Figure 1 compares the normalized organic aerosol mass spectra (MS) obtained from burning the same batch of biomass briquettes and smokeless coal in two different stoves – a conventional stove and an Ecodesign stove. Although sampled from different stoves, the mass spectral signatures as a whole were similar with an uncentred R2 (i.e., ∼ R2) of 0.87 for biomass briquettes and ∼ R2 of 0.97 for smokeless coal. All MS profiles appear to be dominated by fragments of CnH2n+1 (m/z 29, 43, 57, 71, etc.) and CnH2n-1 (m/z 27, 41, 55, 69, etc.), indicating a large contribution from saturated alkanes, alkenes, and/or cycloalkanes. However, the normalized peak intensities at specific m/z (e.g., m/z 41) sometimes varied significantly for the same type of solid fuel in the different stoves. The differences in the MS (discussed in Sect. 3.1.3) could be due to the different burning conditions (e.g., air supply and temperature) employed by the stoves, resulting in different thermal decomposition processes of the solid fuel and the corresponding pyrolysis products (Andreae, 2019; Weimer et al., 2008).

Figure 2 compares the normalized organic aerosol mass spectra (MS) obtained from burning different batches of wood, peat, and smoky coal in a conventional stove (from this study) and a boiler stove (from Lin et al., 2017). The wood-burning OA produced in the two stoves shows the largest variation with ∼ R2 = 0.78, followed by smoky coal (∼ R2 = 0.88) and peat (∼ R2 = 0.95). The large variation in the MS of wood burning was likely associated with the high volatile content (80.8 wt %) in wood (Trubetskaya et al., 2021), which can be sensitive to the burning conditions. Although these solid fuels were purchased from different locations at different times (Dublin, 2019, and Tipperary, 2016) the general signatures were similar for each fuel type and displayed the expected marker ions. The key marker ion in wood-burning OA appears at m/z 60 and m/z 73 (Alfarra et al., 2007). The mass fragment at m/z 60 (mostly from the C2H4O2+) is due to the fragmentation of anhydrosugars (e.g., levoglucosan, mannosan, and galactosan from the combustion of cellulose/hemicellulose; Lee et al., 2010) in the ACSM, and it is, therefore, commonly used as a marker for biomass burning in the AMS/ACSM studies (Cubison et al., 2011; Lee et al., 2010). Therefore, the differences in the content of cellulose/hemicellulose in the test fuels partly contribute to the differences in ion intensity at m/z 60 in ACSM. Specifically, the MS of wood-burning OA has a prominent contribution from m/z 60 (i.e., f60 > 2.9 %; f60 denotes the fraction of m/z 60 in the total organic signal), while the MS of coal burning has a very low contribution from m/z 60 (f60 < 0.1 %). In contrast, f60 in the MS of peat burning was in between coal and wood (1.6 %–1.7 %). This finding is consistent with the cellulose content in each of the solid fuels – wood > peat ≫ coal. While marker ions are important in the identification of specific OA factors during ambient studies, the differences in the intensities at specific m/z are an important source of uncertainty when used as inputs for ME-2 modelling (Canonaco et al., 2013; Canonaco et al., 2021). Therefore, examining variations in the intensities of specific m/z values due to the use of different stoves has great implications for factor analysis of an ambient dataset.

The MS obtained using different stoves are compared by plotting the relative differences in individual m/z values (calculated by (fm/z,stovey-fm/z,stovex)/fm/z,stovex, where fm/z represents the fraction of the measured m/z to the total organic signal, while stove y represents the Ecodesign or the boiler stove, and stove x represents the conventional stove; Figs. S3 and S4 in the Supplement). For wood burning in the conventional and boiler stoves, large differences (0.84 or 84 %) were associated with the intensity of the marker ion m/z 60 (Fig. S3). In addition to m/z 60, other fragments also showed large variations (Fig. S3). For example, the relative difference was 0.5 (or 50 %) for m/z 44, a marker ion for aged or more oxidized OA (Canonaco et al., 2015). In ambient studies, the triangle space between m/z 44 and m/z 60 is often used to study the ageing of biomass burning, in which a decreasing f60 and an increasing f44 are usually associated with the atmospheric ageing process (Ng et al., 2011a). However, the results from our study suggest that variations in f60 and f44 could also be due to different burning conditions (i.e., in different stoves) and do not necessarily correspond to primary OA ageing or atmospheric processing. For peat-burning OA, there is a lower f60 (0.016–0.017) than that for wood (0.029–0.053) due to the lower content of cellulose (Brown et al., 1988; Mikucioniene et al., 2019). Compared to wood, the peat MS appeared to be less affected by the stove type with a difference ratio of -0.06 (or 6 %) for m/z 60 (Fig. 2). But for other fragments (e.g., m/z 29, 41, 43, etc.), a difference ratio of up to 0.46 is also indicative of significant variations caused by the type of stove.

f60 in the MS of biomass briquettes was 0.005–0.008, which was 4–10 times lower than wood and 2–3 times lower than peat. This can be explained by the manufacturing process for the biomass briquettes, which involves torrefaction at a temperature of > 250 ∘C that causes thermal decomposition of the raw biomass. The briquettes thus contain less cellulose and produce a lower f60 as a result. The difference ratios for the MS of biomass briquettes burnt in the conventional and Ecodesign stoves were in the range of -0.4 to 0.6 for the major fragments (e.g., m/z 41, 43, 55, 57). For some minor fragments (e.g., m/z 71 and 85), the difference ratios were even higher with values of up to 1.4 (i.e., 140 %; Fig. S4), again suggesting the large impact of burning conditions on the MS profiles.

For the MS of smoky coal, f60 was reduced to 0.00071–0.00081 (< 0.1 %) while for smokeless coal f60 was 0.0027–0.0045, both of which were lower than that for wood/peat. The reduced f60 in the normalized mass spectra for smoky/smokeless coal is likely due to the breakdown of, for example, cellulose during coal formation over millions of years (Höök, 2012), resulting in a relatively low content of cellulose, while accumulating other carbon-rich content, leading to the observed ions at other m/z's. As a comparison, the large contribution from the fragments at m/z 77, 91, and 115 suggests a high content of aromatic/polycyclic aromatic hydrocarbon (PAH) compounds in the smoky coal burning emissions. Specifically, f77, f91, and f115 were in the range of 0.015–0.016, 0.014–0.015, and 0.019–0.026, respectively, for the MS of smoky coal. However, for the MS of smokeless coal, f77, f91, and f115 were lower with values of less than 0.015. Compared to smoky coal, the lower levels of aromatic/PAH-related fragments (i.e., m/z 77, 91, and 115) in the MS of smokeless coal are associated with its production process, which removed most of the volatiles in the raw coal during torrefaction at high temperatures (Trubetskaya et al., 2021), resulting in lower emission factors.

Figure 3 shows the time series of PM2.5, PM1 components, and OA factors (discussed in Sect. 3.2.2) in suburban Dublin from 1 November 2016 to 31 January 2017. During the sampling period, PM1 (sum of ACSM and BC measurements) showed large variations with 30 min averaged concentrations ranging from < 0.5 to 302.0 µgm-3. In particular, PM1 concentrations of 25 µgm-3 were often (roughly 1 in 3 d) exceeded during the sampling period. The PM1 at the sampling site showed a strong correlation with PM2.5 at the Rathmines station (R2 of 0.87, a slope of 0.74; Fig. S1), thus confirming that the ACSM sampling site was representative of the residential areas in southern Dublin (Fig. 3a). The slope of 0.74 for the linear relationship between PM1 and PM2.5 suggests that PM1 on average accounts for 74 % of PM2.5. However, during pollution events, the values of PM1 and PM2.5 are very similar, indicating that most PM is in the submicron size ranges.

Consistent with the overall trend of PM1, all the measured PM1 components showed similar temporal variation with enhanced concentrations during the evening. The diurnal cycle of OA showed an increase from 16:00 LT (local time; all times in this paper are given in local time unless stated otherwise) which peaked during 20:00–22:00 (Fig. 4). However, OA decreased sharply overnight and remained at a low concentration during the day (08:00–16:00). Similarly, BC concentrations were over 4 times higher in the evening than during the day. The very similar diurnal patterns of OA and BC suggest that they have common emission sources (i.e., heating) during the evening and night hours. Meteorological parameters like temperature and wind speed were also partly contributing to the elevated concentrations in the evening. The scatter plot (Fig. S5 in the Supplement) between OA and temperature or wind speed suggests that the high OA concentrations were coupled with low temperatures (< 7 ∘C) and low wind speed (< 5 ms-1). Also, the shallower planetary boundary layer was an important factor for the increased OA concentration in the evening (Lin et al., 2018). In addition, the diurnal pattern of ammonium, nitrate, and sulfate all showed peaks at the same evening hours with the BC (Fig. 4a), suggesting that they may also be related to heating emissions coupled with low temperatures in the evening (Fig. S6 in the Supplement). Sulfate was likely associated with the primary emissions from solid-fuel combustion given that, in the stove emission, sulfate was found to contribute to < 1 % of PM1 for wood burning but the fraction of sulfate was up to 21 % of PM1 for smokeless coal burning, reflecting the higher content of sulfur in the raw fuel (Trubetskaya et al., 2021). This is consistent with our previous study (Lin et al., 2019), where we demonstrated that sulfate, nitrate, and ammonium can be locally emitted or formed, as well as regionally transported, through the comparison of the ACSM measurement at the same Dublin sampling site and at Carnsore Point, a regional background site.

To evaluate the contributions of solid-fuel burning to the ambient OA in suburban Dublin, the MS of wood, peat, and smoky coal were used as the anchoring profiles for ME-2 modelling (Lin et al., 2017; Lin et al., 2018; Trubetskaya et al., 2021). In the ME-2 analysis, the individual m/z values in the MS for wood, peat, and coal were allowed to vary between the reference profiles (i.e., the limits approach; see Sect. 2; Figs. S7–S10). The time series of solid-fuel-burning factors (Fig. 3c) were very similar (i.e., all peaked during the same evening hours; Fig. S8) due to the similar emission time from the domestic heating activities. This is the reason why ME-2 was used to separate these factors since unconstrained PMF led to highly mixed and not physically meaningful factors (Lin et al., 2017). In addition to solid fuels, a hydrocarbon-like OA (HOA) factor and an oxygenated OA (OOA) factor were also resolved (Fig. S7). HOA was associated with the emissions from oil heating during the evening, while OOA was related to regional transport and/or secondary processes (Lin et al., 2020). Increasing the number of factors during the ME-2 analysis (i.e., the six-factor solution; Fig. S9) identified an additional OOA factor (OOA2) which, however, featured a very low signal at m/z 41 in the normalized mass spectra but a similar signal level at m/z 43 with the already identified OOA factor (Fig. S7). The unambiguous separation of two OOA types requires further research. Nevertheless, for the five-factor solution, the good correlation (R2 = 1, slope = 0.99; Fig. S10) between the time series of the explained fraction and the PMF input suggests that the five-factor solution explained the input matrix well.

Figure 4b shows the diurnal cycle of the averaged contribution of the resolved factors over the entire period. On average, solid-fuel burning (the sum of peat, wood, and coal) was the major contributor (> 50 %) to the total OA during the evening, while during the day OOA was the dominant factor. Therefore, primary emissions from solid-fuel burning were the dominant sources of pollution in the evening, while regional transport and/or secondary processes of OA were the major source during the day. The oil heating factor was contributing, on average, 22 %–25 % of the total OA in the evening. Even though the overall results from the limits approach were consistent with those from the conventional a-value approach in Lin et al. (2018), a detailed comparison of results between the two approaches as well as the corresponding uncertainties are provided below.

Figure 5 shows the comparison of the time series of OA factors resolved by the limits approach and the conventional a-value approach (Lin et al., 2018), while Table 2 shows the corresponding uncertainties. The statistical uncertainty of the limits approach was evaluated through the bootstrap-based resampling strategy (See Sect. 2), while the model uncertainties for the a-value approach was the variation (1 standard deviation) of the accepted ME-2 solutions with the combination of different a values (Lin et al., 2018). The model uncertainty for the limits approach was in the range of 2 %–7 %, considerably lower than 8 %–29 % for the a-value approach. The low uncertainty for the limits approach suggested that the solution was relatively stable. In contrast, the relatively large uncertainty for the a-value approach suggested that the degree of variation from the anchoring profiles could cause uncertainties in the solution of up to 29 %. In addition to the model uncertainties, the dilution and cooling of the aerosol samples after mixing in the ambient atmosphere, as well as atmospheric processes (e.g., night-time chemistry with NO3 radical; Kiendler-Scharr et al., 2016) are also important sources of uncertainties in OA source apportionment since these factors could also cause variation in the mass spectra. In this study, our combustion experiment deployed a Dekati diluter (See Sect. 2) to simulate the dilution and cooling of the raw flue gas samples through mixing with compressed clean air.

Overall, the time series of the OA factors for peat, HOA, and OOA were well correlated with R2 > 0.95 and slopes in the range of 0.95–1.10 (Fig. 5), suggesting excellent agreement between the two approaches despite the difference in uncertainties (Table 2). Although the wood-burning factor time series from the two approaches had a high correlation coefficient (R2 of 0.99), a slope of 1.33 indicated the quantification of the OA factor of wood burning varied to a larger extent (i.e., 33 %). In contrast, the OA factor for coal burning showed the poorest correlation with an R2 of 0.37 and a slope of 0.44. The poor correlation for the coal burning factor was likely due to the low contribution to the total OA (< 10 %; Fig. 5c) and the large uncertainty from the a-value approach (29 %; Table 2). Nevertheless, both approaches pointed to an important source of solid-fuel burning in wintertime Dublin, with the sum of peat, wood, and coal factors, on average, contributing over 50 % of the total OA during the evening hours (20:00–23:00). Specifically, both approaches showed peat burning being the largest OA factor (39 %, Fig. 5f vs. 41 %, Fig. 5g), followed by HOA (24 % vs. 25 %), OOA (20 % vs. 18 %), wood (14 % vs. 11 %), and coal (4 % vs. 5 %) during the evening hours (20:00–23:00) when stove emissions dominate. Therefore, high variations in specific m/z contributions to OA mass spectra from different fuel and stove types do not translate into high source apportionment uncertainties owing to the robust ME-2 approach. Moreover, the time series of OOA showed spikes concurrent with primary factors (Fig. S8) during the evening and night-time, suggesting that OOA was probably associated with the condensation of semi-volatile species and/or ageing of primary emissions in the real atmosphere (Tiitta et al., 2016). As a result, the contribution from solid-fuel burning could be higher than solely represented by the POA fraction given that OOA, on average, accounted for approximately 20 % of the OA in the evening (Fig. 4b).