Iberian Peninsula (seasonal averages above 40 μg m−3), while the highest DJF averages are found over the eastern Mediterranean (25 μg m−3). It is noteworthy that for PM10, the spread associated to these concentrations reaches up to 70% of their mean values. For O3, however, the spreads are much lower, reaching up to 30% as maximum over eastern Europe during both summertime and wintertime.
[12] Projected O3 and PM10 mean changes and accompanying spreads depict diverse signals across the target region, since climate change impacts air pollution as a consequence of a variety of processes that vary from one region to another, namely temperature increase, changes in precipitation, decrease of the mixing heights (hampering the dilution of pollutants), and favored stagnant conditions [Katragkou et al., 2011; Jiménez‐Guerrero et al., 2011]. The largest projected increases of O3(CHNG signal) are found for summertime over southern Europe, especially in eastern Spain (the only region where projected changes are noticeable in DJF, +5 μg m−3, although with the spread being the same order of magnitude), the Italian Peninsula, and the Balkans. In these regions, projected changes for summer can add up to +10 μg m−3, although these CHNG signals have associated spreads of 8 μg m−3, that is, around 80% of the changing signal.
[13] Regarding PM10, the enhanced future secondary activity in southern Europe related to increased temperatures provokes changes over the Iberian Peninsula and the Balkans region up to +5 μg m−3, especially during JJA. Contrary, northern areas show even slight decreases during DJF (−1μg m−3). Usually, large spreads of values are associated with the highest PM10 CHNG signals. However, this is not always the case. For instance, over the western Mediterranean basin or over France during DJF, there is no noticeable increase in the concentration of aerosols from the EM, but the spread is 1 to 2 μg m−3, highlighting the strong uncertainty associated to PM10 projections in this area/season. The high spreads associated to a low mean change also indicate a high uncertainty in the sign of the projected change, with some ensemble members projecting increases and other decreases. Actually, in most parts of the domain, the spread represents above 100% of the ensemble mean‐projected change for PM10, and hence, air quality patterns related to particulate matter show a great sensitivity to the physical configuration of the RCM.
[14] Figure 2 (top) depicts the leading processes (LPs defined as those parameterizations whose change provokes the largest spread) regarding the uncertainty associated to the seasonal projections presented. If LPs are evaluated over the areas showing the largest spreads, some can be easily distinguished. For O3 (and most of gas‐phase pollutants, nonshown, like SO2), the largest spreads over land are related to changes of the microphysics scheme in DJF (north of France and southern Germany), the cumulus scheme over the Mediterranean countries, and the PBL scheme over the ocean. However, for JJA, the spreads are dominated by the cumulus scheme over most of the target domain (PBL governs over some mountain areas), while the change of the microphysics scheme seems less relevant. For PM10 over land, the cumulus scheme determines the spread for both DJF and JJA (except in some northern areas, with PBL determining the spread), with the KF scheme leading to the largest variation of the precipitation amount in the future. However, the huge spread affecting the positive signals over some Mediterranean ocean areas is controlled by the PBL scheme alone in DJF.
[15] Besides mean values, another aspect highlighting the uncertainty associated to the election of a physics ensemble member is related to the occurrence of extreme episodes of air pollution. In this sense, Figure 2 (bottom) indicates that the configuration responsible for simulating the maxima of a given period over a target area is not the same for present‐day conditions (CTRL), future scenarios (SCEN), and projected changes (CHNG). For instance, maximum O3summertime concentrations are generally provided by MRF‐GR‐SI ensemble member over land and MRF‐GR‐MP over the ocean, for both present‐day and future climate simulations. However, the maximum‐projected increases in these maximum concentrations are led by a diversity of members, depending on the area (e.g., ETA‐GR‐MP over southwestern Iberian Peninsula, MRF‐KF‐MP over northern Europe, and ETA‐GR‐SI over southeastern Europe, including also most of the central and western Mediterranean). This dependence becomes more evident for PM10. Maximum levels for PM10 are simulated generally with MRF‐KF‐MP for CTRL, MRF‐KF‐SI for SCEN, while maximum‐projected changes are mostly led by ETA‐KF‐SI over northern Europe and ETA‐KF‐MP over more southern areas. Therefore, the use of the KF scheme combined to the ETA PBL scheme generally leads to the largest projected increases in the concentrations of PM10 over all Europe. It is, however, not relevant for the objectives of this work and beyond the scope of this paper to analyze the physical reasons behind this spread, but it is noteworthy that maximum concentrations (associated to extreme episodes of air pollution) are modeled with different configurations for CTRL, SCEN, and CHNG, and this may condition modeling strategies when assessing future episodes of air pollution.
4 Discussion
[16] This assessment further reveals the great sensitivity of Europe to future climate change and how this propagates into the regional projections of air pollution. A valuable conclusion drawn is that uncertainties affecting atmospheric variables also affect the air quality patterns and associated spreads, which show a great sensitivity to the physical configuration of the regional climate model selected. Although it is beyond the scope of the paper to analyze in depth which are the physical causes behind the spreads for present and future air pollution (results further claim for future studies aimed at deepening the knowledge about the processes and characterizing uncertainties), the work tried to identify the processes leading to the highest spreads (namely, the highest uncertainties, mainly related to the election of the cumulus scheme for concentrations over land) when projecting air quality by the end of the XXI century. This is of special relevance when designing air quality modeling strategies aimed at assessing the uncertainties in future air pollution, since this contribution may help identifying which processes we should pay attention to in order to have the largest spread with the minimum number of simulations (and then reducing the needed computational time, which is importantly demanding in this type of simulations).
[17] As commented before, both spreads and leading processes are different in CTRL and CHNG (and also for SCEN simulations, not shown). This suggests in particular that while some processes could deserve little attention in the CTRL case (at least the way in which they are modeled), their influence grows for future projections. This may have important implications for the future‐minus‐present approach applied generally when projecting changes in air pollution under future scenarios. This method is widely used in air quality projections (the reader is referred, for instance, to Jacob and Winner[2009, and references therein]), based on the assumption that biases cancel when estimating the difference between future and present‐day concentrations. However, our results indicate that under different climatic conditions, as they may be in the future, some processes gain relevance and the cancelation of errors works only partially (otherwise no spreads would appear in the ensemble of air quality projections). So the future‐minus‐present approach widely adopted for characterizing the changes in air quality under future scenarios should be, at least, carefully taken into account.
Acknowledgments
[18] This work was funded by the Portuguese Foundation of Science and Technology (FCT) (project ENAC PTDC/AAC‐CLI/103567/2008), the Spanish Ministry of Economy and Competitiveness (project SPEQTRES), and the “Fondo Europeo de Desarrollo Regional” (FEDER) (project CORWES CGL2010‐22158‐C02‐02). Pedro Jiménez‐Guerrero acknowledges the Ramón y Cajal programme.
[19] The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.
