Quantifying snow darkening and atmospheric radiative effects of black carbon and dust on the South Asian monsoon and hydrological cycle: experiments using variable-resolution CESM

Black carbon (BC) and dust impart significant effects on the South Asian monsoon (SAM), which is responsible for ∼ 80 % of the region’s annual precipitation. This study implements a variable-resolution (VR) version of the Community Earth System Model (CESM) to quantify two radiative effects of absorbing BC and dust on the SAM. Specifically, this study focuses on the snow darkening effect (SDE), as well as how these aerosols interact with incoming and outgoing radiation to facilitate an atmospheric response (i.e., aerosol–radiation interactions, ARIs). By running sensitivity experiments, the individual effects of SDE and ARI are quantified, and a theoretical framework is applied to assess these aerosols’ impacts on the SAM. It is found that ARIs of absorbing aerosols warm the atmospheric column in a belt coincident with the May–June averaged location of the subtropical jet, bringing forth anomalous uppertropospheric (lower-tropospheric) anticyclogenesis (cyclogenesis) and divergence (convergence). This anomalous arrangement in the mass fields brings forth enhanced rising vertical motion across South Asia and a stronger westerly low-level jet, the latter of which furnishes the Indian subcontinent with enhanced Arabian Gulf moisture. Precipitation increases of 2 mm d−1 or more (a 60 % increase in June) result across much of northern India from May through August, with larger anomalies (+5 to+10 mm d−1) in the western Indian mountains and southern Tibetan Plateau (TP) mountain ranges due to orographic and anabatic enhancement. Across the Tibetan Plateau foothills, SDE by BC aerosols drives large precipitation anomalies of > 6 mm d−1 (a 21 %–26 % increase in May and June), comparable to ARI of absorbing aerosols from April through August. Runoff changes accompany BC SDE-induced snow changes across Tibet, while runoff changes across India result predominantly from dust ARI. Finally, there are large differences in the simulated SDE between the VR and traditional 1 simulations, the latter of which simulates a much stronger SDE and more effectively modifies the regional circulation.

We compare the control simulations against a level-3 1 o monthly-gridded product from the moderate resolution imaging spectroradiometer (MODIS) collection-6 suite, retrieved from MODIS Terra and available at https://ladsweb.nascom.nasa.gov/api/v1/productPage/product=MOD08_M3. We make use of 14year AOD averages from 2001 -2014 (Platnick et al., 2015). The label "level-3" (L3) refers to the datasets being monthly statistical summaries of bulk AOD that combine multiple satellite passes into a global dataset at monthly intervals. Specifically, this product contains over 700 derived variables such as atmospheric optical properties, ozone burden, atmospheric water vapor, cloud optical and physical properties, and atmospheric stability parameters. The retrieval algorithm for 550nm AOD makes use of a merged dark target and deep blue algorithm. The algorithm is suitable for this study due to the high spatial variability of vegetation and brightness factor across the TP region and the rest of south Asia.
Aerosol properties are only retrieved by Terra under clear sky conditions, so the L3 gridcells are only filled when the number of monthly cloud-free pixels within a 1 o x×1 o cell exceeds 6. Since MODIS orbits ~15 times per day, and cloud cover is continually varying, different numbers of pixels go into each 1 o gridcell AOD calculation. The selection of these pixels is based on algorithms that rely on the surface characterization, aerosol model, quality checks, and cloud masking. These algorithms contribute to an expected error for MODIS AOD measurements of ± (0.05 + 0.15×AOD AERONET ) over land and ±(0.04 + 0.1×AOD AERONET ) over oceanic regions compared to surface-based observations from AERONET (Levy et al., 2013). Uncertainties in observed AOD specific to certain regions are also noted in Levy et al. (2013), as retrieval processing algorithms vary depending on surface albedo and other regional properties.

S1.2 MISR
Simulated bulk AOD is compared against monthly multi-angle imaging spectroradiometer (MISR) L3 AOD measurements (available at https://l0dup05.larc.nasa.gov/MISR/cgi-bin/MISR/main.cgi) that are binned onto a 0.5 o grid and averaged over a 13-year period from 2002 through 2014 (MISR Science team 2015). MISR is mounted aboard the MODIS Terra satellite, and data products from MISR include surface albedo properties, vegetative indices, and global radiance information. Specifically, AOD is derived for aerosols of differing sizes in blue (443 nm), green (555 nm), red (670 nm), and infrared (865 nm) channels. For this study, we utilize MISR's retrieved bulk AOD in the green (555nm) channel.
Similar to MODIS AOD retrievals, MISR has to contend with uncertainties resulting from cloud contamination, false aerosol detection, and algorithm inadequacies (Witek et al., 2018). Singh et al. (2016) found that the root mean square error (RMSE) between MISR and AERONET AOD to be slightly lower (0.11 -0.20) than the RMSE between MODIS and AERONET AOD (0.15 -0.27) across the Indo Gangetic Plain (IGP). More generally, Kahn et al. (2010) reported that MISR AOD retrievals fall within 0.05 or 20% of AERONET-measured AOD values.

S1. Aerosol reference data comparison
We validate the capabilities of the CESM experiments in simulating specific and bulk aerosol properties against various datasets. Here, "bulk" refers to AOD contributions from all aerosols (sulfate + BC + dust + ammonium + organics + sea salt + nitrates).

S1.1 MODIS
MACv2 AOD estimates were derived as data from AERONET were merged onto background maps from global models that participated in the aerosol model intercomparison (AeroCom) project, effectively making MACv2 an aerosol reanalysis dataset. Data from the maritime aerosol network (MAN) are also included in this version of MAC (Smirnov et al., 2009). The aerosol plumes in this dataset are designed to fit the distribution of mid-visible AOD for the year 2005.

S1.5 AOD validation
Globally averaged, CONT-vr and CONT-un simulate annual AOD values of 0.120 and 0.133, respectively. MODIS and MISR AOD values of 0.172 and 0.148 are observed, respectively, making CONT-un closer to MISR and MODIS observations. MACv2 estimates a global AOD of 0.122, a value more similar to the CESM experiments and less than is observed by satellites. Finally, MERRA-2 estimates a global AOD of 0.141, a value more similar to satellite observations. While both CESM simulations reasonably capture the global annually averaged AOD compared to satellite observations, they do not capture the generally larger annual AOD values in the region bounded by 0 o N-60 o N and 60 o E-140 o E.
CONT-un is better correlated with AERONET AOD than CONT-vr, simulating a Pearson correlation (r) value of 0.451, 0.503, and 0.672 for the total, fine-mode, and coarse-mode AOD, respectively, while CONT-vr simulates r-values of 0.335, 0.497, and 0.467. This means that, although the CESM experiments simulate similar mean biases compared to AERONET, the nature of the AOD bias (high or low) changes from site to site between the UN and VR experiments.

Figure S2. Bar charts at various sites across the Himalaya Mountains discussed in H2014 for (a) total BC deposition, (b) dry BC deposition, (c) wet BC deposition, (d) SWE, (e) elevation, and (f) total precipitation for CONT-vr (red) and CONT-un (black).
a. b. c. d. e. f.
d. e. Figure S8. Same as in Figure S5, but for cloud fraction (%).
a. b. c.
d. e. f. Figure S13. Same as in Figure S12, but for 850 hPa u.