Edinburgh Explorer Modelling surface ozone during the 2003 heat-wave in the UK

. The EMEP4UK modelling system is a high resolution (5 × 5 km 2 ) application of the EMEP chemistry-transport model, designed for scientiﬁc and policy studies in the UK. We demonstrate the use and performance of the EMEP4UK system through the study of ground-level ozone (O 3 ) during the extreme August 2003 heat-wave. Meteorology is generated by the Weather Research and Forecast (WRF) model, nudged every six hours with reanalysis data. We focus on SE England, where hourly average O 3 reached up to 140 ppb during the heat-wave. EMEP4UK accurately reproduces elevated O 3 and much of its day-to-day variability during the heat-wave. Key O 3 precursors, nitrogen diox-ide and isoprene, are less well simulated, but show generally accurate diurnal cycles and concentrations to within a factor of ∼ 2–3 of observations. The modelled surface O 3 distribution has an intricate spatio-temporal structure, governed by a combination of meteorology, emissions and photochemistry. A series of sensitivity runs with the model are used to explore the factors that inﬂuenced O 3 levels during the heat-wave. Various factors appear to be important on different days and at different sites. Ozone imported from outside the model domain, especially the south, is very important on several days during the heat-wave, contributing up to 85 ppb. The effect of dry deposition is also important on several days. Modelled isoprene concentrations are generally best simulated if isoprene emissions are changed from the base emissions: typically doubled, but elevated by up to a factor of ﬁve on one hot day. We found that accurate modelling of the exact positions of nitrogen oxide and volatile organic compound plumes is crucial for the successful simulation of O 3 at a particular time and location. Variations in temperature of ± 5 K were found to have impacts on O 3 of typically less than ± 10 ppb.


Introduction
In the UK, episodes of increased concentrations of groundlevel ozone often occur during periods of elevated temperatures associated with summertime anticyclonic conditions (e.g. Jenkin et al., 2002). During the first two weeks in August 2003, a blocking area of high atmospheric pressure centred over Scandinavia caused very high temperatures (>35 • C) for several consecutive days over parts of the UK and central Europe. This exceptional heat wave has been the focus of several studies (e.g., Schär and Jendritzky, 2004;Ordóñez et al., 2005;Trigo et al., 2005;Vautard et al., 2005;Solberg et al., 2008;Tressol et al., 2008;Andreani-Aksoyoglu et al., 2008;Chaxel and Chollet, 2009). This event was associated with a series of afternoon ozone peaks, reaching above 90 ppb, in the south of England (Lee et al., 2006). The heat-wave period was coincident with the Tropospheric ORganic CHemisty (TORCH) field campaign (Lee et al., 2006), which provided detailed measurements of ozone concentrations and its precursors, including isoprene, at a site in Writtle about 70 km NE of London. The high temperatures and high levels of ozone experienced during the 2003 heat-wave had a substantial effect on human health (Stedman, 2004). For our work, we assumed that the beginning of the heat-wave was on the 4 August and the end was on the 12 August. Between these dates, observed and simulated daily maximum temperatures satisfied the UK Climate Impacts Programme (UKCIP) definition for a heat-wave. These dates match closely the 5-11 August period defined as heatwave in the work of Lee et al. (2006).
In this study we investigate the causes of the elevated ozone levels using a high resolution (5×5 km 2 grid) chemical transport model system over the UK domain (EMEP4UK). This system comprises the EMEP chemical transport model (Simpson et al., 2003a), the Weather Research and Forecast model, and fine-scale UK emissions from UK national databases. This paper represents the first demonstration of the abilities of the EMEP4UK model for photochemical oxidant modelling in the UK. We first show that the model system is able to simulate hourly ozone measurements realistically from a range of sites over SE England during 2003, including measurements made as part of the TORCH campaign. We then conduct a series of sensitivity runs to investigate the influences of a variety of different meteorological and chemical factors (temperature, anthropogenic volatile organic compounds, emissions of biogenic isoprene, anthropogenic emissions of NO x (NO+NO 2 ), ozone dry deposition, and transport) that contributed to the high ozone episodes in this region during the August 2003 heat-wave.

Model description and set-up
The EMEP4UK model framework is a nested regional chemistry-transport model (CTM) driven by high-resolution meteorology and national emissions that is used to produce a detailed representation of the physical and chemical state of the atmosphere over Europe and, in particular, over the UK (Vieno et al., 2009). The underlying CTM is the EMEP Unified Model (Simpson et al., 2003a), which has been modified in recent years to enable application on spatial scales ranging from the 5×5 km 2 grid used here for the UK to the global scale (Jonson et al. 2007(Jonson et al. , 2010. For this study, the EMEP4UK model was driven by the Weather Research Forecast (WRF) model (www.wrfmodel.org) with a resolution of 5×5 km 2 . The WRF model included data assimilation (Newtonian nudging) of the numerical weather prediction (NWP) model meteorological reanalysis from the US National Center for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) Global Forecast System (GFS) at 1 • resolution, every 6 h.
The WRF/EMEP4UK model was applied here using a one-way nested domain approach, with an outer domain resolution of 50×50 km 2 covering the official EMEP domain (Simpson et al., 2003a), an intermediate domain resolution of 10×10 km 2 and an inner domain with a resolution of 5×5 km 2 . Simulations were performed over each of these domains, the results from the outermost domain being used as boundary conditions to the intermediate domain and so on. The intermediate domain is required by the WRF model due to the complexity of simulating atmospheric dynamics, with stability criteria imposing a maximum nesting factor of 5. For the chemical transport (EMEP) modelling there are fewer numerical restrictions and we make use of just the outer and inner domains. As the inner domain covers all of the UK, this approach simplifies the interpretation of the model tests.
The innermost domain covers the whole British Isles, plus adjacent parts of France, Denmark, Holland and Belgium. Both WRF and EMEP4UK models use 20 vertical layers, with terrain following coordinates, and resolution increasing towards the surface. The vertical column extends from the surface (centre of the surface layer ∼45 m) up to 100 hPa (∼16 km). Modelled species are calculated at 3 m above the surface plant or other canopy by making use of the constant-flux assumption and definition of aerodynamic resistance (Simpson et al., 2003b). The WRF coarse grid of 50×50 km 2 resolution was used to drive the EMEP model across the European domain to calculate the chemical initial conditions and boundary conditions (one-way nesting approach) for the EMEP4UK model (driven by the inner WRF domain 5×5 km 2 ). The EMEP model itself (50×50 km 2 ) was initialised with climatologically-derived ozone boundary and initial conditions (Logan, 1999). To simulate the import of ozone realistically in a specific year, the so called "Mace Head" adjustment was applied (Simpson et al., 2003a). This adjustment uses monthly "clean-air (Atlantic)" observations from the Mace Head site on the west coast of Ireland, adjusting the monthly Logan climatology to match Mace Head data, and it was only applied to the EMEP Unified Model at 50×50 km 2 resolution.
The current EMEP Unified model is a development of the 3-D chemical transport model of Berge and Jakobsen (1998), extended with photo-oxidant chemistry (Simpson et al., 1995(Simpson et al., , 2003aAndersson-Sköld et al., 1999) and the EQSAM gas/aerosol partitioning model (Metzger et al., 2002). Two types of emissions are present in the model: anthropogenic and natural. For the UK, anthropogenic emissions of NO x , NH 3 , SO 2 , PM 2.5 , PM CO (coarse particulate matter), CO, and non-methane VOC (NMVOC) are integrated from the UK National Atmospheric Emissions Inventories (NAEI) 1 × 1 km 2 emissions to the required 5×5 km 2 Hellsten et al., 2008). Elsewhere and for international shipping, EMEP 50×50 km 2 emissions are used (www.emep.int). NMVOC are speciated into 10 reactive and one unreactive species, using emission-sector specific values as shown in Simpson et al. (2003a). Biogenic emissions of isoprene are based on Guenther et al. (1993) and Simpson et al. (1999), driven by EMEP 50×50 km 2 landuse, temperature and light. Emissions of monoterpenes are not included in this version of the model; this is discussed in Sect. 4.2.3. Biogenic emissions of dimethlysulphide (DMS) are input as monthly average emission data, derived from Tarrason et al. (1995), and treated as SO 2 on input to the M. Vieno et al.: Modelling surface ozone during the 2003 heat-wave in the UK 7965 calculations. Emissions of NO x from lightning are included as monthly averages (Köhler et al., 1995). Seasonally averaged aircraft emissions are included for NO x from Gardner et al. (1997). Both aircraft and lightning emissions are provided as 3-D fields for the whole model domain. Natural soil NO x emissions and non-anthropogenic biomass burning are not included. For CH 4 a constant mixing ratio over the whole domain is prescribed (Simpson et al., 2003a).
Sixteen basic land-use classes are used in the dry deposition module of the EMEP4UK model. For those vegetative landuse categories for which stomatal modelling is undertaken, the start and end of the growing season is specified and the development of leaf area index within this growing season is also modelled (Simpson et al., 2003a, b). Dry deposition is calculated using a resistance analogy combined with stomatal and non-stomatal conductance algorithms (Emberson et al., 2001;Simpson et al., 2003a, b), whereas wet deposition uses scavenging coefficients applied to the 3-D rainfall.
Full details of the EMEP model are given in Simpson et al. (2003a) and Fagerli et al. (2004).

Methods
A full year simulation was performed for 2003 using EMEP4UK/WRF in the configuration described in Sect. 2. Thirteen further one-month sensitivity experiments were carried out to investigate the contributing factors to the elevated ozone in the southern UK during the 2003 August heat-wave. These were identical to the base experiment in all respects except that in each case a single meteorological or chemical variable was changed in the EMEP4UK 5×5 km 2 grid inner domain. Use of a one-way nesting algorithm means that changes to fields within the inner domain do not influence fields in the outer domains. With this set-up, any air that recirculates (i.e. exits the inner domain, then re-enters) will lose the original influence of the inner domain also the coarse domain used to calculate the boundary and initial condition was not modified. This approximation is not expected to lead to significant problems in the simulations reported here.
The first factor investigated was surface temperature, which was either increased or decreased by 5 K. This affected ozone by changing both emissions of biogenic isoprene, and dry deposition of ozone through the surface exchange scheme. The 3-D potential temperature has also been increased by +5 K and +10 K, to investigate the effect of temperature on the chemistry. The imposed change in temperature did not affect the dynamic meteorology, as the influence was limited to the chemical transport (EMEP) part of the code, and no feedbacks operate from the EMEP model to WRF. Furthermore, those dispersion parameters which are calculated in the EMEP model (mixing height, eddy diffusivity) rely on gradients in potential temperature rather than absolute temperature. As these gradients are preserved with a uniform 5 or 10 K change in potential temperature, this test only affects the chemical scheme and avoids unphysical (and NWP inconsistent) effects on the EMEP dispersion rates. Several experiments then varied emissions of specific species: biogenic isoprene (zero (no emissions), 2×, and 5× base case emissions), anthropogenic VOC (±50%), or anthropogenic NO x (−10% and −50%). The focus of the three isoprene experiments, generally the most important biogenic VOC with regard to ozone formation, was to investigate the importance of UK-generated isoprene on surface ozone formation. As an extreme test of the importance of dry deposition, a further experiment was conducted in which ozone dry deposition (both stomatal and non-stomatal) was entirely switched off. The final experiment fixed ozone at the EMEP4UK boundary to the monthly climatological value from Logan (1999) rather than using 3-hourly values from the EMEP 50×50 km 2 model. Hereafter we refer to this as the "O 3 import" experiment. This allowed the influence of import from the outer domain to be isolated. The sensitivity experiments are summarised in Table 1.

Surface temperature
To demonstrate that surface temperatures simulated by WRF with data assimilation produce a realistic representation of the August 2003 heat-wave, we compare model output with independent measurements (i.e., data that were not used in the WRF assimilation) during the TORCH campaign. Figure 1a shows hourly surface temperatures calculated by WRF from the 5× km 2 grid cell containing Writtle (51 • 44 12 N, 0 • 25 28 E), together with data from two instruments deployed during the TORCH campaign. Figure 1b shows a similar comparison for observed versus modelled temperature at a nearby UK Met Office weather station in Wattisham (52 • 07 22 N, 0 • 57 43 E). WRF is able to simulate the diurnal and longer timescale variations of temperature. August hourly temperatures during the heat-wave are generally well simulated ( Fig. 1a compared to the Univ. Leicester sensor: R 2 =0.9, slope 0.9 intercept of 0.7 K; Univ. Leeds sensor: R 2 =0.9, slope 0.8 and intercept of 0.6 K, Fig. 1b, R 2 =0.8, slope 0.9, intercept 0.8 K). However, the model underestimates some peak temperatures (by up to 5 K), particularly in the period of 9-11 August. Possible reasons for this discrepancy might include: (a) limitations in the driving analysis and assimilated data, (b) significant sub-grid variation that the model cannot resolve i.e., for scales less than 5 km in the horizontal or less than 90 m in the vertical, or (c) that the landuse input to the model, and WRF's handling of this, does not perfectly reproduce the local area. This latter factor could lead to an erroneous calculation of surface temperature, especially in cases where stagnating air is present ac.uk/data/ukmo-midas). Overall the bias was found to be 1.5 • C as for the Wattisham site (Fig. 1b); hence the bias at Writtle site (2 • C) is larger than at other UK sites for the corresponding period.

Surface wind speed and direction
Wind speed and direction are also important parameters that may influence surface ozone at a given location. Figure 2 shows the comparison between the WRF model and observation for the first 15 days of August at Wattisham for surface wind speed and direction. The low winds speeds associated with the slow-moving anticyclone over Europe are well represented, although the magnitude of simulated low wind speeds is underestimated on average by 1.3 m s −1 (±1.2 m s −1 one standard deviation). Moreover, the wind direction (generally from the south east or south west) is well captured by the model (R 2 =0.9, slope 0.8 and intercept of 18 degrees). We note that the highest wind speeds occur on the 6 August when imported ozone made the largest contribution to the ozone simulated at Writtle as compared to other days in the 15-day period (see Sect. 4.2.6). For the 9-10 August when wind speed are lowest the model suggests a smaller  easterly component of the wind than is observed (Fig. 2). An important point we note is that wind speeds on the 9-10 August were amongst the lowest of the period, and overall during the whole heat-wave period the wind speeds were low (Lee et al., 2006) as expected due to anticyclonic conditions. We expect considerable variability in wind direction with low wind speed. , (see Fig. 7 for locations). The model closely simulates the seasonal variation of surface ozone at the two sites. Moreover the model is able to capture ozone peaks (>50 ppb) for the whole of 2003. It is interesting to note that the August episode is not exceptional -there are several episodes of similar magnitude, from late March to mid-September. The fact that comparatively high concentrations of ozone occur in England at other times not associated with exceptional temperatures is probably linked to the fact that many ozone episodes can be attributed to long-range transport, arising from precursors over continental Europe and with multi-day processes controlling ozone formation (e.g. Cox et al., 1975, Guicherit andvan Dop, 1977;Simpson et al., 1995). We found that there is not a clear direct link between high UK temperature and high UK ozone, since high UK ozone episodes may occur at relatively low UK temperatures when ozone and/or precursors are imported from outside of the inner model domain.
Modelled and observed hourly ozone from the same two AURN stations (Wicken Fen and London Eltham) in August 2003 are shown in Fig. 4a and b. Similarly, model simulations are compared with observations at Writtle (TORCH) in Fig. 4c. Scatter plots of these data are shown in Fig. 5. In terms of R 2 for all the hourly August data, the model performs best at Wicken Fen (R 2 =0.7), London Eltham (R 2 =0.6) and worst at Writtle (R 2 =0.5). Nevertheless, the model accurately simulates many of the high ozone days dur- ing the heat-wave in comparison to cooler days with lower ozone, and the typical diurnal variation of ozone at the three sites. At Writtle, the discrepancy between observed and simulated surface ozone is greatest between the 8-11 August.
There are several potential meteorological drivers that may help to explain differences between observed and simulated surface ozone. Firstly, underestimated peak temperatures would yield lower isoprene emissions, which under high NO x conditions would mean less local ozone production. Lower temperatures would shift equilibrium towards more peroxyacetyl nitrate formation, tying up some NO x and radicals thus lowering local ozone production (Sillman and Samson, 1995;Baertsch-Ritter et al., 2004). Secondly, a lower frequency of easterly wind components on the 9-10 August could lead to lesser background concentrations of "high" ozone since the highest ozone concentrations during the heatwave period were found over France and Germany in both model and observations (not shown), although we note that winds were light. Further reasons for the discrepancy between the model and observations at Writtle are discussed in Sect. 4.2.
The spatial and temporal variability of simulated surface daily maximum ozone for the first 15 days of August 2003 is shown in Fig. 6. During this period a clear feature of elevated ozone building up after the 3 August is visible across southern England. The feature shows strong spatial and dayto-day variability. The detailed structure in the simulated O 3 field clearly illustrates how difficult it is to simulate every site accurately, particularly those close to emissions sources or with other strong local influences on ozone.
The modelled monthly mean distribution of surface ozone for August 2003 is shown in Fig. 7, together with the location of the observation sites included in this study. The influence of surface NO x emissions on these ozone fields is clearly evident along road corridors and over cities such as London, Birmingham and Manchester. This highlights the importance of detailed emissions to simulate properly the spatial pattern of ozone over the UK and, more generally, wherever discrete emissions are present (i.e., road, point sources etc.).

Which factors contributed to the high surface ozone during the 2003 heat-wave?
In this section we present and discuss results from the sensitivity experiments described in Table 1, with the aim of determining the key factors that led to the high values of ozone during the 2003 heat-wave. We focus in particular on O 3 at Writtle, but also consider the influences on O 3 across the EMEP4UK model domain. emissions scheme is based on that of Guenther et al. (1995) and Simpson et al. (2003a). At 30 • C (the average max temperature during the heat-wave period) a 5 • C increase in temperature corresponds to a doubling of isoprene emissions. This similarity suggests that the major parameter modified by the temperature experiment in this range of temperatures is the isoprene emissions. Over the 15-day period a +5 K and +10 K increase of the 3-D potential temperature monotonically increased surface O 3 by up to 10 and 15 ppb, respectively. The results for the +5 K increase of 3-D potential temperature are shown in Figure 8 (Changes in O 3 for 10 K, not shown, are essentially double those of the 5 K experiment). Considering that 5 or 10 K changes are a large perturbation to 3-D temperature as compared with model biases, we suggest that O 3 is biased only by a few ppb due to the effects of WRF model temperature biases on the EMEP model. The change of potential temperature throughout the atmosphere affects all chemical conversion rates only, whereas the effect of surface temperature change is limited to isoprene emission and dry deposition rates only. As mentioned in the methods section, the changes to 3-D temperature were applied only to the chemistry, leaving the dynamic meteorology unchanged.

Anthropogenic NMVOC emissions
The effects of the NMVOC sensitivity experiments on modelled surface ozone are shown in Fig. 9. When the UK anthropogenic emissions of NMVOC were modified by ±50%, the model response was to change surface ozone at Writtle by typically ±4 ppb. Larger responses occurred on the 2, 6,  9 and 13 August, when an increase in NMVOC increased surface ozone by as much as 30 ppb, while reduced NMVOC emissions decreased it by as much as 16 ppb (Fig. 9). Some of the days showing great sensitivity (6, 9 August) coincide with days when the temperature-induced changes were also important, but other days (2, 13 August) seem specific to NMVOC. As temperature changes in the model impact BVOC emissions but not anthropogenic VOC, these different periods likely reflect days when isoprene did and did not play a large role in ozone formation.
Sensitivity of ozone to NMVOC is a classic sign of high-NO x chemistry (Sillman et al., 1990;Simpson et al., 1995) and indeed the model results for surface NO 2 (see Sect. 4.2.4) show an abundance of NO 2 on 2 and 9 August. For the 6 August the absolute difference of maximum surface ozone is ∼5 ppb (Fig. 9a), but the timing of the peak is altered by perturbing emissions of NMVOC by ±50%. Surface ozone decrease is limited to 16 ppb when the NMVOC emissions have been reduced by half. The implication is that a possible UK policy aiming to decrease ozone by controlling NMVOC emissions will have non-linear and limited effects, and in  general both NO x and VOC control must be considered together. Such non-linearities are expected from earlier studies (e.g. Sillman et al., 1990;Simpson et al., 1995;Baertsch-Ritter at al., 2004), but quantifying the magnitude of such effects is essential to assess the expected impacts of such policy. Figure 10 shows the comparison between observed isoprene and model-simulated isoprene at Writtle for the base run and for a 2× and 5× increase in UK emissions of biogenic isoprene. The model-simulated isoprene is, in general, in better agreement with observations for the model simulation with double isoprene emissions. In terms of impact on surface ozone (Fig. 11), the models indicates that UK biogenic isoprene emissions contribute up to ∼10 ppb ozone on some days in the base run case as compared to the zero isoprene emissions experiment. Doubling UK isoprene emissions enhances surface ozone concentrations by a further 10 ppb, and with 5× emissions the effect is ∼5 times higher (up to ∼45 ppb). An approximately linear dependency of surface ozone to zero, 2× and 5× UK biogenic emissions during this period is therefore found with this experiment (Fig. 11). The maximum change of ozone due to isoprene (5× scenario) is 45 ppb which occurred on the 6 August 2003 at 17:00, but when the ozone concentration was at its maximum at 15:00 this difference was 23 ppb, which is not the major factor when compared with import on that day (see Sect. 4.2.6). On the 10 August the 5× scenario fits the observed isoprene concentrations better and the ozone attributable to isoprene emissions is then around 30 ppb. Over the whole heat-wave period, it is only on the 10 August that UK isoprene emissions appear to be the dominant cause of elevated ozone concentrations. Taken over an extended period and the whole UK, isoprene emissions had relatively modest effects on simulated UK ozone. However, the modelling suggests that isoprene may play a substantial role for the warmest day at Writtle, when emissions are greatly enhanced.

Biogenic isoprene emissions
Unfortunately, emissions of biogenic VOC are notoriously uncertain, with isoprene emissions estimates for the UK exhibiting substantial variability. The emissions estimates of Guenther et al. (1995), Simpson et al. (1999, as used in this work), and Stewart et al. (2003), suggested annual European biogenic isoprene emissions of 110, 48 and 8 Gg C y −1 , respectively. There are many reasons for the large differences in inventories and their underpinning emission factors, including limitations in the number of measurements, assumptions concerning extrapolation of emission data and characterisation of the effects of environmental and biogeophysical variables (e.g. temperature, light, soil moisture, canopyeffects, diversity between and among vegetation species). Uncertainties for short time-periods and at specific locations can be expected to be larger than for national averages, and the suggestion of Simpson et al. (1999) that overall biogenic isoprene emissions may be uncertain to within a factor of 3 to 5 may even underestimate the uncertainty of UK emissions during this episode. Moreover, due to the high reactivity of isoprene within this intense photochemical episode, a strong vertical gradient of isoprene is present, as shown in Fig. 12 and the vertical resolution of the model may not be adequate to simulate fully the vertical distribution of isoprene. Despite the above-noted complexities, Fig. 10 shows that the EMEP4UK model was able to simulate isoprene at the Writtle site to within a factor of 3 with respect to observations. It should also be noted that the EMEP4UK model currently does not include estimates of any anthropogenic emissions of isoprene. The vertical resolution of the model (lowest level thickness ca. 90 m) also has strong implications for the comparison of modelled versus observed isoprene concentrations. However, the timescale for mixing in unstable boundary layers is typically much less than the oxidation-lifetime of isoprene to OH (order 1 h during daytime), so the model resolution should be adequate for the task. Similar issues apply to NO x also, which also has mainly surface sources and chemical loss slower than mixing times.
It should also be noted that the biogenic emission inventory available to this study has a resolution of 50×50 km 2 , which likely leads to uncertainties in the spatial allocation of isoprene concentrations as applied here. However, isoprene inventories are inherently uncertain, requiring species-level coverage of vegetation which is rarely available (even in the UK), and with different studies suggesting widely different emission factors to be applied Rinne et al., 2009;Stewart et al., 2003). An interesting feature of surface isoprene was a double peak in the morning and evening, with the latter peak generally higher. This feature (also found by Steinbacher et al., 2005) was present in both observations and simulated isoprene concentrations, as seen in Figs. 10 and 12. OH is understood to be the cause of the mid-day dip in isoprene concentrations while the afternoon decline in OH concentration (and hence isoprene loss rate), and increased afternoon temperatures (hence higher isoprene emissions) are the cause of the higher evening peak in isoprene concentrations. Another potentially important contribution to the second peak may be the reduction of the mixing height after sunset (Fig. 12), which will act to limit vertical mixing and dilution. The mixing height and OH-levels are decreasing at the same time in the evening, thus a combination these two factors may be the cause of the evening isoprene peaks. An interesting day in the period under study here is the 1st August. Here the modelled surface isoprene concentration is still high around midnight. This is likely related to almost complete depletion of ozone in the nocturnal boundary layer in the model (i.e. Fig. 11). When surface ozone is depleted there is no loss of isoprene through the isoprene + ozone reaction or the NO 3 + isoprene reaction.
Our results can also be compared to those of Curci et al. (2009), who estimated that BVOC emissions contribute 0-4 ppb towards the maximum daily ozone for the summer (June-July-August) of 2003 in the UK. This is reasonably consistent with our results: we find an EMEP4UK domain average contribution for August of ∼1 ppb for the base simulation, and ∼3 ppb for the case with 5× isoprene emissions (Fig. 11).
Finally, it should also be noted that emissions of other BVOC, including monoterpenes but also a whole host of oxygenated species (e.g. Guenther et al., 1995;Seco et al., 2007) are not included in the standard EMEP model. Such emissions will affect ozone, often with similar dependencies on temperature to isoprene. However, tests with a research version of the EMEP model (Simpson et al., 2007a) which includes monoterpene emissions, as well as the study of Curci et al. (2009), both suggest that isoprene emissions are a much more important factor than monoterpene emissions for ozone formation in NW Europe, and in any case the effects of monoterpenes should fall within the range of uncertainty we have explored here for isoprene.
The large uncertainties in isoprene emission estimates clearly affect model calculations, and emphasise the need for improved inventories of this important compound as well as of other BVOC compounds.

Anthropogenic NO x
The impact of decreasing UK anthropogenic NO x emissions by 10% and 50% on modelled surface ozone at Writtle is shown in Fig. 13. This impact varies substantially in both cases across the 15 days in August. As with NMVOC emissions, a decrease in UK NO x emissions affects a few days on which high ozone concentrations were simulated (Fig. 4c).
Reducing NO x emissions by 10% and 50% enhances the surface concentration of ozone by up to 9 ppb and 65 ppb, respectively, on 9 August, and up to 4 ppb and 32 ppb, respectively, on 2 August, whilst on other days it has less impact (Fig. 13a). On some days both 10% and 50% reduction in NO x emissions leads to increased ozone, a result of the wellknown titration effect and NO x -VOC relationships in high-NO x conditions (Sillman et al., 1990;Simpson et al., 1995). The impact on O 3 at Writtle acts in the same direction in both cases (10% and 50% reduction in NO x emissions) for all the days showing little evidence of non-linearity between the 10% and 50% experiments. Figure 14 show maps of the impacts of these NO x reductions on changes in daily maximum surface O 3 for the 4 and 9 August 2003 across the UK. For most locations across the UK, and on most days, reducing NO x emissions leads to higher O 3 levels (red colours on Fig. 14). This is especially true for locations downwind of large NO x sources (e.g., large urban centres, such as London and Birmingham). This is indicative of the VOC-limited O 3 production regime (e.g., see Fig. 3.3a of Royal Society, 2008). In this regime, increases in NO x lead to reductions in O 3 production. For typical mid-latitude conditions, peak O 3 production occurs at around 1 ppb NO x . This NO x level for peak O 3 production increases as VOC levels increase. This is particularly relevant during the August 2003 heatwave, as on hotter days, biogenic VOC emissions increase, pushing the position of peak ozone production to higher NO x levels, and potentially moving some parts of the UK out of the VOC-limited regime and into the NO x -limited regime. These regions are the blue regions of Fig. 14   Scotland on 4 August - Fig. 14). These regions of NO xlimitation are also slightly more widespread in the 50% reduction case -this is simply because more regions pass over the peak in the O 3 production curve when there is a larger NO x reduction. More regions of NO x -limitation emerge on hotter days during the heatwave (compare 4 and 9 August on Fig. 14 -the 9 was hotter -see Fig. 1); as explained above, these are days with higher VOC levels.
We also show wind speed and wind direction in Fig. 14. The low wind speed and variable wind directions are clearly seen in red region where reducing NO x leads to higher O 3 .
Further results from the 50% anthropogenic NO x emission reduction experiment are shown in Fig. 15. The figure shows hourly modelled values for the whole of August of O 3 / NO x (where is the change in mixing ratio between the base experiment and the NO x reduction experiment), plotted as a function of NO x (from the base experiment), for three sites (Wicken Fen, Writtle, and London Eltham). The three sites broadly represent the gradation from relatively rural (Wicken Fen), with ∼0.5-5 ppb NO x , to urban (London Eltham), with ∼5-50 ppb. Writtle is generally less polluted than London, but has similar upper values, when directly within the London plume (NO x ∼2-50 ppb). At all sites (and especially in London), for most of the time, Figure 14: Change in simulated surface daily maximum ozone (ppb) relative to the base case scenario for the sensitivity experiment with 10% (left) and 50% (right) reduction of the UK anthropogenic NO x emissions, for two days, 4 th (upper panels) and 9 th August (lower panels).
The 12:00 10 m wind is also shown.

41
Fig. 14. Change in simulated surface daily maximum ozone (ppb) relative to the base case scenario for the sensitivity experiment with 10% (left) and 50% (right) reduction of the UK anthropogenic NO x emissions, for two days, 4 (upper panels) and 9 August (lower panels). The 12:00 10 m wind is also shown. ozone declines as NO x increases (i.e. most O 3 / NO x values are negative), indicative of a VOC-limited regime. At times of lower background NO x , additional NO x sometimes leads to increases in O 3 -this is seen most often at the more rural site -indicating a NO x -limited regime. There is not a single value for background NO x where the switch from NO x -limited to VOC-limited occurs (this will be a function of several other variables, e.g. VOC levels), but the regime is clearly VOC-limited above ∼8 ppb NO x , and NO x -limited below ∼0.5 ppb NO x . This is broadly consistent with the schematic figure presented in the Royal Society report discussed above.
These sensitivity tests further support the conclusion that the deviations between modelled and observed O 3 were particularly related to uncertainties in local patterns of calculated NO x concentrations, which may be related to local uncertainties in the NO x emission data. Figure 14 highlights the area affected by the London plume. The location of Writtle is on the edge of the London plume and is therefore highly sensitive to small errors in modelled location of the plume. This is illustrated by the fact that on 9 August the model performed well for the two sites Wicken Fen and London Eltham, which were well outside and inside the London NO x plume, respectively. The model does not agree well with observed NO 2 at Writtle for the first 9 days of August, but shows better agreement for the remaining days included in this study (Fig. 16).
In general the EMEP4UK model captures the concentration of NO 2 quite well (fine-scale models typically have trouble simulating NO 2 ), and with a reasonable diurnal variation on most days. Nevertheless, significant over-predictions are seen on the nights of 2, 3, 9 and 10 August. The discrepancy between modelled and observed NO 2 is consistent with the larger standard deviation of the observed averaged NO 2 concentrations from the high frequency observations during the first week compared with the second week of August (data not shown). Large standard deviations imply the existence of fast small-scale variations of concentration due to local factors which are much more difficult to represent in models. Other studies (e.g., Baertsch-Ritter et al., 2003) also highlight difficulties in simulating urban NO x plumes as a result of emission uncertainties.

Ozone dry deposition
Dry deposition of ozone is a major factor controlling the magnitude of surface ozone concentrations, and during the extreme conditions of August 2003 there is a possibility that uptake to vegetation was severely restricted, as stomatal deposition is a strong function of temperature, humidity, and sunlight (Emberson et al., 2001;Simpson et al., 2007b). The impact of switching off UK ozone dry deposition (both stomatal and non stomatal) is also shown in Fig. 13. This model change had a comparatively large impact on surface ozone throughout the simulation period, particularly at night time when surface ozone increases up to 50 ppb. Suppressing dry deposition in the model generally increased surface ozone, as expected, although there were two points on 2 August when ozone was reduced in the late afternoon/early evening (Fig. 13). This must have been due to the earlier, enhanced levels of O 3 influencing the abun-  dance of O 3 precursors to such an extent that although deposition was switched off, O 3 levels fell below those in the control simulation. Our results show clearly that in general, turning off ozone dry deposition increases modelled ozone concentrations. In the case of the anomalous two-hour period, NO 2 was also overestimated compared with the measurements. This suggests a temporal interaction between O 3 dry deposition and NO concentrations may have occurred, whereby previous higher O 3 concentrations (as a result of no model dry deposition), had depleted the modelled NO levels, thereby briefly limiting the potential for O 3 formation. Vautard et al. (2005) suggest that due to the exceptionally hot weather of August 2003 over Europe, dry deposition calculations in their model needed to be modified to reduce dry deposition of ozone. The present study however retains the unmodified dry deposition calculation for the full year simulation of 2003 suggesting that for the UK the parameterisation used in the EMEP model for dry deposition is, in general, adequate for the range of temperature and extreme weather modelled here. Nevertheless, we do find on some specific occasions (e.g., night of 10/11 August) switching off deposition improves the comparison with observations.

UK import
In the final sensitivity experiment, the ozone boundary conditions for the inner domain (5×5 km 2 region) of the EMEP4UK model were fixed to climatological values (32 ppb - Logan, 1999) for the whole month, rather than using output from the larger scale 50×50 km 2 runs. The results of this sensitivity test (Fig. 17) show that in August 2003 surface ozone concentrations were strongly influenced by import on most days, especially in SE England (Fig. 18). Import contributed up to 85 ppb. For example, most of the ozone present at Writtle on the 6 August was generated and imported from outside the EMEP4UK inner domain.
Correct boundary conditions are very important to accurately calculate UK surface ozone and previous work has demonstrated that European transport and trans-Atlantic transport are well simulated by the EMEP model (e.g. Jonson et al., 2006). Figures 17 and 18 show that the import of ozone from outside the UK was typically the most important factor contributing to the very high surface ozone in SE England during August 2003. Import is important on different days at different locations (Fig. 18). This is evident for the 6 and 10 August (Fig. 18) when localised incursions of European-emitted precursors and/or ozone itself were present. This agrees with Solberg et al. (2008), who reported that the higher values of ozone observed over SE England were often the result of import from the continent. However, one of the conclusions in Solberg et al. (2008) was that Portuguese forest fires were a possible cause for the unusually high surface ozone over Europe. The EMEP4UK model was able to simulate high surface ozone without emissions from forest fires, further development of the EMEP model to include forest fire emissions should be undertaken in order to properly assess their impact on ozone over the UK.
Overall, however, it has been shown that, during the TORCH campaign at Writtle, the high level of ozone observed was not created within the model domain of the British Isles, but imported from continental Europe. Figure 18 highlights this clearly on the 6 of August where an incursion of European ozone was present in SE England. The simulations of Solberg et al. (2008) showed a cluster of high ozone concentrations (>90 ppb) near the border between France and Germany on this date. The implication is that ozone produced in this region was thereafter advected over the UK. High resolution modelling is also critical as this type of incursion may influence a small area (<100 km 2 ) as can be seen in Fig. 18 for the 10 August.

Summary and conclusions
For the first time a derivative of the EMEP Unified model (EMEP4UK) has been successfully applied to the UK at an enhanced horizontal resolution of 5×5 km 2 to simulate surface ozone, and been driven by the WRF model instead of the HIRLAM model used to drive the Unified model. Particular attention has been given to the site at Writtle, where the TORCH campaign made extensive atmospheric measurements, and at two nearby rural and urban background sites.
Modelled meteorology shows some biases compared to observations (Figs. 1 and 2). Daily maximum surface temperatures during the heat-wave are underestimated by up to 5 K at some sites, but averaged across all UK sites, average surface temperature model bias is −1.5 K. Most diurnal and day-to-day meteorological variability is well captured. Seasonal, day-to-day and diurnal variations in ozone are also well simulated (Figs. 3 and 4). Model performance at Writtle, the site with a large suite of campaign measurements during August 2003, is worse than at neighbouring long-term monitoring sites (Figs. 1, 4 and 5). Model results indicate that Writtle is a relatively difficult site to simulate, because it is intermittently exposed to the plume of pollution emanating from greater London and its proximity to the coast. Both these factors contribute to steep spatial gradients in meteorology and air pollutants near the site (Figs. 6 and 7). Nevertheless, model performance is sufficiently good for us to use model results for the site, together with the detailed campaign measurements, to investigate the origins of ozone during the heat-wave.
A series of sensitivity experiments were performed with the model, repeatedly simulating the heat-wave period (Table 1). In each experiment an individual model parameter/input was varied across the EMEP4UK domain, in order to isolate and quantify its influence on ozone. Uniformly increasing surface temperature by 5 K led to increases in ozone at Writtle of up to 9 ppb on certain days; decreasing temperature by the same amount induced similar magnitude ozone reductions (Fig. 8).
The main influence of temperature on ozone is via biogenic isoprene; doubling isoprene emissions produced a similar response to increasing temperature by 5 K (Fig. 11). Removing isoprene emissions, or multiplying them by five, induced ozone responses that indicate a broadly linear response of ozone to the magnitude of isoprene emissions on specific days (Fig. 11). Isoprene concentrations at Writtle show significant day-to-day variability, and this is partly captured by the model (Fig. 10). Overall during the heat-wave, we found that doubling baseline isoprene emissions produced the best fit to observations, although on some days the 5× experiment was best.
Days with the highest sensitivity of ozone to isoprene (6, 9 and 10 August) were days with high isoprene levels, although other days with similarly high isoprene levels (4 and 5 August) showed much lower sensitivity. This sensitivity is a function of coincident NO x levels; relative amounts of NO x and VOC determine whether ozone production is NO xlimited or VOC-limited. These changes in ozone production regime are clearly shown with results from sensitivity experiments that reduced anthropogenic NO x emissions. Figure 14 shows the influence of these NO x reductions on surface ozone across the model domain for 4 and 9 August. On the 4th, winds at Writtle were from the East, and the ozone production regime is not strongly VOC-limited, hence the low sensitivity of ozone to isoprene emissions on this day. In contrast on the 9th, winds are lighter, Writtle is within the London plume, the ozone production regime is VOC-limited, and consequently there is a higher sensitivity of ozone to isoprene emissions. The ozone production regime at Writtle is generally VOC-limited, like London, but occasionally is NO x -limited, like the more rural Wicken Fen site (Fig. 15). Experiments varying anthropogenic NMVOC emissions also show strong day-to-day variations in the sensitivity of ozone at Writtle (Fig. 9), again illustrating the importance of the prevailing ozone production regime.
Switching off the biophysical process of dry deposition across the UK increases ozone at Writtle by up to 50 ppb, and improves the fit to observed ozone on some days (Fig. 13). The influence on ozone on most heat-wave days is ∼20-35 ppb, although on some days there is much less impact (e.g. 9 August).
Setting ozone at the model boundaries to a climatological value, rather than allowing it to vary, shows that import of ozone from outside the EMEP4UK domain typically contributes ∼15-20 ppb to ozone levels at Writtle, but up to 85 ppb on 6 August (Figs. 17 and 18).
In summary, we find that multiple important influences contributed to the elevated ozone over SE England during the August 2003 heat-wave. Our simulations indicate that different processes dominated at different times, with local biogenic and anthropogenic emissions important on some days, whilst import from Europe and suppression of dry deposition were important on other days. All these processes need to be simulated accurately in order to fully understand the episode.