Dynamics

Abstract. Climate–aerosol interactions in India are studied by employing the global climate–aerosol model ECHAM5-HAM and the GAINS inventory for anthropogenic aerosol emissions. Model validation is done for black carbon surface concentrations in Mukteshwar and for features of the monsoon circulation. Seasonal cycles and spatial distributions of radiative forcing and the temperature and rainfall responses are presented for different model setups. While total aerosol radiative forcing is strongest in the summer, anthropogenic forcing is considerably stronger in winter than in summer. Local seasonal temperature anomalies caused by aerosols are mostly negative with some exceptions, e.g., parts of northern India in March–May. Rainfall increases due to the elevated heat pump (EHP) mechanism and decreases due to solar dimming mechanisms (SDMs) and the relative strengths of these effects during different seasons and for different model setups are studied. Aerosol light absorption does increase rainfall in northern India, but effects due to solar dimming and circulation work to cancel the increase. The total aerosol effect on rainfall is negative for northern India in the months of June–August, but during March–May the effect is positive for most model setups. These differences between responses in different seasons might help converge the ongoing debate on the EHPs and SDMs. Due to the complexity of the problem and known or potential sources for error and bias, the results should be interpreted cautiously as they are completely dependent on how realistic the model is. Aerosol–rainfall correlations and anticorrelations are shown not to be a reliable sole argument for deducing causality.


Introduction
Aerosols in India have a significant impact on human health, the environment and through their effects on regional climate, on agriculture and other aspects of society. By scattering and absorbing radiation and modifying cloud properties, aerosols can affect heating at different height levels and the hydrological cycle.
vection, cloud formation and rainfall. Despite having some observational support (Lau and Kim, 2006Bollasina et al., 2008), the EHP hypothesis has received also controversial responses (Kuhlmann and Quaas, 2010;Bollasina, 2010, 2011), a large part, but not all, of them involving aerosol and rainfall properties during March-May, before the monsoon season. The sea surface temperature (SST) gradients take time to adjust to adjust to aerosol forcing and based on simulations coupling an atmospheric model to a slab ocean model, it has been suggested that the slow response through the ocean would be more important for precipitation (Ganguly et al., 2012a, b). Investigating the links from aerosol loading to effects on precipitation is a complex task due to a multitude of relevant, interplaying effects, many of which are 15 non-local in space an time (Lau et al., 2008;Bollasina et al., 2009).
In this article, we study aerosol-climate interactions with help of the global climateaerosol model ECHAM5-HAM. Our goal is to provide a more detailed breakdown of the spatial distributions and seasonal cycles of the aerosol effects than before. We will separate direct and indirect effects as well as absorbing from scattering and modification of 20 cloud properties. We will investigate the SDM and EHP hypotheses and provide our input to the ongoing discussion, especially regarding the strength of different competing effects and their relative importance during different seasons.
The article is organised as follows: Sect. 2 describes the model, the emission inventory and the simulations done. Section 3 describes aerosol climatology and the 25 comparison with measurements. Section 4 presents radiative forcing in the simulations. Section 5 presents aerosol effects on rainfall, followed by conclusions in the last section. 18033

The model, the emission inventories and simulations
The ECHAM5-HAM model (Stier et al., 2005;Roeckner et al., 2006b) is an atmospheric general circulation model (GCM) coupled with an aerosol model simulating five aerosol species in seven log-normal modes. The simulated species are sulfate, black carbon, organic carbon, mineral dust and sea salt. Aerosol transport, chemistry and removal 5 are simulated. The aerosols effect the climate through their impact on shortwave radiation. Optionally, cloud activation by aerosols can be simulated. The model also includes an option of nudging. The model is described in more detail in (Stier et al., 2005) and simulated large-scale aerosol distributions in India and China in (Henriksson et al., 2011), where the model was evaluated against MODIS AOD seasonal cycles and spa-10 tial distributions and other measurements and shown to qualitatively reproduce largescale aerosol properties in India and China. The Ganges valley with large amounts of biomass burning aerosol containing absorbing material was one of the two areas where correspondence with MODIS results was not that good, which could imply that either the simulation or MODIS results or both are inaccurate. For additional model 15 evaluation, we will compare the modeled BC concentrations with those measured in Mukteshwar on the slopes of the Himalayas (29 • 26 N, 79 • 37 N), an important area where biomass burning aerosols emitted in the Ganges valley get transported. The Mukteshwar measurement results have been analysed earlier (Hyvärinen et al., 2009;Komppula et al., 2009;Hyvärinen et al., 2011a,b;Neitola et al., 2011) but have not 20 been compared to climate model results before.
The model was run at horizontal resolution T42 (grid-spacing of 2.8 • ) and 19 levels in the vertical in a hybrid sigma/pressure coordinate system, with the top level at 10 hPa. Sea surface temperatures were prescribed using data for years 2005-2010 from a simulation with the coupled model ECHAM5-MPIOM (Roeckner et al., 2003, SST changes might influence the climate response, a simple modification of the SST distribution in the Indian Ocean is included in two of the experiments described below. A model coupled with a mixed layer ocean model would have been available, but it has not been evaluated against aerosol observations and, more importantly, SSTs in such models react only locally and possibly too strongly to radiative forcings (Anonymous 10 Referee #2 (2011) tributed into RCP (Representative Concentration Pathways) sectors (energy, industry, solvent use, transport, agriculture, cropland burning, residential combustion, and waste treatment) and further spatially allocated into 0.5 • × 0.5 • longitude-latitude using RCP consistent proxies as used and further developed within Global Energy Assessment project (GEA, 2012).

5
The energy and domestic sector emissions are resolved on monthly scale. Monthly temporal patterns for the domestic heating and cooking were developed as well by combining the stove use assumptions presented by Streets et al. (2003) with the global gridded temperature fields from the CRU3.0 archive of monthly mean temperatures in 2005 (http://badc.nerc.ac.uk/data/cru/) (Brohan et al., 2006). 10 The international shipping emissions were developed using two data sources: (1) the global shipping from RCP (Eyring et al., 2010;Lamarque et. al, 2010), and (2) a separate arctic area emissions database developed by Corbett et al. (2010). Some overlaps between the spatial grids were observed and in those cases the Corbett et al. (2010) emission values were used. 15 The global and Indian emissions used as input for the climate modelings are presented in the Supplement.
The wildfire emissions are based on GFED 3 emission database . To be consistent with the GAINS model emissions, we use the wildfire data for the year 2005. GFED 3 emissions include an agricultural wasteland burning sector, which is 20 also present in the GAINS emissions. However, in this work, the GAINS agricultural waste burning emissions are used. Of natural emissions in the model, mineral dust, sea salt and ocean DMS emissions are calculated online and others are prescribed. Natural emissions from wildfires were taken from the GFED3 inventory (van der Werf et al., 2010), terrestial biogenic DMS emissions in the model are prescribed according 25 to Pham et al. (1995). SO 2 emissions from volcanoes are based on Andres and Kasgnoc (1998) and Halmer et al. (2002). SO 2 and biogenic POM emissions on Guenther et al. (1995).
A total of nine simulations were made in this study ( Table 1). Five of the simulations (MAIN, ZERO, NOABS, SSTMODIF and NUDGE) were made without aerosol cloud activation. Thus, in these simulations the indirect effect of aerosols is not included, while the direct radiative effects and semidirect effect on cloudiness are considered. The MAIN simulation included anthropogenic emissions based on the GAINS 5 inventory, while in the ZERO simulation, anthropogenic emissions were excluded. The NOABS, SSTMODIF and NUDGE simulations included anthropogenic emissions from the GAINS inventory but differed from the MAIN simulation as follows. First, in NOABS, aerosol single-scattering albedo was set to 1, which eliminates aerosol absorption. The difference between the MAIN and NOABS simulations thus represents the impact of aerosol absorption. Second, in SSTMODIF, SSTs were modified to study the effect of a potential aerosol-induced cooling of the Northern Indian Ocean (NIO) compared to the equatorial Indian Ocean. Based on the observation that equatorial Indian Ocean has warmed, since 1950s, by roughly 0.5 K more than the Northern Indian Ocean (Ramanathan et al., 2005) a negative SST perturbation increasing linearly from 0 K at the equator to 0.5 K at 20 • N was added to the baseline SST field. Finally, in the NUDGE simulation, simulated vorticity, divergence, temperature, and surface pressure were nudged towards ERA-INTERIM reanalysis data (Dee et al., 2011).
The remaining four simulations MAIN_ACT, ZERO_ACT, NOABS_ACT, and SST-MODIF_ACT included aerosol cloud activation according to Lin and Leaitch (1997). 20 Otherwise, these simulations were similar to MAIN, ZERO, NOABS and SSTMODIF, respectively. Conducting these simulations both with and without aerosol activation is useful for assessing which aspects of the climate response are robust to changes in model formulation. All simulations were run for the years 2005-2010 (the years are not that important as emissions are at 2005 levels during all years). The first year of each 25 simulation was discarded as a spin-up period when analyzing the results.

BC concentrations in Mukteshwar
Emissions of carbonaceous aerosols are more uncertain that those of sulfate aerosols (Ohara et al., 2007;Klimont et al., 2009) and validating the concentrations is thus an essential part of raising confidence in the model results. Figure 1 shows a comparison of simulated daily surface BC concentrations interpolated to Mukteshwar with 5 measurements. The simulation is NUDGE and the year is 2006. The time series follow each other quite well with a minimum in the winter, maxima in the spring and fall and a drop in the monsoon months, although the decrease starts later in the simulation. The precipitation field is not nudged and the discrepancy is most likely due to significantly more wet removal in the real situation. Figure 2 shows multi-year monthly 10 averages of the measured and modeled concentrations for the MAIN_ACT simulation including aerosol cloud activation. In this case, the drop in concentrations in the monsoon months of June and especially July and August are seen also in the simulation, though not as strongly as in the measurements. Modeled concentrations are larger in the winter months January and especially December, while spring and fall maxima 15 are seen both in the simulation and in measurements. In general, BC and OC concentrations are smaller than in the previously published simulations using the REAS emission inventory (Henriksson et al., 2011) due to different emissions and a slightly different model version (we did not track down the detailed reasons behind the discrepancy, but a comparison between the AOD seasonal cycle in previously and presently 20 presented simulations can be found in Fig. S1b).

Radiative forcing and temperature response
In this section, we present estimates for aerosol radiative forcing and the surface temperature response in the model. The seasonal cycle of radiative forcing in the simulations with GAINS emissions is shown in Fig. 3  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | on radiation; the effect including feedbacks will be discussed below. The anthropogenic forcing can be estimated as the difference between the radiative forcing in a simulation containing anthropogenic aerosols and that in the simulation with natural aerosols only. The natural aerosols are causing total negative radiative forcing to be strongest in the summer, but anthropogenic forcing is much stronger in the winter than in the summer. 5 In the summer months, the Indian average negative forcing is only a few tenths of watts per square meter. Figure 5 shows the spatial distribution of annual-mean anthropogenic forcing. Figure 6 shows the total radiation (shortwave plus longwave) anomalies, which include also the effect on cloud cover through semi-direct and other feedbacks affecting 10 the radiative balance (neglecting those adopting slowly e.g. due to ocean thermal inertia), in different simulations with GAINS emissions with reference to simulations without anthropogenic emissions. The surface anomalies are negative in all cases and so are the TOA anomalies in the simulation with aerosol cloud activation included, but in the simulation without aerosol cloud activation, the TOA radiation anomalies in the summer 15 are positive. Figures 7 and 8 show anomalies of the 2 m air temperature in different seasons for the simulations with GAINS emissions both without and with aerosol cloud activation, respectively, with the simulations with no anthropogenic emissions used as reference. Anomalies are similar with and without aerosol activation included and mostly negative, 20 except for Northern India in March-May. A warming tropospheric temperature trend has been observed for these months in the western Himalayas (Prasad et al., 2009), could partly be explained by aerosols according to our results (also see height-resolving plots in the next section) and is argued to be consistent with the EHP hypothesis (Gautam et al., 2009;Lau and Kim, 2010 Figure 9 shows the seasonal cycle of rainfall in the area 65-90 • E, [20][21][22][23][24][25][26][27][28][29][30][31][32][33][34][35] • N in the different simulations. It can be seen that the model simulates significantly more rainfall in Northern India when aerosol cloud activation is included compared to when it is not included. Without cloud activation, rainfall in the monsoon months June-August is larger 5 with anthropogenic emissions included than without, smaller than either in the simulation with anthropogenic emissions but without absorption and smallest in the simulation with modified SSTs. With cloud activation included, turning the anthropogenic emissions to zero does not seem to reduce monsoon rainfall, whereas turning off absorption reduces it a lot and modifying the SSTs reduces it somewhat. With cloud acti- troposphere seemingly warms enough to cause a positive anomaly to the mean vertical motion there. Specific humidity in the troposphere decreases in June-September in the MAIN_ACT and SSTMODIF_ACT simulations, probably largely due to reduced evaporation, but increases during the pre-monsoon months and also in the monsoon months for all the other cases considered. Thus, with enough dimming present in these simu- 25 lations with cooling by the aerosol indirect effect, reduced humidity due to decreased evaporation could also be an important part of the aerosol effect on rainfall. Relative humidity anomalies follow the cloud cover anomalies quite strongly. Introduction  Figure 12 shows rainfall in more southern parts of India: 10-20 • N, 75-80 • E. In these parts, there is high rainfall for a longer time during the year and, in contrast to Northern India, more rainfall in the simulations without aerosol cloud activation. Effects of aerosols are somewhat similar, but not the same as in Northern India. Absorption increases rainfall in most months with higher rainfall both with and without cloud activa-5 tion. Without anthropogenic emissions, the rainfall is clearly decreased without cloud activation and not changed much with cloud activation included. Modifying SSTs decreases the rainfall in both cases, especially when cloud activation is not included.

Aerosol effects on rainfall
Yet, one more point we want to make that to our knowledge has not been presented in the literature before: interannual variations in aerosol load and precipitation can cause 10 correlations of different sign in different months. In the area 20-35 • N, 65-90 • E in some years, winds in May are more westerly, bringing more dust and at the same time, drier air, implying a higher AOD but a lower precipitation, indicating a negative correlation between AOD and precipitation shown in Fig. 13a. At the same time, the correlation is strongly positive in July, shown in Fig. 13b, explained by aerosol hygroscopic growth 15 during years with more precipitation accompanied by a high relative humidity in general and a resulting increased AOD. This is illustrated further in Fig. S7. The interannual variability in aerosol parameters and precipitation is in this case driven by prescribed SSTs and the weather produced by the model.

20
The Indian climate was simulated in a series of nine simulations applying the aerosolclimate model ECHAM5-HAM and the GAINS emission inventory. The model has been evaluated against observed aerosol optical properties and surface concentration measurements earlier. Simulations were performed with and without aerosol cloud activation separately, with and without absorption of shortwave radiation, with and without 25 artificially cooled sea surface temperatures in the Northern Indian Ocean and one simulation was performed using the ERA-Interim reanalysis weather for nudging. Total Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | negative aerosol forcing at the TOA was strongest in the summer and anthropogenic forcing, including that of light-absorbing BC, is strongest in the winter. While the Indian average forcing was found negative in all months and all model setups, in the simulation with GAINS emissions and no aerosol cloud activation there were locations with positive mean forcing. In the summer months, the Indian average radiation anomalies 5 at the TOA were positive due to the semi-direct aerosol cloud effect and other effects turning the small negative forcing in to a positive anomaly in the radiative balance. Seasonal temperature anomalies were mainly negative and locally of the order of up to 2 • , but especially in the pre-monsoon months of March-May there were local positive anomalies north of 25 • N both with and without aerosol cloud activation. 10 As for rainfall, the model serves as a tool for separately studying different, opposing effects and to study separate mechanisms with, of course, scope for model development and evaluation remaining. Our results provide support for the elevated heat pump (EHP) mechanism, which increases vertical velocity and monsoon rainfall in Northern India. However, circulation changes as well as reduced surface evaporation caused by 15 the aerosols seem to nearly or more than cancel out these effects and the total effect of aerosols on rainfall is negative, reaching ∼ 20 % with cloud activation considered and sea surface temperatures modified. A small increase or decrease in precipitation in March-May was observed due to absorption, depending on the model setup, connected with decreased cloudiness due to semi-direct aerosol effects during these 20 months. This could help resolve apparent contradictions between different earlier results in the literature. We hope to have contributed significantly to the ongoing discussion about the EHP and SDM mechanisms by making simulations with setups allowing to separately study the effects of aerosol light absorption, total atmospheric effects of antropogenic aerosols as well as relative cooling of the Northern Indian Ocean as- 25 sumed to have happened mostly because of aerosols during the past decades. In all, we have presented a model analysis of seasonal cycles and spatial distributions of aerosol radiative forcing and aerosol effects on temperature and precipitation in India. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Bollasina, M. and Nigam, S.: Absorbing aerosols and pre-summer monsoon hydroclimate variability over the Indian subcontinent: the challenge in investigating links, Atmos. Res., 94, 338-344, 2009. 18033 Bollasina, M., Nigam, S., and Lau, K. M.: Absorbing aerosols and summer monsoon evolution over South Asia: An observational portrayal, J. Climate, 21, 3221-3239, 2008. 18033 5 Brohan, P., Kennedy, J. J., Harris, I., Tett, S. F. B., and Jones, P. D.: Uncertainty estimates in regional and global observed temperature changes: a new data set from 1850, J. Geophys. Res., 111, 12106, doi:10.1029/2005JD006548, 2006 Bond, T. C., Ramanathan, V., Jamroensa, A., and Marrapu, P.: Asian aerosols: current and 10 year 2030 distributions and implications to human health and regional climate change, Environ. Sci. Technol., 43, 5811-5817, 2009