Precipitation response to Aerosol-Radiation and Aerosol-Cloud Interactions in Regional Climate Simulations over Europe

The effect of aerosols on regional climate simulations presents large uncertainties due to their complex and nonlinear interactions with a wide variety of factors, including aerosol-radiation (ARI) and aerosol-cloud (ACI) interactions. These interactions are strongly conditioned by the meteorological situation and the type of aerosol. Despite increasing, there is nowadays a very limited number of studies covering this topic from a regional and climatic perspective. Hence, this contribution aims at quantifying the impacts on precipitation of the inclusion of ARI and ACI processes in 5 regional climate simulations driven by ERA20C reanalysis. A series of regional climatic simulations (years 1991-2010) for the Euro-CORDEX domain have been conducted including ARI and ACI, establishing as reference a simulations where aerosols have not been included interactively (BASE). The results show that the effects of ARI and ACI on mean spatially averaged precipitation are limited. However, a spatial redistribution of precipitation occurs when introducing the ARI and ACI processes in the model; as well as some changes in 10 the intensity precipitation regimes. The main differences with respect to the base-case simulations occur in central Europe, where a decrease in precipitation is associated with a depletion in the number of rainy days and low clouds. This reduction in precipitation presents a strong correlation with the ratio PM2.5/PM10, since the decrease is specially intense during those events with high values of that ratio (pointing to high levels of anthropogenic aerosols) over the aforementioned area. The precipitation decrease occurs for all ranges of precipitation rates. On the other hand, the model produces an increase in precipitation over 15 the western Mediterranean basin associated with an increase of clouds and rainy days when ACI are implemented. Here the change is caused by the high presence of PM10 (low PM2.5/PM10 ratios, pointing to natural aerosols). In this case, the higher amount of precipitation affects only to those days with low rates of precipitation. Finally, there are some disperse areas were the inclusion of aerosols leads to an increase in precipitation, specially for moderate and high precipitation rates. 1 20 https://doi.org/10.5194/acp-2020-381 Preprint. Discussion started: 18 May 2020 c © Author(s) 2020. CC BY 4.0 License.

aerosols in the climate system (Jiménez-Guerrero et al., 2013), only a small number of scientific papers consider the analysis of climatic events using simulations that include ARI and ACI interactions, which may strongly condition the representation and definition of events associated with precipitation and cloudiness (Prein et al., 2015;Baró et al., 2018).
Traditionally, in regional climate models the representation of the radiative effect of aerosols (ARI) is established by a constant aerosol optical thickness (AOD) value and a predetermined and abundant number of cloud condensation nuclei (CCN) (Forkel et al., 2015) high enough for clouds to form without this variable being a limiting factor. To obtain a more realistic 60 model, ARI and ACI interactions, which require models in which meteorology-climatology, radiation, clouds and aerosol atmospheric chemistry are coupled in a fully interactive way, must be included in the simulation (Grell and Baklanov, 2011;Baklanov et al., 2014). Fully coupled climate-chemistry models (on-line) provide the possibility to explain the feedback mechanisms between simulated aerosol concentrations and meteorological variables.
In simulations including ARI, the number of CCN remains unchanged, but the concentration of aerosols and their impact on 2006). The surface layer is parameterized using the Jiménez et al. (2012) scheme. Finally, the land-soil model chosen to simulate the land-atmosphere interactions was the NOAH model (Tewari et al., 2004).
The gas-phase chemical mechanism used in WRF-Chem is RACM-KPP (Stockwell et al., 2001;Geiger et al., 2003) coupled 120 to GOCART aerosol scheme (Ginoux et al., 2001a;Chin et al., 2002). The photolysis module Fast-J (Wild et al., 2000) was used for feeding photochemical reactions. Biogenic emissions were online calculated using the Model of Emissions of Gases and Aerosols from Nature model (MEGAN) (Guenther et al., 2006). Dust and marine spray are simulated with GOCART (Ginoux et al., 2001b;Chin et al., 2002). Simulated aerosols include five species: sulfate, mineral dust, sea salt, organic matter and black carbon. Anthropogenic emissions are taken from the Intercomparison Project of Atmospheric and Climate Chemistry The simulated historical period for the three simulations covers from 1991 to 2010. Boundary and initial conditions were extracted from the ECMWF reanalysis: ERA20C (ECMWF, 2014;Hersbach et al., 2015), which has a horizontal resolution

Methods
This contribution focuses on the impacts of ARI and ACI on precipitation. Hence, the climatologies for precipitation amount, number of days with precipitation over a given threshold and cloudiness of the different experiments have been intercompared for BASE, ARI and ACI simulations. The data used to evaluate the added value of the aerosol experiments was the ERA5 The statistical significance of the differences among the climatologies reproduced by the simulations is checked by using a Bootstrap method with 1000 repetitions and a p-value < 0.05 was applied. More details about the method can be found in Milelli et al. (2010).
In order to assess the relationship between the obtained changes in precipitation and different variables representing the 145 aerosol load: PM10 (Particulate Matter <10µm), PM2.5 (Particulate Matter <2.5µm), AOD at 550nm (hereinafter AOD) the ratio between PM2.5 and PM10 (hereinafter called PMratio), several events (days) are grouped according to its intensity and extension. The intensity of an event is defined as the minimum value given by a threshold variable that the simulation cells must meet. The extension of the event is defined as the number of cells meeting the previous condition.
The relative differences among the experiments are represented in a two-dimensional heat map, where the axes denote the 150 extent and intensity. The number of days on which the criteria are met is indicated inside each element of the matrix. The total On the other hand, the effect of aerosols could depend on the area, and affecting in a different way weak and strong precipitation events (Rosenfeld et al., 2008). The series of relative differences between the ACI-BASE simulations have been generated for common and non-common days with rainfall exceeding a certain threshold for all points in the domain. The threshold ranges from 0 to 20mm/day on a non-regular basis (with a higher density of values near 0) with a total of 41 values.
In order to investigate areas where the effect of aerosols on precipitation could be different, a clustering method was applied 160 to the constructed series. The algorithm used for the spatial classification is similar to that used in other works (Jiménez et al., 2008;Lorente-Plazas et al., 2015). First, an analysis of principal components (Von Storch, 1999) is made, which is applied to the correlation matrix. Second, a two-step clustering method to a number of the retained principal components is applied. A hierarchical method is applied on a first basis; in this case, the Ward's algorithm (Ward Jr, 1963). This classification provides the number of clusters and the initial seeds (also called centroids) for the subsequent no-hierarchical method K-means which 165 optimizes the grouping (Hartigan and Wong, 1979). More details about the algorithm can be found in Lorente-Plazas et al. (2015). Finally the mean regional series are calculated as the average of time series belonging to a cluster (which corresponds to a spatial region in this study).

Results and discussion
The sensitivity of precipitation to the aerosol treatment in climate simulations is analyzed by comparing BASE, ARI and ACI 170 simulations over Europe during a 20 year period. The differences between ACI-BASE in spatially-averaged total precipitation are small, around 0.5%. Figure 2a shows the differences (percentage respect BASE) in the mean annual rainfall. The results depict a great spatial variability with differences ranging from 10% to -10%. Two zones with opposite behaviors are identified: (1) the central and eastern part of Europe, with a precipitation decrease up to 8% (statistically significant, p<0.05), and the Eastern Mediterranean area, with increases up to 10% (although changes are not significant, p> 0.05). In the rest of the domain, 175 there are other areas, such as the Iberian Peninsula, with a strong spatial variability (e.g. increasing rainfall on the Mediterranean coast and decreasing in the northeastern areas). Overall, the role of introducing ACI interactions leads to a spatial redistribution of precipitation. The differences between ARI and BASE simulations present a similar pattern (not shown).
In order to investigate the variations in the regimes of precipitation, the changes in the number of rainy days is estimated. Figure 2b shows the relative differences in the days with precipitation > 0.1mm. The patterns of differences are similar to 180 those of averaged precipitation, implying that the reduction in precipitation is mainly caused by the decrease in the number of rainy days. However, there are some noticeable exceptions. The relationships in the two large areas mentioned above are direct; that is, higher rainfall is linked to a larger number of precipitation episodes. However, there are areas where the relationship is inverse, more (less) number of days implies less (more) precipitation. The analysis of the low clouds in the domain (Figure 2c) shows a pattern similar to the aforementioned patterns. This may indicate that both the ARI and ACI effect can play very different roles on cloud properties and therefore on precipitation depending on the target area. This issue is addressed later.
Regarding the added value of incorporating aerosol physics into the model has been evaluated by analyzing the differences in precipitation, number of rainy days and low clouds between the simulations and the re-analysis of the European center ERA5 (Figure 2d-f). Overall, WRF-Chem (both in the BASE and ACI simulations), tends to underestimate precipitation over the European Mediterranean region and along the coasts of the Nordic countries, while overestimates rainfall in the rest of 190 the domain. These patterns are analogous for all the analyzed variables. If looking only at the areas where the differences are significant, ACI simulations slightly reduce the differences in the spatial distribution. However, the differences between ERA5 and ACI are much larger than the differences between ACI and BASE (not shown).
Despite this, as previously noted (Figure 2a-c), the ACI experiment introduces significant differences with respect to the BASE simulation over central Europe. These differences reach values about the 5% in the number of rainy days. Therefore, 195 a relationship between aerosols in these areas and the changes aforementioned might be expected. In order to understand the contribution of the different types of aerosol the differences in precipitation have been assessed by choosing a set of episodes.
The episodes were selected attending to the value of variables representative for the aerosols size and concentration (PM10 and PM2.5), their ratio (PMratio) and their impacts on radiation (AOD), as well as the spatial extension of the event.  accomplishing that condition (PMratio > 0.64 in more than 180 cells of the domain), the differences in rainy days over those cells is around 4%. Thus, e.g., the number of days in which PMratio is > 0.75 in more than 280 points is 1030 and the reduction in the number of rainy days is 8%. Following with the case of the PMratio (Figure 3e), the higher the intensity the greater the reduction in the number of rainy days; and the greater the extent of the event, the larger the reduction in rainy days (e.g. 210 reaching the maximum reduction around 15%).In fact, the multiple regression coefficient between the different variables is R = 0.80.
For AOD550 (Figure 3b), the results show that higher AOD550 values lead to a lower reduction in the number of rainy days.
However, in this case the changes are small (under 2%) and the relationships are not clear (R = 0.40). Results are analogous for PM2.5 (Figure 3c). However, relationships with the PMratio (Figure 3e) are important and significant. Therefore, an important 215 conclusion is that the variable with the largest impact on the number of rainy days is the PMratio in this area.
The physical explanation for this behavior in this area is that the higher the PMratio, the greater the concentration of small particles that change the properties of the clouds, mainly the low clouds (Figure 2c, reduction of low cloudiness over Central Region 5, both simulations give us a reduction in the number of days of precipitation. Therefore, both ARI and ACI affect precipitation in the same direction. ARI causes less radiation to reach the surface (Figure 6d). This inhibits convection and 255 therefore, a reduction in cloudiness. On the other hand, the higher concentration of small particles modifies the properties of the clouds, inhibiting precipitation processes again. Moreover, in the area of the Eastern Mediterranean ARI have hardly any impact on cloudiness (Figure 6d), but on the number of rainy days. Therefore, the effects in that area will be mainly due to the interaction of aerosols with clouds when acting as CCN. This effect can be clearly seen in other areas of Region 4, such as the Atlantic Coast of Scandinavia. Finally, there are areas where the effects of ARI and ACI tend to cancel each other, or have 260 different effects on small or large rainfall. For example, in the case of the southern Iberian Peninsula, the inclusion of aerosols leads to a reduction in the number of days of precipitation > 1 mm due to purely radiative effects (Figure 6c).
Finally, Figure 6f shows the relative differences between PM10 concentrations between ACI and ARI. The spatial pattern shows an area of positive differences over Central Europe and the western Mediterranean, except for the western Iberian Peninsula. Conversely, negative differences prevail in the rest of the domain; that is, the ARI simulation has lower concentrations 265 of PM10. Therefore, in Region 4 the increase in precipitation and cloudiness is associated with a decrease in PM10. In order to clarify this analysis, Figure 6g shows the relative difference in the concentration of PM10 between ACI and ARI, and the differences in the number of rainy days with precipitation > 1mm/day. The points are distributed in a quasi-random way with showing an increase in precipitation undergo a decrease in PM10. A plausible explanation is that, in these areas, the PM10 load is high due to the intrusion of desert dust and sea-salt aerosols. The difference between the ACI and ARI simulation is the activation of the aerosol-cloud interaction mechanism, using the aerosols calculated online as CCN to form clouds while in ARI, the CCN are a prescribed at a fixed value. The PM10 used to form clouds in ACI will no longer be counted in PM10 since of in-cloud scavenging. Therefore, a decrease in PM10 occurs and this decrease coincides with an increase in cloudiness.

275
In addition, the increase of precipitation will also decrease PM10 due to wet deposition. Note that the patterns are not completely coincident, with the precipitation pattern shifted slightly to the north (see the comparison in Figures 6e & f). This can be attributed to the displacement of the cloud masses in such area, which under conditions of heavy PM10 loads can have an important southern component.

280
The effect of aerosols on regional climate simulations still presents many uncertainties due to their complex and non-linear interactions that depends on a wide variety of factors. The quantity, size and color of aerosols modify the radiative balance and, therefore, many other derived variables such as local temperature, cloudiness or precipitation. In addition, the amount of moisture available will determine the size of the water droplets based on the amount and type of aerosols available. The size and color of the clouds will be affected, which will once again affect the radiative budget. In addition, this can spatially 285 redistribute precipitation regimes, allowing it to rain in different areas or provoking rainfall intensity to change. However, there is a lack of contributions that have studied these problems and the large increase in computational time needed to include ACI and ARI interactions in regional climate simulations have traditionally hampered the works covering this analysis from a climatic perspective.
In order to address these issues, a set of regional climate simulations have been conducted for the period 1991-2010 without 290 aerosol-atmosphere interactions (BASE), with ARI and with ACI parameterizations in an on-line coupled model. All simulations cover the domain of Europe defined by the Euro-CORDEX initiative. This analysis has focused on average precipitation, number of precipitation days larger than a certain threshold and cloudiness.
When introducing the ACI and ARI interactions, the spatial average of the total rainfall does not vary too much from the BASE scenario. However, there is a spatial redistribution of such precipitation. Although there are changes in many places in Precipitation in the Indus Watershed, Frontiers in Earth Science, 7, 210, 2019. Da Silva, N., Mailler, S., and Drobinski, P.: Aerosol indirect effects on summer precipitation in a regional climate model for the Euro-