Use of filter radiometer measurements to derive local photolysis rates and for future monitoring network application

. Production of hydroxyl ( OH ) radicals is frequently dominated by the photolysis of tropospheric ozone ( O 3 ). However, photolysis of nocturnal radical reservoirs, such as nitrous acid ( HONO ) and nitryl chloride ( ClNO 2 ), also produces radicals ( OH and Cl atoms) that contribute to the oxidising capacity of the local atmosphere, and initiate many radical-chain reactions that lead to the formation of harmful secondary pollutants. Photolysis of nitric acid ( HNO 3 ) is also a minor radical production mechanism. In this paper, locally representative photolysis rate constants ( j -values) for these molecules are shown 5 to be critical for quantifying and understanding the rate of radical production in a local atmosphere. The ﬁrst long-term 4- π ﬁlter radiometer dataset in the UK (21 November 2018-20 November 2019) available for direct atmospheric model validation is reported. Measurements were made at Auchencorth Moss, a Scottish rural background site, and j ( NO 2 ) is used to generate a measurement-driven adjustment factor (MDAF) for calculated j -values that accounts for local changes in meteorological variables without signiﬁcantly increasing computational cost. 10 Modelled clear-sky -values and for Moss the Tropospheric Ultraviolet and Visible radiation model (TUV; v.5.3.1). Applying the MDAF metric resulted in the calculated photolytic production rate of OH radicals, from all sources considered, being ∼ 40% lower over the year. Photolysis of HONO resulted in an increased rate of OH production compared to that from O 3 in low-light conditions, such as sunrise and sunset (Solar Zenith Angle >80 ◦ ). Hydroxyl radical production from HONO photolysis exceeded that from O 3 consistently throughout the day during 15 the winter and autumn (by a factor of 5 and 2.1, respectively). Radical production rates from HONO and ClNO 2 reached maximum values during the early morning hours of summer (06:00-09:00 UTC), with OH produced at a rate of 1 . 03 × 10 6 OH radicals cm − 3 s − 1 , and Cl radicals at 3 . 20 × 10 4 Cl radicals cm − 3 s − 1 , with the MDAF metric applied. This ﬁrst application of the MDAF j -values demonstrates an efﬁcient measurement and computational approach to improve modelling of the local atmospheric photochemistry that drives NO 2 , O 3 and PM pollution levels. The incorporation 20 of local radiation measurements in measurement networks, and the consequent greater spatial resolution of locally-relevant photolysis coefﬁcients in model photolysis parameterisations, will improve the accuracy of assessment of air pollution and policy-intervention impacts. quantify radical production rates from photolysis pathways. It aims to assess the temporal 105 distribution of local OH and Cl radical formation at a single site, using the ﬁlter radiometer measurements applied to clear-sky model outputs to account for changes in local meteorological variables. The inclusion of local values of such would considerably increase model computational cost. The dataset provides the ﬁrst accurate long-term UK point of comparison for validation of model photolysis coefﬁcients. The scientiﬁc, and ultimately policy, value of incorporating ﬁlter radiometer measurements alongside existing long-term and spatially-resolved air quality monitoring networks is emphasised.


25
Atmospheric chemistry is largely driven by solar radiation. The photolysis of nitrogen dioxide (NO 2 ) has significant impact on air quality, contributing to the formation of harmful pollutants like tropospheric ozone (O 3 ;Reactions R1 and R2).
Photodissociation of many trace gases produce highly reactive radical species such as hydroxyl (OH) and chlorine atoms 30 (Cl), which dominate the oxidising capacity of the atmosphere. Since reactions with these radicals are often the rate-determining step in the chemical loss of many species emitted into the atmosphere, such as volatile organic compounds (VOCs), these radicals dictate the atmospheric lifetimes and concentrations of these other species (Atkinson and Arey, 2003). Products of these reactions include a variety of harmful secondary pollutants like tropospheric ozone (O 3 ), and atmospheric reservoirs of NO 2 such as peroxyacetyl nitrate (PAN), that contribute to long-range and regional pollution (Spataro and Ianniello, 2014).  versely, the high reactivity of these radicals means they have very short atmospheric lifetimes (e.g. <1 s for OH; Monks, 2005), so their concentrations and potential oxidation impacts are only relevant on a local scale.
The OH radical is the most dominant radical diurnally, known to be produced through the photolysis of ozone and subsequent reaction with H 2 O (Reactions R3 and R4).

50
However, formation of ClNO 2 is not limited to chloride in sea spray. Its presence has been repeatedly measured at sites up to 1400 km from the nearest coastline (Thornton et al., 2010;Bannan et al., 2015;Osthoff et al., 2018;Sommariva et al., 2018), with the particulate chloride precursor generated from anthropogenic sources such as fossil fuel combustion and biomass burning (Ahern et al., 2018).
reactions with OH (Monks, 2005;Young et al., 2012). Therefore, even at low concentrations, Cl has a significant effect on local tropospheric oxidation (Riedel et al., 2014;Bannan et al., 2015). As a result, quantifying local photolysis rates is as critical to understanding the local air quality as measuring trace gas concentrations.
Photolysis is a unimolecular reaction characterised by a simple rate equation, demonstrated by Eq. (1) for the photolysis of The rate constant for this photolysis reaction, j(NO 2 ), can be measured in absolute terms using a chemical actinometer.
This estimates j(NO 2 ) by measuring the concentration of NO following photodissociation of a known concentration of NO 2 inside a reaction chamber exposed to ambient light (Bahe et al., 1980;Shetter et al., 1992;Lantz et al., 1996;Bohn et al., 2005). Spectroradiometers measure actinic flux independent of the angle of incidence in a 2-π sr field of view, as a function 65 of wavelength (Kraus and Hofzumahaus, 1998). Photolysis frequencies are determined from Eq. (2) for any molecule whose absorption cross-section (σ) falls within the measured wavelength (λ) range (Shetter and Müller, 1999). For instruments at ground level, the lower-bound wavelength limit (λ 1 ) is typically set to 290 nm, while the upper-bound is dependent on the molecule (e.g. λ 2 = 420 nm for j(NO 2 ), and λ 2 = 340 nm for j(O 1 D); Edwards and Monks, 2003).
The actinic flux (F ) is the quantity of light available to molecules at a given point in the atmosphere, that upon absorption results in photodissociation . It is a spherical integration of solar radiant energy over a surface plane, differing by a cosine of the angle of incidence (Madronich, 1987). The absorption cross-section (σ) quantifies the ability of the molecule to absorb radiation, and the quantum yield (φ) is the probability that the photodissociation leads to the product channel under consideration, in this case O( 3 P) and NO. These are molecule-specific parameters that vary with respect to wavelength 75 and temperature, the latter of which is not often assessed for molecules. Hence, uncertainty in these values propagates through to uncertainty in calculated photolysis frequencies (Shetter et al., 1996;Kraus and Hofzumahaus, 1998).
Filter radiometers use band-pass filters to measure broadband actinic flux, between λ 2λ 1 in Eq.
(2), designed to simulate the σ × φ product of the molecule of interest at its strongest absorbing region. A 4-π sr filter radiometer provides a 360 • field of view of the surrounding environment by utilising two identical optical inlets, designed by Junkermann et al. (1989) and modi-80 fied by Volz-Thomas et al. (1996). The two 2-π sr domes face in opposite directions, separating the total measured actinic flux into down-and up-welling components (F ↓ and F ↑, respectively). These instruments exhibit good linear detector responses, and long-term stability in calibration factors (Bohn et al., 2008), and measurements can be recorded with high time resolution (1 s). These instruments are also easy to deploy and maintain, making them excellent candidates for routine measurements.
However they remain reliant on absolute calibrations to quantify j(NO 2 ) from recorded voltages, and measurements are only 85 applicable for specified reactions, with limited potential to estimate the photolysis frequencies of other atmospheric species.
Solar global irradiance (G) measurements are more widespread than j(NO 2 ), as this is a common meteorological parameter included in multiple global monitoring networks, such as the World Meteorological Organisation (WMO) Global Atmospheric Watch (GAW) programme. Extensive efforts have been made to derive non-linear parameterisations that utilise existing measurements of G to quantify downwelling j(NO 2 ), thereby improving spatial coverage of j-value estimations (Bahe et al., 1980;90 Brauers and Hofzumahaus, 1992;Webb et al., 2002;Trebs et al., 2009). G is typically measured using a pyranometer, which has a horizontal sensor measuring total downwelling radiation (direct + diffuse), weighted by a cosine response subject to the angle of incident light (Webb et al., 2002). Trebs et al. (2009) extensively discuss this relationship, and present good agreement between co-located G and j(NO 2 )↓ measurements for 7 ground-based field sites up to 800 m above sea level despite measurements being from different years, seasons and continents. The authors acknowledge that this parameterisation is not a 95 substitute for measurements of j(NO 2 ), but a viable alternative to radiation transfer calculations where input parameters (like cloud cover) are inadequately known.
Most atmospheric and radiative transfer models use Eq.
(2) to estimate j-values, but often oversimplify local meteorological conditions due to the computational cost of their inclusion, or the absence of necessary measured input parameters (Shetter et al., 2003). A myriad of variables impact the photolysis frequency, such as cloud cover, aerosol optical depth (AOD; Wild 100 et al., 2000), surface albedo and ozone column. All of these variables can redirect incident radiation by absorption and/or reflection. Palancar et al. (2013) demonstrate that when measured AOD and NO 2 concentrations are included in model inputs, actinic flux is reasonably well predicted and the main source of uncertainty is then attributable to σ and φ.
This study presents the first long-term filter radiometer dataset in the UK (Auchencorth Moss, SE Scotland) co-located with relevant ancillary measurements to quantify radical production rates from photolysis pathways. It aims to assess the temporal 105 distribution of local OH and Cl radical formation at a single site, using the filter radiometer measurements applied to clearsky model outputs to account for changes in local meteorological variables. components are measured using standard methodology (Tørseth et al., 2012). Collectively, EMEP monitoring sites are used to understand long-range transport of air pollution around Europe, verify regional modelling approaches and advise governmental bodies on concentration and deposition of pollutants (UNECE, 2004;Aas et al., 2012).
Auchencorth Moss is also a site for a number of national and local monitoring networks (see Twigg et al., 2015), hosting an extensive array of instrumentation to measure trace species concentrations and meteorological parameters (Coyle et al., 2019).

120
Long-term instrumentation deployed at Auchencorth Moss utilised in this study is detailed in Table 1. It is acknowledged that both the HONO and HNO 3 concentrations reported by the Monitor for AeRosols and Gases in Ambient Air (MARGA; Metrohm Applikon, NL) have potential interferences from other NO y species. These interferences have been shown to result in both over- (Phillips et al., 2013) and underestimates (Rumsey et al., 2014) in reported concentration. Since these interferences have not yet been formally quantified, this study uses the reported values as they are with caution.

Filter radiometer
The 4-π filter radiometer (Metcon, Meteorologie Consult GmbH, DE) was mounted ∼ 3 m above the ground, recording measurements at 1 s time resolution for a full year (21 November 2018-20 November 2019). The inlet optic of each dome is designed to have a near-uniform angular response through use of a quartz diffusor (Bohn et al., 2004). Each optical inlet is surrounded by a light shield to provide an "artificial horizon", restricting the field-of-view for each dome to one hemisphere 130 (Volz- Thomas et al., 1996) and preventing overlap. Transmitted light is guided through a set of optical filters (2-mm UG3, 1-mm UG5, Schott GmbH) that restrict transmitted photons to wavelengths of interest prior to their detection by a Hamamatsu photodiode, which proportionally converts incident radiation into an output voltage.
Output signals from the filter radiometer were calibrated against a Bentham DTM300 scanning spectroradiometer between 13 and 25 June 2019. This mid-summer period was selected to provide calibration over the maximum range of ambient inci-135 dent radiation. The direction of the filter radiometer was turned 180 • mid-way through this period, in order to calibrate each dome separately. j(NO 2 ) was calculated from the actinic flux spectra measured by the spectroradiometer using σ(NO 2 ) from Mérienne et al. (1995) and φ(O 3 P) from Troe (2000). Calibration factors for each dome were quantified using data presented in Fig. 1.
The limits of detection (LOD; 3σ of background signal) were 9.40 × 10 −6 s −1 and 1.15 × 10 −5 s −1 for the down-and 140 upwelling domes, respectively. Background signals were determined from averaged measurements made after sunset and before sunrise (solar zenith angle (SZA) ≥ 96 • ) and removed prior to data analysis. On a few occasions of peak solar irradiance (noon in summer), recorded voltage of the downwelling dome exceeded the range of the detector and was reported as ∼ 10 V, observed during the calibration in Fig. 1. These incidences of high j(NO 2 )↓ (< 9 V) comprised 1.4% of all data collected, and were removed from the dataset before further analysis. As a consequence, maximum j(NO 2 )↓ presented in this study is 145 an underestimate (7 − 22% based on calibration). During the calibration, measurements made by the filter radiometer were averaged to match the duration of each spectroradiometer scan (3 mins). The standard deviation associated with each of the averaged filter radiometer measurements were used to remove calibration data points where actinic flux was highly variable, and not comparable between the instruments during the 3 min scan.
The uncertainty of the filter radiometer measurements were estimated as a combination of instrumental error (e.g. non-ideal 150 inlet optic responses; Larason and Cromer, 2001), error from calibration, and errors due to external factors (e.g. temperature stability). For six of the same filter radiometers, Shetter et al. (2003) quoted overall error as 9.6 − 11%, the range dependent on whether conditions were clear or cloudy. To provide a conservative estimate of error for both domes of the filter radiometer used here, the upper bound (11%; cloudy conditions) is combined with the calibration error of each dome. Overall errors for down-and upwelling domes were 13% and 12%, respectively. Measurements of actinic flux made using radiometers intrinsically capture variations in solar flux caused by local meteorology and other factors, such as the presence of aerosols. Since it is computationally intensive to replicate these variations in radiation, which cannot be well represented if input parameters are poorly known, it has been demonstrated that radiometer measurements could be used to adjust j-values in chemical transport models (Elshorbany et al., 2012;Sommariva et al., 2020). Equation (3) 175 illustrates how this can be done by defining a measurement-driven adjustment factor (MDAF), using j(NO 2 ) as a reference.
Correction factors like the MDAF have only occasionally been applied both spatially and temporally (Stone et al., 2012;Bannan et al., 2015), with measured and modelled environments often not co-located (e.g. Sommariva et al., 2020). The MDAF metric, or equivalent, can only be used for model validation where radiometer measurements to complement the model 180 study can be obtained.

Production rate of OH radicals
The application of a locally-derived MDAF to model-calculated j-values is demonstrated here for calculations of the rate of production of OH radicals at Auchencorth Moss from O 3 and two NO y species, HONO and HNO 3 . For both the latter, OH radicals are produced by a direct photodissociation as shown in Reaction (R5) for the case of HONO. The rates are calculated 185 using Eq. (4), and an equivalent for the photolysis of HNO 3 .
These are compared with the production rate from O 3 photolysis (Reactions R3 and R4) as calculated by Eq. (5) and (6).  (Kleffmann et al., 2003;Young et al., 2012;Ryan et al., 2018). As HONO is measured at a height of 3.55 m at Auchencorth Moss (Twigg et al., 2015), p(OH) HONO presented here are applicable for ground level, and would be an overestimate for higher altitudes (>1 km).

Production rate of Cl atom radicals 200
The MDAF is also demonstrated for calculation of the production rate of Cl atoms from the photodissociation of ClNO 2 , shown in Eq. (7), to the products in Reaction (R6).
Values of σ(ClNO 2 ) used to determine j(ClNO 2 ) were compared between those used in the TUV model (Sander et al., 2006) and the more recent IUPAC preferred values from Ghosh et al. (2012). concentration, and as the mean concentration of ClNO 2 for an approximate diurnal cycle that might be expected at Auchencorth Moss (Fig. 2). The shape of the diurnal profile follows that reported by Sommariva et al. (2018) at Weybourne.
3 Results and discussion

Filter radiometer measurements
A direct comparison between the co-located downwelling filter radiometer and G measurements for the duration of the study is presented in Fig. 5. The relationship between the two measurements is linear until G ≈ 500 W m −2 , above which a slight curvature is observed. A quadratic fit to this data yields predicted values of j(NO 2 ) that lie between two previously published parameterisations for this relationship. Predicted j(NO 2 ) values in this study are lower (for G > 150 W m −2 ) than those was that they did not include measurements at SZAs < 30 • . However Auchencorth Moss is at a higher latitude than any of 250 these sites, with a minimum SZA of ∼ 32 • in June. Curvature of the plot of j(NO 2 ) against G is still observed, although there is still a moderate amount of scatter.
The time series of measured and modelled j(NO 2 ) derived from this study is presented in Fig. 6. In general, model results are larger than the measured counterpart, which is expected as meteorological conditions at Auchencorth Moss are typically overcast. One clear discrepancy with this generalisation occurs between 30 January and 2 February 2019. This peak in measured 255 data is not reflected in the solar irradiance time series in Fig. 3 as it was largely caused by snow cover at Auchencorth Moss substantially increasing the surface albedo.
The MDAF values calculated as per Eq. (3) from the measured and modelled j(NO 2 ) was largest during the sunrise and sunset hours, where SZA exceeded 80 • , due to the model predicting very small values of j(NO 2 ). In general, for the rest of the year the adjustment factor was greatest in the morning and steadily decreased throughout the day until sunset. The range of 260 values was greatest in the winter months, and the smallest in the summer.

Estimates of j(HONO) and j(ClNO 2 )
Parameterisations of j(HONO) and j(ClNO 2 ) based on other more-commonly available variables have been proposed before from both measurement intercomparisons and modelling studies. For example, Kraus and Hofzumahaus (1998) noted that molecules which photodissociate in a similar spectral region display a higher correlation between rate coefficients. The em-265 pirical parameterisation that they derived from spectroradiometer measurements to calculate j(HONO) from measured j(NO 2 ) (Eq. (10)) has been implemented in further studies requiring j(HONO) estimates (Alicke et al., 2003;Kleffmann et al., 2005;Acker and Möller, 2007).
A comparison between annual mean diurnal cycles of j(HONO) at Auchencorth Moss parameterised using Eq. (10) and  Fig. 7.
Overall, the TUV model yields the greatest annual mean j(ClNO 2 ) values (Fig. 7, bottom panel), exceeding the others by 280 11−26%, with the maximum difference at noon. While the parameterisations from Young et al. (2014) and Riedel et al. (2014) demonstrate reasonable agreement to the updated j(ClNO 2 ) results, they are closest at different periods during the diurnal cycle. The Riedel et al. parameterisation matches best during peak solar hours (11:00-14:00 UTC), but overestimate by ∼ 15% in the morning (06:00-09:00 UTC) and early evening (17:00-18:00 UTC). In contrast, results from the Young et al. (2014) parameterisation reveal a closer match to updated j(ClNO 2 ) during the non-peak solar hours, and underestimate by ∼ 5% at 285 noon.

Production rate of OH radicals
Average seasonal production rates of OH radicals (p(OH)) from the photodissociations of O 3 , HONO and HNO 3 are presented in Table 3, calculated using j-values directly from TUV model output, and with the MDAF applied. In general, the application of the MDAF metric results in ∼ 40% decrease in p(OH) from all considered sources at Auchencorth Moss. In all seasons, The diurnal variations of these seasonal averages are shown in Fig. 8, where the ∼ 40% reduction due to the MDAF application is clear in daylight hours. The maximum MDAF decrease in hourly measurements is 71%, corresponding to summertime O 3 photolysis, resulting in an adjusted p(OH) O3 of 1.90 × 10 6 radicals cm −3 s −1 . In comparison, p(OH) HONO in the same 295 month reaches a maximum of 1.03 × 10 6 radicals cm −3 s −1 at 09:00 (UTC; ∼ 5 hours after sunrise). Conversely, over the daylight hours in winter, p(OH) HONO consistently exceeds p(OH) O3 by a factor of ∼ 5. A similar pattern is observed in autumn, but at a lower magnitude (factor of ∼ 2.1). This is likely to be a consequence of the generally overcast conditions typical to Scotland, resulting in the shorter wavelengths of light necessary for O 3 photolysis (≤ 340 nm) readily scattering before reaching ground-level, c.f. longer wavelengths for HONO photolysis (305 ≤ λ ≤ 420 nm). This is also observed in spring 300 and summer (Fig. 8), where p(OH) HONO is greater in the early morning (∼04:00-09:00 UTC) and the evening (∼16:00-20:00 UTC). This is is surpassed by p(OH) O3 in the middle of the day, closely following the diurnal cycle of light intensity and peaking when shorter wavelengths are more prevalent. The same diurnal pattern of p(OH) O3 is observed in all seasons ( Fig.   9), but a difference of ∼ 25 is observed between peak p(OH) O3 in summer and winter (1.90 × 10 6 radicals cm −3 s −1 and 7.4 × 10 4 radicals cm −3 s −1 , respectively) despite similar concentrations of O 3 (61.0 µg m −3 and 57.1 µg m −3 , respectively).

305
The photolytic production of OH radicals presented here at Auchencorth Moss is lower than those reported in Melbourne, Australia for March 2017 (Ryan et al., 2018). Compared to the summertime p(OH) rates presented here, peak values of ClNO 2 concentration profile shown in Fig. 2. When using the former to estimate p(Cl), those calculated using j(ClNO 2 ) directly from TUV model output provide the largest estimation for all seasons. When the updated j(ClNO 2 ) values accounting for ambient temperature are used (by means of updated σ(ClNO 2 ); Ghosh et al., 2012), total p(Cl) decreased by ∼ 30% 320 compared to the TUV output. As with calculations of p(OH), application of the MDAF metric resulted in ∼ 40% reduction of p(Cl) when using both TUV and updated j(ClNO 2 ). An exception is in winter, where during the lowest rates of Cl production, the adjustment factor and updated j(ClNO 2 ) decrease p(Cl) calculated from the TUV output to an approximately equal quantity in the morning hours. This could be due to cooler temperatures in winter (3.99 ± 3.6 • C) leading to decreased σ(ClNO 2 ) c.f.
other seasons (6.76 ± 4.3 • C, 13.5 ± 3.7 • C and 7.47 ± 4.3 • C in spring, summer and autumn, respectively), so that it becomes 325 comparable to TUV results with the MDAF applied.
When the assumed diurnal cycle of ClNO 2 concentrations are used to derive p(Cl) (bottom panels of Fig. 10), the maximum rate of Cl radical production decreases by 59% to 7.98 × 10 4 radicals cm −3 s −1 compared with when the diurnal cycle of ClNO 2 is unaccounted for. When ClNO 2 concentrations drop close to zero (first occurring between 10:00-11:00 UTC), there is obviously little photolytic production of Cl. Early morning peaks in p(Cl) in all seasons are staggered, due to the between Cl and VOCs can be up to an order of magnitude larger than comparable rate coefficients with OH, Cl atoms have potential significant impacts on the radical budget at Auchencorth Moss.

Implications of the MDAF metric
Previous sections describe how application of the MDAF metric at Auchencorth Moss leads to a ∼ 40% reduction in the j(HONO), j(ClNO 2 ) and j(O 1 D) clear-sky values calculated by the TUV model, and consequently also in the radical produc-350 tion rates from these photolyses. In contrast, Sommariva et al. (2020) report measurements of j(NO 2 ) and j(O 1 D) made in Boulder, Colorado that were 25 − 30% greater than model estimates using the Master Chemical Mechanism (MCM) parameterisations. The difference was attributed to discrepancies between the measurement site and model output location (the latter is > 1 km lower in altitude, and further north by a latitude of 5 • ) and to temporal variation in measurements (different seasons and years). Taken together, however, these authors' findings combined with this study highlight the importance of matched 355 measurement and model location to generate a MDAF to derive molecule specific j-values.
Most local air quality modelling, e.g. for assessment of compliance and mitigation measures, include only simple chemical schemes with highly parameterised photochemistry (e.g. ADMS). In contrast to model outputted gas and particle concentrations, the values of the model photochemical variables often go unverified. In more detailed instances, atmospheric models typically use separate radiation models, such as the TUV model used in this study, Fast-JX, or less explicit photolysis schemes The implication of the MDAF approach illustrated in this study is that long-term multi-site radiometer measurements could fill a major gap in measurement and model knowledge. Implementation of such measurements into existing atmospheric trace 365 gas monitoring networks (e.g. AURN in the UK or EMEP in Europe), would provide invaluable data for understanding tropospheric photochemistry and radical production rates. Accurate j-values are integral to accurate assessment of air quality, in particular, to photolytic production of secondary pollutants which negatively impact both human and environmental health, including tropospheric O 3 and particulate matter. Radiometer measurements will consequently have immense value in supporting atmospheric chemistry measurement and modelling.

Conclusions
This study presents the first year of a long-term 4-π filter radiometer j(NO 2 ) measurement dataset at Auchencorth Moss, colocated with relevant ancillary measurements required to determine local radical production rates. Through illustration with the TUV model, it is demonstrated that long-term filter radiometer measurements can be used (1) to evaluate calculated j-values in current radiation models and parameterisations; and (2) to generate a straightforward measurement-driven adjustment factor (MDAF) for correcting clear-sky modelled j-values for variations in local meteorology (e.g. cloud, aerosol and surface albedo) that would otherwise be both computationally intensive and costly to model.
When the 4-π filter radiometer measurement is used to capture these changes in solar flux through the MDAF at Auchencorth Moss, it was found that clear-sky modelled j-values at Auchencorth Moss were reduced by ∼ 40% throughout the year.
Quantified production rates of OH radicals in this study differ by approximately one order of magnitude between summer and 380 winter, with the maximum rate in summer reaching 1.90 × 10 6 radicals cm −3 s −1 (Auchencorth Moss; 55 • N). At this rural background site, rate of OH radical production from photolysis of HONO exceeds that from photolysis of O 3 during low-light hours, particularly during the sunrise and sunset hours of spring and summer (SZA > 80 • ), and through the full diurnal cycle in winter and autumn (by a factor of 5 and 2.1, respectively). The enhanced contribution from HONO in the colder months is likely due to Scotland's considerably more overcast conditions in these seasons reducing the shorter wavelengths contributing 385 to O 3 photolysis relatively more than the longer wavelengths contributing to HONO photolysis.
Implementing radiometer measurements within existing long-term monitoring networks would provide a higher spatial density of model-measurement j-value comparison points in the UK. The consequent more accurate estimations of radical production rates would improve the quantification of subsequent radical-driven chemistry and secondary pollutant generation, and of all their associated impacts on human health, crops, ecosystems and radiative forcing.

390
Data availability. Filter radiometer measurements for the study year are available in the CEDA Archive (Walker et al., 2020). Concentrations of O3, HONO and HNO3 used were downloaded from UK-AIR (Defra, 2020a). Meteorological measurements used were provided by M.
Coyle et al. and are available for previous years in the CEDA Archive . The TUV model (NCAR, 2019) source code is available for download at: https://www2.acom.ucar.edu/modeling/tuv-download. All figures were created using the ggplot2 package for R (Wickham, 2016).

395
Author contributions. HLW, MRH, CFB and MMT devised the study. HLW and IS set up the filter radiometer. SRL, IS, and MRJ collected concentration and other ancillary measurements at Auchencorth Moss. RK ran the spectrophotometer used for calibration. MC provided meteorology data. HLW performed all TUV model runs and analysed the data, with help from MRH, CFB and MMT. The paper was written by HLW with contributions from all co-authors.  20 https://doi.org/10.5194/amt-2020-219 Preprint. Discussion started: 18 June 2020 c Author(s) 2020. CC BY 4.0 License. and number of measurements (n).