Biases in absorption coefficients measured using a
filter-based absorption photometer (Tricolor Absorption Photometer, or TAP)
at wavelengths of 467, 528 and 652 nm are evaluated by comparing to
measurements made using photoacoustic spectroscopy (PAS). We report
comparisons for ambient sampling covering a range of aerosol types including
urban, fresh biomass burning and aged biomass burning. Data are also used to
evaluate the performance of three different TAP correction schemes. We found
that photoacoustic and filter-based measurements were well correlated, but
filter-based measurements generally overestimated absorption by up to 45 %. Biases varied with wavelength and depended on the correction scheme
applied. Optimal agreement to PAS data was achieved by processing the
filter-based measurements using the recently developed correction scheme of
Müller et al. (2014), which
consistently reduced biases to 0 %–18 % at all wavelengths. The biases
were found to be a function of the ratio of organic aerosol mass to
light-absorbing carbon mass, although applying the
Müller et al. (2014) correction
scheme to filter-based absorption measurements reduced the biases and the
strength of this correlation significantly. Filter-based absorption
measurement biases led to aerosol single-scattering albedos that were biased
low by values in the range 0.00–0.07 and absorption Ångström
exponents (AAEs) that were in error by
Aerosol–radiation interactions are estimated to contribute a global mean
effective radiative forcing of
The main types of absorbing aerosol include black carbon (BC) and
light-absorbing organic carbon, commonly referred to as brown carbon (BrC)
(e.g. Myhre et al., 2013a). On a global
scale, aerosol absorption is dominated by BC, a carbonaceous product formed
during incomplete combustion, which may exert the next largest positive
radiative forcing after carbon dioxide (Stocker
et al., 2013). BC absorbs strongly across visible wavelengths and
contributes an estimated 0.71 (0.09 to 1.26) W m
Over the course of several decades, filter-based absorption photometry has been used to measure aerosol absorption coefficients. The approach has considerable benefits including that it is relatively inexpensive, portable and capable of unattended measurements for long periods of time (Baumgardner et al., 2012). Filter-based instruments measure the light transmittance across a filter continuously, which changes as particles are deposited onto the filter, providing a measure of aerosol absorption (see Sect. 2.1) (e.g. Bond et al., 1999). Absorption coefficients determined using filter-based absorption photometry can be subject to measurement artefacts due to (i) scattering of light away from the light detector leading to erroneous apparent absorption and (ii) enhanced absorption as particles are deposited onto the filter (Bond et al., 1999). The latter leads to multiple scattering between the particles and the filter medium, providing multiple opportunities for absorption. The enhancement is complex to characterise and depends on the filter loading such that an increase in the number of deposited absorbing particles reduces the multiple scattering between the filter and particle layers (Bond et al., 1999; Liousse et al., 1993; Weingartner et al., 2003) leading to lower absorption coefficients for highly loaded filters (Weingartner et al., 2003). The sensitivity of filter-based absorption photometers is also affected by the penetration depth of particles within the filter matrix, which depends on particle size (Moteki et al., 2010; Nakayama et al., 2010). A number of empirical and semi-empirical correction schemes have been derived to correct for the aforementioned artefacts. This study will focus on correction schemes derived for use with glass-fibre Pallflex E70-2075W filters that have been used widely with the Particle Soot Absorption Photometer (PSAP, Radiance Research) (Bond et al., 1999; Müller et al., 2014; Virkkula, 2010; Virkkula et al., 2005). These correction schemes are also valid for similar instruments using this filter substrate, for example the Tricolor Absorption Photometer (TAP, Brechtel Manufacturing) used in this study and described in Sect. 2.2.2 (Ogren et al., 2017).
Another potentially significant measurement artefact is due to liquid-like organic aerosols spreading across the filter fibres (Lack et al., 2008). The mechanisms proposed for this artefact include a change in the physical shape and therefore optical properties of deposited particles, or a coating effect whereby deposited particle absorption is enhanced via a lensing effect (Cappa et al., 2008; Lack et al., 2008; Subramanian et al., 2007). Although recognised as potentially significant, there are no empirical corrections to account for these artefacts.
Previous work has examined the magnitude of biases in filter-based
absorption measurements. For example, Lack et al. (2008) found PSAP absorption coefficients were biased high in the range 12 % to over 200 % at 532 nm compared to photoacoustic spectroscopy (PAS)
measurements for aerosols over the Gulf of Mexico, which included BC,
nitrate, sulfate and organic aerosols from shipping emissions. The PSAP
biases were found to be positively correlated to the organic aerosol mass
concentration and even more strongly correlated to the ratio of the organic
aerosol to light-absorbing carbon mass. To verify these measurements,
Cappa et al. (2008) performed laboratory experiments using secondary organic aerosol
(SOA) derived from the ozonolysis of
More recently, Subramanian
et al. (2010) derived the BC mass absorption coefficient (MAC) at 660 nm for
fresh and 1–2 d old aerosol emissions in and around Mexico City by
dividing the absorption coefficients measured using a PSAP by the refractory
BC mass concentrations measured using a single-particle soot photometer (SP2, Droplet Measurement Technologies). For the fresh emissions, they found
a
Using a similar methodology, McMeeking et al. (2011) derived the BC MAC at 550 nm using PSAP and SP2 measurements for
urban pollution aerosols around the UK, reporting organic aerosol mass
concentrations in the range 1–7
Biases in filter-based absorption photometry measurements can limit the accurate determination of key climate-relevant parameters including, for example, the aerosol SSA and absorption Ångström exponent (AAE) (e.g. Sherman and McComiskey, 2018). Mason et al. (2018) compared PAS to filter-based absorption measurements of wildfires and agricultural fires over the continental United States during August and September 2013, which included a PSAP and a continuous light absorption photometer (CLAP) (Ogren et al., 2017). All PSAP and CLAP data were corrected using the Bond et al. (1999) and Ogren (2010) corrections. Biases in filter-based measurements were evaluated by comparison to PAS measurements, which were in the range 0.61 to 1.24, dependent on measurement wavelength (405, 532 and 660 nm). Mean SSA and AAE values derived using filter-based absorption photometry were found to be in error by up to 0.03 and 0.7, respectively, compared to PAS.
Further, Backman et al. (2014) assessed the sensitivity of the PSAP-derived AAE to the Bond et al. (1999) and Virkkula (2010) correction schemes for measurements recorded on the central Highveld in South Africa, where emissions were dominated by fossil-fuel burning activities including from coal-fired power plants. They found that the AAE varied between 1.34 and 1.96, dependent upon the PSAP correction scheme applied, which led to different conclusions regarding the aerosol composition and source (Backman et al., 2014).
Despite this body of previous work, there remains significant uncertainty related to the magnitude of biases in filter-based absorption measurements, particularly regarding dependence on source type and the correction scheme applied. The aim of this study is to address this gap. We assess biases by comparing absorption coefficients determined using multi-wavelength TAP and photoacoustic instruments during a series of research flights aboard the UK Facility for Airborne Atmospheric Measurements (FAAM) BAe-146 aircraft. Aerosol sources sampled include urban aerosol emissions over London, fresh biomass burning aerosol (BBA) over West Africa and aged BBA over the Southeast Atlantic Ocean (see Fig. 2). We follow the methodology of Lack et al. (2008) by looking at the absorption biases as a function of organic aerosol concentration, extending their study by looking at a greater range of wavelengths and aerosol sources as well as evaluating additional correction schemes, namely those developed by Virkkula (2010) and Müller et al. (2014). We then assess the impact that biases in filter-based absorption photometry have on the aerosol SSA and AAE. This is the first study to simultaneously evaluate the Bond et al. (1999), Virkkula (2010) and Müller et al. (2014) correction schemes for ambient aerosol sampling across multiple aerosol types.
Filter-based absorption photometers measure the light transmitted through a
filter as particles are deposited onto the filter such that the attenuation
can be defined as
The Bond et al. (1999) correction scheme was
developed empirically by comparing PSAP absorption coefficients to reference
absorption coefficients determined using the difference between extinction
as measured by an optical extinction cell and scattering coefficients
measured using a nephelometer. Calibration aerosols included polydisperse
nigrosin and ammonium sulfate. This correction scheme was updated by
Ogren (2010). Bond et al. (1999) found that
The Virkkula et al. (2005) correction scheme and the
subsequent Virkkula (2010) erratum were derived for the
PSAP wavelengths 467, 530 and 660 nm, which is reflected by the
The values of the constants used in the Virkkula (2010) correction scheme (Virkkula, 2010).
The constrained two-stream (CTS) algorithm developed by
Müller et al. (2014) includes a
two-stream radiative transfer model that explicitly accounts for the optical
properties of the filter substrate and deposited particles and is
constrained by either the Bond et al. (1999),
Virkkula et al. (2005) or Virkkula (2010) parameterisations. This section
covers the key equations from Müller et al. (2014) to show how they have
been implemented in this study and the reader is referred to Müller et al. (2014) for a full derivation. The M2014 correction scheme makes use of
the relationship between the absorption coefficient and the change in
particle absorption optical depth,
To confirm the accuracy of the implementation of the M2014 algorithm used during this analysis, Eqs. (16)–(23) were used to reproduce the results in Fig. 6 of the Müller et al. (2014) study, which were verified against intermediate results (Thomas Müller, personal communication, 2016).
All measurements presented in this study were made aboard the UK's
BAe-146-301 large Atmospheric Research Aircraft (ARA) operated by the
Facility for Airborne Atmospheric Measurements (FAAM;
An important strength of this dataset is that the TAP, PAS and cavity
ring-down spectrometer (CRDS) instruments used to sample aerosol optical
properties all shared a common sample inlet and were subject to the same
flow conditioning. Aerosols were drawn into the aircraft through a modified
Rosemount inlet (Trembath et al., 2012). The aerosol-laden
stream was first dried to < 20 % relative humidity (Permapure,
PD100T-12MSS) and then passed through a scrubber (MAST Carbon) to remove
absorbing gaseous impurities such as ozone and nitrogen dioxide. An impactor
removed particles with aerodynamic diameter > 1.3
Schematic diagram highlighting the flow conditioning and how the aerosol-laden stream was distributed between the PAS and CRDS cells and the TAP. All PAS and CRDS wavelengths were centred at 405, 514 and 658 nm respectively.
The TAP is a commercially available (Brechtel) version of the continuous
light absorption photometer (CLAP), described by Ogren et al. (2017). The TAP comprises eight sample filter spots and two reference
filter spots. The aerosol-laden air passes through one sample spot at a
time, which allows for 8 times the filter lifetime compared to
single-spot photometers. The filtered air is recirculated through one of
the reference spots to enable the attenuation calculation (see Eq. 1)
(Ogren et al., 2017). Upon reaching a predefined filter
transmittance set point, the TAP automatically changes to the next available
sample filter spot. We used 47 mm diameter Pallflex (E70-2075W) glass-fibre
filters, which were nominally identical to the filters used to derive the
correction schemes applied in this study (see Sect. 2.1.1–2.1.3). The TAP
provides measurements at three wavelengths with peaks centred at 467, 528
and 652 nm, which allows the spectral dependence of climate-relevant
parameters such as the SSA and AAE to be evaluated (Sect. 3.3). The LEDs are
cycled through each wavelength once per second, providing absorption
measurements at 1 Hz at all wavelengths. The inlet of the TAP is heated to
To determine the areas of the spots resulting from particle deposition onto
the filter, nigrosin (Sigma Aldrich, product number 198285-100G) was
atomised from solution, dried to < 10 % relative humidity using a
silica gel diffusion drier (Topas, DDU-570) and sampled by the TAP. The
areas of the eight sample spots were determined by measuring the number of
pixels corresponding to the diameters in a magnified digital photograph,
which yielded areas in the range 32.4–36.8 mm
The photoacoustic and cavity ring-down spectrometers used in this study were based on the designs by Lack et al. (2012) and Langridge et al. (2011), respectively, and are described in detail in Davies et al. (2018) and Szpek et al. (2019). PAS measures absorption directly for aerosols in their suspended state (Arnott et al., 1999). The PAS principle relies on converting energy from a light source into sound. Light-absorbing media, such as aerosol, transfer electromagnetic energy into thermal energy that heats the surrounding air. This gaseous heating generates a pressure wave, which is detected by a microphone located within the PAS cell. The amplitude of the microphone signal is related to the sample absorption coefficient through calibration (Arnott et al., 1999; Davies et al., 2018; Moosmüller et al., 2009).
Much of this analysis relies on accurate PAS absorption measurements and
thus we focus here on describing the uncertainty associated with these
measurements. The total PAS measurement uncertainty is comprised of the
measurement precision and accuracy. The PAS measurement precision was
derived by evaluating the minimum sensitivities of the suite of PAS
instruments in a similar way to the TAP, as described in Sect. 2.2.2, and were in the range 0.01–0.06 Mm
The accuracy of PAS absorption measurements was determined primarily by three factors: (i) uncertainty in the ozone calibration, (ii) uncertainty in corrections applied to account for the PAS microphone pressure sensitivity and (iii) uncertainty in subtraction of background noise which arose primarily from laser heating of the PAS cell optical windows. We consider each of these in turn below.
The accuracy of the PAS ozone calibration has previously been evaluated in laboratory experiments that compared measured and modelled absorption and extinction cross sections of strongly absorbing nigrosin aerosol. This analysis showed the PAS calibration accuracy to be better than 8 % and the accuracy of the CRDS instruments used in this study to be better than 2 % (Davies et al., 2018). Moreover, our recent work has demonstrated that the calibration accuracy of PAS using ozone is optimal when the gas-phase composition closely resembles that of ambient air (Cotterell et al., 2019), as is the case for calibrations performed for this work.
The second source of PAS measurement uncertainty was due to the PAS
microphone sensitivity to pressure, which was evaluated by performing ozone
calibrations at several pressures in the range 600–1000 mbar (typical of
those encountered during airborne operation). The measured PAS microphone
sensitivities were fit to a linear trend across this range and normalised to
yield a correction factor that varied from 0.83 (600 mbar) to 1.00 (1000 mbar).
The uncertainty introduced by applying this pressure-dependent correction to
PAS calibrations was estimated by propagating the
The third source of PAS measurement uncertainty was due to subtraction of
window-generated background noise, which is unstable for airborne operation
due to its dependence on pressure. To account for this, in-flight background
noise is typically characterised by periodically measuring a filtered-air
stream for 30 s every 300 s. These measurements are then used post-flight to
derive a background correction as a function of pressure. To evaluate the
uncertainty introduced by this background noise correction, we took
continuous PAS measurements of filtered air in the laboratory and varied the
pressure within the PAS cells over the range encountered during airborne
operation. This laboratory PAS dataset was then processed to mimic in-flight
conditions, with 30 s windows of data every 300 s being used to derive a
continuous pressure-dependent background correction. Examining the
difference between the continuous filtered-air measurements and the
synthetically generated background data series provided the uncertainty in
the background noise correction under variable pressure conditions. The
uncertainty in the background noise correction was found to be normally
distributed, with a
The total uncertainty in PAS measurements is the combination of the
measurement precision and accuracy, including the PAS calibration accuracy,
the pressure-dependent calibration correction uncertainty and the background
noise correction uncertainty. These factors were combined in quadrature,
leading to total PAS measurement uncertainties of 29.0 %–55.0 % for 1 Mm
Nephelometer measurements (TSI 3563) were used to derive the aerosol asymmetry parameter needed to apply the Müller et al. (2014) correction scheme (see Sect. 2.1.3) and were corrected according to Müller et al. (2011). A time-of-flight aerosol mass spectrometer (TOF-AMS) (e.g. Drewnick et al., 2005) measured the aerosol composition. The TOF-AMS was run as described in previous publications (e.g. Morgan et al., 2010).
All absorption, scattering and extinction coefficient data measured using
the PAS, TAP, CRDS and nephelometer were recorded at 1 Hz. Data were
subsequently averaged to 30 s during post-flight analysis to reduce
the noise in these measurements and to aid temporal alignment of the PAS and
TAP for direct comparisons. In the case of TAP measurements, the intensities
of light transmitted through a filter were first averaged to 30 s and
then input into Eqs. (1)–(9) to determine the corresponding absorption
coefficients. To account for time lags between the PAS and TAP, an
optimisation routine was run that maximised the correlation coefficient
(
This study uses data collected aboard the FAAM aircraft during 30 research
flights (each 3–4 h duration) in three distinct regions: London (three
flights, 17 to 20 July 2017, from 1.7
FAAM research aircraft flight tracks (red) over London in the United Kingdom (July 2017), West Africa (February and March 2017) and the Southeast Atlantic (August and September 2017). For each of the geographical areas highlighted in the white boxes, the mean aerosol optical depths (AODs) measured using the Moderate Resolution Imaging Spectroradiometer (MODIS) satellite instruments are displayed. A time series of Aerosol Robotic Network (AERONET) data shows AODs at 500 nm corresponding to each measurement period. Note the discontinuous AERONET AOD time axis. AERONET sites are shown on the MODIS AOD plots by arrows.
The primary result of this study is that the absorption coefficients
determined using a TAP and PAS are linearly correlated and that the slope
(
A summary of the slopes (
For the B1999 correction scheme, the range of TAP biases across all aerosol sources was 1.18–1.45. The smallest biases were consistently associated with 467 or 652 nm wavelength measurements and the largest for 528 nm wavelength measurements. An interesting feature of this result is that the B1999 scheme led to the largest biases at 528 nm, which is the wavelength closest to that at which the scheme was derived.
For the V2010 correction scheme, the range of TAP biases across all aerosol sources was 1.08–1.38. The largest biases were consistently at 467 nm and the smallest at 652 nm. Relative to the B1999 correction scheme, the V2010 scheme reduced the biases at 528 and 652 nm by 5 %–15 % while it increased the bias at 467 nm by 3 %–5 %, dependent on the aerosol source. The sensitivity of TAP biases to the wavelength-dependent constants used in the V2010 scheme was investigated due to the mismatch in the TAP wavelengths and those for which the V2010 correction scheme was derived. Applying the single-wavelength V2010 correction scheme (i.e. applicable at all wavelengths) decreased TAP biases by 7 %–9 % at 467 nm, increased biases by 1 % at 528 nm and increased biases by 6 %–8 % at 652 nm.
For the M2014 (B1999 parameterisation) correction scheme, the range of TAP biases across all aerosol sources was 1.04–1.26, and for the M2014 (V2010 parameterisation) the range of TAP biases was 1.01–1.18. The M2014 (V2010 parameterisation) scheme reduced TAP biases relative to the B1999 and V2010 schemes by 7 %–38 % and 7 %–25 %, respectively, dependent on the aerosol source and wavelength. The most significant reductions in TAP biases were for urban aerosol emissions and had the most impact on measurements at 652 nm. As discussed in Sect. 2.1.3, the M2014 (V2010 parameterisation) correction scheme applied here used the wavelength-dependent Virkkula (2010) parameterisation, in contrast to Müller et al. (2014), who applied the Virkkula et al. (2005) parameterisation. Although not shown, applying the Virkkula et al. (2005) parameterisation to TAP data in this study would act to decrease TAP biases by 3 %–4 % at 467 nm and increase biases by 1 %–2 % at 528 nm and by 3 % at 652 nm.
Absorption coefficients measured by PAS versus TAP for urban
emissions around London in July 2017. The columns correspond to
467 nm (column 1), 528 nm (column 2), and 652 nm (column 3) wavelengths, and the rows
correspond to the B1999, V2010 and M2014 corrections. All absorption
coefficients correspond to > 1 Mm
The
As Fig. 3 but for fresh biomass burning aerosol over Senegal in February and March 2017.
An analysis of the dependence of TAP bias as a function of filter loading revealed no point-by-point dependence but potentially a weak signal in the large-scale mean such that the difference in absolute filter transmittance associated with the highest 10 % of TAP biases compared to the lowest 10 % of biases across all channels and wavelengths was up to 0.12. The filter transmittance changed over the course of a flight by a maximum of 0.21.
As Fig. 3 but for aged biomass burning aerosol over the Southeast Atlantic Ocean in August and September 2017.
The TAP biases exhibited a strong wavelength dependence. In general, the lowest biases were seen at 652 nm and the largest biases at 467 nm when the V2010 and M2014 (V2010 parameterisation) schemes were applied to TAP measurements for all aerosol sources. The exceptions to this trend were when the M2014 scheme (V2010 parameterisation) was applied to urban aerosol measurements, which led to the largest biases at wavelength 528 nm. The M2014 scheme (B1999 parameterisation) also led to the largest biases at 528 nm for all aerosol types.
As highlighted in the introduction, filter-based absorption photometers are sensitive to the particle penetration depth, which is dependent on particle size. Indeed, this sensitivity may have contributed in part to the variation in TAP biases observed for the three types of aerosol investigated during this study.
Perhaps the most important and robust observation is that the M2014 scheme consistently led to the lowest biases across all measurement wavelengths and aerosol sources investigated. The largest biases were associated with TAP measurements corrected using the B1999 scheme at wavelengths 528 and 652 nm and when using the V2010 scheme at wavelength 467 nm for all aerosol sources.
The biases of 1 %–45 % observed in this study are at the lower end of
those measured by Lack et al. (2008) and Cappa et al. (2008), who reported biases of 12 % to
The ratio of TAP to PAS absorption coefficients at 528 nm as a
function of the organic aerosol mass concentration using the B1999
correction scheme (
Figure 6a–c show how TAP biases vary with OA mass concentration for TAP measurements corrected using the B1999 correction scheme, for direct comparison with the Lack et al. (2008) study. The linear relationship between the PSAP biases and OA observed by Lack et al. (2008) is superimposed for reference. For urban emissions (Fig. 6a), TAP biases and OA mass are positively correlated, and the trend is broadly consistent with that observed by Lack et al. (2008). There is however no correlation for fresh (Fig. 6b) or aged BBA (Fig. 6c).
Campaign-mean single-scattering albedo (SSA) derived using PAS and CRDS measurements and TAP and CRDS measurements.
TAP biases were also plotted as a function of the ratio of the mass
concentrations of OA to light-absorbing carbon (LAC), denoted by
Probability density functions of the single-scattering albedo
derived using (i) PAS and CRDS and (ii) TAP and CRDS for the range of TAP
correction schemes outlined in Sect. 2.1.1–2.1.3 at wavelengths 467, 528
and 652 nm. All absorption coefficients correspond to > 1 Mm
This analysis was repeated at wavelengths of 467 and 652 nm. For
measurements at 652 nm, where BrC absorbs relatively weakly (e.g. Andreae and
Gelencsér, 2006), stronger correlations between TAP biases and
A key result of this analysis is to show that biases observed in filter-based aerosol absorption measurements are strongly dependent on the type of aerosol being sampled. Correlating biases to aerosol composition information may provide tight constraint for a single source study, such as that observed by Lack et al. (2008) for aerosol emissions over the Gulf of Mexico, but care must be taken when applying these findings more broadly to other aerosol types.
We now assess the impact that the observed TAP biases may have on climate-relevant parameters including the aerosol single-scattering albedo and
absorption Ångström exponent. Figure 7 shows histograms of the SSA
derived using PAS or TAP absorption data together with CRDS extinction data
for the aerosol sources described in Sect. 2.2.6 and for the TAP corrections
described in Sect. 2.1.1–2.1.3. The SSA is biased towards lower values when
derived using TAP measurements, consistent with the results in Figs. 3–5
which typically show a
Campaign-mean absorption Ångström exponent (AAE) derived using PAS and TAP measurements.
Probability density functions of the absorption Ångström
exponents derived for PAS and TAP measurements using the range of TAP
correction schemes as outlined in Sect. 2.1.1–2.1.3. All absorption
coefficients correspond to > 1 Mm
The SSA values were most different for urban aerosols, which were biased low by 0.01–0.07, dependent on wavelength and the TAP correction scheme applied. This is consistent with the results in Table 2, which highlights that TAP biases were largest for urban aerosol measurements. The wavelength dependence of the TAP-derived SSA values depended on the correction scheme applied. SSA values derived using the M2014 correction scheme agreed most closely with those derived using PAS measurements for all measurement wavelengths and correction schemes.
Similarly, Fig. 8 shows histograms of the AAE values derived by performing linear regressions between the logarithms of the PAS-measured absorption coefficients and the PAS measurement wavelengths (405–658 nm) (Moosmüller et al., 2011). It also shows the same information for the TAP-derived AAE values. The AAE values were calculated for the aerosol sources outlined in Sect. 2.2.6 and TAP correction schemes outlined in Sect. 2.1.1–2.1.3.
The AAE values were strongly dependent on the TAP correction scheme applied.
Campaign-mean AAE values are summarised in Table 4, which highlights that
the highest mean AAE values were associated with fresh BBA emissions and the
lowest for aged BBA emissions. TAP-derived AAE values were in absolute error
by
Measurement artefacts in a commercially available filter-based absorption
photometer (TAP) were evaluated as a function of wavelength and aerosol
source. A range of correction schemes have been proposed in the literature
to account for these artefacts and thus to maximise the accuracy of aerosol
absorption coefficients determined using this technique, although biases can
remain. Three correction schemes were evaluated, which all reduced the TAP
mean bias to within 1 % to
Biases in filter-based absorption measurements were strongly source dependent. On no occasion were the very large biases of over 200 % noted in the Lack et al. (2008) study observed. However, we note that the aerosol types measured in the Lack et al. (2008) study were very different to those studied here, and therefore this result may well be consistent with the strong source dependence observed in the current study.
The positive bias in filter-based absorption measurements resulted in a low bias in determinations of single-scattering albedos of up to 0.07. The largest biases in SSA values were for urban aerosol measurements at wavelength 652 nm. The M2014 scheme consistently led to SSA values that were closest to those derived using PAS measurements across all wavelengths and aerosol sources.
Large discrepancies were seen between AAE values derived from PAS versus TAP measurements, the latter depending strongly on the correction scheme applied. The largest discrepancies in AAE values were for TAP measurements of urban aerosols corrected using the B1999 scheme, which were biased low by a mean absolute value of 0.54. The best agreement with AAE values derived using PAS measurements was obtained when TAP measurements were corrected using the M2014 (B1999 parameterisation) correction scheme and when (i) urban aerosol measurements were corrected using the V2010 scheme, (ii) fresh BBA measurements were corrected using the M2014 scheme and (iii) aged BBA measurements were corrected using the B1999 scheme. This highlights that the AAE is strongly source and correction scheme dependent.
The strong aerosol source dependence of biases observed in this study cautions against extrapolating results more widely to other aerosol types. Further analyses exploring biases in filter-based absorption coefficient measurements may help to address this issue. However, given the empirical nature of filter-based correction schemes and strong source and wavelength dependencies, even this is unlikely to fully bound uncertainties associated with filter-based absorption measurements to the high level of confidence that can be achieved using alternative methods, such as photoacoustic spectroscopy.
For data related to this paper please contact Justin Langridge (justin.langridge@metoffice.gov.uk).
NWD, JMH and JML designed the research. All co-authors collected the data used for the research. NWD, CF, KS, JWT, JDA, PIW, JT, and JML prepared the data for analysis. NWD performed the analyses and prepared the paper with contributions from all co-authors.
The authors declare that they have no conflict of interest.
We thank the Met Office, Natural Environment Research Council (NERC), Norwegian Research Council, and the Royal Society of Chemistry for funding this work. Airborne data were obtained using the BAe-146-301 Atmospheric Research Aircraft operated by Directflight Ltd and managed by FAAM, which is jointly supported by NERC and the Met Office. The authors acknowledge the dedicated work of FAAM, Directflight and Avalon during the aircraft campaigns. We thank the MODIS Science Team for the freely available Terra and Aqua level 2 AOD MODIS data at
This research has been supported by the Natural Environment Research Council (grant nos. NE/L013797/1, NE/L013584/1, and NE/N015835/1), the Natural Environment Research Council/Met Office (grant no. 640052003), the Research Council of Norway (ACBC and NetBC grants), and the Royal Society of Chemistry (Analytical Chemistry Trust Fund, Tom West Fellowship).
This paper was edited by Omar Torres and reviewed by John Ogren and one anonymous referee.