Near-surface elevated pollution: what we don’t know doesn’t hurt? A numerical study over Mt. Carmel

Many air pollution events are occasionally difficult to explain. While most monitoring-based air pollution assessment studies deal with surface analysis, the near-surface elevated pollutants are challenging. The lack of data and understanding of those elevated layers, leaves us ‘blind’ and with no clue where, when and how intensively these pollutants may hit the surface. Here, this challenge at the specific domain of Mt. Carmel is addressed. The atmospheric numerical models RAMS and HYPACT were employed on Haifa Bay in the Eastern Mediterranean with nested horizontal grids down to 0.5 km, in order to resolve the fine-scale flow, along an air pollution episode which serves as a case study. Sixteen locations were determined, representing monitored and non-monitored sites in the complex terrain sub-domains. Results show multi-inversion profiles, which are consistent with an earlier observational study over the region. Concentration differences up to an order of magnitude between adjacent sites (∼2 km) were found, often associated with near-zero surface values, while some simulated peaks were at elevations of 100–400 m above ground level (AGL). The current event offers a view on the near-surface elevated layers, and points at limitations of ground-level monitoring as an indicator of air pollution. This study highlights the importance of near-surface pollution, which is often an unknown source for surface pollution. Overall, steep vertical gradient of pollution as shown here is associated with a combination of deep inversion (or multi-inversion profile), vertical circulation due to topography or synoptic flow, and small scale circulation induced by the complex topography. Since monitoring of the elevated layers is limited by the technology, it is suggested that high resolution advanced models should be used for further exploration of the near-surface pollution.

. (a) And (c) as in figure 6 (a) and (c). TKE profiles over sites 1, 4 are presented in (b), (d) respectively. Over site 1 there are low concentrations (a) versus high TKE values (b), while over site 4 the concentrations are 5 to 10-fold higher (c) and the TKE shows 10fold lower than the values over site 1 (d).

Introduction
Air pollution episodes strongly depend on synoptic to local scale meteorological conditions (e.g., Kallos et al 1993, Astitha et al 2008, and may occur either in the vicinity of emission sources or farther away. Most pollutants that are emitted into the boundary layer remain and disperse within it, and may occasionally be caught and accumulated under a strong or long-lived inversion layer (Stull 1988, Kotroni et al 1999. However, some pollutants may be emitted above the capping inversion and/or be transported to higher layers where they may accumulate and persist, especially in coastal areas where diurnal mesoscale circulation dominates (e.g., Soriano et al 2001). Coastal mesoscale recirculation of sea-land breeze may also cause accumulation of air pollutants due to plumes return flow, as was found at the Israeli Mediterranean coast by Alper-Siman-Tov et al (1997) and Levy et al (2009). Simulated mountainous coastline flow suggest that sea breeze and topographic forcing produce a more intense circulation when combined (e.g., Pielke 1977, Perez-Landa et al 2007). Such circulations could be suggested as another mechanism for a buildup of high concentrations in some locations. Slope flows and convergence zones in complex terrain may also induce significant vertical and circular transport of pollutants (e.g., Lang et al 2015. Reuten et al (2004)  Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. boundary layer (CBL), and their calculated mass transport suggested a closed slope-flow circulation within the CBL, in which air pollutants maybe trapped.
Atmospheric modeling over complex terrain requires adequate resolution in order to represent a realistic and accurate flow. Studies of air pollution in regions of complex terrain achieved improved accuracy and were able to better explain local flow and pollution dispersion when high resolution (1-4 km) models were employed (e.g., Grell et al 2000, Jazcilevich et al 2003, Schmitz 2005, Ritter et al 2013, Valverde et al 2016. High resolution modeling allows effective and realistic comparison between simulation results and local ground monitoring sites within the study domain (e.g., Perez-Landa et al 2007, Balanzino and Trini Castelli 2018), which is also important for an efficient model validation. In this study the numerical atmospheric models RAMS and HYPACT were employed with high resolution (0.5 km) to resolve spatial and temporal patterns in the fine-scale complex domain of Haifa, addressing the pollution dispersion within the near-surface elevated layers. The Haifa region in northern Israel is characterized by a complex terrain on the Mediterranean coast (figure 1). Previous air pollution studies over the Haifa basin addressed health effects (Goren et al 1990, Paz et al 2009, Eitan et al 2010, identification and analysis of pollutants (Mamane and Mehler 1987, Mamane and Gottlieb 1995, Garty et al 2001, two-dimensional analysis based on monitoring data (Yuval andBroday 2006, Eitan et al 2010), and modeling of potential SO 2 concentrations (Doron and Kinrot 1989). Although Haifa densely populated area is relatively quite well-covered by monitoring stations (about 1 monitoring station for every 20 km 2 ), there are flow patterns which the observations are insufficient to capture (e.g. Pielke et al 1983). Most air pollution assessments are based on surface data (e.g. Here, we focus on some major variability between the pollution patterns at the surface and at elevated layers, over the Haifa complex domain. An air pollution episode was chosen as a case study to challenge the question whether it is justified to ignore near-surface elevated pollution. The mesoscale flow over the domain was investigated with a focus on the pollutants dispersion at surface and near-surface levels. The prevailing synoptic system along the event was a Red Sea Trough with easterly flow in the region, which expectedly would result with low pollution concentrations over the study domain due to the off-shore flow. However, at that specific day, high concentrations of SO 2 were recorded in a few monitoring stations. The formation mechanism and the timing of the peak concentrations could not be explained by the ground observations alone, and atmospheric numerical modeling was employed for exploration of the event. The atmospheric conditions, the pollutant sources and receptors, and model description are given in section 2; local flow patterns with selected upper-air analysis are discussed in section 3.

Data and methodology
Following are descriptions of the domain characteristics (section 2.1), the numerical model (section 2.2), and the emission sources and receptors which were used in this study (sections 2.3, 2.4, respectively).

Regional characteristics
The Haifa region in northern Israel is characterized by a complex terrain adjacent to the Mediterranean Sea, with Mt. Carmel ridge rising from the coastline up to an altitude of 450-500 m (ASL). Haifa Bay (35.1E, 32.5N) is bordered by easterly hills 10 km away from the coastline and the north-easterly Upper Galilee summits of 600-1200 m, 20 km away. The city of Haifa is spread over the ridge, and a few densely populated towns are located on the coastal plain of the bay, adjacent to industrial zones (figure 1).
2.2. Numerical models 2.2.1. RAMS The numerical Regional Atmospheric Modeling System RAMS 6.0 (Cotton et al 2003) was employed with four nested grids from synoptic to fine-scale resolution (32, 8, 2, 0.5 km). A total of 42 vertical layers were defined, from 50 m AGL. up to 20 km, out of which 11 levels are within the lower 1 km of the troposphere, hereby referred by rounded values (105 rounded to 100 m, etc). RAMS was initiated with 2.5°NCEP reanalysis data (Kalney 1996) and updated every 6 hr. RAMS simulation results were used as an input to the HYPACT simulation.

HYPACT
The HYbrid PArticle and Concentration Transport Model HYPACT (Walko et al 2001) version 1.9 was employed in the current study. HYPACT simulations were employed on the inner grid of RAMS, at a domain size of 60 km×60 km. Emissions of SO 2 from the largest emission sources were simulated from 0100 UTC to 2200 UTC, with steady continuous emissions. No pollutant removal from the plume was included. The results were extracted in a 10 min time resolution.

Emission sources
The SO 2 pollutant was chosen as a tracer for this study, due to well established database of emission sources and ambient monitoring of this pollutant in the region. Emission data was obtained from the Ministry of the Environment, based on their 2006 inventory. Emission rates of the largest SO 2 sources in the region were used, and steady continuous emission rates were assumed. No pollutant removal from the plume was considered, since SO 2 residence time in the troposphere is in the range of a day (Hobbs 2000), while the simulation lasts hours.

Observations and simulation receptors
About 20 monitoring stations are spread over the region, out of which seven major sites were chosen for this study. Sixteen receptors were defined for the simulation, representing the different geographic sub-domains: the eastern foothills of Mt. Carmel (1,3,7,8,9,16), the ridge top and slopes (5, 6, 13, 14, 15), and the coastal zone (2, 4, 10, 11, 12 in figure 1). Seven of the receptors represent locations of the monitoring stations (1, 2, 3, 5, 7, 8, 9 in figure 1), while the others are located at non-monitored sites. In the following section 3 simulation results at these receptors are discussed.

Results and discussion
The numerical simulation provided spatial information of the flow and dispersion patterns along the event.
Here the prevailing synoptic system along the event is described in section 3.1, mesoscale and local atmospheric patterns are presented in section 3.2, and selected near-surface profiles of pollutants are discussed in section 3.3.
3.1. Synoptic sea-level pressure system The dominating synoptic system was a low pressure penetration from the Red Sea towards the Eastern Mediterranean coast, called a Red Sea Trough (RST) with a central axis over Israel (figure 2(a)). Such a system forces easterly to north-westerly winds over the Haifa domain. The RAMS simulated synoptic flow shows similar pressure gradient, however, with more of an eastern RST (figure 2(b)), in which case the pressure gradient forces more northerly to north-westerly winds in the domain ('central' or 'eastern axis' refer to the axis position relative to Israel) . The RST incidence over Israel is about 9% (based on Alpert et al 2004), mostly occurs in spring or fall.

Mesoscale and local atmospheric features
The atmospheric flow on 6 May 2007 started as easterly synoptic flow from ground to about 2.5 km AGL and higher. Eeasterly flow in this region is continental and causes a transport of the largest plumes towards the sea. The model showed a development of deep east-west vertical circulation along the day, which resulted with the near surface wind turning into westerly. Flow patterns such as horizontal and vertical local-scale circulations were identified, apparently affecting the pollution dynamics at surface and at near-surface elevated layers. A brief discussion on the simulated surface flow is presented in the next section 3.2.1, and selected simulated multiinversion thermal profiles are discussed next (section 3.2.2).

Surface flow pattern
Under the easterly synoptic flow, the simulated morning surface flow over Haifa Bay shows off-shore horizontal circulation cells of 10-20 km diameter within Mt. Carmel wake zone, as demonstrated in figure 3. The local wind direction associated with the evolution of the off-shore cells is generally easterly over the ridge and south-easterly below the northern slopes of Mt. Carmel. These circulation cells must have served as pollutant trap, as corroborated by peak SO 2 concentrations of 400 and 700 μg m −3 in near-surface elevated layers above sites 10 and 2 respectively (figure 5). These circulation cells disappear when the coastal flow veer into northerly flow (figure 3, 1000 UTC).
Surface concentrations were extracted from the simulation results for eight receptors over the domain, and compared with observations from six sites (1, 3, 5, 7, 8, 9 as in figure 1). Receptors 4 and 16 were set for this study at non-monitored sites, while all the others represent monitoring stations in the domain. Maximum half-hourly values of 300-400 μg m −3 SO 2 were recorded along the study day, an order of magnitude higher than the background level. The maximum observed values were compared with the simulated maxima and they showed a fairly similar pattern ( figure 4). Then, for each sub-period of the day (based on the peaks pattern) we verified which receptor presented the most significant concentrations, as an indicator of the pollution plume. We compared and concluded the deviation of the simulated from observed ground level indicators. We found that the simulation receptors 4 and 16 had a significant impact, suggesting these locations are unique in terms of spatial distribution of pollutants in the domain. At the Carmel foothills we found sideways deviation (Site 8 versus 16 or 9), vertical deviation (site 5 versus 7) and even significant downwind deviation (namely site 9 versus 4). These results are summed in table S1 and available online at stacks.iop.org/ERC/1/085003/mmedia. The deviations between the indicators may hint at the uncertainty of the simulation, or the uncertainty of the plume location.

Near-surface stable layers
The Haifa region is subjected to multi-inversion thermal profiles, due to topographic effects, mesoscale marine or subsidence inversions, and radiation inversions. Each inversion layer may have a different inner-structure and life-time, some break into sub layers, as presented by Haikin et al (2015). In general, the simulated thermal profiles in this case study are multi-inversion and show surface and near-surface inversion layers throughout the day. During most of that day, the temperature at 0.6 km ASL remains higher than at surface, associated with strong inversion beneath. The deepest and strongest inversions were found over the bay and the coastal sites, some as strong as 10 to 12 K (100 m) −1 (figure 5). It is assumed that the relationship between the type of the    (1,3,5,7,8,9) (circles) and eight simulation receptors (1,3,4,5,7,8,9,16) (triangles).
inversion and the pollution dispersion or accumulation beneath or within the layer, is of relevance for pollution dispersion. See elaboration in the supplementary material. Next (section 3.3), two cases are analyzed.

Pollutants near-surface concentration profiles
In the following sections vertical concentration profiles are addressed, with a focus on spatial variation. In many profiles we found zero (observed and simulated) concentration at ground level while high concentration values were found at 100-400 m AGL. This finding significantly differ from other studies which showed constant or decreasing values with elevation in the PBL (e.g. Li et al 2018, Sajani et al 2018. The nature of the emission sources may partially provide an explanation for the differences, where in this study we simulated high sources (stacks) while some of the other studies look at ground level sources. The scope of this paper does not allow presenting a thorough analysis of all sites, hence examples are presented and discussed in sections 3.3.1, 3.3.2.
3.3.1. Case A: morning thermal profiles over the coast The three coastal sites 2, 10, 11 (figures 1, 3) are about 2 km apart. While sites 10 and 11 are close to sea level, site 2 is on Mt. Carmel at about 200 m ASL. Morning thermal profiles over these sites showed significant inversion layers, based at ground level or at 50-100 m AGL (figures 5(b), (d), (f)). The ground level temperature at the coastal site 10 is in the range of 15°C-30°C ( figure 5(d)), while at the elevated site 2 it is 20°C-29°C ( figure 5(b)), and at site 11 only 17°C-18°C ( figure 5(f)). The inversion layers over the three sites are as deep as 300-400 m, and all peak concentrations are within the inversion levels, besides the peak at 0900 UTC ( figure 5(a)) having its maximum at the top of the ground-based inversion. The concentration profiles over sites 2 and 11 show zero at the surface. The patterns in figure 5 suggest some terrain-following features of the thermal layers. Focusing on the 0900 UTC profiles one may see, that the highest peak is found at 100 m above site 2 (figure 5(a)), a much smaller peak at 150 m above site 10 (figure 5(c)), and no peak over site 11 (figure 5(e)). A close examination of the simulated flows at 100 m AGL showed a local circulation over site 2 with a northerly component from the zone of the emission sources, and a northerly flow over the coastal sites 10 and 11 (flow from the sea). However, over site 11 there is also a (north-easterly) component of a flow from the ridge top, which may explain the source origin for the small peak in figure 5(e). Also, as suggested in section 3.2.1 above, circulation cells over the edge of the ridge and into the sea, might have lead to the high concentrations over sites 2 and 10. After 0900 the emitted elevated plumes changed direction with the near-surface north-westerly flow. At the same time, the surface and near-surface levels above site 2 cleared out when the pollutants were transported towards site 16. There, high concentrations were built at the surface (see table 1) and even higher concentrations occurred at near-surface levels (not shown).

Case B-evening profiles over the bay
Sites 1 and 4 (as in figure 1) are located at about sea level, ∼2.5 km apart each other. While site 1 is in an industrial area, site 4 is at the waterfront. Evening profiles over these two sites show no pollution at ground-level, yet they present a significant concentration at near surface elevated levels, with an order of magnitude difference between the sites (figure 6). Temperature profiles over both sites present a total of 400 m inversion depth, based on top of a very strong ground inversion. Since the inversion layers over both sites have similar dimensions, evidently, the temperature behavior (and wind, not shown) alone cannot explain the order-of-magnitude difference between the two sites. A further investigation revealed a TKE fit with the concentration pattern: the TKE over site 1 is five to ten times stronger than over site 4, which may explain the 10-fold lower concentrations over site 1, with respect to site 4. The TKE values over site 1 are within the scale of midday values, and may be the major reason for the strong suppression of the pollutant concentrations there. TKE pattern suggests a shallower ABL over site 4 compared to site 1 (figures 7(d) versus (b)), an additional parameter which also supports the formation of higher concentrations. Note, that in this case TKE is negligible at ground level while it is significant at elevated levels.

Conclusions
Danger may come from the least expected direction, as for instance from the 'dead zone' in a car's mirror. Therefore, when it comes to strong pollution, it is most important to reduce the uncertainty as much as possible, and gain the best information on the mechanism and pattern. Analysis of adjacent vertical profiles of pollution showed numerous occurrences where no pollutant was observed or modeled at ground level, while high concentrations were found at a nearby site and/or at near-surface elevated levels, such outstanding cases were discussed above. This finding highlights the importance of gaining knowledge of near-surface elevated pollution, which is often not addressed in surface pollution studies.
High variability of pollution concentrations was found between adjacent sites, both at ground and at elevated layers, emphasizing the strong uncertainty of the pollutants dispersion in the study domain, with its most complex features. We presented local circulation cells over non-monitored sites, which were associated with high concentrations (section 3.3). An inverse relation between TKE and concentration peaks was found in elevated levels over adjacent sites, where weak TKE was associated with an order-of-magnitude higher concentrations at elevated levels over one site, compared to the nearby site. The day-lasting deep inversion structure along the event in this study, had an impact on the formation of high concentrations, both at ground and elevated levels.
The simulated thermal profiles revealed multi-inversion structures, which agree with earlier observations over the region. An attempt to find (co)relation between the thermal profiles and the vertical structure of the pollution concentrations showed that persistent low-level inversions were associated with high pollutant concentrations close to the ground and at near surface elevated levels. However, higher vertical resolution is required, in order to explore the finer structures of the inversions, and their association with the pollution dynamics and patterns.
Results of this study highlight the importance of gaining a better understanding of the near-surface elevated layers as potential and unexpected pollution sources to the ground level. It is suggested that further modeling efforts with advanced models should be used to elaborate and further explore near-surface elevated pollution layers and their impact and interaction with surface layer. Remote sensing technology may also add useful spatial information of pollution dispersion near and over the local topography, and support the simulation effort. As expected, simulation results in this study illustrate the limits of even highly dense ground monitoring to indicate  occasional air pollution episodes: they may either miss concentration peaks or even totally miss the existence of air pollution plumes, when the plumes are slightly shifted from the monitoring station or above.
In summary, the near-surface steep vertical gradient of pollution as indicated here is associated with a combination of deep inversion (or multi-inversion), vertical circulation due to topography or synoptic flow, and small scale circulation induced by the complex topography.