The impact of cloudiness and cloud type on the atmospheric heating rate of black and brown carbon

The impact of cloudiness and cloud type on the atmospheric heating rate of black and brown carbon Luca Ferrero1,*, Asta Gregorič2,3, Grisa Močnik3,4, Martin Rigler2, Sergio Cogliati1,5, Francesca Barnaba6, Luca Di Liberto6, Gian Paolo Gobbi6, Niccolò Losi1 and Ezio Bolzacchini1 5 1 GEMMA and POLARIS Research Centers, Department of Earth and Environmental Sciences, University of Milano-Bicocca, Piazza della Scienza 1, 20126, Milan, Italy. 2 Aerosol d.o.o., Kamniška 39A, SI-1000 Ljubljana, Slovenia. 3 Center for Atmospheric Research, University of Nova Gorica, Vipavska 11c, SI-5270 Ajdovščina, Slovenia. 4 Department of Condensed Matter Physics, Jozef Stefan Institute, SI-1000 Ljubljana, Slovenia. 10 5 Remote Sensing of Environmental Dynamics Lab., DISAT, University of Milano-Bicocca, P.zza della Scienza 1, 20126, Milano, Italy 6 ISAC -CNR, Roma Tor Vergata, Via Fosso Del Cavaliere 100, 00133, Roma, Italy.


Introduction
The impact of aerosols on climate is traditionally investigated focusing on their direct, indirect and semi-direct effects (Bond et al., 2013;IPCC, 2013;Ferrero et al., 2018Ferrero et al., , 2014Bond et al., 2013;Ramanathan and Feng, 2009;Koren et al. 2008;Koren et al., 2004;Kaufman et al., 2002). Direct effects are related to the sunlight interaction with aerosols trough absorption and scattering, indirect effects are related to the ability of aerosol to act as cloud 55 condensation nuclei affecting the clouds' formation and properties, semi-direct effects are those related to a feedback on cloud evolution affecting other atmospheric parameters (e.g. the thermal structure of the atmosphere) (Ramanathan and Feng, 2009;Koren et al. 2008;IPCC, 2013;Koren et al., 2004;Kaufman et al., 2002).
Both the direct and indirect radiative effects on the climate caused by anthropogenic and natural aerosols still represent major sources of uncertainty (IPCC, 2013); for example the aerosol direct radiative effect (DRE), on a 60 global scale, may switch from positive to negative forcing on short (e.g. daily) time-scales (Lolli et al., 2018;Tosca et al., 2017;Campbell et al., 2016).
This is due to the fact that aerosol is a heterogeneous complex mixture of particles characterized by different size, chemistry, and shape (e.g., Costabile et al., 2013), greatly varying in time and space both in the horizontal and vertical dimension (e.g., Ferrero et al., 2012). On the global scale, most of the values reported for the DRE, used 90 of absorbing aerosols (i.e. Black Carbon, BC; Brown Carbon, BrC; or mineral dust) might have important effects on the radiative balance. It is estimated that, due to its absorption of sunlight, BC is the second most important positive anthropogenic climate-forcing agent after CO2 (Bond et al., 2013;Ramanathan and Carmichael, 2008), while BrC contributes ~10-30% to the total absorption on a global scale (Ferrero et al., 2018;Shamjad et al., 2015;Chung et al., 2012;Kumar et al., 2018). As a main difference compared to CO2, absorbing aerosols are short-lived 95 climate forcers, thus representing a potential global warming mitigation target. However, the real potential benefit of any mitigation strategy should also be based on observational measurements, possibly carried out in all sky conditions.
It also noteworthy that the HR induced by absorbing aerosol can trigger different atmospheric feedbacks. BC and dust can alter the atmospheric thermal structure, thus affecting atmospheric stability, cloud distribution and even 100 synoptic winds such as the monsoons (IPCC, 2013;Bond et al., 2013;Ramanathan and Feng, 2009;Koch et al., 2009;Ramanathan and Carmichael, 2008;Koren et al. 2008;Koren et al., 2004;Kaufman et al., 2002). Even in this case, the feedbacks should be quantified on the basis of HR measurements carried out in any sky conditions.
In agreement with the aforementioned points, both Andreae and Ramanathan (2013) and Chung et al. (2012) called for model-independent, observation-based determination of the absorptive direct radiative forcing (ADRE) of 105 aerosols. Since cloudiness and cloud type change on short time scales, long-term, highly time-resolved measurements covering different conditions, are necessary to unravel the role of absorbing aerosol on the HR.
Some satellite-based studies investigated the role of cloudiness and cloud type on the HR of aerosol layers above clouds (Matus et al., 2015). To our knowledge, there has been no experimental investigation on the impact aerosol layers laying below the clouds, where conversely most of the aerosol pollution resides. This study was performed 110 in Milan (Italy), located in the middle of the Po Valley (section 2), this region representing a pollution hotspot in Europe due to the high emissions coupled to a complex topology of the landscape. In fact, similarly to a multitude of basin valleys surrounded by hills or mountains in Europe, low wind speeds and stable atmospheric conditions are common, thus promoting high concentrations of aerosol and BC (Zotter et al., 2017;Moroni et al., 2013;Moroni et al., 2012;Ferrero et al., 2011a ;Carbone et al., 2010;Rodriguez et al., 2007). At the same time, cloud 115 https://doi.org/10.5194/acp-2020-264 Preprint. Discussion started: 18 May 2020 c Author(s) 2020. CC BY 4.0 License.
presence cannot be neglected considering that in the last 50 years annual mean cloudiness, expressed in oktas, is estimated to be ~5.5 over Europe (Stjern et al., 2009) and ~4 over Italy (Maugeri et al., 2001). This is in agreement with 80 years of data of cloud cover in the United States (Crock et al., 1999). Moreover, recently, Perlwitz and Miller (2010) reported a counterintuitive feedback linking the atmospheric heating induced by tropospheric absorbing aerosol to a cloud cover increase.

120
Due to the aforementioned reasons, this study attempts to experimentally unravel for the first time the impact of different cloud types on the HR exerted by aerosol layers. To this purpose we use a methodology, previously developed in Ferrero et al. (2018), and further extended the analysis to explore the effects of different cloud types on BC and BrC on HR.. More in detail, with respect to the preliminary results by Ferrero et al. (2018), this work introduces the following novelties: 1) the introduction of a cloud type classification; 2) the determination of the seasonal variation of aerosol concentrations within the mixing layer, well visible even from satellites (Ferrero et al., 2019;Di Nicolantonio et al., 2009;Barnaba and Gobbi 2004). A full description of the aerosol behavior in Milan at the University of Milano-Bicocca and the related aerosol properties (vertical profiles, chemistry, hygroscopicity, sources, and toxicity) are reported in previous studies (Diemoz et al., 2019;D'Angelo et al., 2016;Curci et al., 2015;Ferrero et al., 2015Ferrero et al., , 2010Perrone et al., 2013;Sangiorgi et al., 2011). Within the framework 140 of the present work is important to underline that the U9 experimental site is particularly well suited for atmospheric radiation transfer measurements, in fact it is characterized by a full hemispherical sky-view equipped with the instruments described in Section 2.1. The measurements assembly allow the experimental determination of the instantaneous aerosol HR (K day -1 ) induced by absorbing aerosol (e.g. BC and BrC) as detailed in Section 2.2. The methodological approach used to quantify the cloud fraction and to classify the cloud type is instead 145 reported in Section 2.3.

Instruments
At the U9 sampling site in Milan, the aerosol, cloud and radiation instrumentation ( Figure S1) needed to determine the HR (section 2.2), the cloud fraction and the cloud type (section 2.3) has been installed since 2015.

150
In particular, measurements of the wavelength dependent aerosol absorption coefficient babs(λ) in the UV-VIS-NIR region were obtained using the Magee Scientific Aethalometer AE-31. The reason of this choice (detailed in Ferrero et al., 2018) is related to the number and range of spectral channels (7-λ: 370, 470, 520, 590, 660, 880 and 950 nm) not available in other instruments (e.g. MAAP, PSAP, photoacustic) (Virkkula et al., 2010;Petzold et al., 2005). This spectral range is needed for the HR determination (section 2.2). It noteworthy that the Aethalometers take also the advantage of global long-term data series (Ferrero et al., 2016;Eleftheriadis et al., 2009;Collaud-Coen et al., 2010;Junker et al., 2006) that should allow in the future to derive historical data of the HR.
To account for both the multiple scattering (the optical path enhancement induced by the filter fibers) and the loading effects (the non-linear optical path reduction induced by absorbing particles accumulating in the filter), the AE-31 data were corrected applying the Weingartner et al. (2003) procedure (Ferrero et al., 2018(Ferrero et al., , 2014(Ferrero et al., , 2011 160 Collaud-Coen et al., 2010

165
Overall, the multiple scattering parameter C was 3.24±0.03 as obtained by comparing the AE31 data at 660 nm with a MAAP at the same wavelength ( Figure S2). This value lies very close to that suggested by GAW (2016)
The MRI resolves the UV-VIS-NIR spectrum (350 -1000 nm) in 3648 spectral bands for both the downwelling and the upwelling radiation fluxes. The MRI was equipped with a rotating shadow-band enabling to measure separately the spectra of the direct, diffuse and reflected radiation.
The reflected radiation originated from the Lambertian concrete surface(due to its flat and homogeneous 175 characteristics which well represents the average spectral reflectance of the Milano urban area; Ferrero et al, 2018).
Details of the MRI are reported in Cogliati et al. (2015).
Broadband downwelling (global and diffuse) and upwelling (reflected) radiation measurements were also collected using LSI-Lastem radiometers (DPA154 and C201R, class1, ISO-9060, 3% accuracy; 300-3000 nm). Diffuse radiation was measured using the DPA154 global radiometer equipped with a shadow band whose effect was 180 corrected (Ferrero et al., 2018) to determine the true amount of both diffuse and direct (obtained after subtraction from the global) radiation.
In addition to radiation measurements, temperature, relative humidity, pressure and wind parameters were measured using the following LSI-Lastem sensors: DMA580 and DMA570 for thermo-hygrometric measurements (for T and RH: range -30 -+70 °C and 10% -98%, accuracy of ± 0.1 °C and ± 2.5% sensibility of 0.025°C and 185 0.2%), the CX110P barometer model for pressure (range 800-1100 hPa, accuracy of 1 hPa) and the combiSD anemometer (range of 0 -60 m/s and 0-360°) for wind. noise ratio. The Nimbus 15k lidar-ceilometer is able to determine cloud base heights, penetration depths, mixing layer height and, with specific processing, vertical profiles of aerosol optical and physical properties (e.g., Haeffelin et al., 2011, Dionisi et al., 2018Diemoz et al., 2019aDiemoz et al., , 2019b. For the specific purpose of this study, exploitation of the U9 ALC data has been limited to cloud layering and relevant cloud base height as the system can reliably detect multiple cloud layers and cirrus clouds (Boers et al., 2010;Martucci et al., 2010;Wiegner et 200 al., 2014) within its operating vertical range (up to 15 km).
Global and diffuse radiation measurements, coupled with the ALC data were used to determine the sky cloud fraction and to classify the cloud types by following the methodology presented in the Section 2.3.

205
The instantaneous aerosol HR (K day -1 ) induced by absorbing aerosol is experimentally obtained following Eq. 1 using the methodology reported and validated in Ferrero et al. (2018). Here we briefly summarize the method and the reader is referred to the aforementioned publication for the physical demonstration of the approach.
The integral over the whole shortwave solar spectrum and over the whole 2π hemispherical sky of the interaction between the radiation (either direct from the sun, diffuse by atmosphere and clouds and reflected from the ground) 210 and the absorbing components of aerosol (BC and BrC in Milan, as detailed in Ferrero et al., 2018) gives the HR as: where r represents the air density (kg m -3 ), Cp (1005 J kg -1 K -1 ) is the isobaric specific heat of dry air, n is the index indicating the n th type of radiation (direct, diffuse or reflected) impinging the absorbing aerosol, λ and θ 215 represent the wavelength and zenith angle of the radiation, Fn(λ,θ) is the n th type (direct or diffuse or reflected) monochromatic radiation of wavelength λ that strikes with an angle θ the aerosol layer, μ is the cosine of θ (cosθ), babs(λ) is the wavelength dependent aerosol absorption coefficient.
Considering that the absorptive DRE (ADRE), i.e. the radiative power absorbed by the aerosol for unit volume of the atmosphere (W m -3 ), is equals to: Eq. 1 can be also re-written as: Both Eq. 1 and 2 can also be solved for each of the three components of radiation (direct, diffuse, or reflected), i.e.:
Eq. 4 and 5 allow to split the total ADRE and HR into the three components of radiation. As the intensity of these radiation components is a function of cloudiness and cloud type (section 2.3), Eqs. 4 and 5 enable to assess the 230 impact of the latter components on the aerosol absorption of shortwave radiation and thus on the corresponding HR (sections 3.2 and 3.3).
In addition, as the spectral signature of babs(λ) reflects the different nature of absorbing aerosol (BC and BrC), babs (λ) and thus the HR can be apportioned to determine the contributions of BC and BrC (HRBC and HRBrC), respectively.
This result can be achieved considering that BC aerosol absorption is characterized by an Absorption Angstrom 235 Exponent, AAE ≈1 (Massabò et al., 2015;Sandradewi et al., 2008a;Bond and Bengstrom, 2006). Conversely, BrC absorption is spectrally more variable, with an AAE from 3 to 10 (Ferrero et al., 2018;Shamjad et al., 2015;Massabò et al., 2015;Bikkina et al., 2013;Yang et al., 2009;Kirchstetter et al., 2004). This is due to the negligible BrC absorption in the infrared compared to UV. In this study we determined AAEBrC following the innovative apportionment method proposed by Massabò et al. (2015). This allows to apportion babs(λ) from BC and BrC at the 240 same time and to determine the AAEBrC assuming that the whole BrC is completely produced by biomass burning.
The method by Massabò et al. (2015) was successfully applied to the Milan U9 measurements leading to an average AAEBrC (over a full solar year) of 3.66±0.03, and to an associated HRBrC explaining 13±1% of the total HR (Ferrero et al., 2018). The apportionment of absorption coefficient also enables to investigate the role of clouds on different absorbing aerosol species. As already pointed out in Ferrero et al. (2018), it is worth recalling that in 245 the present method (equation 1), both the ADRE and the HR are independent from the thickness (Δz) of the investigated atmospheric aerosol layer. At the same time, BC and HR vertical profiles data previously collected both at the same site and in other basin valley sited (Ferrero et al., 2014) revealed that ADRE and HR were constant inside the mixing layer, The methodology is therefore believed to be valid for applications in atmospheric layers below clouds, assuming that near-surface measurements are representative of the whole mixing layer. Main 250 advantage of the new method to quantify the impact of clouds on the LAA HR is that it allows to obtain experimental measurement (not estimations) of ADRE and HR, which are continuous in time and resolved in terms of sources, species of LAA, cloud cover, and cloud types.

Cloud fraction
The cloud fraction was determined following the approach reported in Ehnberg and Bollen (2005). In particular, radiometer measurements were used to calculate the fraction of the sky covered by cloud in terms of oktas (N), overall leading to 9 classes, corresponding to values of N ranging from 0 (clear sky) to 8 (complete overcast situation). As reported in Ehnberg and Bollen (2005), the amount of global radiation (Fglo) can be related to the 260 solar elevation angle (π/2-θ) and to the cloudiness condition following the Nielsen et al. (1981) equation: where N represents one of the possible 9 classes of cloud fraction and a, a0, a1, a3 and L are empirical coefficients that enable to compute the expected global radiation for each oktas class (0 56789 ), at a fixed solar elevation angle (π/2-θ). Their values, extracted from the original work of Ehnberg and Bollen (2005), are summarized in Table   265 S1. Eq. 6 allows to determine the unique oktas value N by comparing the measured global radiation (Fglo) with Fglo-N at any given time.
Still, the so-derived cloud fraction can be used to evaluate the interaction between incoming radiation and light absorbing aerosol in cloudy conditions but without the possibility to discriminate between cloud type. The https://doi.org/10.5194/acp-2020-264 Preprint. Discussion started: 18 May 2020 c Author(s) 2020. CC BY 4.0 License.

Cloud classification
Cloud classes and cloud cover is by common practice still largely determined on the basis of human observations based on the reference standard defined by the World Meteorological Organisation (WMO). However, these 275 observations lack high time resolution information and are subjective. Due to high spatial and temporal variability of clouds, determination of cloud classes can be improved by measurements, adding highly temporally resolved and observer-independent information on the cloud base height and the magnitude of solar radiation.
In this study, clouds were classified coupling measurements of broadband solar radiation (global irradiance, Fglo) and lidar-ceilometer measurements. The full methodology is described below.

280
As first introduced in the study by , measurements of the magnitude of global solar irradiance and its deviation in 20-minute intervals can be used for cloud classification. Irradiance is used to calculate two quantities: 1) the ratio (R) between observed global irradiance (Fglo) and the modelled clear sky from the SD-R diagram alone. Therefore, the cloud classification was further improved in this study by including information from the automated Lidar-Ceilometer measurements on the cloud base height and the number of cloud layers. First of all, to avoid misclassification cases due to the presence of multiple cloud layers, we limited the analysis to those cases where only one cloud layer was detected by ceilometer (ALC). In this respect, the ALCderived cloud base height information allowed us to cluster clouds according to their altitude and distinguishing 310 between low level clouds (<2 km), mid-altitude clouds (2-7 km) and high-altitude clouds (>7 km). The cloud https://doi.org/10.5194/acp-2020-264 Preprint. Discussion started: 18 May 2020 c Author(s) 2020. CC BY 4.0 License.
altitude of each analyzed data is reported in Figure 3 within the SD-R diagram. It shows that, on average, low level clouds are located on the left side of the SD-R diagram(stratiform clouds), high-altitude clouds are conversely on the opposite side (this being the the region of Ci and Cu clouds); finally, mid-altitudes clouds density in the diagram mostly cover its the central part describing all the possible transitions/combinations from St to Cu and Ci, e.g.

315
altostratus (AlSt) altocumulus (AlCu). Figure 3 further shows that use of the clouds base height as a third source of information (in addition to R and SD) allows to better separate overlapping cases in the bi-dimensional, SD-R diagram alone.
Overall, coupling the SD-R plot and cloud base height, enabled us to identify seven classes: St (stratus), Cu (cumulus) and StCu (stratocumulus) as low level class; AlSt (altostratus) and AlCu (altocumulus) as mid-altitude 320 clouds; and Ci (cirrus) and CiCu-CiSt (cirrocumulus and cirrostratus) as high-altitude clouds. The final overview of the parameters (R, SD, cloud level) and their threshold values used for cloud classification is presented in Table   1. The final SD-R diagram with presentation of mean value and 99% confidence interval for R and SD of each cloud class, plus the clear sky (CS) case, is presented on Figure 4 while the same SD-R diagram with presentation of mean value and the standard deviation of each cloud classes, plus the clear sky (CS) case, is presented on Figure   325 S3. Note in particular that the overlapping in the standard deviation of each classes shown in the SD-R plot in

Average photon energy
The relative distribution of energy over the solar spectrum in the measured range of the MRI (350 -1000 nm) was also investigated for each cloud type calculating the average photon energy (APE) which describes the spectral characteristics of direct and diffuse radiation modulated by clouds. In fact APE quantifies the spectral shape of solar irradiance and represents the average energy of photons impinging upon a target, in this case the aerosol 335 layer close to the surface. Thus, single APE can identify a unique spectral irradiance distribution which describes the light available for absorption in different spectral regions. APE (expressed in eV) is calculated dividing the total energy in a spectrum by the total number of photons it contains (Norton et al., 2015), i.e.: where q represents the electron charge, Fn,λ is the n th type (direct, diffuse) radiation at wavelength λ (W m -2 nm -1 ),

340
and Φλ (photons m -2 s -1 nm -1 ) is the photon flux density at wavelength λ determined using the Plank-Einstein where h is the Plank constant and c the speed of light.
From Eq. (8) it follows that APE is normalized for the total amount of radiation, becoming thus independent from 345 the absolute intensity of light at each λ and indicating only the average distribution of light across the spectrum.
Particularly, higher APE values describe the shift of a radiation spectrum towards UV-blue region ( Figure S4). It has to be noted that the APE index depends on the range of the investigated spectrum (lower and upper limits of https://doi.org/10.5194/acp-2020-264 Preprint.

350
Characteristic APE values of diffuse (APEdiff) and direct (APEdir) irradiance measured from U9 site for different sky condition are presented in Section 3.4 together with a discussion concerning the relationship between APE and HR.

355
HR values considered in this study were measured over Milan from November 2015 to March 2016 are, as this period covers the simultaneous presence of radiation, lidar and absorption measurements fundamental for the analysis presented here (section 2). These data are presented in Section 3.1. The role of cloudiness and its influence on the HR is discussed in section 3.2 while section 3.3 describes the impact of each cloud type on the HR. In Section 3.4, the clouds impact on the HR is discussed with respect to the light absorbing aerosol species: BC and 360 BrC. All the data are reported everywhere as mean±95% confidence interval.

HR, eBC and radiation data
Monthly average values of eBC and HR are presented in Figure 5a while the corresponding numerical values of these and additional parameters (e.g. ADRE, babs) are also summarized in Table 2. Corresponding high time 365 resolution data (5 minutes) are shown in Figure S5.
The highest values of eBC (and babs(880nm) ) were found, as expected, in the middle of the winter, in December and confirm that eBC is the main driver for the behavior of HR and ADRE on the seasonal time scale.
However, in agreement with Eq. 1 (section 2.1), the interaction of absorbing aerosol with the impinging radiation cannot be neglected as heating rate varies differently than anticipated from the concentrations alone. In fact, during the investigated period, the ratio between maximum and minimum eBC monthly mean concentration (December to March, eBC ratio: 4.10±0.12) was higher than that of HR (2.65±0.16). This is because the incoming radiation These considerations introduce the importance of both amount and kind (direct, diffuse and reflected) of the radiation that interacts with light absorbing aerosol. In brief, any process able to influence the total amount and the kind of impinging radiation (e.g. presence/absence of clouds, cloudiness and cloud type) will result in a 390 different HR, even keeping constant eBC levels. The investigation of this aspect is the main focus and added value of this study and is reported in the next sections.

Cloud fraction impact on the hating rate
The first indication of the important role played by clouds on the HR can be derived from the contribution of the 395 diffuse radiation to the HR (HRdif) as reported in Figure 5a. It shows the monthly average values of HR, HRdir, HRdif and HRref revealing that the diffuse contribution accounted for 40±1% of the total HR. On a monthly basis, this was comparable or even higher than HRdir. The only exception was in November 2015 were a lower fraction of diffuse radiation was measured (Figure 5b) compared to the other months. In fact, in November, the average okta value was 2.91±0.06, lower than that observed in the other months (3.75±0.03), due to the highest frequency 400 of clear sky conditions. The aforementioned data demonstrate the importance of diffuse radiation and thus of cloudy days in determining the HR induced by the absorbing aerosol. In order to investigate the role of cloudiness, it is necessary to decouple the variability of the HR induced by radiation from that due to eBC concentrations. In HRdif/eBC and the diffuse radiation reached their minimum due to the capability of clouds to effectively attenuate the radiation. Yet, differently from the direct radiation, the HRdif /eBC is still not null (0.08±0.01 K m 3 day -1 μg -1 ) becoming the highest contributor of the total atmospheric HR, with a percentage of 84±1%.

415
The absolute values of the HR and its components as a function of cloudiness is shown in Figure 6b. conditions were present only 23% of the time, the remaining time (77%) being characterized by partially cloudy (35%, 1-6 oktas) to totally cloudy (42%, 7-8 oktas) conditions.

Cloud type impact on the heating rate
The previous section showed the importance of cloudiness in determining both the kind of the active radiation and 435 the suppression of HR with increasing the cloud cover. This is relevant as it was found that cloudy conditions are dominant in terms of frequency. Here we will further investigate the clouds-HR relationships by exploring the effect of different types of clouds on this relationship. Figure 7a shows how the overall 77% of cloudy conditions encountered during the observational period was composed by the different cloud types, revealing that these were mainly St (42%), followed by StCu (13%) Ci, CiCu-CiSt (7% and 5%, respectively). The contribution of each 440 cloud type to the cloudiness (expressed in oktas) of the sky is reported in Figure 7b. This clearly shows that, while

455
Exploring the relationship between cloud type and HR, we found a strong linear relationship between the mean cloudiness (in oktas) and the percent decrease of HR due to each cloud type with respect to the clear sky (CS) case ( Figure 10). These results were obtained by averaging the cloudiness (in oktas) for each cloud type (as detected in section 3.3) and computing the cloud-type resolved percentage decrease of LAA HR with respect to clear sky conditions. Overall, the derived linear regression (R 2 =0.96) indicates a HR decrease of about 12% per okta.

460
Knowledge of the dominant cloud types associated to the different cloud cover also allows us to associate this decrease to specific cloud types ( Figure 10). In particular, Ci are found to produce a modest impact on cloudiness (0.50±0.05 oktas) decreasing the HR by ~3%, while Cu (1.76±0.09 oktas) decrease the LAA HR by -26±8%.
CiCu-CiSt (oktas of 3.56±0.14) were responsible for a -49±6 decrease of the HR. Their impact was comparable to that of StCu (4.68±0.10 oktas, -48±4% of HR It is also worth to mention that not only the absolute value of the HR changes as a function of clouds in the atmosphere, but the presence of clouds also alters its diurnal pattern. In fact, as introduced in section 3.1, Fdir is scaled by μ in Eq. 1 (section 2.1) and thus it is perfectly constant along the day only in clear sky conditions.

470
Conversely, even when scaled by μ, the diffuse and reflected radiation linearly follow the behavior of irradiance Fdif and Fref (under the assumption of isotropic and Lambertian surface, Ferrero et al., 2018).
Thus any influence of clouds on Fdir, Fdif and Fref will reflect into the interaction between the radiation itself and the absorbing aerosol, changing the HR diurnal pattern . To illustrate this effect, Figure 11 shows the average diurnal pattern of the HR in both clear sky (blue) and cloudy conditions (red; oktas=7-8, dominated exclusively 475 by St and AlSt). This clearly shows that, while in clear sky conditions the HR exhibits an asymmetric diurnal pattern with a maximum around 10:00 LST, in cloudy conditions it shows a bell shape curve similar to that of Fglo (which is driven by the diffuse only component, which peaks at midday). As explained in more detail in Ferrero et al. (2018), the presence of the asymmetrical peak in clear sky conditions is due to the coupling between the eBC daily pattern (characterized by a morning rush hour peak) and that of Fdir/μ, that is constant in CS. This is not the 480 case in cloudy conditions when the most important radiation is Fdif.
A further important consequence of that change in the diurnal pattern of HR is that it reflects into related atmospheric feedbacks, such as the influence on the liquid water content (Jacobson et al., 2002), planetary boundary layer dynamics (Ferrero et al., 2014;Wang et al., 2018), regional circulation systems (Ramanathan and Carmichael, 2008;Ramanathan and Feng, 2009) and finally on the cloud dynamic and evolution itself (Koren et 485 al., 2008;Bond et al., 2013). Thus, any inappropriate use of clear sky assumption in models will also reflect on the modelled HR-triggered feedbacks.

The impact of clouds on the absolute and relative BC and BrC heating rates
As mentioned in the introduction, one of the key uncertain factors in climate change evaluations is the role played 490 by different species of absorbing aerosol, the two most important species being BC and BrC. In this work we thus investigate the contribution of these two species to the HR at our measuring site. In the previous sections we discussed the absolute intensity of HRBC and HRBrC. They varyis function of four main variables, namely: 1) the absolute absorption coefficient values (babs(λ)) of both BC and BrC, 2) the absolute magnitude of the impinging radiation (Fn(λ,θ)), 3) the different spectral absorption of BC and BrC, described by their AAE, and 4) the spectral 495 features of the impinging radiation (Fn(λ,θ)) described by the APE (section 2.3.3). Among these factors, the first two are the dominant ones. However, the presence of clouds influences both the absolute magnitude and the spectral feature of the impinging radiation (sections 3.2 and 3.3).
We first present the impact of cloudiness and cloud type on both HRBC and HRBrC considering the absolute values of babs(λ) and Fn(λ,θ) measured during the campaign (section 3.4.1). Then, in Section 3.4.2, we discuss the influence 500 of different sky conditions and cloud type on HR due to both BC and BrC, focusing on the radiation APE through a HRBC and HRBrC data normalization with respect to the absolute magnitude of the babs(λ) of both species.

The role of cloudiness
To complement the results in Figure 5a, the contribution of BC and BrC to the monthly averaged HR is reported 505 in Figure 12. On average, the HRBrC accounted for 13.7±0.2% of the total HR, the BrC being characterized by an AAE of 3.49±0.01, thus fully within ranges previously observed in other studies (e.g., Yang et al, 2009;Massabò Ferrero et al., 2018). In Figure 13, HRBC and HRBrC are reported as a function of the oktas (total HR in Figure 13a and the contribution of direct, diffuse and reflected HR in panels b-d, respectively). As expected, Figure 13a shows that both HRBC and HRBrC decreased with increasing oktas, going from the clear sky maxima  Figure   13b shows both HRBC,dir and HRBrC,dir to decrease as a function of cloudiness to negligible levels (HR<10 -4 K day -1 ) in overcast conditions. Conversely, HRBC,dif and HRBrC,dif increased for increasing oktas (Figure 13c), reaching 520 their maximum in partially cloudy conditions (at oktas=6, 0.51±0.01 and 0.09±0.01 K day -1 ) when also the maximum of Fdif was registered (section 3.2 and Figure 6a). Then, for further increasing cloudiness, they dropped Figure 13 also clearly shows that HRBC is always greater than HRBrC, as expected. However, a deeper investigation of the data reported in Figure 13 allows us to better describe the interaction between radiation and LAA in heating the surrounding atmosphere. To this purpose, it is particularly useful to compare the relative decrease of HRBrC 530 from clear sky to complete overcast situation to that of HRBC. The clouds, going from 0 to 8 oktas, decrease the HRBrC 12±6% more compared to HRBC. The same happened to HRBC,dir and HRBrC,dir. The diffuse component of the HR behaves differently: the clouds decrease HRBrC,dif 38±6% more compared to HRBC,dif.
At a first glance, Figure 13 could give the impression that BrC is more efficient in heating the surrounding atmosphere (with respect to BC) in clear sky conditions, compared to cloudy ones. Note however that, as stated at 535 the beginning of this section, any change of both BC and BrC babs(λ) in different sky conditions has to be accounted for to avoid any misinterpretation of the results.
In fact, we observed that at all wavelengths and for both BC and BrC, babs(λ) was not constant during periods with different cloudy conditions ( Figure S7). However, while the variability of babs(λ) BC with varying oktas was limited, this was not the case for BrC ( Figure S7a).Values of babs(λ) BrC in high cloud cover conditions were statistically 540 lower than the one in clear sky/moderate cloudy conditions (at oktas=8 the babs(λ) of BrC was on average -23±3% lower than in clear sky, Figure S7b). The full understanding of this behavior, perhaps linked to the formation of secondary BrC at high radiation in clear sky compared to cloudy ones (Kumar et al., 2018), is beyond the aim of the present paper. Here we focus the attention on the fact that the magnitude of babs(λ) of BC and BrC changed differently with cloudiness. This behavior explains why, at a first glance, the relative decrease of the HRBrC, from 545 0 to 8 oktas, was higher compared to that of HRBC. At the same time, the fact that the diffuse component of the HRBrC (HRBrC,dif) experienced a higher relative decrease (from clear sky situation to overcast ones) than those https://doi.org/10.5194/acp-2020-264 Preprint. Discussion started: 18 May 2020 c Author(s) 2020. CC BY 4.0 License.
observed for the total HRBrC asks for further investigation. Some insights into this behavior are given in the next Section. 550

The role of the average photon energy and cloud type
In order to decouple the variability of the HR induced by radiation from that due to babs(λ), both HRBC and HRBrC were normalized for the adimensional integral of the babs(λ) over the whole aethalometer spectrum. In this way, the magnitude of babs(λ) is accounted for along the whole spectrum avoiding the choice of an arbitrary λ as a reference for the normalization. Figure S8 reports the same data present in Figure 13a after the normalization for babs(λ) and

555
for the corresponding CS HRBC and HRBrC values. Results first show that the relative decrease of the HRBrC, from 0 to 8 oktas, was 12±6% lower compared to that of HRBC, or, in other words, it was the decrease the HRBC that was 12±6% higher compared to that of HRBrC A counter-intuitive consequence of this analysis is that, compared to CS, cloudy conditions suppress much more the HRBC with respect to HRBrC. The diffuse component of the HR was the only one that kept an opposite behavior after the normalization for babs(λ). The decrease of HRBrC,dif was 560 21±6% higher compared to that of HRBC,dif (from clear sky situation to overcast ones); however this value is lower than the 38±6% reported in section 3.4.1 before the normalization for babs(λ) meaning that, even at equal absorption, the diffuse component of radiation plays a role in affecting the BrC response. This means that cloudiness and clouds not only affect absolute values of both HRBC and HRBrC but they markedly affect their ratio.
This pattern can be related to the different APE (section 2.3.3) that the direct and diffuse radiations feature in 565 different sky conditions ( Figure 14). Higher APE values describe the shift of a radiation spectrum towards UVblue region and vice versa (section 2.3.3). Figure 14 shows that while APEdir slightly increases towards overcast conditions, APEdif strongly decreases going from clear sky to 8 oktas. The APEdif,dir behavior can easily be explained considering the features of the direct and diffuse radiation spectra ( Figure S9). In fact, in clear sky conditions, the diffuse radiation is characterized by a high density in the UV-blue high energy region with respect 570 to the direct radiation, which indeed is depleted in that region by the molecular Rayleigh scattering. APEdir in clear sky conditions is in fact 1.89±0.01 eV, lower than the 2.20±0.01 eV of APEdif ( Figure 14). Conversely, in cloudy conditions ( Figure 14 and Figure S9) Fdif,λ and Fdir,λ behave similarly and the APEdif values equals that of APEdir: 1.99±0.01 eV. The BrC has the capacity to absorb much more radiation in the UV-blue region (featuring higher AAE of 3.49±0.01, compared to ~1 of BC). It follows that, depending on sky conditions, different parts of the 575 absorption spectra are important for BrC relative to BC. In this respect, ∆APEdir (cloudy-CS) was 0.11±3*10 -3 eV while ∆APEdif was 2 times higher (0.22±2*10 -3 eV). This explains the behavior of HRBC,dir and HRBrC,dir and of HRBC,dif and HRBrC,dif after the normalization for babs(λ). However, they do not explain the behavior of the total HRBC and HRBrC with respect to cloudiness: the absolute amount of direct and diffuse radiation Fdir, Fdif (and not only their spectral feature) has to be accounted for. Thus, the APE for the total sky radiation was determined as a 580 weighted average with respect to the absolute amount of Fdir and Fdif in function of cloudiness expressed in oktas; results are reported in Figure 14 and clearly show an increasing APEtot from clear sky to cloudy conditions, approaching APEdif at okta=8. This APEtot feature explain the counter-intuitive property that cloudy conditions suppress much more the HRBC with respect to HRBrC, as shown above.
We have shown that different cloud types are responsible for the different cloudiness (Section 3.3 and Figure 7b).

585
It is worth to explore the relationship between cloud type and both HRBC and HRBrC as previously done for the total LAA HR ( Figure 10). Also in this case, the variability of the HR induced by radiation was decoupled from https://doi.org/10.5194/acp-2020-264 Preprint. Discussion started: 18 May 2020 c Author(s) 2020. CC BY 4.0 License.
that due to babs(λ) by normalizing HRBC and HRBrC for the adimensional integral of babs(λ) over the whole aethalometer spectrum. We found a strong linear relationship between the mean cloudiness (in oktas) and the percent decrease of both (BC and BrC) HRs with respect to those in clear sky conditions (Figure 15). These results

590
were obtained by averaging the cloudiness (in oktas) for each cloud type (as detected in section 3.3) and combining them with percentage decrease of HRBC and HRBrC (again averaged for each cloud type) with respect to clear sky conditions. Overall, the derived linear regression indicates for both HRBC and HRBrC a decrease of about 12% per oktas (with high R 2 ). Knowledge of the dominant cloud types associated to the different cloud cover also allows us to associate this decrease to specific cloud types. In particular, Ci were found to produce a modest impact on 595 cloudiness (0.50±0.05 oktas) decreasing the HRBC and HRBrC by ~1-6%, respectively. Instead, Cu (1.76±0.09 oktas) decreased the HRBC and HRBrC by -31±12% and -26±7%, respectively. CiCu-CiSt were associated to an averaged oktas of 3.56±0.14, and were responsible for a -60±8% and -54±4% decrease of the HRBC and HRBrC.