Assessment of the aerosol optical depths measured by satellite-1 based passive remote sensors in the Alberta oil sands region 2

Several satellite aerosol optical depth (AOD) products are assessed in terms of their data quality in the 7 Alberta oil sands region. The instruments consist of MODIS (Moderate resolution Imaging Spectroradiometer), 8 POLDER (Polarization and Directionality of Earth Reflectances), MISR (Multi-angle Imaging SpectroRadiometer), 9 and AATSR (Advanced Along-Track Scanning Radiometer). The AOD data products are examined in terms of 10 multiplicative and additive biases determined using local AERONET (AEROCAN) stations. Correlation with 11 ground-based data is used to assess whether the satellite-based AODs capture day-to-day, month-to-month, and 12 spatial variability. The ability of the satellite AOD products to capture interannual variability is assessed at Albian 13 Mine and Shell Muskeg River, two neighbouring sites in the northern mining region where a statistically significant 14 positive trend (2002-2015) in PM2.5 mass density exists. An increasing trend of similar amplitude (~5%/year) is 15 observed in this northern mining region using some of the satellite AOD products. 16


Introduction
Fine-mode aerosols can be harmful to the respiratory system in large doses and are thus a critically important constituent with regard to air quality.For this reason, particulate matter with median aerodynamic diameter less than 2.5 m (PM 2.5 ) is one of the atmospheric observables used to calculate the Air Quality Health Index (AQHI) in Canada (Stieb et al., 2008).Similar indices are used in other countries (Kelly et al., 2012).Tropospheric aerosols are also a major source of uncertainty in estimating the radiative forcing of climate (Myhre et al., 2013).Many satellitebased instruments can provide information about atmospheric aerosols in the form of aerosol optical depth (AOD), a measure of the vertically integrated extinction of the solar beam by aerosols.Measurements of AOD tend to be proportional to particulate matter mass density measured at the surface when the boundary layer aerosol concentrations are elevated (e.g.Tian and Chen, 2010).
The Alberta oil sands region (AOSR) has been under rapid industrial development during the past decade (Foote, 2012).Satellite measurements already indicate a significant increasing trend in nitrogen dioxide between 2005and 2014(McLinden et al., 2012;;McLinden et al., 2016).Additionally, the AOSR is being deforested as part of expanding surface mining operations.This inevitably increases levels of dust, which partly arises from Atmos.Chem.Phys. Discuss., doi:10.5194/acp-2016-748, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 12 October 2016 c Author(s) 2016.CC-BY 3.0 License.
by Diner et al. (2008).The aerosol retrieval for AATSR is described by Bevan et al. (2012) and references therein.Deuzé et al. (2001) detail the approach used to retrieve aerosol information from POLDER observations over land.MODIS Terra is not considered since it is highly similar to MODIS Aqua but, for collection 6, the former is less reliable for trend studies in spite of improvements relative to collection 5 (Levy et al., 2015).The MODIS-based Multiangle Implementation of Atmospheric Correction (MAIAC) (Lyapustin et al., 2011) product is not currently available in the AOSR (van Donkelaar et al., 2016).VIIRS (Visible Infrared Imaging Radiometer Suite) (Hillger et al., 2013) is not considered in this study because of its shorter data record relative to the MODIS sensors.Active remote sensing instruments are not considered because of the long revisit time and poor spatial coverage of the relatively small AOSR.
For validation of satellite-based AOD data, AERONET (Holben et al., 1998) is the ideal choice since the same quantity is measured by this ground-based network of direct-sun multiband photometers and the ~3 minute typical sampling interval generally ensures a good temporal coincidence during clear sky conditions.Quality-controlled AERONET data (Level 2, version 2) are used (http://aeronet.gsfc.nasa.gov).CIMEL (French manufacturer) CE318 sensors used by AERONET measure at several wavelength, some of them (e.g.500 and 870 nm) are close to the wavelengths at which the selected satellite instruments report AOD (e.g.470, ~550, and 865 nm).There are two AERONET sites in the oil sands region: Fort McMurray (56.752°N,111.476°W)August 2013 meaning that there is no temporal overlap with Advanced ATSR (AATSR) and only seven coincidences with POLDER/PARASOL using coincidence criteria of ±12 minutes and 10 km.The spatial coincidence criterion corresponds to the smallest AOD footprints of the selected data sets (Table 1).A larger spatial coincidence criterion is not considered since, as shown below, strong spatial gradients in AOD exist in this aerosol source region.Furthermore, as mentioned in Sect. 1, the surface type also changes on such spatial scales.The temporal coincidence criterion was set to limit the number of independent AERONET measurements used in the statistical analysis.There can be multiple AERONET observations that are temporally coincident with a satellite observation and there can be up to four spatial coincident satellite AODs during a satellite overpass of an AERONET site.All of these coincidences are treated as independent data points in the validation and correlation analyses.In order to properly validate satellite AOD bias, AERONET 500 nm AODs are interpolated to the satellite AOD wavelengths (see Table 1) using the coincident AERONET Ångström exponent derived from 440 and 675 nm measurements, except for POLDER/PARASOL, for which no scaling of the AERONET was applied.
The ability of each satellite-based sensor to capture the AOD seasonality in snow-free months is determined at Fort McMurray using the correlation of monthly averaged AODs (using all overlapping years) with AERONET.A minimum of 20 coincident data points per calendar month must be available for that month to be included in the correlation.
In order to assess the ability of the satellite data to capture the spatial variability in this region, spatial correlation is determined for hourly in-situ surface-level PM2.5 from the 10 NAPS (National Air Pollution Surveillance) stations (Table 2) and satellite AODs averaged over all coincidences within their temporal overlap period.NAPS stations Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-748, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 12 October 2016 c Author(s) 2016.CC-BY 3.0 License.
continuously monitor PM 2.5 mass density using tapered element oscillating microbalances (TEOMs).The NAPS network is reviewed by Demerjian (2000), although recently there has been a gradual shift in technology since 2011 to a SHARP (Synchronized Hybrid Ambient Real-time Particulate) monitoring system, which is a hybrid of a nephelometer and a beta attenuation monitor (Hsu et al., 2016).The same 10 km spatial coincidence criterion is used but temporal coincidence limit is extended to ±1 hour to match the temporal resolution of the selected NAPS datasets.
Similar to the spatial and seasonal variability, the ability of the satellite instruments to capture interannual variability can be assessed by correlating yearly satellite-based AODs averaged over all coincidences with NAPS PM 2.5 measurements over the overlap period.20 coincidences in a calendar year are required for the year to be included in the correlation calculation.As an example, for MISR, 14 sufficiently sampled years (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015) are used in the correlation with NAPS data at Millennium mine.
For temporal trends in AOD, a simple linear regression is performed on annual averages and medians.Similarly, for PM 2.5 , the annual average of daily average values are used since the PM 2.5 auto-correlation timescale is on the order of 6.5 hours, based on analysis of Albian mine PM 2.5 data from 2002.The extra step of daily averaging prior to annual averaging yields more conservative annual standard error (s.e.) estimates.Partial years at the start and the end of a data record are removed.Trend periods are given below for each sensor.The area over which the satellitebased AOD trend maps are calculated is 0.1°×0.1°by default.This default setting is used to determine the AOD trend for both MODIS/Aqua 10 km products (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015).The trend domain considered in this work spans from 56-58°N and 111-112°W.For sensors with poorer spatial coverage (MISR, AATSR, POLDER/PARASOL), the spatial binning is expanded in latitudinal and longitudinal increments of 0.1° until there are ≥20 observations in each calendar year within at least one grid cell in the domain.The trend maps are ultimately generated at 0.3°×0.3°for AATSR (2003( -2011( ) and MISR (2000( -2015) ) whereas a 0.4°×0.4°area is required for POLDER/PARASOL (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013).Outlying individual data points (>4 standard deviations above the climatological average in the domain) are recursively filtered mainly to reduce the influence of forest fires on trends.The same filtering is applied to the PM 2.5 datasets.Interannual consistency in the month-to-month sampling is checked for any location with a positive satellite AOD trend significant at the 95% confidence interval by calculating the average day-of-the-year for each calendar year.Such temporal sampling anomalies occur for MISR AOD data at some locations if a 0.1°×0.1°grid were used, for example.The Albian mine (2001-2008) and Shell Muskeg River (2009-2015) forest-fire-filtered PM 2.5 datasets were merged for trend analysis since the sensor was relocated from the former to the latter site in January 2009 and these sites are separated by <5 km.

Results
First, the general spatial distribution of AOD is illustrated for some of the aforementioned data sets.In Fig. 1, the climatological average POLDER AOD on a 0.1° × 0.1° grid is shown.This is the default grid used for climatological maps of all satellite AOD datasets.The POLDER sample size per grid cell is 90 to 170 in the AOSR over the discontinuous period from 1996 to 2013 (see Table 1).There is a clear hotspot in 865 nm AOD in the Atmos.Chem. Phys. Discuss., doi:10.5194/acp-2016-748, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 12 October 2016 c Author(s) 2016.CC-BY 3.0 License.
AOSR region, roughly double the surrounding background values.Note that for POLDER and MISR, there are expected voids in their spatial coverage (Fig. 1) due to the spatial sampling of these instruments, whereas MODIS and AATSR footprints can be centered on any geolocation within the AOSR.
The AOD hotspot in the AOSR seen by POLDER is less obvious with MISR (Fig. 1).The ability to capture spatial variability with MISR is generally much worse than the other instruments based on spatial correlations of average satellite-based AOD versus average NAPS PM 2.5 mass density over the ~10 available sites (Table 3).Table 4 provides the number of coincidences for each satellite with the both Fort McMurray and Fort McKay AERONET observations to provide a relative sense of the sample sizes.
The climatological AOD maps for the MODIS/Aqua collection 6 DT andDB products (2002-2014) are also shown in Fig. 1 however there is a major issue with the confidence as shown in Fig. 2. Near the Syncrude facility at Mildred Lake (57.05°N, 111.6°W), the confidence approaches 0 in both MODIS products in the two adjacent 0.1°×0.1°cells (Fig. 2).In the western cell, the inadequate confidence in MODIS Aqua collection 6 DT data is due to failure of the AOD retrieval algorithm due to the 2.1 m reflectance exceeding the allowed upper limit of 0.35.This is a fundamental weakness of the Dark Target retrieval strategy (see sect. 2).In the adjacent eastern cell, the low confidence stems from the low number of 0.50.5 km 2 pixels (see Table 1) used in the AOD retrieval.The number of pixels used in the AOD retrieval is reduced by high 2.1 m reflectance (>0.35), but also by cloud masking and an independent test for optically thicker cirrus, diagnosed using the 1.38 µm channel (Levy et al., 2013;Hubanks, 2015).The high reflectance in the near-infrared affecting the western cell and possibly the eastern cell is typical of desert or sandy loam.The higher spatial resolution of the MODIS-Aqua 3 km DT data clarifies the importance of this issue: key areas in the AOSR are simply not monitored with confidence by the current MODIS/Aqua DT product.For example, there are 0.01° × 0.01° areas with no AOD measurements of the highest confidence in 12 years, whereas surrounding, equal areas have tens of observations.The lack of confidence is not unique to the AOSR.Low confidence is also observed in urban areas within the province (e.g.Calgary, not shown).
The low confidence in the MODIS DB product is due to the spatial heterogeneity of the surface between vegetated and non-vegetated area, which leads to pixels falsely identified as cloudy (N.Christina Hsu, NASA, priv.communication).Li et al. (2009) identified the need for improved AOD measurements using the DB algorithm over transitional land covers.
A similar issue exists for AATSR (Fig. 3) and ATSR-2 (not shown), which both have an exceedingly small number of successful retrievals in a 0.1° × 0.1° area containing the Mildred Lake Syncrude facility (e.g.N<10) during their respective missions (Table 1).Similarly to MODIS, this is probably caused by falsely identifying bright patches in otherwise vegetated scenes as clouds (P.North, Swansea University, priv.communication).Cloud fraction for successful AOD retrievals tends to be as high as 0.18 within the oil sands region, including the northern mining region, yet drops to 0.02 in the surrounding region (Fig. 3).Note that cloudy 1  1 km 2 pixels are not used during the AATSR AOD retrieval.The spatial correlation coefficient between sample size and cloud fraction as illustrated in Neither POLDER nor MISR show a sampling void in the AOSR.Table 1 shows that these two sensor types have coarser AOD spatial resolution by a factor of 3-4 than MODIS, ATSR-2, and AATSR.Note that some of the PM 2.5 sites are located in the periphery of the industrial and mining areas and thus spatial coincidences exist for MODIS and AATSR in spite of the aforementioned issues, given the 10 km coincidence criterion.
In terms of the validation using AERONET data (Table 4), MISR has a large multiplicative bias (i.e.small slope), which is consistent between both sites in the AOSR.Excluding Fort McMurray coincidences for which the AERONET AODs interpolated to 558 nm are >0.4,the slope improves to 0.74 and is of a similar value to the slope found in previous studies for inland (Liu et al., 2004), dusty (Kahn et al., 2005), and urban environments (Jiang et al., 2007).MODIS DB tends to yield more data than the DT product, but the correlation is lower with AERONET on individual coincidences and in terms of the seasonal variation.At both AERONET sites, the MODIS products behave oppositely in terms of multiplicative and additive biases (discussed in Sect.4).AATSR and POLDER/PARASOL show no major deficiencies, with the latter exhibiting the closest slope value to unity of all of the satellite sensors at Fort McMurray.

Trends
Before considering trends in the AOSR, it is useful to look at whether the different satellite data products capture the (not shown), where the largest and most significant MODIS DB AOD trend occurs and also within the larger area of the spatially coherent POLDER/PARASOL AOD trend.
In order to quantitatively compare trends in AOD and PM 2.5 , the ratio of the average AOD to average PM 2.5 mass density over all coincidences between each satellite instrument and a given NAPS site is used to convert the AOD trends from the satellite instruments to PM 2.5 trends.This implicitly assumes that the ratio of PM 2.5 to AOD is constant over time.This ratio is determined for the merged Albian mine / Shell Muskeg River dataset.Since aerosol optical depth histograms indicate a skewed distribution, it is also useful to verify trends using annual medians.For that purpose, the ratio of median AOD to median PM 2.5 is used instead.This approach is particularly important for POLDER/PARASOL because of the very low 865 nm AODs (Fig. 1) and the negative offset (Table 4) that do not allow a relative trend to be meaningful.
A significant positive trend of 0.24±0.06(±1 standard error) (Figs.5-6) and 0.24±0.07µg/m 3 /year is detected in the Albian mine/ Shell Muskeg River merged annual average and median PM 2.5 mass densities (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), respectively.Limiting the merged PM 2.5 dataset to the warm season (April-October) to mimic the temporal coverage of the satellite data (Table 4), the trend (0.25±0.07 µg/m 3 /year) does not change significantly from the trend using year-round data (Fig. 6).A consistent trend of 0.21±0.09µg/m 3 /year is found in annually-averaged PM 2.5 at Albian mine (2002)(2003)(2004)(2005)(2006)(2007)(2008) alone, and the trend there during the warm season is also statistically significant and not different (0.24±0.06 µg/m 3 /year).Furthermore, there is no indication of a discontinuity between 2008 and 2009 when the monitoring site was relocated.The trend in PM 2.5 at the surface is in quantitative agreement with the PM 2.5 trends derived from MODIS/Aqua Deep Blue and POLDER/PARASOL annually averaged AOD data over similar, yet shorter periods.For both MODIS/Aqua Deep Blue and POLDER/PARASOL, trends using annual medians agree with trends determined using annual averages within their respective standard errors (1 s. e.).The low bias of POLDER/PARASOL AOD near these two Shell mines is expected from the validation with AERONET at Fort McMurray (Table 4) and previous work on larger spatial scales (Deuzé et al., 2001).
Contrary to the localized, significant AOD trend in satellite data records in the eastern portion of the Muskeg River region, a statistically significant trend is found at two other ground-based stations within the AOSR for the period 2002-2014, namely Syncrude UE1 and Millennium mine (Fig. 6).The largest trend occurs at Millennium mine, the closest NAPS station to the southeast of the Shell Muskeg River region (see Table 2 and Fig. 4 for location).The trend is insignificant using either annual means or median PM 2.5 data at CNRL Horizon and Anzac where data records are shorter, while the trend at Wapasu (2013-2015) was not evaluated.The PM 2.5 trends at the remaining sites in the AOSR, namely two sites at Fort McMurray and one at Fort McKay are discussed below.Note that POLDER/PARASOL does not measure at Syncrude UE1 (see Table 3) and there is insufficient sampling at Millennium Mine over an area of 0.4°×0.4° in each of the years (2005-2013) for trend analysis.For POLDER/PARASOL, the trend, while mostly insignificant in the AOSR, is always positive.For AATSR, the AOSR has regions of statistically insignificant negative and positive trends.For MISR, the trend is positive in 56% of the trend domain and even more so (83%) in the northern half of the domain (57-58°N).For MODIS DB and DT, some of the AOSR is not sufficiently sampled with high confidence (see Sect. 2), but where confidence is ≥1, the trend tends to be negative in 69% and 77% of this area, respectively.Bari and Kindzierski (2016)  using MODIS/Aqua DB data (2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), insignificant at the 2 s. e. level.

Discussion and conclusions
In this section, the advantages and limitations of the various data products are summarized.As shown in Table 4, all of the satellite sensors capture the temporal variability in AOD over Fort McMurray, based on correlations with AERONET, in spite of the low AODs there (e.g.Fig. 1).This temporal variability is largely driven by day-to-day variability as forest fires lead to episodes with large AODs (>3) in summer months that strongly influence the calculated correlation.
The two MODIS AOD data products (Deep Blue and Dark Target) have low confidence in the AOSR due to issues relating to elevated surface reflectivity in the vicinity of the Mildred Lake Syncrude facility.However, the MODIS dark target product is the best at capturing temporal variability in terms of the correlations with AERONET AOD at Fort McMurray and in terms of capturing the month-to-month variability.This is likely due to MODIS's combination of spatial resolution (Table 1) and higher signal-to-noise ratio (SNR): its radiances have SNR > 1000 (Xiong et al., 2003) whereas the other instruments have SNR of 1000 or less (Deschamps et al., 1994;European Space Agency, 2007;Diner et al., 1989).MODIS DT clearly has a slope slightly greater than unity over the AOSR, in contrast to MODIS DB (Table 4).Focussing on Fort McMurray, where there is a longer AERONET data record than at Fort McKay, the MODIS DT slope changes insignificantly when coincident AERONET AOD is limited to <0.7.The same pattern of consistently high and low slope values for the MODIS Aqua DT and DB (collection 6) products, respectively, was found over two sites in Pakistan, namely Lahore and Karachi, by Bilal et al. (2016) and during non-fire summertime periods over semi-arid Nevada and California as shown in Table 4 of the work of Loría-Salazar et al. (2016).A high slope may be related to the use of the 2.1 µm channel to determine the reflectivity in the visible over non-vegetated surfaces as suggested by Bilal et al. (2016).High-biased AODs result because the surface reflectance in the visible assumed by the retrieval algorithm is less than the actual value as the relationship between the visible and 2.1 µm was developed for vegetated land for which a stronger spectral variation exists than for barren land.Li et al. (2005) have shown that the spectral reflectance relationship is much different even for dry vegetation than green vegetation.Note that high day-to-day variability can be captured in spite of biases in assumed surface reflectance since the latter changes slowly with time over the warm season, when successful measurements occur more frequently.A MODIS algorithm designed to function over inhomogeneous surfaces such as the AOSR region, and which would also likely be applicable to urban areas, is being investigated to exploit the many benefits of MODIS radiance data.One such benefit is the twice-daily revisit over the AOSR that the current multi-angle  4), but struggles to capture the local spatial variability including the AOD hotspot in the AOSR as discussed in Sect.3. The MISR low bias may be related to the need for darker spherical particles (Kahn et al., 2005) given that forest fire smoke plays a significant role throughout the western Canada in the warm season (O'Neill et al., 2002).Spherical particles with lower single scattering albedo (SSA) may also be required to properly represent local anthropogenic pollution (Kahn et al., 2005) in the AOSR.
The 3×3 superpixel averaging that is used when the MISR retrieval fails for the central superpixel could also contribute to a low bias (Jiang et al., 2007), particularly at Fort McKay as background AODs to the west could be lowering the average.
AATSR has a major spatial sampling issue in the heart of the AOSR, but also captures month-to-month variability from late spring to early autumn (Table 4) as well as short-term (Table 4) and spatial variability (Table 3).Based on a previous analysis (Che et al., 2016), the AATSR AOD underestimation of the Swansea University product (also used here) is larger over barren surfaces or sparse vegetation.Such land cover types are present in the AOSR.The slight bias (Table 4) is not strongly AOD-dependent as removing coincidences with AERONET 500 nm AOD of >0.35 does not significantly change the slope of the regression equation (Table 4).
POLDER has a known negative offset in AOD (Deuzé et al., 2001), confirmed using coincident Fort McMurray AERONET AOD data.However, POLDER/PARASOL is the most accurate satellite-based aerosol sensor at Fort McMurray during periods of the higher AODs (e.g.≥0.31,Table 4), when its negative offset becomes rather trivial.
Overall, the POLDER AOD product is without a major weakness relative to the other instruments, although it is provided at a relatively coarse spatial resolution (Table 1) and the fixed spatial sampling pattern of this sensor inhibits the application of spatial oversampling techniques.The use of polarized radiances reduces the sensitivity of the retrieved AOD to surface reflectance (e.g.Deuzé et al., 2001).The trend in POLDER/PARASOL AOD at the Shell mines (Albian and Shell Muskeg River) is probably not driven by a trend in surface reflectance since agreement with AERONET tends to be independent of surface type (e.g.Chen et al., 2015).A future sensor of POLDER heritage, namely the Multi-viewing, Multi-channel, Multi-polarisation Imager (3MI), offers higher spatial resolution, the availability of longer wavelength channels, and the potential for accurate monitoring of the local aerosol loading in the decade to come.
While AODs in the AOSR are relatively small according to POLDER/PARASOL (Fig. 1), the significantly positive trend in AOD from this satellite sensor and the similar trend in observed surface-level PM 2.5 in the region of the Muskeg River mine points to the need to continue monitoring of this region with a combination of surface and satellite-based aerosol observations.7) Minimum scattering angle is larger than a threshold or maximum scattering angle is smaller than a threshold.
8) Aerosol optical thickness is larger than a threshold such that surface reflectance cannot be estimated adequately.
9) A large difference between measured and modeled reflectance exists for 443 nm.
10) Differences are too large between measured and modeled reflectance (risk of glitter).
11) Meteorological data indicate the presence of snow at ground level.
12) The quality index is 0.00 for viewing geometry conditions 13) The quality index is 0.00 for polarized reflectance fit.

Fig. 3
Fig. 3 is -0.73, indicating that the spatial variation in AATSR sample size is mostly related to cloud flagging.
AOD interannual variability at Fort McMurray, where a sufficiently long record(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015) of 500 nm AOD exists.All of the products capture the interannual variability of the annual mean AOD observed by AERONET at Fort McMurray (Table5).Correlation coefficients for forest-fire-filtered annual means tend to be only slightly lower.In general, very few of the 200 grid cells in the trend domain (56-58°N, 111-112°W) indicate a statistically significant (2 s. e.) positive trend that is consistent from one satellite to the next.In fact, there are no points in the domain for which MODIS/AquaDT (2003DT ( -2013)), AATSR, orATSR-2 (1996ATSR-2 ( -2002, 0.3°×0.3°), 0.3°×0.3°)show a significant positive trend in AOD.Similarly, POLDER/PARASOL only shows a significant positive trend in three adjacent grid points at 57.3°N between 111.3 and 111.5°W (see Fig.4) and MISR also finds a significant positive trend at only two locations in the domain.Finally, MODIS/Aqua DB has two points with the largest and most significant positive AOD trend in the region of the Muskeg River mine at 57.25°N, 111.25°W (Fig.4).In fact, two satellite data products, namely POLDER/PARASOL and MODIS/Aqua DB, exhibit a significant positive trend in this mining area.Although not statistically different from zero, the AOD trend in both AATSR and MISR data is positive in the area of the positive POLDER/PARASOL trend (Fig.4), whereas MODIS DT tends to show an insignificant negative trend.Changes to the surface may be at the root of the increasing AOD trend in this area, either since clearing of vegetation could lead to higher concentrations of dust, or by biasing the AOD retrieval.Trends in surface albedo were determined from the combined MODIS Terra/Aqua MCD43C3 albedo data product at four wavelengths relevant to the MODIS or POLDER AOD retrievals: 470, 645, 860, and 2130 nm (see Appendix A).For all four wavelengths, neither the largest nor the most significant trends in surface reflectivity occur at 57.25°N, 111.25°WAtmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-748,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 12 October 2016 c Author(s) 2016.CC-BY 3.0 License.
found no indications of a positive trend in PM 2.5 at Fort McKay and the Fort McMurray Athabasca Valley site, using a longer period (1998-2014), although, as shown in Fig. 2 of Bari and Kindzierski (2016) for Fort McKay, there is an abrupt decrease in PM 2.5 mass densities that occurs between 2001 and 2002 that has a profound effect on the trend and its uncertainty.This discontinuity is observed at all sites in the AOSR that extend back to 2001.An earlier study by the same authors (2015) also indicated no trend between 1998-2012 at the same sites and at the Fort McMurray Patricia MacInnes site as well.Li et al. (2016) find a small positive trend in AOD over Athabasca(56-58°N, 110-113°W) Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-748,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 12 October 2016 c Author(s) 2016.CC-BY 3.0 License.sensors, namely MISR and SLSTR (Sea and Land Surface Temperature Radiometer) (Coppo et al., 2010), cannot offer.SLSTR, onboard the recently launched Sentinel-3a satellite, is the next generation in the ATSR series.MISR clearly captures the short-term and month-to-month AOD variability at Fort McMurray based on correlations at the individual coincidence level and the monthly time scale (Table Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-748,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 12 October 2016 c Author(s) 2016.CC-BY 3.0 License.6) Stratospheric aerosol correction is uncertain or imprecise (i.e.stratospheric AOD larger than a certain threshold).

Figure 3 .
Figure 3. Map of sample size (left) and average cloud fraction within AOD superpixels when the AOD retrieval is successful (right), compiled from the entire AATSR data record.Smaller sample sizes are expected over Moose Lake and Gordon Lake (56.5°N, 110.5°W).

Figure 4 .
Figure 4. Areas with a significant positive trend in AOD in the POLDER/PARASOL, and MODIS/Aqua DB data records.The area over which the AOD time series is determined for MODIS/Aqua DB (0.1×0.1°), and POLDER/PARASOL (0.4×0.4°) is outlined in pink and blue, respectively.Locations of 10 NAPS PM 2.5 monitoring sites are also shown as small green squares.The central one of 3 adjacent (overlapping) grid cells at constant latitude is plotted for POLDER/PARASOL (see Sect. 3 for details).The grid cell with the largest trend in the domain is plotted for MODIS/Aqua DB (see Sect. 3 for details).Note that the Albian mine site (57.2808°N,111.526°W) was replaced by the nearby Shell Muskeg River site (57.2491°N,111.509°W) in 2009 (both station symbols are filled in red).The two AERONET instruments are co-located with NAPS monitors and those sites are filled in blue.

Figure 5 . 1 Figure 6 .
Figure 5. Annual average PM 2.5 mass density for the merged Albian mine and Shell Muskeg River dataset, along with PM 2.5 annual averages derived from satellite AOD data records (see Sect. 3 for details and Fig. 4 for satellite trend areas).Each satellite time series is plotted at the average decimal time for each calendar year.Trend lines are fitted to each time series using a matching colour.Vertical error bars indicate ±1 standard error of the annual mean.There are, on average, 33 and 50 observations per year for POLDER/PARASOL and MODIS/Aqua DB, respectively.The secondary ordinate applies to the MODIS DB observations, but not POLDER/PARASOL (for which the 865 nm AODs are in the 0.01 to 0.03 range).
and Fort McKay (57.184°N, 111.64°W).Measurements at Fort McMurray started in 2005.The Fort McKay site has only been in operation since

Table 1 .
Spatial resolution of AOD data products from selected satellite instruments.The third column contains the wavelength at which aerosol optical depth is reported in each satellite data product.MISR and both MODIS instruments are currently operating.Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-748,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 12 October 2016 c Author(s) 2016.CC-BY 3.0 License.

Table 3 .
Spatial correlation between PM 2.5 mass density and AOD using means of coincident data over the entire overlapping period at 10 sites in the AOSR.Wapasu has insufficient or no temporal overlap with POLDER/PARASOL and AATSR.Syncrude UE1 is not spatially coincident with any of the POLDER locations given the 10 km criterion (see Sect. 2)..Chem.Phys.Discuss., doi:10.5194/acp-2016-748,2016Manuscriptunder review for journal Atmos.Chem.Phys.Published: 12 October 2016 c Author(s) 2016.CC-BY 3.0 License.

Table 4 .
Statistical comparison of coincident AODs observed by satellite-based sensors and AERONET CIMEL sun photometer.For each satellite AOD product, the upper row is for Fort McMurray and the lower row is for FortMcKay.The CIMEL 500 nm AOD is used for comparison with all satellite sensors except POLDER/PARASOL, for which the CIMEL 870 nm AOD is more appropriate (see Table1).The simple linear regression equation used to obtain the slope and offset assumes AERONET AOD and satellite-based AOD are the independent and dependent variables, respectively.The number of MISR-Fort McKay coincidences is insufficient to assess the month-to-month variability.Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-748,2016Manuscriptunder review for journal Atmos.Chem.Phys.Published: 12 October 2016 c Author(s) 2016.CC-BY 3.0 License.

Table 5 .
Correlation of annual mean AODs with Fort McMurray AERONET AODs during the respective overlap periods of the various satellite AOD products.In the rightmost column, the contribution of large forest fires has been removed from AERONET data and satellite datasets using +4 standard deviations () as a cutoff.