HTML
--> --> --> -->2.1. First indirect effect in RegCM4.4
In RegCM4.4.1, all aerosols have only direct effects on the climate, except SO42-, which has direct and first indirect effects. The optical properties of the clouds depend on the r e, which is calculated as a function of temperature and the type of liquid-phase cloud (e.g., maritime versus continental) (Giorgi et al., 2012). To represent the first indirect effect, r e (μm) is represented as a function of cloud droplet number concentration (N c; cm-3), as in the formula of (Martin et al., 1994) [Eq. (1)], which is related to the total mass mixing ratio of SO42- using the empirical relationship derived by (Hegg, 1994) [Eq. (2)]: \begin{eqnarray} r_{\rm e}&=&\left(\dfrac{3w_{\rm L}}{4\pi \rho_{\rm w}kN_{\rm c}}\right)^{\frac{1}{3}} ,\ \ (1)\\ N_{\rm c}&=&10^6[90.7(10^9\rho_{\rm a}x_{\rm tot})^{0.45}+23]\rho_{\rm w} ,\ \ (2) \end{eqnarray} where ρ a and ρ w are the densities of air and water, respectively (kg m-3); x tot is the mass mixing ratio of total aerosols (kg kg-1); w L is liquid water content (kg m-3); and k=0.80 for maritime air masses and 0.67 for continental air masses. Although the above parameterization is for SO42- aerosol, we assume it is equally applicable to hydrophilic BC and OC as well. This parameterization has been tested in RegCM previously by (Qian and Giorgi, 1999) and (Huang et al., 2007).2
2.2. Implementation of the second indirect effect in RegCM4.4.1
To represent the aerosol second indirect effect in RegCM4.4.1, the parameterization of cloud microphysics in the model is altered so that the rate of precipitation is affected by the aerosol concentration. In RegCM4.4.1, SUBEX (Pal et al., 2000) calculates the cloud cover fraction based on the relative humidity. In the cloud fraction, a Kessler-type bulk formulation (Kessler, 1969) is used to parameterize the auto-conversion and accretion processes. The Kessler-type formula ("KS" scheme) assumes that precipitation is formed at any model level when the cloud water mixing ratio (q L=w L/ρ a) exceeds the threshold value (q L,th), as in the following relation: \begin{equation} P=C_{\rm ppt}\left(\dfrac{q_{\rm L}}{f_{\rm c}}-q_{\rm L,th}\right)f_{\rm c} , \ \ (3)\end{equation} where P is the rain drop formation rate (kg kg-1 s-1), 1/C ppt is the characteristic time for which cloud droplets are converted into raindrops, and f c is the cloud fraction. The threshold value is obtained as a function of temperature according to the following relation derived by (Gultepe and Isaac, 1997): \begin{equation} q_{\rm L,th}=C_{\rm acs}10^{-0.49+0.013T} ,\ \ (4) \end{equation} where T is temperature in °C, and C acs is the auto-conversion scale factor. Also, in SUBEX (Pal et al., 2000), the amount of accreted cloud water (P acc) and evaporated precipitation (P evap) are expressed as follows: \begin{eqnarray} P_{\rm acc}&=&C_{\rm acc}q_{\rm L}P_{\rm sum} ,\ \ (5)\\ P_{\rm evap}&=&C_{\rm evap}(1-{\rm RH})P_{\rm sum}^{\frac{1}{2}} ,\ \ (6) \end{eqnarray} where C acc is the accretion rate coefficient, P sum is the accumulated precipitation from above falling through the cloud, and C evap is the evaporation rate coefficient.Several prior studies have found that the second indirect effect is very sensitive to the parameterizations of auto-conversion and cloud cover in models (Lohmann and Feichter, 1997; and Huang et al., 2007). Here, we implement three different auto-conversion schemes in RegCM4.4.1:
The first parameterization depends on Beheng (1994) (referred to as the "BH" scheme), which is based on Lohmann and Feichter (1997): \begin{equation} P=\dfrac{6\times 10^{28}\gamma_1n^{-1.7}(10^{-6}N_{\rm c})^{-3.3}(10^{-3}\rho_{\rm a}q_{\rm L}/f_{\rm c})^{4.7}}{\rho_{\rm a}} , \ \ (7)\end{equation} where γ1=150 is a tunable parameter, and n=10 is the width parameter of the initial cloud droplet spectrum. All parameters are in SI units.
The second parameterization depends on (Tripoli and Cotton, 1980) (referred to as the "TC" scheme): \begin{equation} P=\dfrac{0.104gE_{\rm c}\rho_{\rm a}^{\frac{4}{3}}q_{\rm L}^{\frac{7}{3}}}{\mu(N_{\rm c}\rho_{\rm w})^{\frac{1}{3}}}H(N_{\rm c20}-10^3) , \ \ (8)\end{equation} where g is gravity, E c=0.55 is the collision/collection efficiency of cloud droplets, μ=1.83× 10-5 kg m-1 s-1 is the dynamic viscosity of the air, and H is the Heaviside function. Since cloud droplets convert to rain drops when the N c of larger than 20 μm in radius (N c20) is more than the 103 m-3 (Rogers and Yau, 1989), where \begin{equation} H=\left\{ \begin{array}{l@{\quad}l} 1, & N_{\rm c20}>10^3\\ 0, & N_{\rm c20}\leq 10^3, \end{array} \right. \ \ (9)\end{equation} N c20 is calculated assuming a gamma cloud droplet size distribution according to the Khrgian and Mazin distribution (Pruppacher and Klett, 1997).
The third parameterization of auto-conversion (referred to as the "R6" scheme), based on (Liu and Daum, 2004), accounts for the dispersion effect of cloud droplets (Liu and Daum, 2004, 2007): \begin{equation} P=\left(\dfrac{3}{4\pi\rho_{\rm w}}\right)^2\dfrac{k_{2}\beta_6^6}{N_{\rm c}}w_{\rm L}^3H(R_6-R_{\rm 6c}) , \ \ (10)\end{equation} where R6 is the mean radius of the sixth moment of the droplet size distributions in (Rotstayn and Liu, 2005), k2=1.9× 1011 cm-3 s-1 is a constant describing the increase in the collection efficiency of cloud droplets with increasing collector drop size, β6 represents the dispersion effect of cloud droplets assuming a gamma distribution for the cloud-droplet spectrum, \begin{equation} \beta_6=\left[\dfrac{(1+3\in^2)(1+4\in^2)(1+5\in^2)}{(1+\in^2)(1+2\in^2)}\right]^{\frac{1}{6}} ,\ \ (11) \end{equation} and R 6c is the critical radius in μm, \begin{equation} R_{\rm 6c}=4.09\times 10^{-4}\beta_{\rm con}^{\frac{1}{6}}\dfrac{N_{\rm c}^{\frac{1}{6}}}{w_{\rm L}^{\frac{1}{3}}} ,\ \ (12) \end{equation} where =1-0.7exp(-α N c), is the relative dispersion of the droplet size distribution, α=0.003 (Rotstayn and Liu, 2005), w L is in g m-3, N c is in cm-3, and β con=1.15× 1023 s-1 is the mean value of the condensation rate constant.
These three auto-conversion schemes differ in their dependence on the total aerosol mixing ratio (x tot), which relates to the r e and w L, as shown in the following proportionalities. These are derived by eliminating N c with x tot and r e using Eqs. (1) and (2) in the raindrop formation rate (P) for the BH [Eq. (7)], TC [Eq. (8)] and R6 [Eq. (10)] schemes: \begin{equation} \left. \begin{array}{l} {\rm BH}:P\propto w_{\rm L}^{4.7}x_{\rm tot}^{-1.5}\propto w_{\rm L}^{1.4}r_{\rm e}^{9.9}\\[1mm] {\rm TC}:P\propto w_{\rm L}^{2.3}x_{\rm tot}^{-0.15}\propto w_{\rm L}^2r_{\rm e}\\[1mm] {\rm R6}:P\propto w_{\rm L}^3x_{\rm tot}^{-0.45}\propto w_{\rm L}^2r_{\rm e}^3 \end{array} \right\}\ \ (13) \end{equation}
The precipitation rates simulated by the KS, BH, TC and R6 auto-conversion schemes, with different values of r e and q L are shown in Figs. 1a and b for r e=10 and 7.5 μm, which represent large and small cloud droplets, respectively. Because the auto-conversion rates depend on the f c in KS and BH, we show the range of values for two f c values (f c=1 and f c=0.5; Figs. 1a and b). The cloud fractional cover has an effective influence on the KS auto-conversion rate at low in-cloud liquid water (q L≤ 0.6 g kg-1), where the lower f c (f c=0.5) increases the auto-conversion rate faster than the one (f c=1) (Figs. 1a and b). For larger droplets (r e=10 μm), the auto-conversion rate is enhanced by the BH scheme with more efficiency than TC, R6 and KS, respectively. On the other hand, with smaller cloud droplets (r e=7.5 μm), the auto-conversion in the BH scheme is faster than in the R6 scheme only at extremely low q L (≤ 0.2 g kg-1), whereas the precipitation rate produced by TC is more than that of BH at q L≥ 0.1 g kg-1.
Figure1. Auto-conversion rates (P) (units: 106 kg kg-1 s-1) as a function of liquid water mixing ratio (q L) (units: g kg-1) for the different auto-conversion schemes of KS, BH, TC and R6. The calculations assume an r e of (a) 10 μm and (b) 7.5 μm. Note that for the purpose of these figures, the calculations of the KS and BH schemes assume a cloud fraction cover of f c=1 and f c=0.5; KS is unaffected by changing the r e, and its calculation for these figures assumes q L,th=0.2 g kg-1 in Eq. (3).
2
2.3. Experimental design
The simulations in this study are conducted over a region extending from tropical Africa to northern Africa and the Mediterranean, as shown in Fig. 2. This domain has a complex mixture of aerosols from various origins, such as desert dust, urban pollution and biomass-burning/smoke aerosol. The model domain is centered at (19.0°N, 12.0°E), with a grid of 84× 116 points at a horizontal grid spacing of 60 km, and 18 vertical sigma levels with the model top at 10 hPa. For our analysis, the region extending from the equator to 15°N is divided into two sub-regions, referred to as the western region (Domain1) and the central region (Domain2), as shown in Fig. 2a. In all simulations, the global data of NCEP-2 provide the meteorological initial and lateral boundary conditions. For the SST, OISSTv2 weekly data are used. For the chemical boundary conditions, we use the global output from the Model for Ozone and Related Chemical Tracers (Emmons et al., 2010). We conduct a one-year simulation (1 October 2005 to 1 December 2006) with the first two months used as model spin-up, and focus on the season of the WAM (June-July-August; JJA).We simulate online aerosols for the chemical species of SO42- and hydrophobic and hydrophilic BC and OC to investigate the effects of the aerosols from biomass and anthropogenic sources. Aerosol emissions are based on the Emission Database for Global Atmospheric Research (EDGAR) (Olivier et al., 2001) for anthropogenic and biomass-burning BC and OC and biogenic SO2, and the Reanalysis of the Tropospheric Chemical Composition Inventory (RETRO) (Schultz et al., 2007) for anthropogenic SO2. Figures 2a-c show the spatial distributions of emissions during summer for SO2, derived from the anthropogenic emissions of RETRO and biomass-burning emissions of EDGAR, and the BC and OC derived from the anthropogenic and biomass-burning emissions of EDGAR. Figure 2a shows that the total emission rates of SO2 are concentrated around the Mediterranean basin, especially in the large cities due to anthropogenic activities, with emission rates of up to 4× 10-10 kg m-2 s-1. Emissions are also high in West Africa near the Gulf of Guinea, due to biomass burning and anthropogenic activities. The spatial distributions of the total BC emission rates are similar to those of SO2, as shown in Fig. 2b; the emissions rate reaches 3× 10-13 kg m-2 s-1 over the large cities in the Mediterranean and Arabian Peninsula. The emissions of OC (Fig. 2c) follow the same patterns as BC.
To validate and intercompare the simulations, we use gridded (0.5°× 0.5°) observations from the CRU (Mitchell and Jones, 2005) for the monthly surface air temperature and precipitation data over land. The Level-3 (version 5) global-gridded 1°× 1° data product retrieved from MODIS onboard Terra is used to evaluate the total cloud cover distribution over the entire simulation domain. The Level-3 Terra/MODIS AOD, retrieved using the Dark-Target (Levy et al., 2007) and Deep Blue (Hsu et al., 2006) aerosol algorithms, is used to evaluate the simulated regional AOD.
Figure2. Emissions rate (units: kg m-2 s-1) of (a) SO2× 10-10 derived from the EDGAR and RETRO emissions inventories, and (b) BC× 10-13 and (c) OC× 10-13 derived from EDGAR only. The two selected domains are outlined by the dashed black lines in (a): Domain1 is West Africa and Domain2 Central Africa.
We conduct nine sensitivity simulations with varying treatments of the aerosol indirect effect. Four control runs (CTRL, CTRL_BH, CTRL_TC and CTRL_R6) are performed with the different auto-conversion schemes of KS, BH, TC and R6, respectively. In the control simulations, the r e is constant (at 10 μm) and no aerosol effects are considered. The simulation called "DIRECT" includes the direct effect of all types of aerosols in RegCM4.4.1 (SO42-, hydrophobic and hydrophilic OC and BC) with r e=10 μm. The first indirect effect of SO42-, hydrophilic OC and BC, and the direct effect of all aerosols, are evaluated with the simulation called "INDIR1", in which the size of cloud droplets changes according to the aerosol mass concentration. The effect of the auto-conversion scheme is discussed in section 3.4, and the combined effects of the aerosols (direct, first and second indirect) are included in the runs of "ALL_BH", "ALL_TC" and "ALL_R6" with different auto-conversion schemes. A description of all sensitivity experiments is provided in Table 1.
Here, we focus on the indirect effects of aerosols on the regional climate by changing only the parameterization of the large-scale precipitation processes without changing the convective precipitation parameterizations, because the convective parameterizations implemented in RegCM4.4.1 do not include cloud microphysics that can be directly connected with cloud condensation nuclei and hence aerosols. In addition, we only consider warm cloud processes, as we do not explicitly permit aerosols to act as ice nuclei in these simulations. However, it is possible that the properties of ice cloud can be affected through interaction processes between liquid and ice phases.
-->
3.1. Control simulations with different auto-conversion parameterizations
Figures 3a and b show the spatial distributions of the total cloud cover from MODIS and total precipitation from CRU, respectively, over the studied domain during JJA 2006. It can be seen that the cloud and rainfall concentrated in the region south of 15°N, especially in West Africa, are associated with a cloud fraction greater than 80%, and precipitation rates exceeding 200 mm month-1 in several regions across Central and West Africa.First, we examine the influence of changing the auto-conversion scheme on the mean cloud cover and precipitation in the control runs without including the effects of aerosols. The area-averaged values of CLWP, cloud cover (low, medium, high and total), and total precipitation, as well as the ratio of convective to total precipitation, simulated by different control runs with different auto-conversion schemes, are described in Table 2. These mean values are calculated over West Africa (Domain1) and Central Africa (Domain2) over land only. The CLWP of TC and R6 is greater than that of KS by about 18% and 43%, respectively. This enhancement in cloud liquid water content results in higher values of different cloud types [low (LCLD), medium (MCLD), high (HCLD)] and total cloud, by 18% and 21% for TC and R6, respectively. The surface air temperature (T) averaged over Domain1 with CTRL_R6 and CTRL_TC decreases by approximately 1°C compared to CTRL and CTRL_BH. The CTRL_R6 run produces higher total precipitation (T precip) than CTRL by about 16%. In addition, the ratio of the precipitation produced by convection processes (C precip/T precip) is over 80% of the total precipitation in all simulations.
Figure3. Spatial distribution of (a) total fractional cloud cover from MODIS/Terra, (b) total precipitation (units: mm month-1) from CRU, and (c) AOD at 550 nm from MODIS/Terra, during JJA 2006.
In Domain2, the CLWP maximum is produced by the R6 parameterization (CTRL_R6) and the minimum is simulated by the BH parameterization (CTRL_BH). The R6 scheme simulates a larger cloud fraction for all cloud types (LCLD, MCLD, HCLD), and total cloud cover, than the other schemes, with CTRL_R6 resulting in a 25% increase in total cloud with respect to the reference control run (CTRL) (Table 2). Again, CTRL_R6 and CTRL_TC produce lower surface air temperature than CTRL_BH and CTRL. This reduction can be attributed to the increased LCLD in these experiments. These enhancements of CLWP and cloud cover simulated by R6 result in a 10% increase in total precipitation compared to CTRL. It is worth noting here that using different auto-conversion parameterization schemes for large-scale precipitation generally increases the percentage of convective to total precipitation over the two domains compared to the KS scheme. The exception to this is the BH scheme, which reduced this ratio over Domain1, which may be attributable to the enhancement in liquid water content in cloud with the different schemes.
Figure 4a illustrates that, over Domain1, the CTRL, CTRL_BH and CTRL_TC simulations underestimate the total cloud cover compared to MODIS by over 15% with CTRL_BH, while CTRL_R6 overestimates the cloud cover by less than 1%. In addition, over Domain2, the simulations of CTRL, CTRL_BH and CTRL_TC show negative bias (greater than Domain1) compared to MODIS, whereas CTRL_R6 results in lower positive bias than in Domain1.
By comparing the simulated total precipitation based on the control runs with CRU data as shown in Fig. 4b, we find that all the runs result in overestimations, ranging between 30% and 80% with CTRL_BH and CTRL_R6, respectively, over Domain1, and >50% and >100% with CTRL_BH and CTRL_R6, respectively, over Domain2.
2
3.2. AOD
The spatial distribution of the AOD over the studied domain during JJA 2006 observed from Terra/MODIS (Dark Target and Deep Blue combined data) at the mid-visible wavelength (550 nm) is shown in Fig. 3c. Here, the AOD from Dark Target and Deep Blue is averaged using the method of (Gautam et al., 2011). Higher values of AOD, mainly due to dust, are noted in Central Africa extending to the west. Figure 4c shows the bias in AOD simulated by the five different model sensitivity tests (DIRECT, INDIR1, ALL_BH, ALL_TC, ALL_R6), relative to that detected from Terra/MODIS. We note that these simulations do not include dust emissions; however, these large differences in AOD are reduced significantly with the inclusion of the combination of all aerosol effects, especially with using the auto-conversion schemes of R6 and TC, respectively.2
3.3. Aerosol direct and first indirect effects
Here, we discuss the changes in cloud cover and precipitation due to the aerosol direct and first indirect effects, based on the differences between DIRECT and INDIR1 from the CTRL simulation during JJA 2006. Note that all these three simulations use the same KS auto-conversion scheme, so the differences in the simulations are due primarily to the treatment of the aerosol direct (DIRECT) and first indirect effect (INDIR1).By focusing on the aerosol effects over Domain1 and Domain2, the results (Table 3) show that, over Domain1, the aerosol direct effect can be linked to a slight suppression in the CLWP. Furthermore, by adding the first indirect effect, this suppression increases to more than -44 g m-2, which is similar to the values published by (Costantino and Bréon, 2013). Generally, it is found that the CLWP at all atmospheric levels decreases slightly in the direct simulations (generally, <0.02). The DIRECT run leads to an increase in surface air temperature by 0.2°C relative to CTRL, and this increase enhances in the INDIR1 run.
Figure4. Relative errors in JJA 2006 (a) total fractional cloud cover and (b) total precipitation (units: %), with respect to MODIS and CRU observations, respectively, for the different simulations, averaged over Domain1 (dark gray) and Domain2 (light gray). The average is calculated over land only. (c) Bias ratio of AOD (units: %) calculated for the simulations with aerosols only, with respect to MODIS. The error bars are plotted at 5%.
Over Domain2, the situation is slightly reversed; the aerosol direct effect is also linked to the CLWP, where the CLWP increases (>1 g m-2). Small changes in LCLD are found relative to CTRL, with a slight increase in the DIRECT run, and reduction of about 0.02, when adding the first indirect effect. Also, the DIRECT simulation results in a reduction of MCLD and HCLD, but the total fractional cover increases; whereas, the INDIR1 simulations suggest an increase corresponding MCLD and a decrease associated HCLD and total cloud cover. Also over Domain2, the INDIR1 run leads to an increase in mean surface temperature by 0.5°C compared to CTRL, but the DIRECT run leads to a slightly weaker decrease.
In terms of precipitation changes, results indicate a net reduction in total precipitation over Central Africa in both DIRECT and INDIR1, where ? P is reduced by 4.0 and 42 mm month-1, respectively. In summary, the reduction in precipitation in West Africa (Domain1) is greater than that in Central Africa (Domain2). This is primarily due to the greater emissions of SO42- aerosols over Domain1. SO42- aerosols cause a reduction in the r e, which results in an enhancement of cloud albedo, in turn resulting in enhanced cooling at the surface. It is well known that a cooler surface is associated with suppressed convection processes that reduce the CLWP, in turn reducing the overall precipitation (Lohmann and Feichter, 1997).
2
3.4. Combined effects (direct, first and second indirect)
To study the combined aerosol effects (direct, first and second indirect), we quantify changes in cloud cover and precipitation (as illustrated in Table 3) simulated by ALL_BH, ALL_TC and ALL_R6, relative to their control runs (CTRL_BH, CTRL_TC and CTRL_R6, respectively). With the combined aerosol effects, and consistent with the hypothesis of aerosol inhibition of precipitation (Albrecht, 1989), CLWP is found to increase over West Africa with the three auto-conversion schemes relative to their control runs, with the greatest positive change from the TC scheme (?CLWP >1.8 g m-2) (Table 3). The combined aerosol effects increase the LCLD in all the sensitivity tests, with the greatest change in the R6 scheme (an increase of 0.06). The MCLD reduces consistently across all schemes, with the maximum change in the R6 scheme (0.06); however, there are some differences for HCLD among the different schemes. Whereas the combined aerosol effects reduce the HCLD in BH and R6 (-0.06), the TC scheme shows a slight increase in cloud cover (by 0.004) compared to the cases without aerosols. The mean total cloud cover increase over Domain1 was found to be small, by approximately 0.003, in all schemes. Interestingly, the decrease in air temperature at the surface seen in all experiments, especially ALL_R6 (? T=-0.4°C), is a characteristic resulting from the aforementioned overall increase in total cloud cover. The combined aerosol effects suppress the total precipitation over Domain1 with the schemes of BH and R6, with a reduction of precipitation by 46 mm month-1 with R6, whereas TC increases the precipitation by less than 3 mm month-1. This increase may be attributable to an enhancement in HCLD.Over Domain2, the CLWP decreases with the two auto-conversion schemes of BH and TC (?CLWP = -2 g m-2 and -4 g m-2, respectively), but the R6 scheme causes an increase in CLWP to more than 24 g m-2. The domain-average changes in cloud are relatively small, with the greatest changes for MCLD being an increase of 5% with the R6 scheme. Similar to Domain1, the total cloud cover increases over Domain2 in each experiment, especially for the ALL_TC and ALL_R6 simulations (?TCLD >0.007 and 0.009, respectively), albeit these changes are relatively small. Furthermore, the air temperature, averaged over Domain2, is associated with a decrease in the three experiments relative to their control runs (? T=-0.2°C with ALL_TC and ALL_R6). The total precipitation averaged over Central Africa decreases in the three runs with different percentages, with the maximum suppression in the ALL_R6 simulation (? T precip>-57%) relative to its control run (CTRL_R6).
2
3.5. Radiative forcing
The radiative forcing (RF) of aerosols represents the influence of aerosols on the Earth's energy balance, where a positive RF indicates that the energy of the surface-atmosphere system increases, leading to a warming of the system. In contrast, negative RF corresponds to a cooling of the system. Here, the RF is estimated as the difference in the net radiative flux (downward minus upward) between the present-day total aerosol loading (natural and anthropogenic) and the simulation with no aerosols (control simulations).The net RF (shortwave and longwave) at the surface and top of the atmosphere (NRF_SRF and NRF_TOA, respectively) due to the different aerosol effects are averaged over Domain1 (Fig. 5a). There is a small positive NRF_SRF over west Africa due to the DIRECT simulation (NRF_SRF = 1 W m-2; standard deviation of 2.5 W m-2), which becomes negative (-7 W m-2) by including the first indirect effect (INDIR1). Also at the TOA, the warming caused by the direct aerosol effect is transformed to negative RF in the INDIR1 simulation (NRF_TOA = 3 3 W m-2 and -8.5 5 W m-2, respectively). The ALL_BH run (combined aerosol effects with the BH scheme) further decreases the cooling at the surface to -1 2 W m-2 and leads to a warming at the TOA of 0.8 3 W m-2. However, using the TC and R6 auto-conversion schemes, the combined aerosol effects (ALL_TC and ALL_R6) lead to a cooling of -4 2 W m-2 and -23 2 W m-2, respectively, at the surface, and -1 0.9 W m-2 and -21 1.5 W m-2, respectively, at the TOA.
Figure5. Net RF (shortwave and longwave) at the surface (dark gray) and TOA (light gray) during JJA 2006 simulated by the different experiments (DIRECT, INDIR1, ALL_BH, ALL_TC, and ALL_R6) averaged over (a) West Africa and (b) Central Africa, with error bars of standard deviation.
Figure6. Effect of the first indirect effect and R6 auto-conversion scheme on monsoon circulation (JJA 2006): (a) difference in circulation between the ALL_R6 and CTRL_R6 simulations at 850 hPa and the MSLP; (b) difference in circulation between the INDIR1 and CTRL simulations at 850 hPa and the MSLP. The shading shows the difference in the MSLP in units of hPa. Dark gray indicates positive anomalies and represents a strengthening of the MSLP, while the lightest gray shows below-zero anomalies and presents a weakening of the MSLP. Arrows represents the difference in circulation, i.e., the direction and relative intensity of the change in the wind field (units: m s-1) due to the auto-conversion.
Figure7. Effects of aerosols on convection with the different auto-conversion schemes. Specifically, the panels show meridional cross sections of the heating rate due to convection (units: K d-1), where negative in dashed lines (positive in solid lines) values indicate cooling (heating) of the atmosphere: (a) ALL_BH minus CTRL_BH; (b) ALL_TC minus CTRL_TC; (c) ALL_R6 minus CTRL_R6; (d) the INDIR1 minus CTRL.
The average NRF_SRF and NRF_TOA over Domain2 are shown in Fig. 5b. The aerosol direct effect (DIRECT) causes negative RF at the surface (-4 1 W m-2) and very low positive RF at the TOA (0.2 1 W m-2). The RF over Domain2 could be due the low emissions of BC and OC over this domain, as shown in the spatial distribution of their emissions in Figs. 2b and c.
After adding the aerosol first indirect effect, the aerosol-induced cooling at the surface is found to be NRF_SRF = -10 2 W m-2 and NRF_TOA = -14 3 W m-2, at the surface and TOA, respectively. With the combined aerosol effects, the sign and magnitude of RF differ according to the different auto-conversion schemes, since the three runs keep the cooling at the surface, with the maximum caused by the ALL_R6 run (NRF_SRF = -23 3 W m-2). However, the ALL_BH and ALL_TC runs cause very low warming (<0.5 1 W m-2) at the TOA, with the ALL_R6 run cooling reaching a large negative forcing of -25 3 W m-2.
2
3.6. Influence on WAM circulation
Here, we investigate the dynamic and thermodynamic responses to the aerosol indirect effect in the cases of the INDIR1 and ALL_R6 simulations, which show pronounced change in precipitation, through analyzing the average WAM circulation anomalies——namely, the changes in the mean SLP (MSLP) and wind field at 850 hPa due to the indirect effect. Figure 6 shows the change in the MSLP and wind field during WAM for both simulations. The continental pressure increases in both schemes but to different degrees. Pressure increases in West Africa and decreases over ocean, which results in a reduction in monsoon pump intensity (Konare et al., 2008). The indirect effects weaken the monsoon's circulation, where the differential wind field has totally reversed its direction (all inflow becomes outflow). ALL_R6 shows a strong reduction in average monsoon circulation (Fig. 6a) compared to the INDIR1 simulation (Fig. 6b). The manifestation of pressure system reduction is shown in the reduction in the wind field.Intensification of continental pressure is only possible if the atmospheric column has been cooled aloft. This hypothesis can be shown by the analysis of the vertical heating rate owing to latent heat due to convection. Figure 7 shows the vertical zonal average convective heating rate. During JJA, deep convection is inhibited at the middle and high levels of the atmosphere. Comparison between vertical cloud cover (not shown) and the convective heating rate shows that the reduction in vertical cloud extension is due to cooling in the middle and upper levels. Between 5°N and 15°N, the vertical extension of the difference in the convective heating rate shows a dipole structure, where a positive (negative) change in the heating rate is observed in the lower (middle and upper) atmosphere up to 850 hPa (200 hPa). The strength of this dipole is an indication of deep cloud suppression and hence precipitation reduction. The dipole strength is very weak in BH and TC, which agrees with the results in Table 3. The R6 auto-conversion scheme shows a strong reduction in the heating rate (-0.9 K d-1) in the middle and upper troposphere. This reduction in the convective heating rate is comparable to the direct effect made by dust aerosol, as shown in (Solmon et al., 2012). On the other hand, the INDIR1 simulation shows a strong yet different signal where the reduction in the convective heating rate has a narrow meridional extension in the middle troposphere and, contrary to R6, the positive convective heating rate extends to the upper troposphere but north of 10°N. The first indirect effect reduces the CLWP, leading to a reduction in LCLD and yet enhances cloud albedo. This pathway results in a reduction in MCLD and a stabilization of the atmospheric column. On the other pathway of the ALL_R6 that increases the low cloud cover, which results in reduction in surface temperature, which in turn reduces monsoon circulation and inhibits deep convection.