Surface aerosol in-situ data from two sites in Colorado was used: Storm Peak Laboratory and the Table Mountain Field Site. Both sites are part of the NOAA Federated Aerosol Network (NFAN), a collaborative network of sites that make long-term measurements of surface in-situ aerosol optical properties worldwide (Andrews et al., 2019). The NFAN sites have similar inlet and aerosol system infrastructure, operational and data processing protocols, and calibrations that follow established Global Atmosphere Watch Programme (GAW) aerosol sampling protocols to ensure consistent high-quality measurements that are directly comparable across sites in the network (Andrews et al., 2019; WMO, 2016). The location of the sites, their proximity to population centers in Colorado, and topographical information is shown in Fig. 1. For the analysis presented herein, the primary time ranges used are 2011–2024 for SPL and 2019–2024 for BOS; when different time ranges are used it is noted.

Aerosol particles at both SPL and BOS are sampled from 10 m above the surface to minimize contamination from local activities. At BOS, when needed, the sample air is gently heated (< 40 °C) to achieve a sample RH ≤ 40 %. Malm et al. (2024) suggests this level of heating would not lead to volatilization of the BOS aerosol. At SPL, bringing the cool outside air into the heated observatory is enough to ensure the sample RH ≤ 40 %. A switched impactor system provides size-segregated measurements using Berner-type multijet cascade impactors for 10 and 1 µm size cuts, which have aerodynamic cut diameters of ∼ 7 and ∼ 0.7 µm under our sampling system (Berner et al., 1979; Hillamo and Kauppinen, 1991). Switching between PM10 and PM1 occurs every 6 min, yielding a 30 min measurement of each size-cut per hour. Optical measurements at both sites are made at 1 min resolution; here hourly averaged data are used. Prior to averaging, all measurements underwent a quality assurance process to remove contaminated values (e.g. spikes from ski area operations at SPL and mowing operations at BOS that are not representative of the wider region) and invalid values (e.g. when instruments failed).

Aerosol scattering (σsp) and backscattering (σbsp) coefficients are measured at both sites using an integrating nephelometer (Model 3563, TSI Inc.). The nephelometer measures aerosol σsp and σbsp coefficients at three wavelengths: 450, 550, and 700 nm. Total scattering coefficient is measured over an angular range of 7–170° and total backscattering coefficient is measured over an angular range of 90–170°. The nephelometer measurements were corrected for angular truncation and other instrument non-idealities using the method described by Anderson and Ogren (1998). The nephelometer data are adjusted to standard temperature and pressure (0 °C, 1 atm). The overall uncertainty in the nephelometer measurements has previously been evaluated to be ∼ 9.2 % under average conditions at continental sites (Sherman et al., 2015). Error was calculated for the instrument and measurement conditions at SPL and BOS and were found to be generally comparable under both extremely clean (∼ 10 % error) and heavy aerosol loading (∼ 9.2 % error) conditions (Table S2 in the Supplement).

At SPL, aerosol absorption coefficients (σap) were measured using both a Particle Soot Absorption Photometer (PSAP; Radiance Research), from 2011–2018, and a Continuous Light Absorption Photometer (CLAP; NOAA; Ogren et al., 2017), from 2011–2024. Both instruments measure absorption at three wavelengths: the PSAP at 467, 530, and 660 nm, and the CLAP at 467, 528, and 652 nm. The two instruments have previously been shown to be comparable measurements (Ogren et al., 2017), and an intercomparison of the PSAP and CLAP data from SPL was done to confirm this (Sect. S1 in the Supplement). For the analysis presented in this work, only the CLAP data from SPL were used. The work done by Japngie-Green et al. (2019) used data from both the PSAP (2011–2013) and the CLAP (2013–2016), but the assessment in Sect. S1 indicates that the CLAP absorption data presented here is comparable to the combination of PSAP and CLAP data presented by Japngie-Green et al. (2019). Only a CLAP was deployed at BOS and those are the data used. Absorption measurements were corrected for scattering and filter artifacts, sample area, flowrate, and non-idealities in the manufacturer's calibration (Bond et al., 1999; Ogren, 2010). Uncertainty in the CLAP σap measurement is ∼ 20 %, though it increases substantially under extremely clean conditions to ∼ 40 % (Sect. S2; Ogren et al., 2017; Sherman et al., 2015). The absorption data are reported at standard temperature and pressure, so no STP correction is needed. In the analysis presented, the wavelengths of all absorption coefficients (σap) were adjusted to match the nephelometer scattering wavelengths (450, 550, and 700 nm) using the measured spectral dependence of absorption (Sect. S2).

Using the measured scattering and absorption coefficients, several variables can be calculated: single scattering albedo (SSA), backscattering fraction (BFR), scattering Ångström exponent (SAE), and absorption Ångström exponent (AAE). The equations for these variables are given below: 1SSA=σsp(σsp+σap),2BFR=σbspσsp,3SAE=-log⁡σsp,λ1σsp,λ2/log⁡λ1λ2,4AAE=-log⁡σap,λ1σap,λ2/log⁡λ1λ2,

In Eqs. (3) and (4), λ represents wavelengths, and the subscripts indicate the specific wavelengths used in the calculation – both SAE and AAE were calculated using the 450 and 700 nm wavelength. These calculated parameters were only considered when constraints were met: σsp > 1 Mm-1 and/or σap > 0.5 Mm-1. These constraints were applied to the parameters that used the σsp and/or σap values in their calculation (e.g. for SAE only the σsp constraint was applied but for SSA both constraints were applied). Below those values the calculated parameters are more heavily influenced by instrument noise and are less reliable. Average fractional uncertainties for SSA, BFR, SAE, and AAE at these σsp and σap constraints are 3.5 %, 8.2 %, 6.3 % and 32.2 %, respectively (Table S2). This filtering predominately affected parameters related to absorption (SSA and AAE). The SPL data were more heavily impacted as the site has lower overall aerosol loadings, resulting in filtering out over 50 % of the potential SSA and AAE values. Despite this filtering, the overall statistical values for the variables were not significantly affected (Table S3). This was also shown by Schmeisser et al. (2017) who applied the same constraints to SPL data for the time period 2012–2013 (their supplemental Table S5).

The measured σsp and σap parameters depend on aerosol loading while the calculated parameters (SSA, BFR, SAE, and AAE) are independent of aerosol loading and instead describe characteristics of the aerosol. For atmospheric aerosol particles SSA typically ranges from 0.3 to 1 and provides information on aerosol composition and reflectivity (Laj et al., 2020). SSA values near 1 are representative of low/non-absorbing “white” aerosol particles, while low SSA values < 0.85 indicate darker, more-absorbing aerosol particles (Bond et al., 2013). BFR is sensitive to the smallest accumulation mode particles (diameter < 0.3 µm; Collaud Coen et al., 2007). If the accumulation mode shifts to larger sizes, BFR will decrease and vice versa. Typically, BFR ranges from 0.08–0.2. SAE also yields information about the size distribution of aerosol particles (Schuster et al., 2006). Low values (SAE < 1) indicate the presence of coarse particles, while high values (≥ 2) correspond to an aerosol dominated by fine-mode aerosol particles. The absorption Ångström exponent can yield information about particle composition (Russell et al., 2010) with values near 1 associated with black carbon and fossil fuel burning, while larger AAE values (≥ 2) are representative of biomass burning, light-absorbing organic carbon, and dust.

Following a similar approach to McClure and Jaffe (2018), quantile regression was used to assess trends in extreme events at SPL (Koenker et al., 2025). This method was developed by Koenker and Bassett (1978), and the software package used herein follows Koenker and D'Orey (1987). Quantile regression (QR) applies asymmetric weighting at different quantiles and regresses the entire dataset at specific quantiles to estimate the rate of change across the entire distribution of data (Koenker, 2005; Koenker and Hallock, 2001). The regressions presented are not for a single quantile or subset of the data, but for the entire dataset at a specific quantile. For example, in the estimation of a 75th quantile regression a regression line is fit through the data so 75 % of the residuals are negative (below the regression line) and then the weighted distances are minimized. This method is appropriate in situations where a dataset has significant outliers that would not be well characterized using symmetric weighting, such as that done in ordinary least squares regressions. This analysis was performed on the σsp and SAE data, to assess changes in both aerosol amount and type. The same filtering used for the calculated optical properties was applied to σsp and SAE before this analysis to minimize effect of noise on the trends in the lowest quantiles. Trends using this method were considered significant when the p-value was < 0.05, corresponding to a 95 % confidence level. All trends are reported with a standard error, calculated using nonlinear interaction decomposition. This method uses the Huber Sandwich estimate to calculate the variance and uses local sparsity of the data around the percentile being fitted, which helps incorporate the uncertainty of the percentile. Bootstrapping was also performed and produced similar error estimates. The seasonal trend of Mm-1yr-1 was converted to %yr-1 using the appropriate percentile of the seasonal values from the first year with σsp data as the baseline: 2011 for the winter, spring, and summer and 2012 for the fall. Similar to other trend analysis, long term observations are needed and breakpoints in the data caused by changes in measurement conditions (e.g. inlet and instrument changes) must be resolved. An effort was made to identify any breakpoints in the SPL data, though nothing significant was found. Trend analysis was not performed on the BOS data as the time series was too short.

Mean annual PM10 σsp and σap at both field sites follow similar seasonal patterns with differences in overall magnitude. This is visible in both the timeseries of average monthly σsp and σap values, shown in Fig. 2, and the annual climatology, shown in Fig. 3. While measurements were made at both PM10 and PM1 size cuts, the discussion here centers on the PM10 measurement. Average sub-micron fractions for σsp were 0.8 ± 0.2 and 0.7 ± 0.1 for SPL and BOS, respectively, and for σap the sub-micron fractions were 0.8 ± 0.2 and 0.8 ± 0.1 for SPL and BOS, respectively (Fig. S2 in the Supplement) indicating that 70 %–80 % of the σsp and σap are due to the PM1 aerosol. Both σsp and σap exhibited pronounced seasonal patterns with periodic low values occurring every winter (December–February) and peaks occurring in the summer into the early fall (June–September). Observed seasonal patterns for SPL's σsp and σap generally agree with those previously reported by Japngie-Green et al. (2019). While similar seasonal patterns were observed at both sites, BOS exhibited higher σsp and σap values at all wavelengths and across all months for both PM10 and PM1 data (Fig. S3). Specifically, at BOS the σsp and σap values were up to a factor of 2 to 3 higher than those measured at SPL, depending on month and size cut. The largest differences between SPL and BOS in both σsp and σap were observed in the winter (Fig. S4). These differences in magnitude are likely due to both the regional urban influences at BOS, and the elevation gradient between the two sites. Evidence for this can be seen in the diurnal variability at the two sites (Fig. S5). SPL's lower σsp and σap values and weaker diurnal cycles (particularly in winter) are largely due to the site's high elevation causing air masses measured at SPL to frequently be decoupled from the polluted boundary layer (Andrews et al., 2011; Collaud Coen et al., 2018). The slight increase in diurnal variability at SPL in the summer is likely due to thermally driven upvalley and upslope flow, which is also consistent with the overall increase in σsp and σap during those months. BOS exhibits a similar diurnal cycle given its location and proximity to complex terrain, though its cycle also shows influence from some anthropogenic emissions throughout the year. This is expected, as BOS resides within the boundary layer where most anthropogenic emissions take place. This also results in the higher overall PM10 σsp and σap values at BOS.

While the seasonal cycle of σsp and σap (high in summer, low in winter) occurred over the entire period of measurement at both sites, there was noticeable interannual variability with respect to the magnitude of the cycles (Fig. 2). This variability was largest in the summer months, likely associated with episodic wildfires. Often the magnitude of change could be visibly tied to wildfire metrics, such as the total acres burned locally and nationally as shown in Fig. 2c. A notable exception to this was in 2021, where the large increase in aerosol load was driven by transported smoke from the Pacific Northwest region of Canada and the United States, which experienced a severe heat wave that lead to a significant increase in wildfires (White et al., 2023), as well as smoke from central Canada (Bruce et al., 2025). At both sites the influence of extreme wildfire events is not difficult to distinguish from baseline air pollution as aerosol loading is significantly higher during smoke events (ranging from a factor of 2 higher to an order of magnitude higher depending on intensity of the smoke). Increases in aerosol loading associated with intense smoke events align well between the two sites (Fig. 2). Most long-range smoke observed at these two sites is transported eastward from western North America and impacts both sites. However, a notable exception occurred in May 2023 during transport of Canadian wildfire smoke into the US. For this event, the aerosol increase was larger at BOS as a result of the smoke being channelled down the Front Range of the Rocky Mountains with less upward and westward transport to SPL (see 2023 BOS σsp and σap spike in Fig. 2).

The influence of outlier events on the temporal cycles in aerosol loading is apparent in the monthly mean values of σsp and σap, as shown in Fig. 3. In the summer months at both sites, monthly mean values of σsp and σap fall at the edge or above the interquartile range indicating a skew towards higher loading for those months. To investigate the influence of extreme events, simple outlier testing was done for the σsp upper bound, and a value of 50 Mm-1 was determined to be appropriate to use for both sites (Table S4). Overall, July–September had the most hours with σsp exceeding 50 Mm-1 and these months dominated most years in terms of number of hours over the 50 Mm-1 threshold (Fig. S6, Table S5). There were notable exceptions at SPL, where June 2011 and May 2023 had the highest number of hours over 50 Mm-1 which could be the result of dust and/or wildfire events. In the SPL data, there is a large variation in the number of outlier periods per year with 2020 and 2021 having the highest rates of these periods. The median number of instances per year has increased since previous analysis (2011–2016; Japngie-Green et al., 2019) from 90 to 110 hyr-1 (97 hyr-1 if the extreme 2020–2021 period is excluded). While this does not indicate a significant trend, it does show that overall, these periods have become more frequent. This change in frequency is not discernible in the shorter BOS time series.

Qualitative information on the general seasonality of aerosol composition and size can be inferred from calculated aerosol optical properties. For readability, the figures below present just the PM10 data, but the seasonal cycles for all variables are similar for the PM1 size cut (Fig. S7). The seasonal patterns for the 2011–2024 SPL calculated aerosol optical properties are consistent with those previously reported for SPL for the 2011–2016 period (Japngie-Green et al., 2019; their Fig. 5) as discussed in more detail below.

PM10 SAE and BFR both decrease in the spring at SPL and BOS, indicating a shift towards larger particles (Schuster et al., 2006; Fig. 4a and d). The decrease in the sub-micron scattering fraction during the spring also shows increased influence from coarse mode particles (Fig. S2c). These changes are likely due to dust sources in the spring. Transported dust from Asian outflow is a prominent springtime dust source at both sites (Augustine et al., 2008; Hallar et al., 2015; Japngie-Green et al., 2019), though surface fine dust from local deserts and semi-arid regions in North America are also an important source of dust (Hallar et al., 2011, 2015). The influence of these dust sources is also confirmed by chemical data from the region presented by Hand et al. (2024a). The relationship between SAE and BFR during summer months is more complicated. SAE increases in all summer months, however, BFR increases in June at both sites and through July at SPL before decreasing for the rest of the summer. Increases in both SAE and BFR during the early summer indicate an increased contribution from smaller particles. As BFR and SAE are sensitive to different parts of the aerosol size distribution, the higher sub-micron contribution (increased SAE) with increased large accumulation mode particles (decreased BFR) later in the summer and fall could indicate a narrowing of the size distribution (Collaud Coen et al., 2007). Alternatively, the relationship could also indicate changes in the number or relative importance of different aerosol size modes if the particle size distribution is multi-modal (Schuster et al., 2006).

Aerosol SSA tends to be lower in the fall and winter months and higher in the spring and summer months. The high SSA values in July and August are consistent with aged smoke aerosol where gases associated with the smoke have had the opportunity to condense resulting in a larger, more reflective aerosol (Andrews et al., 2004; Bian et al., 2020; Yokelson et al., 2009). Both sites exhibit a local SSA minimum in June – though the minimum is more pronounced at BOS – indicating aerosol particles are slightly more absorbing during the early summer.

The aerosol absorption Ångström exponent is composition dependent, with different aerosol types having distinct ranges. Theoretically, black carbon (BC) aerosol has an AAE value of 1 while dust aerosol generally has an AAE value ≈ 2 (Bergstrom et al., 2007). Wildfire smoke plumes are typically expected to have AAE values ≥ 2 depending on fuel types and transport time (Forrister et al., 2015; Selimovic et al., 2018). At both sites, the PM10 AAE is highest in the spring and fall, with a seasonal low in the summer (Figs. 4b and S8). The average spring and fall AAE values range from 1.4–1.6 (Table 1), which aligns with literature value ranges for dust and large particle/BC mixed aerosol (Lee et al., 2012; Schmeisser et al., 2017). The average AAE summer values range from 1.2–1.3 (Table 1), which is consistent with a mixture of fossil fuel, biomass burning, and BC dominated aerosol particles (Schmeisser et al., 2017; their Tables 1 and 4).

The impact of intense smoke plumes in the summer is not clear in the AAE monthly climatology depicted in Fig. 4b (i.e., AAE is not > 2). This is likely due to the mixing of isolated biomass burning plumes with background air and other emissions. Clarke et al. (2007) were able to sample pollution and biomass burning plumes during flights over North America and reported the distribution of AAE for these plumes along with regional background (their Fig. 6). The distributions for all summer SPL and BOS hourly AAE data (Fig. S9) were most similar to the distribution reported for regional background by Clarke et al. (2007) but were broader indicating that multiple indistinguishable sources could be present. Summer distributions showed more frequent high values of AAE than in the spring. Looking at AAE in relation to month and σsp, AAE was high (> 2) during high aerosol loading in the summertime which is consistent with the presence of smoke plumes (Fig. S10). It can also be seen that in the springtime there were both dust (AAE ≥ 2, for high σsp) and pollution (AAE ≤ 1, for high σsp) plumes, with pollution having a higher impact at BOS than dust during that season (Fig. S10).

While the seasonality of AAE at SPL is similar to that reported in Japngie-Green et al. (2019), median AAE values were higher from April to November relative to the period they considered (Fig. 4b). The AAE shift in the summer and fall is clearly attributable to the increase in extreme events with high AAE values (Fig. S10), however, the AAE increase in springtime could be the result of either an increase in dust events, overall decrease in background aerosol, or some combination of the two. For example, in their analysis of data from 2000–2016, Hand et al. (2019) observed significant decreases in PM2.5 leading to an increase in the relative contribution of coarse aerosol mass over the continental US but also found that mean coarse mass in some regions did increase significantly over time. Without quantitative chemical data it is difficult to discern which process is driving changes in the optical properties at SPO and BOS. Additionally, while Japngie-Green et al. (2019) show SPL PM1 AAE values being consistently lower than those of PM10, this study observed PM1 AAE values that were equal to or higher than those calculated for PM10 from August to November (Figs. S7 and S8). This is likely in response to the overall decrease in aerosol loading at SPL (Sect. 3.3), and a shift in the sub-micron composition towards larger, less absorbing particles. Support for this can also be seen in the overall decrease in sub-micron σsp and σap fractions compared to the 2011–2016 period, indicating an increase in the impact of larger particles (Fig. S2c and f). This kind of shift in size has been previously observed at other sites within the continental US (Sherman et al., 2015). Additionally, PM1 SSA at SPL increased from August to November (Fig. S8) relative to the 2011–2016 time period, indicating that aerosol particles have become less absorbing overall. SSA was lowest during the late fall and winter (October–February), which is attributable to increased anthropogenic emissions during colder months. Both AAE and SSA values were generally lower at BOS than at SPL, as a result of the geographical proximity of urban sources to BOS described above.

Long-term seasonal analysis of PM10 σsp at SPL shows that overall aerosol loading is decreasing, with the majority of quantiles showing negative trends in all seasons except the fall (Fig. 7a). This confirms the observed changes discussed in previous sections and is generally consistent with other trend analyses for aerosol scattering coefficients (Collaud Coen et al., 2020) and aerosol extinction coefficients (Hand et al., 2020) in the US. Differences were observed in the trends between seasons and over different quantiles (Figs. 7 and S16). In all seasons, PM10 σsp values in and below the 70th quantile had significant but small observable changes (less than ± 2 %yr-1, Table S7). At the higher percentiles, statistically significant trends were more pronounced indicating that changes in PM10 σsp are being driven by extreme events (Fig. 7a). For the seasonal analysis of SAE, significant trends were generally negative and larger in the lower quantiles for all seasons except the summer (Fig. 7b). A negative trend in SAE indicates an increased contribution from coarse mode aerosol.

Trends in the winter and spring extreme PM10 σsp events were negative, with average significant trends of -2.4 ± 0.2 and -1.7 ± 0.4 %yr-1 in and above the 80th quantile. This could be the result of a decrease in the intensity of dust events, or a decrease in background aerosol overall given that the winter and spring exhibited similar trends in and below the 70th quantile (average significant trends of -1.7 ± 0.3 and -1.6 ± 0.3 %yr-1). In these seasons, SAE trends above the 80th quantile were minimal and not significant (Fig. 7b, Table S8). In contrast, the SAE in the winter and spring had average significant negative trends in SAE of -1 ± 1 and -0.5 ± 0.2 %yr-1 in and below the 70th quantile that were significant indicating an increase in coarse mode aerosol.

The summer and fall upper quantiles exhibited the largest trends in σsp, corresponding to significant increases in aerosol loading during extreme events (i.e. wildfire smoke). This was also observed by McClure and Jaffe (2018), who reported a positive trend in the 98th quantile particulate matter (PM2.5) concentrations of 1.77 ± 0.68 %yr-1 for the northwest region of the United States. They attribute the increase to wildfire activity based on similar trends in total carbon and AOD observed in their study. Here, a positive trend in PM10 σsp of 10 ± 1 %yr along with a trend in SAE of 1.4 ± 0.2 %yr-1 at the 98th percentile was observed in the summertime at SPL indicating increased loadings of smaller particles attributable to wildfire activity. The fall had a significant trend in the σsp 96th quantile of 2.4 ± 0.4 %yr, but no significant trends in the upper quantiles (> 90th) of SAE to indicate changes in aerosol size (Table S7). The positive fall trend for σsp may be due to wildfire season extending further into the autumn months (Goss et al., 2020). This is consistent with positive trends in organic mass concentration in Colorado that have been previously attributed to wildfires (Hand et al., 2024b).

The summer and fall trend attributable to smoke is also apparent in the fire data for Colorado. Previous source analysis of SPL data (Hallar et al., 2015; Japngie-Green et al., 2019) focused on measurements collected prior to 2016. At that time, the 2012 wildfire season was regarded as one of the most severe seasons impacting Colorado, marked by extreme dry winter conditions, hot summer temperatures and intense wildfire activity (Val Martin et al., 2013). Prior to 2017, the 2012 season also had two of the ten largest wildfires in Colorado history. Since then, there have been significant increases in the size and number of fires in Colorado according to metrics reported by the Rocky Mountain Area Coordination Center (RMACC, Figs. 2c and S17). As of 2024, the 2020 fire season is now regarded as the most severe with the three largest fires in Colorado history occurring during that year. Additionally, much of the smoke impacting Colorado comes from outside of the state and the wider western US has also experienced exponential growth in wildfire frequency and area burned (Burke et al., 2021; Weber and Yadav, 2020).