Significance of the organic aerosol driven climate feedback in the boreal area

Aerosol particles cool the climate by scattering solar radiation and by acting as cloud condensation nuclei. Higher temperatures resulting from increased greenhouse gas levels have been suggested to lead to increased biogenic secondary organic aerosol and cloud condensation nuclei concentrations creating a negative climate feedback mechanism. Here, we present direct observations on this feedback mechanism utilizing collocated long term aerosol chemical composition measurements and remote sensing observations on aerosol and cloud properties. Summer time organic aerosol loadings showed a clear increase with temperature, with simultaneous increase in cloud condensation nuclei concentration in a boreal forest environment. Remote sensing observations revealed a change in cloud properties with an increase in cloud reflectivity in concert with increasing organic aerosol loadings in the area. The results provide direct observational evidence on the significance of this negative climate feedback mechanism.

the air mass had spent over land and the amount of rain the air mass had experienced during the 96hour back trajectory had relatively low importance as explaining variables in the model (Supplementary Table 1).
The model was calculated also by using only data of air masses from one arrival sector at a time (i.e. three independent models, one per sector). Also in this case, for each of the three sectors, difference in BIC when removing temperature from the model was larger (16.5 % for clean, 8.2 % for eastern, 8.0 % for southern) compared to the difference in BIC when BC was removed (8.6 % for clean, 5.9 % for eastern, 7.1 % for southern). The leading role of temperature as an explaining variable for OA mass loading was especially clear for air masses arriving from the clean sector, while for air masses arriving from eastern and southern sectors difference in the importance between temperature and BC was smaller. The temperature dependent increase in OA mass loading estimated with these separate models for each sector were 0.23 μg m -3 °C -1 for clean, 0.27 μg m -3 °C -1 for eastern and 0.21 μg m -3 °C -1 for southern sector. These values are close to the value 0.24 μg m -3 °C -1 obtained with the model including data for all air mass arrival sectors. It should be noted that for the purpose of the statistical model, the eastern arrival sector was split in to two (30-60° and 60-180°) to separate the air masses arriving from northeast above the Kuala peninsula from the rest of the eastern sector and to improve the model performance. The results described above for the model on eastern sector refer to the sector 60-180°. The model results for air masses from the northeast (30-60°) were inconclusive due to low number of data.
These results from the multivariate mixed effects model support the conclusion that the anthropogenic pollution or biomass burning emissions are overall in a minor role in the observed increase of OA mass loading with temperature, and the trend is dominated by the increased biogenic secondary organic aerosol (BSOA) formation.

A.2 Solar radiation and cloudiness
Field data on OA mass loading were compared to aerosol optical thickness which was obtained from remote sensing observation only in clear sky conditions. To address the possible systematic influence of this, we divided the hourly data based on brightness parameter (BP) value (i.e. the ratio between the measured global radiation and the theoretical maximum global radiation intensity) in to "cloudy" (BP < 0.3) and "clear sky" (BP > 0.7) cases. Supplementary Fig. 8 shows that the on average temperature was lower in cloudy cases compared to clear sky cases and also the average concentrations differed between the two cases. However the increasing trend of OA mass loading with temperature is visible also for both brightness class separately.
The division of data in cloudy and clear sky cases demonstrates also the importance of changes in temperature dependent biogenic volatile organic compounds (BVOC) emissions in BSOA formation.
Enhanced BSOA formation can be driven both by increased BVOC emissions and by increased oxidation due to the intensified UVB radiation. While UVB radiation intensity and temperature are positively correlated at the measurement site in summer, our analysis on cloudy and clear sky data (Supplementary Fig. 8) suggests that the increase in OA mass loading is dominated by the increased temperature dependent BVOC emissions over the increased oxidation. At a same temperature, both OA mass loading and number concentration of particles larger than 100 nm (N100) tend to be lower at clear sky conditions compared to cloudy conditions. This is opposite to what would be expected if oxidation was driving the enhanced BSOA formation.

A.3 Cloud condensation nuclei concentration
Number concentration of particles larger than 100 nm calculated from the size distribution data were used as an approximation for cloud condensation nuclei (CCN) number concentration in this analysis.
This approximation was necessary due to the lesser availability of CCN concentration data due to the time resolution of four hours and gaps in the measurements. The CCN concentration measured with a CCN counter at 0.2 % supersaturation shows a temperature dependence similarly to N100 ( Supplementary Fig. 7) although CCN concentration has a less steep slope with temperature compared to N100. The correlation coefficient between the measured CCN concentration and N100 was 0.80.

A.4 Cloud properties and particle concentration
In the main text, cloud albedo feedback was calculated based on cloud properties and OA mass loading. The cloud albedo feedback was calculated in similar manner also based on number concentration of particles larger than 100 nm (N100) instead on OA mass loading ( Supplementary Fig.   6). In this case the data in each cloud water path category was divided to low (< 340 cm -3 ) and high (> 633 cm -3 ) N100 classes. The differences between the N100 classes were statistically significant (t-test with 95 % confidence level) in the six highest cloud water path categories. From the statistically significant changes in cloud optical thickness (τc) we estimated the corresponding change in cloud albedo using Eq. (2). Then, using this calculated change in albedo in Eq. (3) we estimated the CAE which was -2.59 W m -2 (95 % confidence interval -4.51 --0.19 W m -2 ). The albedo change corresponds to a median increase of 568 cm -3 in the N100. The change in N100 was calculated as the average difference between the medians of the low (233 cm -3 ) and high (808 cm -3 ) N100 classes in the cloud water path categories. As the N100 was shown to increase by 88 cm -3 °C -1 , the difference between the low and high N100 (568 cm -3 °C -1 ) corresponds to a temperature difference of 6.4 °C. Consequently, the temperature dependent cloud albedo feedback is -0.40 W m -2 °C -1 (95 % confidence interval -0.70 --0.03 W m -2 °C -1 ) which is in the same range but slightly stronger than the estimate based on OA mass loadings.
Similar investigation was attempted also using measured CCN number concentration instead of N100.
The results based on CCN number concentration were consistent with those derived using N100, however the differences were not statistically significant due to low number of data. Supplementary Table 1. Significances of the explaining variables in variability of OA mass loading. Bayesian information criterion (BIC) value and the relative difference in BIC (diff. %) compared full model when one variable at a time is removed from the model. CO, NOx and black carbon (BC) concentration and temperature are in-situ observations. Cumulated rain, time over land and arrival sector were calculated from the trajectory data. Results are presented for model applied for the whole data and for model applied separately for data from the different air mass arrival sectors.  Supplementary Figure 6. Cloud properties and N100. a) Cloud effective radius and b) cloud optical thickness divided based on level of cloud water path. Data is divided to low (< 33 rd percentile (340 cm -3 ), blue) and high (> 66 th percentile (633 cm -3 ), red) number concentrations of particles larger than 100 nm. The box shows the quartiles of the dataset while the whiskers show the rest of the distribution, except for points that are determined to be "outliers" using a method that is a function of the inter-quartile range. The notch in the box displays the confidence interval around the median. The blue and red numbers above each figure indicate the number of data points in each box.