The impact of temperature inversions on black carbon and particle mass concentrations in a mountainous area

Residential wood combustion is a widespread practice in Europe with a serious impact on air quality, especially in 15 mountainous areas. While there is a significant number of studies conducted in deep urbanized valleys and basins, little is known about the air pollution processes in rural shallow hollows where around 30 % of the people in mountainous areas across Europe live. We aim to determine the influence of ground temperature inversions on wood combustion aerosol pollution in hilly, rural areas. The study uses Retje karst hollow (Loški Potok, Slovenia) as representative site for mountainous and hilly rural areas in central and southeastern Europe with residential wood combustion. Sampling with a mobile monitoring platform 20 along the hollow was performed in December 2017 and January 2018. The backpack mobile monitoring platform was used for the determination of equivalent black carbon (eBC) and particulate matter (PM) mass concentrations along the hollow. To assure high quality of mobile measurement data, intercomparisons of mobile instruments with reference instruments were performed at two air quality stations during every run. Our study showed that aerosol pollution events in the relief depression were associated with high local emission intensities originating almost entirely from residential wood burning and shallow 25 temperature inversions (58 m on average). The eBC and PM mass concentrations showed stronger associations with the potential temperature gradient (R=0.8) than with any other meteorological parameters taken into account (ambient temperature, relative humidity, wind speed, wind direction and precipitation). The strong association between the potential temperature gradient and pollutant concentrations suggests that even a small number of emission sources (total 243 households in the studied hollow) in similar hilly and mountainous rural areas with frequent temperature inversions can significantly 30 increase the levels of eBC and PM, and deteriorate local air quality. During temperature inversions the measured mean eBC and PM2.5 mass concentrations in the whole hollow were as high as 4.5 ± 2.6 μg m and 48.0 ± 27.7 μg m, respectively, which is comparable to larger European urban centres. https://doi.org/10.5194/acp-2021-869 Preprint. Discussion started: 25 October 2021 c © Author(s) 2021. CC BY 4.0 License.


Introduction
One of the biggest emission sources of Particulate Matter (PM) and carbonaceous aerosols is wood burning (Van Der Werf et 35 al., 2010). It is a worldwide problem affecting undeveloped and developed countries as well as rural areas and cities (e.g. Bonjour et al., 2013, Bond et al., 2004Yttri et al., 2005;Caseiro et al., 2009;Fuller et al., 2014;Puxbaum et al., 2007;Stohl et al., 2007;Reis et al., 2019). There is a major concern regarding the impact of PM particles from wood burning on health (e.g. Kocbach Bølling et al., 2009;Fong and Nussbaumer, 2012) and climate (Bond et al., 2013).
Wood combustion represents 51 % of the total PM2.5 emissions in the European Union (EU), which is the highest among all 40 emission sources (European Environmental Agency, 2020). Not only is wood used as a heating fuel in the Nordic and Alpine regions (Puxbaum et al., 2007;Yttri et al., 2005;Herich et al., 2014 and the reference therein), but is becoming a more common practice elsewhere, also in the Mediterranean region (Titos et al., 2017 and the reference therein). Fuller et al. (2014) state that promotion of biomass as a CO2 neutral fuel, higher taxes in heating diesel and the financial crisis have resulted in the widespread use of wood as fuel in Europe in the last years. In addition, actions implemented for reduction of PM emissions in 45 Europe mostly focus on road traffic and cities whereas residential wood combustion and rural areas remain underrepresented with only few studies (for instance, Reis et al., 2009;Wählin et al., 2010;Becerril-Valle et al., 2017) showing that wood combustion in rural regions can induce PM exposures tantamount or often even higher than those from traffic.
Populated relief depressions are particularly vulnerable to air pollution as a combined result of high emissions and unfavourable meteorological conditions for dilution of pollutants. As a result of near-surface radiative cooling, ground temperature 50 inversions frequently and distinctly occur in concave-shaped landforms, especially in the cold season. At the bottom of the surface depressions a strong stable layer is formed preventing vertical mixing of pollutants and ventilation, leading to the accumulation of pollution (Sandradewi et al., 2008;Lyamani et al., 2010). According to Herich et al. (2014) concentrations of carbonaceous aerosols in the Alpine valleys are commonly up to 6 times higher than in urban or rural locations at the foothills of the Alps. 55 The problem of air pollution in deep urbanized valleys and basins was recognized and systematically studied in the 20 th century already (e.g. Whiteman, 1990, 1999 andreference therein). Severe episodes of air pollution and the lack of knowledge about complex atmospheric processes in populated terrain hollows are reasons for several studies of this topic in the last decade as well (e.g. Chazette et al., 2005;Malek et al., 2006;Chuang et al., 2008;de Franceschi and Zardi, 2009;Gohm et al., 2009;Silcox et al., 2012;Green et al., 2013;Rendón et al., 2015). There is a predominant use of atmospheric models that continue 60 to have difficulties predicting dispersion in the strongly stable boundary layer in complex relief (e.g. Nielsen-Gammon et al., 2010;Ritter et al., 2013;Cuvelier et al., 2014;Tartakovsky, Stern, and Broday 2016;Taylor, Hirsch, and Burns 2017). The problem of models is the use of empirical data collected on a flat, homogenous terrain that cannot be simply transferred to a complex relief where atmospheric conditions change on much shorter distances (Holmes et al., 2015). Consequently, air pollution problems in populated relief depressions, especially in small-scale ones, are often underestimated. In order to are needed. Available methods (e.g. remote sensing instruments such as sodars and ceilometers, aerosol lidars, aircraft measurements, ground-based scanning winds) for measuring horizontal and vertical inhomogeneities in a complex relief are often expensive (Baasandorj et al., 2017) and do not provide all the necessary information if operated alone (there is a need for 3D data, continuous measurements and measurements of different aerosol properties) (Gohm et al., 2009;Wang et al. 70 2019). In order to acquire highly resolved spatial and temporal data, mobile measurements have been increasingly used in air quality monitoring, predominantly in urban areas (e.g. Alas et al., 2019aAlas et al., , 2019bAlas et al., , 2020 and the reference therein), yet rarely in hilly/mountainous rural regions (for instance Weimer et al., 2009).
While much attention is given to urbanized deep valleys, only sparse data are available on people's exposure to air pollution in the countryside. According to the population density of Europe (Eurostat, 2011) which includes all EU's countries, the 75 United Kingdom and four European Free Trade Association (EFTA) countries roughly one third (27 %) of the population are living in rural areas. More recent data on population distribution by degree of urbanisation from 2015 (DEGURBA, Eurostat, 2016) which includes EU member countries only, report very similar share of the population in rural areas (28 %) with a gradual increase by 1.7 % from 2010 to 2015 (Margaras, 2019).
The aim of this study is to quantify the influence of ground temperature inversions on spatiotemporal variability of wood 80 combustion aerosol pollution in mountainous regions, an example of which is the model region Loški Potok, Slovenia. The study uses the Retje karst hollow as a representative example of a small-size terrain depression with a village situated at its bottom. Besides the symmetrically shaped topography and smaller size of the hollow, allowing to study generally valid atmospheric and aerosol processes in relief depressions, the example is very convenient for studying the influence of residential wood combustion on local aerosol pollution, as the impact of other emission sources is expected to be rather small. The 85 objectives of the study are: 1) to analyse wood burning aerosol pollution with quality high-resolution aerosol measurements, 2) to assess spatiotemporal variability of equivalent black carbon or eBC (term introduced by Petzold et al., 2013) and PM2.5 during temperature inversion episodes and during mixed atmosphere, and 3) to determine the correlation between meteorological parameters with eBC and PM pollution.
To achieve these, we performed stationary and mobile measurements along the selected hollow. We identified temperature 90 inversion events according to the vertical potential temperature gradients and determined mixing layer heights (MH) during these periods. Afterwards, correlation analysis between meteorological parameters and pollutants levels at the fixed stations and along the hollow with respect to time of day was performed. Subsequently, the obtained results are discussed and finally, we provide conclusions and suggestions for rural relief depressions with limited self-cleaning air capacities.

Measurement site
The area selected for the study is a small, covering approx. 1.5 km 2 , and shallow, less than 150 m deep, karst depression with a topography favourable for the formation of ground temperature inversions or cold air pools (CAPs), especially in winter.
Terrain configuration of the karst depression in northeast direction with marked location of the Retje village is shown in Fig.  100 1. (c).

105
Surrounded by forests in all directions and therefore influenced almost solely by 243 households with 690 residents (SURS, 2018), the site is ideal to study impacts of residential wood burning on local air quality. Due to these characteristics, measurements at Loški Potok can be taken as representative of the atmosphere found in rural hilly regions, where wood is used as a heating fuel. For more information about the measurement site the reader is referred to Glojek et al. (2018) and Glojek et al. (2020). In order to analyse the impact of ground temperature inversions on aerosol pollution, two temporary air quality 110 stations with reference instruments were set up in the study area. One was placed at the bottom of the karst depression in the Retje village (45°42' 34.8''N 14°34'53.8''E, el. 715 m a.s.l.) as a rural village site and one on top of Tabor hill (45°42'01.5"N 14°35'19.7"E, el. 815 m a.s.l) for rural background atmosphere investigation. The Tabor site was selected for a rural background location due to its position, presumably above the ground temperature inversions; with minimal number of possible https://doi.org/10.5194/acp-2021-869 Preprint. Discussion started: 25 October 2021 c Author(s) 2021. CC BY 4.0 License. emission sources in its immediate vicinity (cemetery, rectory building and one residential house); with limited accessibility to 115 vehicles, yet reachable by mobile measurements; and due to possibility to place there the air quality station. Besides two air quality stations deployed in the study area, the 6 km long fixed route for mobile measurements was designed along the entire hollow ( Fig. 1. (b)). The route starts in the capital of the Loški Potok municipality i.e. in the settlement Hrib at south-eastern part of the karst hollow. The agglomeration is located along the regional thoroughfare and involves all the public buildings in the area (school, kindergarten, rest home, grocery store, post office, town hall and health centre). From there the fixed mobile 120 route leads to the top of the hill Tabor passing the weather station in its immediate vicinity. After returning from the rural background air quality station Tabor, route continues in the direction of the village Retje (BR & VB). Leaving the lowest and the widest part of the hollow, the route proceeds towards the chapel of St. Florjan at the north-western most point of the hollow.
This part of the route (CR) is inaccessible to vehicles, and it is dominated by permanent pastures including some arable land.
From there route returns to the residential area of the village Retje, which includes the bottom part of the hollow (BV) and the 125 slopes, facing south (VS). The mobile measurements route covers the lower part of the slopes. The fixed route around the hollow ends in the settlement Hrib.

Aerosol measurements
In this study, the focus is on eBC mass concentrations and size-resolved measurements of particle number (PNC) and mass concentrations (PMC) of fine particle range in winter 2017-18. At the Retje air quality station, additional PM10 mass 130 concentrations are also available. In December 2017 and in January 2018 mobile measurements were carried out as well. The instrumentation used for the fixed stations and mobile platform are summarised in Table 1.

Aerosol measurements at fixed stations
For both measurement stations, the inlet systems were equipped with a PM2.5 inlet head which was connected to a 1.5 m Nafion®Permapure dryer (Wiedensohler et al., 2012) to maintain the relative humidity of the sampled air below 40 %. For 140 eBC measurements, the Aethalometer model AE33 (Magee Scientific / Aerosol d.o.o.) was used. The instrument calculates the absorption and corresponding eBC mass concentrations from measurements of light transmission at 7 different wavelengths (370-950 nm). It uses a "dual-spot" technique to compensate the loading effect artifact in real-time (Drinovec et al., 2015).
The flow rates of Aethalometers at the stations were set to 5 l min -1 and the measurement time resolution to 1 min. The filter tape used was TFE-coated glass fiber filter (no. M8060). The mass absorption cross section (σ) of 7.77 m 2 g -1 was used to 145 convert the optical measurement at 880 nm to eBC mass concentration. The Aethalometer AE33 also served as reference for eBC mass concentration for the mobile platform.
Parameters of PNC and their size distributions (PNSD) were measured using the Mobility Particle Size Spectrometers (MPSSs) following the recommendations given in Wiedensohler et al. (2012). The quality assurance of the instruments and measurements were done according to Wiedensohler et al. (2018). The MPSS classifies electrical particle mobility typically 150 in the range 0.01-0.80 µm. Using the standardized bipolar charge distribution, the PNSD can be calculated (Wiedensohler, 1988 85 Kr neutraliser in bipolar diffusion charger, while the TSI MPSS at the Tabor station had an x-ray source to charge aerosol particles before they enter the DMA. The TROPOS-type MPSS uses high-voltage supply of positive polarity, whereas TSI MPSS has a negative polarity. The sheath flow rate of 5 L min -1 and an aerosol flow of 1 L min -1 were set up for TROPOS MPSS with particle size range 0.01-0.80 µm. At Tabor station the sheath air to aerosol flow ratio was set to 4.1:1 l min -1 , resulting in particle size range 0.01-0.60 µm. Both instruments were operated with a time resolution of 5 min. The data from 160 both stations were processed with the TROPOS software with a linear multiple charge inversion algorithm (Pfeifer et al., 2014) and additional corrections for internal and sampling losses due to diffusion and sedimentation as described in Wiedensohler et al. (2012). In case of the TSI MPSS at Tabor station the inversion for an x-ray matrix was included as well (Tigges et al, 2015).
The quality of the MPSS data was assured by laboratory inter-comparisons with reference instruments performed by the World Calibration Center for Aerosol Physics (WCCAP) at TROPOS in Leipzig, Germany ( Fig. S2 to S6 in the Supplement). From 165 the PNSD measured at the Retje site PM1 mass concentrations were calculated. The results of the procedure described in Sect.  Glojek et al., 2019Glojek et al., , 2020 and in the Supplement.
Additional PM10 measurements were obtained according to EN 12341 standard using a high-volume sampler (Digitel, DHA-80). The PM particles were collected on preheated quartz fiber filters (150 mm diameter) at a flow rate of 500 l h -1 . 12-hour 170 sampling was performed, from 6:00 to 18:00 and from 18:00 to 6:00 local time (LT). The filters were stored frozen until analysis.

Mobile eBC and PM2.5 measurements
The mobile measurements were done using the TROPOS aerosol backpack equipped with portable instruments (Alas et al., 2019a(Alas et al., , 2019bAlas et al., 2020). On top of the backpack, the aerosol is sampled with 1 m long stainless steel inlet. Inside the 175 backpack, eBC mass concentration is measured with microAethalometer AE51 (AethLab). The portable absorption photometer measures the attenuation of light at wavelength of 880 nm through a particle-loaded T60 Teflon coated glass fiber filter and converts it to eBC mass concentrations using σ of 12.5 m 2 g -1 (provided by the manufacturer). Before entering the AE51 the aerosol sample passes through a silica-gel drier which dries the aerosols and dampens the effect of sudden changes in relative humidity and temperature (Cai et al., 2014;Düssing et al., 2019), which could produce false signals in the instrument. 180 Downstream the silica gel dryer, another portable filter photometer, the microAethalometer MA200 (AethLab), is installed.
Due to software problems of the MA200 and the filter tape ran out during the campaign the data set of the instrument is not sufficient for the spatial and statistical analysis of this study. Hence, it was not used. MicroAethalometers were operated with a flow of 150 ml min -1 and time stamp of 1 s. In order to minimize the impacts of single events that do not represent the typical concentration profile of the area and due to high noise of eBC mass concentrations at 1 s resolution, the data were processed 185 to 10 s medians.
To obtain PM mass concentrations, optical particle size spectrometer (TSI OPSS, model 3330) was used. The instrument is connected to another aerosol line inside the backpack, which is not dried. It is an optical instrument that measures the optical particle diameter according to the light scattered and related refractive index of the calibration aerosol, typically polystyrene latex (PSL) spheres with refractive index (ñ) of 1.59. However, the optical properties of ambient aerosols largely differ from 190 the PSL, which results in wrong sizing in terms of the volume equivalent diameter. Furthermore, the OPSS measures an PSL equivalent size range of 0.3-10 μm, while the accumulation mode is only partially measured, and the Aitken mode is not measured at all. This means that PM calculation from the OPSS PNSDs generally lead to an underestimation of the mass concentrations, which can be signifcant. To address this issue, the OPSS data were corrected according to the procedure briefly described in the latter part of this section, which follows the methodology presented in Alas et al., 2019a. During the Loški 195 Potok campaign, the OPSS had a flow of 1 l min -1 and 10 s time resolution. The position of the mobile measurements was obtained by a GPS device placed on top of the inlet. All instruments together with the GPS unit were synchronized and controlled by a microcomputer where the data was also logged. Simultaneous mobile measurements with two instrumented backpacks were carried out 3 times a day along a fixed route in the karst depression Retje. The run in the morning lasted from 6:30 to 9:00, at noon from 12:00 to 14:00 and in the evening from 200 17:00 to 19:00 LT. To obtain representative information, daily measurements along the depression were conducted 107 times, totalling to 642 km. Results of the convergence analysis (see Alas et al., 2020 and the reference therein for the assessment process) for all data points in the Retje village and at Tabor hill are depicted in Fig. S12. The number of runs when the cumulative median concentrations of randomly selected runs deviate less than 50 % from the median of all the runs at the given location was selected for a threshold. According to visual analysis, the data converged faster for the hill Tabor compared 205 to the village Retje. Thus, the minimum number of runs required to obtain a representative image of the air quality on the Tabor hill is 46 for eBC and 52 for PM2.5, whereas for the Retje village it is 65 for eBC and 58 for PM2.5. The route (see Fig. 1. (b)) led past fixed stations with the following reference instruments, the AE33 Aethalometer and the MPSS (Table 2). During each run the carriers of the backpacks intercompared the mobile AE51s with the AE33 Aethalometers and the OPSSs 3330 with the MPSSs. Intercomparison at the Retje site (the rural village station) lasted 20 min and at the Tabor site (the rural background site) 10 min. The methodology used to assure high-quality mobile measurements of eBC and PM mass concentrations is provided in Alas et al., 2019a. 215 In order to obtain accurate eBC mass concentrations, mobile filter photometers were corrected for the filter-loading effect (FLE). The correction algorithm presented by Virkkula et al. (2007) was used, with a loading parameter (k) of 0.005 for the AE51 based on Drinovec et al. (2017). Besides diesel dominated aerosols, the selected value of k is representative for freshly emitted particles from wood burning as well. Detection of the filter-loading effect, correction, and correlation between the microAethalometers AE51 against reference instruments (the AE33 Aethalometers) used in the Loški Potok campaign are 220 published in Alas et al., 2020. As previously stated, PM2.5 mass concentrations were derived from the OPSS measurements that were corrected by the aerosol type-dependent complex refractive index (ñ) and unique fine-mode volume correction factors (CFsf, vol). The real part (ñre) of the OPSS ñ was adjusted to 1.52 and the imaginary part (ñim) to i0.02 for all size bins. Both values are typical for rural, residential areas containing absorbing aerosols (Ebert et al., 2004;Lu et al. 2015). However, applying one corrected ñ for 225 particles in range 0.3-0.8 µm, as well as for the particles larger than 0.8 µm, could lead to high uncertainties due to unknown https://doi.org/10.5194/acp-2021-869 Preprint. Discussion started: 25 October 2021 c Author(s) 2021. CC BY 4.0 License. particle shape and light absorbing particle composition in the supermicrometer range. After the ñ correction had been applied on the OPSS PNSD, using the Mie Theory, the optical diameters were converted to geometric mean volume equivalent diameters for the submicrometer size range (the correction of the data is illustrated in the Supplement Fig. S7, S8 and S9). The CFf, vol corrects the OPSS constraint of measuring only a fraction of the fine mode particles (from 0.3 to 10 µm). Therefore, 230 CFsf, vol were calculated as a ratio between the fine mode of the OPSS (from 0.4 to 0.8 µm) with the fine mode of MPSS (from 0.001 to 0.8 µm) during each intercomparison period, since the variability of PVSD requires the use of unique CFf, vol for each individual run (Alas et al., 2019a). As there are no significant differences in aerosol sources along the hollow (see Glojek et al., 2018) and due to instrument failures at the second measurement location, data points of the whole selected route were corrected with the correction factors derived at the Retje rural village station. The relevance of the use of correction factors 235 obtained at a single measurement site (Retje) for the entire route was confirmed by the comparison of the particle volume size distributions (PVSDs) measured by the MPSS at the Tabor rural background station. Namely, PVSDs at both measurement sites peaked around 0.3 µm (see Supplement Fig. S8). Once all the corrections were applied for each run, PM1 mass concentrations were calculated. Since the organic aerosols prevail in the studied area, a particle density of 1.5 g cm -3 (Turpin and Lim, 2001) was used for the calculation. The absolute value of PM1 was added to the absolute value of PM1-2.5 to obtain 240 PM2.5 mass concentrations. Corrected PM1 mass concentrations of the OPSS are comparable to the reference MPSS.
Correlation between median PM1 mass concentrations obtained from the OPSS and MPSS during each intercomparison time is high, with R 2 > 0.8 and slope of 1.12. Additionally, PM1 mass concentrations derived from the MPSS measurements were compared to the 12-hour PM10 data, measured by the High-Volume Sampler Digitel DHA-80. These two parameters correlate well too, with R 2 =1 and slope of 1.09 (Supplement Fig. S7). The relationship shows that most of the mass concentration of 245 PM10 is already found in PM1. Despite considerably improved calculated PM2.5 mass concentration values of the OPSS after applied corrections, some uncertainties remain, primarily due to the particles non-sphericity, assumed particle density and the use of the same time-dependent correction factors for the whole route. With an increase in particle size, the shape of the particles becomes even more non-spherical and thus the refractive index correction based on the Mie theory, assuming spherical particles, is not possible. Due to this high uncertainty, PM10 mass concentrations from the OPSS data were not 250 calculated for this study.

Meteorological parameters
Hourly and daily averages of meteorological parameters recorded from air quality stations in Retje and at Tabor (ambient temperature, pressure, and relative humidity) and from the weather station of the non-governmental Society for Weather and Climate Research (Društvo za raziskovanje vremena in Podnebja, 45°42′4″ N 14°35′27″ E, el. 775 m a.s.l.) (ambient 255 temperature, pressure, relative humidity, wind speed and direction and the sum of precipitation) were computed. The location of the weather station is marked with the black cross (Hrib in Fig. 1. (b)). To determine temperature inversion episodes, mobile temperature measurements with a temperature sensor (DS1922L iButton, Maxim Integrated) were considered as well. Besides the observed meteorological variables, heating degree days (HDD) were determined from the daily mean temperatures at the https://doi.org/10.5194/acp-2021-869 Preprint. Discussion started: 25 October 2021 c Author(s) 2021. CC BY 4.0 License.
stations. The index was calculated according to Eurostat (2019) as well as ARSO (2020) methodology. The former uses a daily 260 threshold value, defined as the lowest daily mean air temperature not leading to indoor heating, of 15 °C and the latter 12 °C.

Detection and selection of temperature inversion periods
Temperature inversion periods in the selected karst depression were detected by analysing the calculated potential temperature gradient, obtained from ground-based observations of temperatures at different elevations, and the mobile temperature measurements along the hollow. The ratio of temperature difference between stations at different heights (Δθ/ΔZ) is a good 265 indicator of the stability of the boundary layer during temperature inversions as shown in, for instance, Petkovšek, 1978;Whiteman et al., 2000Whiteman et al., , 2014Largeron and Staquet, 2016;Holmes et al., 2015. In our study temperature differences between the Retje station at the bottom of the karst depression, the Hrib station on the slope and the Tabor station on top of the hill were calculated. Mobile measurements of temperatures in December along the Retje karst depression were used as supplementary information to determine atmospheric conditions during the runs. If temperatures during mobile measurements 270 along the karst depression were increasing with altitude (positive gradient), the runs were classified as runs during the temperature inversions. However, if temperatures were decreasing (negative gradient), the runs were classified as runs with a mixed atmosphere. To conform to the defined atmospheric conditions other meteorological variables were considered as well.
Number of classified runs during temperature inversion and during mixed atmosphere is represented in Table 3.

275
The number of runs with strong temperature inversion is shown in brackets.

Time of day (LT)
Number Out of the total 107 runs performed along the hollow, 98 were valid, i.e. without measurements failures (e.g. leak in the system, wrong GPS route tracking, full filter etc.), and classified according to the meteorological conditions. Forty-six runs were classified as runs during temperature inversion, from which 27 with strong temperature inversion (the numbers indicated in brackets). The criteria for further separation of the temperature inversions is determined by the potential temperature gradient 280 above 1 K (100 m) -1 and by the change of the weather regime during the run. Thus, run with strong temperature inversion stands for the run with a potential temperature gradient above 1 K (100 m) -1 during the whole run with no weather changes or intensive mixing of atmosphere during the run. Forty-three runs had a negative temperature gradient during the whole run and were therefore classified as runs with mixed atmosphere. During 9 of the runs atmospheric conditions changed from stable to unstable or vice versa and thus were not taken into account in the analysis according to meteorological conditions.

Determination of the mixing height during temperature inversions
The mixing height (MH) separates the surface boundary layer from the free troposphere. Detection of the MH depends on the method used (e.g. the bulk Richardson number, temperature, humidity or wind and turbulence parameters of vertical profiles).
The method is usually chosen based on data availability (Seibert et al., 2000). In the present study, the MH was determined by a gradient method (Ferrero et al., 2010 and the references therein) from eBC vertical profiles. Previous studies (Kim et al., 290 2007;Angelini et al., 2009;Ferrero et al., 2010;Mocnik et al., 2011;Ferrero et al., 2016) demonstrated the accuracy of particle and eBC derived MH method, namely, atmospheric particles indicate the atmospheric dispersion state. Particles in the mixing layer accumulate due to a very stable atmosphere. Therefore, a significant difference in concentration levels between the surface boundary layer compared to the free atmosphere commonly occurs (Emeis et al., 2008;Summa et al., 2013). This sharp decrease in concentrations is defined as a MH (Balsley et al., 2006;Kim et al., 2007;Yang et al., 2017). 295 Mobile measurements of eBC and temperature along the hollow were used for the determination of mixing layer heights during the selected temperature inversion episodes. MH was calculated by means of the gradient method from 5 m averaged eBC pseudo-vertical profiles (some details on spatial averaging are provided in Sect. 2.4). The profiles were named pseudo-vertical as they were obtained with mobile measurements along the slopes of the hollow and not with commonly used methods for vertical profile measurements, e.g. radio-soundings, balloons, UAVs or lidars. It should be noted that local emission sources, 300 slope winds and surface are expected to have a greater impact on measurements performed along slopes as opposed to the stated standard vertical measurements.
The strongest gradient of the eBC concentration profile was chosen as the mixing height. To confirm the reliability of eBCderived MH, mixing heights for the selected temperature inversion episodes in December were determined by mobile temperature measurements as well. As with the eBC-derived MH, MH estimated from temperature profiles was determined as 305 the height of sudden change in temperature. Comparing the results shown in the Supplement Fig. S13 it is important to consider that time resolution of temperature measurements was lower (2 min) than the resolution of particle measurements (10 s) and this can influence the accuracy of the mixing layer height determination. The agreement between different methods is very good for very pronounced inversions.

Spatial and statistical analysis 310
Concerning the second goal, mobile measurements were spatially averaged according to the method described in Alas et. al., 2018 andAlas et al., 2019a. Firstly, a pre-determined route with 10 m equidistant points in QGIS 2.18.22 was created. These points were used as centre points for spatial aggregation performed in R programming language. Prior to spatial aggregation, mobile runs were separated into two groups according to atmospheric conditions: runs occurring during temperature inversions and runs with a mixed atmosphere (for the classification of runs with respect to meteorological conditions see section 3.3.1). 315 Subsequently, the median of all data points in a radius of 10 m around each centre point was calculated. Results are moving circular medians of eBC an PM2.5 mass concentrations along the route calculated separately for all the runs during strong  (27) and for all the runs during mixed atmosphere (43). In addition, average ambient ratios of eBC to PM2.5 for the whole hollow were calculated.
For the goal to establish association between aerosol pollution and meteorology, correlation analyses between meteorological 320 variables with eBC and PM mass concentrations for the whole measurement period and for the selected days with temperature inversion (temperature inversion days) and days with mixed atmosphere (non-inversion days) was performed. An ordinary least squares (OLS) linear regression was applied using Python programming language with the "scipy" module.  Similar results were found for PM10 as well. There was also a third distinct increase in pollution levels at Retje during the longest temperature inversion period in February. However, the eBC data for this period are partly missing. The first period lasted from 4 December to 8 December, the second period from 17 December to 25 December and the third period from 14 February to 18 February. The second period in December presents the longest period with temperature inversion of the whole 360 campaign. All of these periods were characterized by an anticyclonic weather with none or weak, less than 0.7 m s -1 wind speeds, the absence of precipitation, and low temperatures at the bottom of the hollow, reaching a trough of -18.4 °C on February 15 at the Retje village station. Due to lower temperatures at the bottom of the hollow than on top of the hill, the relative humidity in the hollow was higher than on Tabor hill. Morning and evening temperatures in the hollow dropped regularly below the dew point during temperature inversions, forming radiation fog. Regular diurnal eBC, temperature and 365 relative humidity oscillations could be seen during temperature inversions. The valley floor was covered with snow during all three temperature inversion episodes. The period between the first and the second longer temperature inversion in December i.e., from 8 December to 16 December, was characterized by an intense frontal activity with strong, from 2 m s -1 (25 th percentile) to 4 m s -1 (75 th percentile) south to southeast winds with gusts above 9 m s -1 and high snow precipitation with 61 cm of new snow recorded. After 10 December there was a strong thaw with warm air masses. After the second temperature inversion 370 period in December the temperatures rose again and stayed above the long-term average for the whole month of January and at the beginning of February (see Glojek et al., 2018). Precipitations in the form of rain were above the long-term average in comparison with the usual amount of rainfall in January (Cegnar, T., Knez, 2018). Out of 14 precipitation days in January, only 4 days had snowed. In January and at the beginning of February periods with temperature inversions were shorter compared to December, lasting no longer than 4 days, with higher temperatures and lower relative humidity, respectively, 2.5 375 °C higher and 7.6 % lower mean values at all the stations. Moreover, the PM10 and eBC mass concentration levels during January and February temperature inversions were lower compared to temperature inversions in December, except from 14 February to 18 February when PM10 mass concentrations reached maximum levels. In Retje, mean PM10 and eBC levels were, on average, more than 2 times lower, with mean of 133.5 ± 42.3 µg m -3 and 12.9 ± 10.7 µg m -3 , respectively, for December inversions and mean of 56.2 ± 19.9 of PM10 and 6.3 ± 6.7 µg m -3 of eBC for inversions in January and February. Due to missing 380 data, the longest temperature inversion episode in February was not included in the calculation. At Tabor the difference between mean value of 3.6 ± 3.3 µg m -3 for December inversions and mean of 2.3 ± 2.7 µg m -3 for inversions in January and northeast winds prevailed over the area bringing snow precipitation respectively. In February there were more than 19 days 385 with precipitation, including 17 days of snowfall, and 26 days of snow cover. Cloudy weather prevailed during the whole month. Temperature inversions occurred mainly within the first and the second third of February.

Spatiotemporal variability of eBC and PM2.5 during temperature inversion and during mixed atmosphere
As already seen in time series of eBC and PM10 mass concentrations at the stations in Fig. 2  were observed along the route to the village Retje (BR). Levels of PM2.5 were increasing from 32 to 60 µg m -3 along the route towards the village (BR), which lies at a lower altitude. The eBC trend, however, depends on the time of the day (Section 4. 3., Fig. 4.). In the morning, from 6:30 to 9:00, eBC levels with the mean value of 3 µg m -3 remain constant along the whole bottom route (BR), whereas around noon and in the early evening, eBC mass concentrations change along the route. As seen with the PM2.5 levels, early evening eBC mass concentrations increase from 6.6 µg m -3 to 11.6 µg m -3 in the direction of the 415 village Retje as well. Conversely, around noon eBC mass concentrations are higher on the part of the route leading to the settlement Hrib (SR), which is at a higher altitude, increasing from 4.0 to 7.5 µg m -3 .

Mixing height during temperature inversions
The mixing layer height was determined for the selected temperature inversion events, based on eBC vertical profiles collected   Fig. S13). On average, during strong winter temperature inversion events the MH in the hollow is 58 ± 15 m. For temperature inversions with a stronger temperature gradient of about 4.7 ± 1.7 K (100 m) -1 , prevailing in December, the MH is on average 6 m higher (61 ± 15 m) than during temperature inversions with a weaker temperature gradient of 2.5 ± 2.1 K (100 m) -1 (55 ± 12 m).   profile were 4.0 ± 4.2 µg m -3 , within the MH 5.0 ± 4.5 µg m -3 and above MH 1.7 ± 1.6 µg m -3 . Early evening runs are characterized by the highest eBC mass concentrations with mean eBC levels of 12.3 ± 8.9 µgm -3 within the mixing height, 4.2 ± 3.4 µg m -3 above MH and 9.8 ± 8.5 µg m -3 in the whole vertical profile. During afternoon runs the difference between mass concentrations within the MH and just above the MH was 2 times smaller than during morning and evening runs. The mean eBC value within the MH was 6.9 ± 5.0 µg m -3 compared to 5.1 ± 3.7 µg m -3 of eBC detected just above the MH. On average, 475 eBC mass concentrations in the free atmosphere represented 32 % of the total eBC mass concentrations within the mixing height.
Besides the strong eBC mass concentration gradient, representing the border of the MH, two other layers of high eBC mass concentrations are seen in most of eBC height profiles (Supplement Fig. S13). As seen on the spatial maps (Fig. 3.)

Association between meteorological variables with eBC and PM at the fixed stations 485
Correlation analysis between hourly and daily meteorological variables and eBC mass concentrations at both air quality stations revealed a statistically significant correlation between eBC levels and temperature, HDD, ambient pressure, wind speed and the potential temperature gradient. No statistically significant association was found in the following meteorological variables: wind direction (N=73) and daily sum of precipitation (N=38). The conditions for correlation analysis between MH and pollution levels or any other meteorological parameter have not been fulfilled, as the determination of MH was done only 490 for the runs during temperature inversions (N=47) with a limited height range between 25 to 105 m, mostly from 40 to 75 m.
A stronger correlation between most of the variables is found at the Retje village site, located in the hollow, than at the Tabor rural background station, positioned on top of the hill. The only exceptions were the wind speed measurements performed at the meteorological station Hrib, positioned outside the hollow. At the village station correlation analysis with 24-hour PM10 samples was conducted as well. Very similar correlation coefficients were obtained as for 24-hour average eBC mass 495 concentrations and are presented in Table 6.   at the Retje village station, with R 2 amounting to around 0.4. The same influence is also observed for PM10. The relationship is negative, meaning that eBC and PM10 mass concentrations decrease with higher temperatures, on average by -0.51 µg m -3 505 and 5.79 µg m -3 , respectively. The results, however, are not most reliable as the correlation is moderate. The same relationship is found for HDD (Eurostat, 2019;ARSO, 2020). According to the HDD index there was the need for indoor heating in all winter days.
Relation analysis of wind speeds and eBC was performed, which considered only days with no precipitation and wind speeds higher than 0.09 m s -1 . The location of the wind measurements is not representative of the conditions on the floor of the hollow. 510 Therefore, correlation analysis is adequate only for the station on Tabor hill. However, merely a moderate correlation was found for the winter days. The relation with eBC is negative with a Pearson correlation coefficient R of -0.5. On average there is a reduction of 31 % in eBC per 1 m s -1 at Tabor hill, yet the results cannot be used with confidence due to the relatively low R 2 . Among all meteorological variables the strongest and the highest correlation was found between the potential temperature gradient in the hollow and pollutants mass concentrations (R 2 > 0.75). A particularly high correlation (R 2 =0.8) is determined 515 at the Retje village station during the selected temperature inversion days and non-inversion days. eBC and PM10 mass concentrations increase with a higher potential temperature gradient of the hollow, on average by 1.98 µg m -3 and 20.21 µg m -1 per 1 K (100 m) -1 , respectively. This increment represents the largest induced change in pollution levels of all the meteorological parameters studied.

Correlation analysis inside the cold air pool 520
High time and spatial resolution data of mobile measurements allowed us to explore the spatiotemporal dependency of the eBC and PM2.5 mass concentrations along the hollow to the meteorological variables. The linear correlation between the variables studied for the different run times is shown in Fig. 6.  Temperature dependence of eBC is higher for evening runs (R 2 = 0.5) compared to morning (R 2 =0.44) and afternoon (R 2 =0.26) runs. However, for PM2.5, the correlation is stronger in the morning (R 2 =0.5) than in the evening and in the afternoon (R 2 = 530 0.3). Yet, the gradient of the linear regression line is, like eBC, the steepest in the evening. With temperature decrease of -1 °C, eBC and PM2.5 are reduced by 14 %.
Despite not the most representative location of wind measurements for the atmospheric conditions inside the CAP, wind-speed dependence of eBC and PM2.5 obtained with mobile measurements was confirmed as well. Correlation is better for eBC than for PM2.5 and it is the strongest for the morning runs (R 2 = 0.6 for eBC and R 2 = 0.3 for PM2.5). Nonetheless, the reduction in 535 pollutants levels in the hollow per 1 m s -1 of wind speed is higher in the evening (-40 %) compared to afternoon and morning runs (from -22 to -30 %).
Unlike the other meteorological variables, the correlation between the pollutant levels and the potential temperature gradient of the hollow is the strongest in the early evening with R 2 = 0.8 for eBC and R 2 = 0.5 for PM2.5. During this time, the increase in the pollutants levels is also by 63 to 72 % higher compared to the morning and the afternoon runs. The change in pollutant 540 mass concentrations in the evening is comparable to those induced by wind speed. However, the significance of the wind speed model is low with R 2 from 0.12 to 0.3.

Discussion
We found out that out of 37 M people in Europe who live in rural mountain areas, 80 % of them live in relief depressions, most of them being from Central Europe. Yet only little is known about the real air pollution levels in those areas. Similar to 545 many other locations in the Alps (Herich et al., 2014) and many other rural hilly and mountainous areas in Europe (e.g. Reis et al., 2009;Wählin et al., 2010;Becerril-Valle et al., 2017;Chevrier, 2017), residential wood burning is a major source of airborne particles in the study area. This was assessed with the so-called "Aethalometer approach" (Sandardewi et al., 2008) in previous research (Glojek et al., 2018;, which is further confirmed by the low ratio of eBC to PM2.5 observed in this study. Namely, the calculated ratios with values between 0.062 and 0.094, correspond to other ratios calculated in the areas 550 with residential wood combustion and for forest fires (Herich et al., 2014;Liu et al., 2018). Moreover, the high correlation between eBC and PM2.5 mass concentrations along the whole hollow indicates homogeneous atmospheric conditions with the prevalence of one emission source.
Our measurements confirmed favourable topographical and meteorological conditions for the formation of temperature inversions in the studied hollow, particularly in the cold part of the year. These stable atmospheric conditions have a significant 555 impact on the air pollution levels in the studied area. Therefore, eBC and PM2.5 mass concentrations build-up in the whole hollow during temperature inversions and reached values (mean of 4.5 ± 2.6 µg m -3 of eBC and 48.0 ± 27.7 µg m -3 of PM2.5) that are comparable to larger European centres (e.g. Fuller et al. 2014;Manigrasso et al., 2017;Küpper et al., 2018). Moreover, they were above the EEA yearly limit and WHO daily guideline values for PM2.5 (both account for 25 µg m -3 ). According to the European Air Quality Index (EEA, 2020) for PM2.5 and PM10 mass concentrations levels, air quality in the hollow during 560 temperature inversions was very poor. For the entire winter campaign 2017-2018, however, it was moderate.
During temperature inversion, pollution levels in the hollow were the greatest in early evenings, reaching up to 22 µg m -3 of eBC and 560.7 µg m -3 of PM2.5. This is the result of domestic wood burning, which increases when people return home after work, and a very shallow mixing height of only 55 m on average, limited to the bottom of the hollow alone. However, during mixed atmosphere both eBC and PM2.5 levels in the hollow dropped to less than 1 µg m -3 of eBC and 12 µg m -3 of PM2.5, which 565 is approximately 4 times lower than during temperature inversions and corresponds to European regional background sites values (EEA, 2019;Lilijana Kozlovič et al., 2019;Sun et al., 2019). However, in proximity of combustion sources, namely in the village Retje at the bottom of the hollow (VB) and at the settlement Hrib on the south-eastern slope, pollutant mass concentrations stayed above 1.8 µg m -3 of eBC and 12 µg m -3 of PM2.5.
Despite the size of the hollow, covering an area of 1.5 km 2 only, we demonstrated that an increase in pollutant mass 570 concentrations during temperature inversions is not the same in the whole hollow. Moreover, it strongly differs within a 250m radius (see Fig. 3.). Besides, maximum mass concentrations are not always associated with the location of local emission sources. This coincides with Gohm et al. (2014) findings, which reject the concept of spatial concentration homogeneity under the mixing height in a complex terrain. Moreover, our results demonstrate that spatial inhomogeneity in small-size hollows could be even smaller than 1 km as observed in the study mentioned above. Spatial distribution of eBC and PM2.5 mass 575 https://doi.org/10.5194/acp-2021-869 Preprint. Discussion started: 25 October 2021 c Author(s) 2021. CC BY 4.0 License.
concentrations differ during the day as well, largely because of different atmospheric conditions. During the morning runs with temperature inversion i.e. from 6:30 to 9:00 LT, the maximum of eBC as well as PM2.5 occurred along the road on the lower slope (SR), facing the southern side of the hollow, and not at the bottom of the hollow (BV), where the highest number of emission sources were detected. This might have happened because the morning run along the slope route was conducted between 8:15 and 8:30 LT, when the sun had already risen. Radiative heating of air above the slope exposed to the sun with 580 no or almost no snow, having a lower albedo compared to a snow-covered floor in the hollow, may be strong enough to induce upslope winds. Differences in temperature above different types of land covers in the hollow were previously detected by a thermal heat camera in the study of Vysoudil et al. (2019). This is one of the possible reasons for the highest pollution levels on the first slope road. However, to verify this hypothesis, wind measurements in the hollow would be required. Local morning traffic (people driving to a local school and to their workplace) contribute slightly as well, especially when it comes to eBC 585 mass concentration levels (Glojek et al., 2020). Yet, the lowest eBC and PM2.5 mass concentrations were detected at the end of the path, at the north-western most point of the hollow (C). The temperature and pollutant concentrations, however, remained the same at the bottom of the hollow. Contrary to pollutant maxima, the location of the minima could be explained by downslope winds as crossing this part of the route happened just before the sunrise. Night slope winds meet further down at the narrow part of the hollow where a slight channelized flow seems to appear, leading towards the lower parts of the valley 590 in the southeast direction. This interpretation looks plausible since the location of the minimum pollutant level in the evening is the same. However, in the evenings with a lower mixing height (55 m on average), the maximum of eBC and PM2.5 mass concentrations is limited to the bottom of the hollow alone. During the day thermal forcing, occurring due to solar irradiation, has the greatest impact on dispersion of pollutants. Like the morning runs, maximum pollution levels were reached at the start of the lower slope road (SR), facing south. Notwithstanding, in the afternoon the difference between eBC mass concentrations 595 above the mixing height and the levels within the hollow was not so distinct. Moreover, eBC mass concentrations were increasing with the height, forming an elevated pollution layer. The formation of the elevated pollution layer by cross-valley flow and upslope wind transport was detected by previous researchers studying air pollution and meteorological processes in cold air pools as well (e.g. Gohm et al., 2014;Shapiro, Fedorovich, 2007;Rakovec et al., 2002). Connected to this, it should be noted that the settlement Hrib, positioned mostly above the mixing layer height of the studied hollow, has most likely a 600 greater effect on pollutant levels on top of the hill Tabor than more distant sources from the village Retje at the bottom of the hollow.
The impact of temperature inversions on pollutant levels was additionally confirmed by a strong correlation between the eBC mass concentrations in the hollow and the vertical potential temperature gradient (R 2 >0.7). The correlation was the highest for the selected temperature inversion and non-inversion days (R 2 >0.8). Correlation coefficients (R 2 ) of around 0.7 to 0.9 were 605 also observed in other studies considering PM2.5 or PM10 pollution levels in winter and measures of atmospheric stability in relief depressions (e.g. Whiteman et al, 2014;Largeron, Staquet, 2016). A moderate correlation (R 2 of 0.5) between eBC and ambient pressure was detected as well since it is indirectly connected with the formation of ground temperature inversions.
Because the measurement campaign was performed in the cold part of the year, correlation analysis of mean daily ambient https://doi.org/10.5194/acp-2021-869 Preprint. Discussion started: 25 October 2021 c Author(s) 2021. CC BY 4.0 License. temperature did not show a considerable impact on the pollution levels (R 2 =0.4). Namely, all days in winter fulfilled the criteria 610 to be classified as heating degree days (HDD according to Eurostat, 2019 and ARSO, 2020). However, the impact of temperature differs within a day. In the morning, when temperatures in the hollow drop to their lowest, the highest impact on PM2.5 mass concentrations was observed. The effect on the eBC mass concentration is slightly lower, probably due to the increased morning traffic. As already described in the result section, wind measurements were performed outside the studied hollow and therefore did not show a very strong relation with eBC mass concentration levels in the hollow. Thus, wind speeds 615 in the relief depression are expected to be even lower due to the topographical effect. Nonetheless, the decrease of pollutants mass concentrations with increasing wind speed in the hollow is clearly observed, indicating entirely local origin of pollution.
It should be noted that the meteorological variables considered are often interrelated (e.g. precipitation and wind, temperature and RH) and can often not be treated individually.

Conclusions 620
Pilot study to determine the residential wood combustion air pollution under "real-world laboratory conditions" is presented.
The studied example is representative for rural hilly and mountainous areas in central and southeastern Europe. We provide rare information on eBC and PM air pollution in rural small-scale relief depressions with mainly one emission source, i.e. residential wood burning. Due to its symmetrically shaped topography, smaller size (1.5 km 2 and less than 150 m deep) and remoteness, the selected Retje karst hollow in Slovenia is very convenient for the study of the impact of temperature inversions 625 on local air quality.
Sampling with a mobile monitoring platform was conducted along the whole karst hollow in December 2017 and in January 2018. The mobile monitoring platform was equipped with portable instruments measuring at time resolution of 10 s. The monitoring sites with reference instruments (Aethalometer AE33 for eBC, Digitel DHA-08 and the TROPOS-type MPSS model for PM determination) were selected according to their position in the hollow and land use. The rural village station 630 was located at the bottom of the karst depression whereas the rural background station was positioned at the top of the hill Tabor.
The primary emission source of eBC and PM in the area is residential wood combustion that is also indicated by a low ratio of eBC to PM2.5, ranging from 0.062 to 0.094. Performed measurements confirmed meteorological conditions favourable for the formation of ground temperature inversions in the studied hollow. Namely, temperature inversion was present in more than 635 70 % of all winter nights and mornings. These very stable conditions prevent effective mixing and dispersion of pollutant concentrations in the hollow, leading to elevated pollution levels. Therefore, during temperature inversions eBC and PM2.5 mass concentrations in the hollow increase to levels which are comparable to larger European urban centres and are above the EEA daily limit value for PM10 i.e. 50 µg m -3 as well as above yearly limit and WHO daily guideline values for PM2.5, i.e. 25 µg m -3 . 640 https://doi.org/10.5194/acp-2021-869 Preprint. Discussion started: 25 October 2021 c Author(s) 2021. CC BY 4.0 License.
Compared to larger and deeper basins and valleys, in small-size, shallow hollows the dynamic of temperature inversion seems to be much faster. Due to a smaller volume of the air in the hollow, a considerably lower number of emission sources can generate eBC and PM2.5 pollution, which is comparable to larger urbanised basins and valleys. Our measurements, therefore, highlight that eBC and PM levels in rural shallow terrain depressions with residential wood burning could be much greater than predicted by models. Besides, even in a small hollow as presented in this case study, pollutant concentrations differ 645 significantly depending on its location. During morning and afternoon temperature inversions people living at the lower part of south-facing slopes proved to be the most exposed to the high eBC and PM2.5 mass concentrations, while in the early evenings, when the MH is limited only to the bottom of the hollow, people in the village Retje, living on the floor of the hollow breathe the highest pollutants level. Pollutant concentrations measured during temperature inversions in the rather sparsely populated small relief depression are a cause for concern since similar conditions can be expected in numerous hilly and 650 mountainous regions across Europe where approximately 20 % of the total population lives, out of which 30 % live in rural relief depressions, which is akin to the Retje site.
The results of this study highlight the importance of quality high-resolution air quality measurements for monitoring, assessing the effects of actions on the local air pollution, and reducing pollution by residential wood burning, especially in mountainous areas with limited self-cleaning capacities of the atmosphere. Therefore, we suggest the following: 655