On the Impact of Time-Resolved Boundary Conditions on the Simulation of Surface Ozone and PM10

Air pollution has always been a trans-boundary environmental problem and a matter of global concern for past many years. High concentrations of air pollutants due to numerous anthropogenic activities influence the air quality. There are many books on this subject, but the one in front of you will probably help in filling the gaps existing in the area of air quality monitoring, modelling, exposure, health and control, and can be of great help to graduate students professionals and researchers. The book is divided in two volumes dealing with various monitoring techniques of air pollutants, their predictions and control. It also contains case studies describing the exposure and health implications of air pollutants on living biota in different countries across the globe.


Introduction
The grid-spacing of chemistry-transport models (CTM) is always limited by computational resources and ranges from 100-200 km of global scale models, to 25-50 km of continental scale models, to 1-10 km of regional and local scale models. We push to higher resolution in hope of better reproducing small scale processes that affect our ability to assess the environmental and health impacts of emissions. Running simulations at a resolution less than 50 km is often feasible only using a limited area model, which uses a domain ascribed to the region of interest. However, even the air quality of a single city is in principle affected by all the emission sources at global level: we thus account for such long-range transport of pollutants and oxidants specifying the chemical state of the atmosphere outside the domain through boundary conditions (BC). BC concentrations are usually taken from typical profiles or from larger-scale simulations with a procedure called "nesting". In this chapter, we focus on the latter technique, exploring in particular the effect of different BC timeresolutions (monthly to hourly) on the simulation of ozone and particulate matter on a nested domain at European scale. For the sake of completeness, we point out here that even very high-resolution models cannot explicitly simulate processes at all possible spatialtemporal scales and thus a certain degree of parameterization is always required. For further insights on the "subgrid" issues we refer the reader to the literature (e.g. Galmarini et al., 2008;Qian et al., 2010;Denby et al., 2011;Paoli et al., 2011).
In addition to boundary conditions, chemistry-transport model simulations also require initial conditions (IC) for chemical species. The general aspects of the influences of IC and BC on the simulation may be understood in the simplified framework presented by Liu et al. (2001). The authors consider an Eulerian box model with one chemical species, whose evolution of concentration C is regulated by the species continuity equation: where P and L are the production and loss rates, respectively, C BC is the background concentration, which represent the boundary condition in this case, and τ r is the residence time into the box. The third term on the right-hand side isolates the source term attributable to the BC. The analytical solution of the equation is as follows (eq. 4 in Liu et al., 2001): ( 2) where C IC is the initial condition. The influence of C IC exponentially decreases with time, due both to photochemical loss and deposition (L) and outward-transport (1/τ r ), thus it vanishes if a sufficient "spin-up" time is allowed. On the other hand, the importance of local sources (second term on the r.h.s.) and boundary conditions (third term on the r.h.s.) grow with time and drive the evolution of C after the "spin-up" time. The importance of BC is to be evaluated comparing the relative magnitudes of the local production term Pτ r (photochemical plus emission) against C BC . If the local sources are much larger than C BC , the influence of BC might be ignored. It is expected that boundary influence decreases during downwind transport and that it reaches a maximum when the arrival time is short and the species lifetime is long.
The simple considerations about the influence of IC and BC obtained from the solution in equation (2) of the box model, was demonstrated to be valid also for full three-dimensional Eulerian chemistry-transport models. Regarding IC, Berge et al. (2001) found that their influence a 3-D model decreases more slowly with time with respect to a box model, but still is reduced to <10% in the planetary boundary layer (PBL) after 3 days in a 400 km × 480 km domain covering Southern California. However, the same authors pointed out that the influence of IC might be >10% after 3 days for long lived grouped species (e.g. sum of reservoir species of ozone) and in the free troposphere. A similar spin-up time of 2 days for ozone in the PBL was reported by Jiménez et al. (2007) for a 272 km × 272 km covering North-Eastern Iberian Peninsula. On a larger domain covering Europe, Langmann & Bauer (2002) found that 5 days are needed by ozone in the PBL to "forget" its initial condition. More recently, a further sensitivity test at the North American continental scale, confirmed that a week is the minimum spin-up time recommended for a 3-D simulation of ozone and particulate matter (Samaali et al., 2009). Liu et al. (2001) analyzed the influence of BC on their 3-D ozone simulation over California using the difference of concentrations between a reference run and another with zeroed boundary concentrations. They found that the percentage of ozone concentration attributable to BC is mostly determined by the distance from the domain edges, the influence being inversely proportional to the distance. The influence at a specific location and time is modulated by the characteristics of the local and upwind sources. During night the impact of BC on ozone is less than daytime, because of a less active photochemistry. It was calculated that BC may contribute 30-40% of ozone formation in polluted PBL, while stratospheric BC dominated ozone values in the free troposphere (Langmann & Bauer, 2002;Song et al., 2008). Barna & Knipping (2006) pointed out that a different representation of BC has a great impact on source-apportionment analysis.
While Liu et al. (2001) helped clarifying the general concepts of the influence of BC on chemistry-transport model simulations, several other studies focused on the impact of improved boundary conditions on simulations. Many studies found that increasing both temporal and spatial resolution of BC benefit the ozone (Langmann et al., 2003;Appel et al., 2007;Song et al., 2008;Szopa et al., 2009), carbon monoxide (Tang et al., 2007;Tombrou et al., 2009) and particulate matter simulation (Barna & Knipping, 2006;Borge et al., 2010). On the continental scale, BC have a significant impact on ozone background levels, while having much less impact on the variability and peak values (Tang et al., 2007;Szopa et al., 2009). In the free troposphere, a careful treatment of the variable tropopause is critical for a real advantage in using improved stratospheric ozone BC on model top (Lam and Fu, 2009;Makar et al., 2010).
The chapter is organized as follows. We first briefly describe the models used in the study in section 2. Then we study the impact and time scales of IC and BC on simulated ozone in section 3.1. In the same section, we analyse the difference among surface ozone simulations with the use of boundary conditions alternatively with monthly, daily or hourly update rate. In section 3.2 we analyse the effect of alternative BC on surface PM10, in particular during a Saharan dust event in July 2005. The scientific questions we shall try to address are:  How long should be the spin-up time for simulated surface concentrations to be unaffected by initial conditions?  How much is the contribution of boundary conditions to local ozone and PM10 levels?  Is there any improvement in simulations of ozone and PM10 if the boundary conditions are provided at an higher frequency (up to the model time-step)?
In final section 4 we draw conclusions on these questions.

Models description
In this section we give a brief description of chemistry-transport models used in this study. We use the CHIMERE regional model  to simulate lower atmosphere composition over continental Europe and the GEOS-Chem global model (Bey et al., 2001) to provide CHIMERE with gases and dust boundary conditions.

CHIMERE regional model
CHIMERE is a regional chemistry-transport model developed by a community of French institutions primarily designed to produce daily forecasts of ozone, aerosols and other pollutants and make long-term simulations for emissions control scenarios CHIMERE, 2011). In this study, The model is setup on a 0.5°×0.5° horizontal grid covering Europe (35°-58°N; 15°W-25°E) and 8 hybrid-sigma vertical layers extending to 500 hPa. Meteorological input is provided by PSU/NCAR MM5 model (Dudhia, 1993) run at 36×36 km 2 horizontal resolution and 29 vertical sigma layers extending up to 100 hPa, and regridded on the 0.5° × 0.5° CHIMERE grid. The model is forced by NCEP analyses using the grid nudging (grid FDDA) option implemented in MM5.
Anthropogenic emissions are derived from the Co-operative Programme for Monitoring and Evaluation of the Long-range Transmission of Air pollutants in Europe (EMEP) annual totals (Vestreng, 2003) for NOx, CO, VOC, SOx, NH3 and primary PM species, while carbonaceous aerosol emissions are taken from Junker & Liousse (2008). Biogenic emissions of isoprene and monoterpenes are calculated with the MEGAN model (Guenther et al., 2006). Dust and sea-salt emissions are also simulated on-line (CHIMERE, 2011).
The gas-phase chemical mechanism MELCHIOR (Latuatti, 1997) includes about 50 species and 120 reactions. The aerosol phase is simulated with a sectional approach with 8 size bins from 0.04 to 10 µm of diameter. The main processes governing the production and loss of main inorganic (sulphate, nitrate, ammonium) and organic secondary species are included.
Boundary conditions are implemented in a classical one-way approach. Species concentrations at the boundaries are introduced in the simulation through an outer envelope of model cells having the same resolution of the actual simulation grid. The boundary concentrations are transported inside the domain by the transport operator, i.e. the part of the model that simulates advection (Fig. 1). Regardless of the particular scheme choice (CHIMERE, 2011), the model uses information only from one upwind cells to solve for advection. Numerical stability is warranted by the adaptive time-step adjusted in order to have a Courant number always less than 1, e.g. for the zonal direction: where C x is the Courant number for the x-direction, U is the maximum zonal wind speed, Δt is the time step, and Δx is the grid spacing. By definition, the Courant number measures the influence that upwind concentrations may have on a given grid-cell in a single time-step in units of the grid size. If this number is less than one it means that information from only one upwind cell is needed. In CHIMERE, the time-step is adapted throughout the simulation to ensure this condition always holds. For some more details on the relationship among advection schemes and boundary conditions the reader is referred to the nice discussion given by Wang et al. (2004). The transport operator of the model will use those cells for advection calculations and it will transport BC information into the domain.
In the default configuration, the boundary conditions for CHIMERE regional simulations are taken from monthly mean simulations of the global model LMDz-INCA (Hauglustaine et al., 2004) for species listed in Tab. 1. For aerosol species, size-resolved mass concentrations of global model are redistributed onto regional model size bins, using a simple linear interpolation in logarithmic bin diameters space. In Fig. 2 we show a sample of the monthly static ozone boundary conditions for the month of June.

GEOS-Chem global model
GEOS-Chem is a global chemistry model developed by a large international users community, originally stemming from the Harvard's Atmospheric Chemistry Modelling group (Bey et al., 2001;GEOS-Chem, 2011). The meteorological input is provided by the Goddard Earth Observing System (GEOS) of the NASA Global Modeling and Assimilation Office (GMAO). Mostly relevant to this study is the emission module developed by Fairlie et al. (2007), used here to include a more detailed contribution of Saharan dust emissions into the simulations. Desert dust emissions are lowered by a factor of three according to the study of Generoso et al. (2008).
GEOS-Chem simulations will be used in this study to produce alternative boundary conditions for the CHIMERE regional model at different time scales, from hourly to monthly.

Time-scales and impact of IC and BC
In order to study the effect of initial and boundary conditions on the CHIMERE ozone simulation at the European scale we use the simulations listed in Tab. 2. Basically, we alternatively zero IC and BC to isolate their effect through the difference with a reference www.intechopen.com simulation (Stein & Alpert, 1993). We choose a one month in summer (June 2005) in order to allow enough time to study the effect of IC and to ensure an active ozone photochemistry.

Simulation Label Description CTRL
Control simulation (reference) NIC No Initial Conditions (IC = 0) NBC No Boundary Conditions (BC = 0) Table 2. List of simulations performed to study the effect of initial and boundary conditions on ozone.
The reference ozone simulation of CHIMERE is quickly evaluated against ground based measurements available from the EMEP network (www.emep.int). Let


Pearson's Correlation (r) and coefficient of determination (R 2 ): where X is a generic vector and Z(X) is its standard score, also defined above. R 2 is defined as the square of r and denotes the fraction of variability of observations explained by the model.
Results for current simulation are presented in Tab. 3 and Fig. 3. The model captures the central part of the observed distribution, but overestimates its low end and underestimates the upper end. This is a quite typical behaviour of regional chemistry-transport models (e.g. Appel et al., 2007). The average MNBE and MNGE for ozone hourly timeseries is slightly above the quality thresholds recommended by EPA (15% and 35% respectively), but the indices are well within the suggested limits for the daily maxima timeseries.  The impact of IC on ozone is studied through the difference between CTRL and NIC runs. We arbitrarily define the influence of Ψ IC as the relative difference between CTRL and NIC simulations and the spin-up time τ IC as the time needed to reduce Ψ IC to less than 1%. In Fig.  4 we show the average ozone timeseries as measured and modelled at EMEP stations. After about 4 days of simulations the average Ψ IC becomes negligible. According to our definition the average τ IC is 3.9 days, but ranges from 0.5 to 8.1 days. In Fig. 5 we may appreciate the spatial distribution of τ IC . We find a clear positive gradient from the North-West to the South-East of the domain. The reason may be found in the ozone distribution itself. In Fig. 6 we show the average ozone concentrations simulated by the model and we note that τ IC gradients follows closely ozone gradients: the model just takes more time to build up ozone from the zero concentration starting point when the ozone level to be reached is higher. While this may pose questions on the method we used to estimate the "lifetime" of IC, this test is useful to verify that the model is essentially able to completely forget a whatever "wrong" initial condition after about 9 days of run. The influence of boundary conditions is studied in an analogous way. We define Ψ BC as the difference between CTRL and NBC runs and the time of arrival of BC τ BC as the first time when Ψ BC is larger than 1%. In Fig. 4 we see that the behaviour of Ψ BC mirrors that of Ψ IC . As expected, the influence of BC grows as that of IC decreases. According to our definition, the average τ BC is 16 hours, and ranges from 1 to 29 hours. As shown in Fig. 5, shortest times are found near the domain edges, while longest times are found in the interior of the domain. The variability inside the domain is attributable to the specific meteorological situations, since the time of arrival of BC at a specific location is determined by the winds encountered along the travel from the edges.
Until the spin-up time (τ IC ) is elapsed, Ψ BC continues to ramp until a plateau is reached: the model is "warmed-up", and the ozone production at certain place is determined by the equilibrium between the sources inside the domain and the boundary conditions (eq. 2). For any time and location, Ψ BC quantifies the relative influence of BC with respect to local production. The periods with higher values of Ψ BC in timeseries of Fig. 4, thus indicate periods of less intense photochemical activity.  The spatial distribution of the average Ψ BC at the surface calculated from 10 to 30 June 2005 with CHIMERE is shown in Fig. 7. As also noted above, we find that the maximum influence of BC is around the borders, and it is striking to see that it is higher than 80-90% even in the polluted North-Western Europe. Ψ BC reaches a minimum of less than 50% over Po Valley where the local production is invigorated by the elevated precursors emissions and very active photochemistry. This result imply that an error in BC may be effectively propagated into the simulation domain. For example, an error of 1 in the BC becomes 0.7 in a place where Ψ BC = 0.7. Liu et al. (2001) found that 3-D model results are consistent with analytical solution of a simple one-dimensional model, where the influence of BC may be written as: where x is the distance to boundary and u is the wind speed, so that x/u is the time of BC arrival. S is a step function which is 0 for t < x/u (before BC arrival) and 1 for t ≥ x/u (after BC arrival). If the ozone lifetime τ = 1/L is known, equation (8) can be applied directly to estimate Ψ BC . Since the lifetime is generally unknown, but the BC arrival time τ BC can be easily estimated, equation (8) can be inverted to roughly estimate the local averaged ozone lifetime. Inserting our definition of relative influence of BC on ozone Ψ BC we obtain: The result is shown in Fig. 7. Ozone lifetime estimated with this simple method ranges 1-3 days in the continental boundary layer and is longer than 5 days over the ocean, which is quite consistent with our expectations.

Effect of alternative BC on surface ozone
We analyse the effect of different BC on simulated surface ozone using four simulations as listed in Tab. 4. We choose a longer summer period of two months, June and July 2005, to have a more robust statistics. Consistently with what found in previous section, we spin-up the model for 10 days.

Simulation Label Description CTRL
Control simulation (w/ reference LMDz-INCA BC) BCGM BC from GEOS-Chem monthly output BCGD BC from GEOS-Chem daily output BCGH BC from GEOS-Chem hourly output Table 4. List of simulations performed to study the effect of alternative boundary conditions on ozone.
In Fig. 8 we compare the timeseries of ozone BC in the simulations averaged over the western border (leftmost rectangle in Fig. 2). The GEOS-Chem model simulates lower (higher) ozone values with respect to LMDz-INCA in June (July). One important reason, apart the many differences in models' formulation, is that the latter simulation is an average over five years of run, while GEOS-Chem simulates the "actual" (i.e. assimilated) meteorology of the CHIMERE simulation. The introduction of more detailed in time BC introduces much more variability, with differences up to ±30% with respect to the fixed monthly BC. In Tab. 5 we report the statistical summary, i.e. the average indices over all available EMEP stations, of the four simulations. The comparison with reference simulation over the twomonth period is consistent with the one month simulation without spin-up (Tab. 3). The introduction of GEOS-Chem BC benefits the high CHIMERE model bias, reducing both the normalized bias and the gross error, probably because they are more specific of the simulation period than the LMDz-INCA climatology. The introduction of time resolution into the BC produces further reduction of model error and also significantly increases the correlation with the measurements. In Fig. 9 we compare the simulated ozone with measurements at two selected EMEP stations, one near the border and very sensitive to BC (Mace Head, Ireland), the other about the centre of the domain and much less sensitive to BC (Heidenreichstein, Austria). For the Irish site we note interesting differences among the runs. During the first week of simulation, the monthly GEOS-Chem alternative BC enhances model underestimation with respect to reference, while the time-resolved GEOS-Chem BC slightly alleviates the bias with respect to reference. In the days around June 17 th the time-resolved BC allow the model to capture a low ozone episode, but the subsequent week GEOS-Chem values are even too low. Also in other periods, the time-resolved BC allow the model to go closer to observations (July 7-10, 20-23). The correlation with measurements goes from 0.38 of the CTRL run, to 0.29 of the BCGM, to 0.44 of BCGD and BCGH. The gross error is reduced from 26% to 22% from CTRL to BCGD and BCGH runs. We also point out that the difference of the impact of hourly and daily BC is minimal.
The effect of alternative BC on the Austrian site, as expected is much less evident. However, the statistical indices of comparison with observation constantly get better as we introduce more resolution in time. The correlation increases from 0.56 to 0.61, the bias decreases from 23% to 14%, and the gross error decreases from 39% to 35%. Again, we note that using hourly or daily resolved BC does not significantly impact the simulation.
These results point out that the time resolution of BC may greatly affect the simulation, but an higher resolution may episodically worsen model skills. The big step is between monthly and daily resolved BC, while going down to hourly resolved BC, at the expense of more disk space and pre-processing time, does not yield further significant improvements..

Effect of alternative BC on surface PM10
Similarly to ozone, we study the effect of alternative BC on simulated surface PM10 using the same simulations listed in Tab. 4. For this study, we choose to introduce alternative BC into CHIMERE only for dust, because the inflow of Saharan dust into the European domain is expected to contribute much more to the PM10 simulation than the other species Curci & Beekmann, 2007). While transboundary pollution from anthropogenic sources is expected to impact background levels of PM (Park et al., 2003;, Saharan dust may episodically yield to the exceedance of the PM10 limit for the protection of human health of 50 µg/m 3 (Gobbi et al., 2007;Koçak et al., 2007;Perrino et al., 2009). It is estimated that in Italy the subtraction of natural dust to PM10 may yield to a reduction from 5% to 50% of the number of threshold exceedances depending on the meteorology and the station type (Pederzoli et al., 2010).
In Fig. 10 we compare the boundary conditions to CHIMERE from the Southern border in the four simulations. GEOS-Chem monthly mean dust values are higher than LMDz-INCA, and similarly to ozone the time-resolved BC have a much higher variability, with differences of ±50% and one episode with hourly values three times higher than monthly values. In Tab. 6 we report the average statistical indices over all available EMEP stations of the four simulations. The higher dust values in GEOS-Chem BC drastically reduce the model bias from -40% to -7%, but the "noise" introduced by the large dust variability gradually degrades the correlation and the gross error as we increase the BC time resolution. This fact points out how tricky is the simulation of the Saharan dust contribution on European PM10 levels. In Fig. 11 we compare the PM10 timeseries at two EMEP stations, one near the South-Western border of the domain and more affected by Saharan dust (Barcarrota, Spain), and another to the North (Schauinsland, Germany). At the Spanish site, higher dust values in GEOS-Chem BC reduces the mean bias from -14 µg/m 3 to less than 1 µg/m 3 , but the correlation decreases from 0.77 to 0.56 in CTRL and BCGH runs, respectively. The better resolved BC allow the model to better capture the observed PM10 variability by the end of the simulated period, but they also induce episodic overshoots during the first period of simulation that are completely unrealistic. Very similar features may be also noted at the German site, indicating that the importance of dust BC are not limited to the Southern part of the European domain. We now focus on the dust episode of 27-29 July 2005. In Fig. 12, the daily AOD observed by MODIS/Aqua between 24-29 July clearly tracks a dust cloud swapping Southern Europe from West to East. In Fig. 13 we show the same episode as recorded at ground level by the EMEP network and simulated by CHIMERE. The introduction of time-resolved BC helps the www.intechopen.com model in better reproducing both timing and magnitude of the event. Unfortunately, as we have seen in previous timeseries, this is not a general conclusions and further work is certainly warranted on the dust BC issue. www.intechopen.com

Eye-witness of a Saharan dust event over Central Italy
In a sort of "divine intervention", Sahara desert decided to produce one of its episodes while writing this chapter, during the first days of September 2011. The event was eyewitnessed by the author, and by its fellow citizens of L'Aquila in Central Italy, during the days 2-4. A large "bubble" of dust travelled over the South-Western Mediterranean and it was captured by the MODIS/Terra satellite instrument, as depicted in Fig. 14, which estimated a maximum optical thickness of more than 1.5. In Fig. 15 we show the striking effect on atmospheric visibility as observed from the ground: the mountain peak by the centre of the pictures, having a distance of about 25 km from the shot location, was almost obscured by the dust layer. The latter did actually hit the ground, as witnessed by the PM10 monitoring station in L'Aquila valley, which exceeded 50 µg/m 3 on September 3 rd (Fig. 16), and also by the dust deposited over surfaces at the ground (e.g. the author's car depicted in Fig. 15).    Interestingly, Fig. 17 shows that the arrival of the dust layer was qualitatively predicted by the ForeChem experimental chemical weather forecast system operating at University of L'Aquila (Curci, 2010)

Conclusion
The effect of initial (IC) and boundary conditions (BC) on the simulation of surface ozone and PM10 over the continental scale European domain is evaluated with several sensitivity tests of the CHIMERE chemistry-transport model.
Zeroing alternatively IC and BC in the model, and comparing the results with a reference run, we estimate an optimal model spin-up time of 9 days for the domain used in this study (35°-58°N; 15°W-25°E; 79 x 47 cells at 0.5° horizontal resolution). The BC have a significant L'Aquila www.intechopen.com impact on simulated ozone, especially in a belt of about 1000-2000 km around the domain borders. There, BC dominates ozone variability, while in the interior of the domain they have a weight similar to the local photochemical production. Through the BC test, the surface ozone lifetime is estimated to be 1-3 days over the continent and longer than 5 days over the oceans.
The impact of different time-resolution of BC is studied feeding the CHIMERE model with GEOS-Chem global model simulations. With respect to the reference BC, provided by five years monthly mean LMDz-INCA model simulations, the GEOS-Chem model has generally lower ozone and enhanced dust values during the period of simulation (June-July 2005). The positive ozone bias with respect to EMEP measurements is alleviated by GEOS-Chem BC, and also the introduction of BC better resoled in time benefit model skills. The average correlation increases from 0.55 to 0.59 and the normalized bias decreases from 28% to 17%. The large improvement is noticed when passing from monthly to daily BC, while hourly BC do not produce further improvements. We noticed, however, that time-resolved BC may episodically worsen model skills.
The introduction of different aerosol BC is tested focusing on dust, because of the prominent role that Saharan dust events play on the PM10 levels especially in Southern Europe. GEOS-Chem predicts higher dust concentrations with respect to LMDz-INCA, and its use as BC alleviates the CHIMERE low bias with respect to EMEP measurements. However, the agreement with observations get better during the events by the end of the simulated period (July 2005), but worsen at the beginning (June 2005). In particular, GEOS-Chem has a tendency to episodically overshooting dust at unrealistic levels. The introduction of time-resolved dust BC may allow the model to better reproducing both time and magnitude of Saharan events, but this is not a general conclusion with models' set-up used in this study. Further work on dust BC for European scale models is certainly needed in the future, possibly combining satellite observations with ground measurements of the aerosol composition and sizedistribution that may better constrain the dust contribution to particulate matter.

Acknowledgment
The author was supported by the Italian Space Agency (ASI) in the frame of QUITSAT and PRIMES projects.

References
All figures in this chapter are originals by the author.