Cyclone Fani: the tug-of-war between regional warming and anthropogenic aerosol effects

Before Cyclone Amphan took place in 2020, Cyclone Fani (May 2019) is the strongest pre-monsoon cyclone in the Bay of Bengal (BOB) since 1991, killing 90 people in eastern India and Bangladesh while causing US$1.81 billion of damages. Fani developed during a period of high concentration of anthropogenic aerosols in the BOB with abnormally high sea surface temperature (SST), thereby presenting an opportunity to understand the compound effects of atmospheric aerosols and regional climate warming on a tropical cyclone. A quantitative attribution analysis was conducted using the Weather Research and Forecasting model with chemistry (WRF-Chem) run at the convection-permitting (4 km) grid spacing, accompanied by an ensemble of coarser-resolution simulations to quantify the uncertainty. The removal of post-1990 trends in the tropospheric variables and SST from WRF-Chem’s initial conditions (IC) and boundary conditions (BC, including the lateral and lower boundary conditions) resulted in a reduction of cyclone precipitation by about 51% during the 5 d of April 28-May 2. The removal of tropospheric warming shows approximately twice as strong an effect on Fani (39% reduction in precipitation) as that of SST warming (22% reduction). When aerosol’s direct and indirect effects were removed from the simulations, i.e., no aerosol influence on radiation and cloud microphysics, Fani initially strengthened but later weakened, as measured by geopotential height and precipitation amounts. These results suggest that aerosol and its interaction with the atmosphere acted to mitigate the strengthening effect of anthropogenic warming on Fani, but was not strong enough to entirely counteract it. Although the ensemble of coarser simulations appears to overestimate Cyclone Fani in terms of precipitation, the direction of the effects is in agreement with that obtained from the 4 km simulations. Given the increasing anthropogenic aerosols in the BOB, future attribution studies using more sophisticated dynamical aerosol models on BOB tropical cyclones are urged.


Introduction
Originated from a tropical depression that formed west of Sumatra in the Indian Ocean on 26 April 2019, cyclone Fani eventually developed to the strongest pre-monsoon cyclone (A Category 4-equivalent cyclone in figure 1(a)) in the Bay of Bengal (BOB) since 1991. According to Aon's Weather, Climate & Catastrophe Insight: 2019 Annual Report, Cyclone Fani made landfall in the Indian state of Odisha on 3 May 2019, killing 81 people while causing damages approximated to 8.1 billion US dollars in Bangladesh and India (www.aon.com). Exposed to a high concentration of anthropogenic aerosols during the pre-monsoon season (Wang et al 2013), BOB tropical cyclones are facing a double-punch effect from anthropogenic climate change: persistent warming in the ocean surface and the air, accompanied by a detectable increase in aerosol loading (Bhat et al 2001). Literature shows that local climate warming has increased the intensity of BOB cyclones (Yu and Wang 2009 analysis found that increased aerosols can deepen the short-wave trough that develops in May, subsequently intensifying tropical cyclones during the pre-monsoon season (Wang et al 2013). Cyclone Fani developed around the equator as depression and slowly move upward, then it has a quick intensification as an extremely severe cyclonic storm during the 24 h from 0530Z 30th April to 0530Z May 1st, the maximum sustained surface wind speed increased from 120-130 km h −1 to 180-190 km h −1 (http://www.rsmcnewdelhi.imd.gov.in). Thus, the quick intensification of Fani presents an interesting case in understanding the compound effects of atmospheric aerosols and climate warming in the BOB on a pre-monsoon tropical cyclone. High values of aerosol optical thickness (AOT) 1 due to sulfate extinction were present in the BOB around Cyclone Fani (figure 1(b)), concurrent with intense warm sea surface temperature (SST) 2 anomalies in the northern BOB, with regions of the ocean reaching 32 • C prior to Fani's landfall (figure 1(c)). In the weather timescale, aerosols present both a negative and positive impact on the microphysical and dynamical properties of clouds: a negative impact is caused by processes that reduce precipitation efficiency (Tao et al 2012, Hazra et al 2013 and weaken the cyclone winds (Rosenfield et al 2012 ), while a positive impact is associated with the 'cloud invigoration effect' linking an increase in aerosol loading with deepening of convective clouds. The debate about the invigoration effect and possible cloud responses remains fierce, however, as conclusions vary regionally, and depend on aerosol types (Altaratz et al 2014, Varble 2018, Jiang et al 2018, Fan et al 2018. As reported by Evan et al (2011), anthropogenic black carbon and sulfate aerosols would reduce the vertical wind shear to create an environment more favorable for tropical cyclone intensification, while Lau and Kim (2007) found that the loading of dust aerosol in the atmosphere might cool SST to suppress the development of cyclone.
During the pre-monsoon season in the BOB, the observed increases in anthropogenic aerosols and ocean heat content may have complex effects on Fani, and understanding these effects requires a novel event attribution analysis. This study does not focus on detailing the aerosol effects on cloud microphysics and convective processes, as that itself is an active research topic (see a review by Fan et al 2016). Rather, we seek to provide an attribution analysis that assesses the compound effects of highly concentrated aerosols and persistent SST warming, both of which are present in the BOB during Cyclone Fani.

Background of attribution on individual cyclones
Few studies have discussed the effect of climate change on individual tropical cyclones. Due to the large natural variability and the limited period of consistent observations, there is no consensus on how climate change affects the intensity and track of individual tropical cyclones (Landsea et al 2006, Patricola andWehner 2018). The few diagnostic and modeling studies that attributed the effect of climate change on individual tropical cyclones in the recent years point to a general intensification effect on tropical cyclones but with various degrees of rainfall or wind speed change. For example, statistical analysis and modeling experiments have indicated that the warming trend likely intensifies Hurricane Harvey's precipitation by 15%-20% (Risser and Wehner 2017, Van Oldenborgh et al 2017, enhances Super Typhoon Haiyan by about 20% (Takayabu et al 2015), and strengthens rainfall for Hurricane Florence by 2%-9% (Reed et al 2020). In general, the Clausius-Clapeyron (CC) relation that increases moisture holding capacity under climate warming plays a key role in the overall increase of precipitation. However, most previous studies found that the response of tropical cyclone (TC) rainfall to climate warming exceeds the CC relation. (Lenderink et al 2017) suggested that both higher moisture content and higher vertical velocities contribute to the TC precipitation process. Adding to the challenges in attributing climate change on individual storms, global climate models have a common difficulty in directly representing tropical cyclones. Therefore, recent studies such as Wehner et al (2017), Patricola and Wehner (2018), and  urged the need for applying convection-permitting regional simulations for the attribution analysis of hurricanes.
The impact of aerosol on tropical cyclones has mostly been investigated independent of the studies on regional climate warming effects. Using the assimilation of aerosol optical depth, the effects of dust on increasing static stability produce a less favorable environment to tropical cyclogenesis (Reale et al 2014). By comparing the effects of aerosol radiative forcing and greenhouse gas on the TC potential intensity, aerosol cooling seems to reduce TC intensity more strongly than greenhouse gas warming increases it (Sobel et al 2019). Given the high concentration of anthropogenic aerosols in the BOB, what would these effects be on Fani?

Model experiments and data sources
Following the emerging 'storyline approach' of attribution analysis (e.g. Zappa andShepherd 2017, Wang et al 2018), we adopted a quantitative attribution approach that constrains the boundary and initial conditions of a regional atmospheric model used to simulate this synoptic event. Using the Weather Research and Forecasting model with chemistry (WRF-Chem), we can generate the closest possible influences of the environmental conditions at the time of Cyclone Fani (via initial and boundary conditions) and simulate the same event by assuming the absence of long-term (warming) trends in those conditions and of the aerosol effects, in order to examine their combined effects on a TC. We used the WRF-Chem version 3.9.1 that includes aerosol interactions with the atmosphere (Grell et al 2005, Skamarock et al 2008, Powers et al 2017. WRF-Chem utilizes the Modal Aerosol Dynamics for Europe (MADE) aerosol module and Carbon-Bond Mechanism version Z (CBM-Z) to parameterize aerosol transport and deposition and represent the gasphase chemistry (Ackermann et al 1998, Zaveri andPeters 1999). The domain of simulations is composed of 623 × 539 grid points (outlined in figure 2) with a horizontal grid spacing of 4 km, a resolution that starts to resolve convective precipitation (i.e. convection-permitting) while convective parameterization is turned off. The model was forced by initial conditions (IC) and boundary conditions (BC, including the lateral and lower boundary conditions) using the 0.5 • × 0.5 • resolution Global Forecast System (GFS) initial analysis with nesting. Here, SST is derived from GFS rather than other observed data. As reported by Fu and Wang (2018), SST from GFS is a best lower boundary condition for WRF to represent the intensity of typhoon Nangka compared with the NOAA optimum interpolation SST (OISST), and ECMWF Interim reanalysis (ERA-Interim). We conducted four sets of 4 km simulation based on the combination of Lin (Lin et al 1983 (Hong et al 2006), and subsequently produced the control simulation (CTL, table 1) from their ensemble mean. To quantify model uncertainty, we also conducted multiple simulations with coarser resolution settings to increase the ensemble size. The ensemble spread and members are discussed in section 4.
Based upon Cho et al (2016) and , the regional model experiments were designed under the assumption that any post-1990 trend manifest in the tropospheric and ocean temperatures contains signals that are traceable to the anthropogenic global warming. Since GFS data only started in 2004 (https://www.ncdc.noaa.gov/data-access/model-da ta/model-datasets/global-forcastsystem-gfs), which is not adequate for computing the post-1990 trends, we used the NCEP Reanalysis 2 trend given that NCEP GFS also contributes to the NCEP reanalysis (Campana et al 2019). We first conducted four experiments of anthropogenic global warming effects with the chemistry component engaged (see table 2): a control simulation (1) forced by the original GFS analysis as IC and BC (CTL, as a 4-member ensemble), and 'detrended' simulations in which we removed the linear monthly trends from the IC and BC for (2) SST (denoted as DSST), (3) all tropospheric variables, such as temperature, geopotential height, relative humidity, U wind and V wind (DAIR), and (4) both SST and tropospheric variables (DALL). The trends were first computed from the NCEP-DOE Reanalysis 2 monthly data for each variable (Kanamitsu et al 2002) and then linearly interpolated onto GFS's horizontal and vertical grids. We then subtracted these trends from the GFS's initial analysis (including geopotential height, horizontal winds, air temperature, etc to be consistent), before using it as IC and BC to drive the WRF-Chem. We note that the individual trends revealed in the different IC and BC variables may not be linearly correlated but the nonlinear effect among these different trends is negligible (not shown). More details about the detrend processes can be referred to Cho et al (2016) and Wang et al (2016Wang et al ( ), (2018. Here, the anthropogenic emission was based on the Emissions Database for Global Atmospheric Research (EDGAR) HTAP V2 inventory (Janssens-Maenhout et al 2015). The EDGAR HTAP emission inventory provides total emissions, as well as sectorby-sector estimates for air, energy, industry, residential, shipping, and transportation sectors. To compare the non-anthropogenic aerosol effect on meteorological variables, four additional experiments with non-anthropogenic aerosol effects were designed: (5) a control aerosol simulation forced by the original GFS analysis as IC and BC without aerosols (RAERO; with the chemistry module turned off) and within this setup, a 'detrend' experiment of (6) SST (DSST_RAERO), (7) tropospheric variables (DAIR_RAERO), and (8) both SST and tropospheric variables as above (DALL_RAERO), all of which without simulating the interaction between anthropogenic aerosol and the atmospheric variables. However, due to the absence of aerosols, the direct and indirect effects like aerosol-radiation and aerosolmicrophysics interaction are not present in these (5)-(8) simulations. To evaluate the sensitivity of the climate trend effect, we also included another three 'double-trend' simulations forced by the original IC and BC added with the post-1990 monthly trends (thus doubling the warming effect, i.e. DB run): (9) the atmospheric variables (DBAIR_RAERO), (10) SST (DBSST_RAERO), and (11) both the atmosphere and SST (DBALL_RAERO). We have tested certain runs (RAERO) using both a single domain and a nesting domain with the chemistry component turned off. The results show that both runs can capture the main pattern of the observed precipitation (figures S2 and S3 (available online at stacks.iop.org/ERL/15/094020/mmedia)), which is indicated by the pattern correlation of precipitation between simulations and observation. It is arguable that the lateral boundary interpolated from 50 km to 4 km is not problematic for this simulation domain. Table 2 provides a summary of these experiments, while the detailed model setup is described in . Other physics schemes were configured identically in all simulations, with the following common options: YSU PBL scheme, the NOAH landsurface model (Chen and Dudhia 2001), and the Morrison microphysics scheme. The Fu-Liou-Gu Liou 1992, Gu et al 2011) radiation scheme, which allows for aerosols in long-wave and shortwave radiation, was applied for the simulations. The additional coarser-resolution simulations (to create a large ensemble) used a combination of other physics schemes not listed above. All simulations cover the period from 0000 UTC 26 April to 0000 UTC 04 May 2019.
The observed gridded 12 h and daily mean precipitation, derived from Level-3 half-hourly Integrated Multi-satellite Retrievals of Global Precipitation Measurement (IMERG) final product at the 0.1 • × 0.1 • resolution, was obtained from NASA Global Precipitation Measurement (GPM) data products (https://pmm.nasa.gov/). For observed SST, we use the Reynolds SST Analysis (Reynolds et al 2007).

Results
The observed evolution of Fani from 28 April 2019 to 2 May is displayed in figure 2 (left column) in terms of 850 hPa streamfunction at 00 UTC, overlaid with the daily mean precipitation. Pre-monsoon tropical cyclones in the BOB tend to develop from low-pressure systems in the equatorial Indian Ocean and migrate northward (Krishna 2009, Ng andChan 2012) and this was also the case with Fani. After its genesis, Fani tracked northward into the high-aerosol area and its interaction with the aerosols is visible in the swirling patterns of AOT of sulfate ( figure 1(b)). Meanwhile, the local SST anomaly at Fani's position reached as high as 2.5 • C ( figure 1(c)), which is in line with projections from the post-1980 warming trend of the BOB SST during May (Wang et al 2013).
The control simulation of Fani (CTL) in terms of daily precipitation and the 850hPa streamfunction at 00 UTC is plotted in figure 2 (middle column), alongside the aerosol removal simulation (RAERO, figure 2 right column). The WRF-Chem simulation captures the size and intensity of Fani in general, albeit overestimating the cyclone winds starting on April 29. The position of cyclone rainbands is slightly offset dayby-day, while the accumulated precipitation amounts within 12 • lat. × 12 • long. around the cyclone center averaged to an underestimation of~15% during the 5 d, based on CTL (figure 3(d)). A comparison of Fani's observed and modeled trajectories is plotted in figure 3(a) for CTL and in figure 3(b) with other experiments, following the 850 hPa vorticity center. Despite the sub-daily variation in the trajectory, the simulated cyclone track is generally in good agreement with the observed one, until 12Z April 30 when the CTL track persistently deviated east by about 90 km.
Acknowledging this track bias, we next examined the differences in the 850 hPa geopotential height (HGT) between the control simulation and the other experiments to represent Fani's intensity (figure 3(c)), averaged within 12 • lat. × 12 • long. around the cyclone center. Not surprisingly, the removal of the trends in the air and SST (DALL; olivedrab line) results in the strongest increase in cyclone pressure, signifying a reduction in Fani's strength, with the air temperature's warming trend accounting for approximately two-thirds of the cyclone reduction (DAIR) and the SST trend accounting for one-third (DSST). The removal of aerosol (cyan line; RAERO) intensified the tropical cyclone initially, but then weakened it near landfall (May 1).
This result suggests that the effect of the BOB aerosols on Fani is relatively weak compared to that of the air temperature and SST trends, which could be attributed to greenhouse gases induced warming (Wang et al 2013). In terms of precipitation (figure 3(d); averaged within the 12 • lat. × 12 • long. box following Fani), the outcome of the simulations agrees with that of the 850 hPa HGT. The removal of both the air temperature and SST trends (DALL) resulted in a reduction of cyclone precipitation by about 51% during 4/28-5/2, in which the tropospheric warming shows almost twice as strong an effect on Fani (39% reduction) as that of SST warming (22% reduction). These results are consistent with Trenberth et al (2018) in that the increased SST not only enhances fuel available to sustain and intensify a hurricane but also increases its rains upon landfall. As shown in figures 3(d) and (e), DSST and DBSST_RAERO (detrend SST and double trend SST) generally decreased and increased precipitation, respectively. Note that the precipitation response to the detrended IC and BC is much quicker than the pressure field, showing differences at the first forecast hour of the simulations. The removal of aerosols (RAERO), on the other hand, does not show a robust net effect from 4/27-5/3. Without aerosol and its direct and indirect effect with radiation and microphysics, the simulations with removed climate trends show a consistent tendency of a weakened Fani, with increased geopotential height and decreased precipitation of smaller amplitude. Comparison between DALL and DALL_RAERO suggests that aerosol interactions with the atmosphere act to mitigate the warming effect on Fani by reducing the 155 m increase in 850 hPa geopotential height to a smaller level of 138 m, and the 51% decrease in precipitation to 39% from the fiveday average. In other words, Fani might have grown noticeably stronger if the BOB aerosols were not present. Likewise, by doubling the warming trends in the air and SST with the absence of aerosols (DBALL_RAERO), Fani would intensify with a drop of height by 11 m accompanied by a precipitation increase of 15% during 4/28-5/2; this sensitivity test shows the effect of amplified warming and lends support to the detrend simulations.
Any change in the cyclone intensity can affect its trajectory. As shown in figure 3   to different parameterization schemes (table 1) was assessed using the 4 km simulations, as shown in figures 4(a) and (b). By changing the PBL schemes to MYNN and the microphysics schemes to LIN, the different members produce a difference in the 850 hPa geopotential height that is more noticeable in the later period (after 4/30) than before, up to 100 m. The precipitation fluctuation is also more pronounced after 4/30 with a difference up to 5 mm 12 h −1 . The ensemble size of TC simulations is important. Reed et al (2020) and Patricola and Wehner (2018) argued that an ensemble size of at least 10 is necessary for TC attribution experiments. Here, we conducted twelve more simulations using a 24 km grid spacing by following the experiments in table 2 through the combination of three cumulus schemes and two different microphysics schemes and two different PBL schemes, which are described in table 3 and named with 24 km (figure 4). Together with the 4 km cloud-resolving simulations, we produced an ensemble size of 13 for each experiment. Based on the median of 850 hPa geopotential height difference in CTL_24 km relative to CTL (figure 4(c)), the ensemble of simulations appears to overestimate Fani by about 58 m in terms of the 850 hPa HGT. Meanwhile, other 24 km simulations uniformly experienced an intensification of Fani's geopotential height within the range from −11 m (DSST_RAERO_24 km) to −45 m (DBALL_RAERO_24 km). In other words, the coarser-resolution simulations are in general agreement with the convection-permitting ones, suggesting aerosol invigoration achieved by the two-way interaction between convection and microphysics (d) Precipitation ratio (c) Geopotential height difference (m) (b) Precipitation ( a ) Geopotential height (m) Figure 4. (a) domain averaged (12 • lat. × 12 • long. around the CTL cyclone center) 850 hPa geopotential height (HGT, m) (instantaneous value at each 12 h) at 4 km grid spacing; (b) domain averaged simulated precipitation at 4 km grid spacing (mm/12 h, accumulation within the past 12 h; box and whisker plots of (c) the ratio of domain averaged simulated 850 hPa geopotential height and (d) domain averaged simulated precipitation relative to CTL with ensemble members from april 28 to may 2; noted that ensemble size of 13 contains twelve simulations with 24 km grid spacing and 4 km cloud-resolving simulations. Grell-3 YSU Lin 11 Grell-3 MYNN Morrison 12 Grell-3 MYNN LIn interaction might be weaker in the case of Fani. It is likely due to the highly saturated aerosol in the BOB compared to other regions. In terms of the accumulated precipitation, figure 4(d) shows the precipitation ratio compared to CTL during the 5 d. According to the median, CTL_24 km increased precipitation by 22%. By comparing the detrend + aerosol experiments (DAIR_24 km, DSST_24 km, and DALL_24 km) with the aerosol removal simulations (DAIR_RAERO_24 km, DSST_RAERO_24 km, and DALL_RAERO_24 km), the accumulated precipitation was increased up to 6%. These ensemble results suggest that aerosol plays a negative effect on cyclone Fani by weakening its strength and precipitation. The double trend experiments profoundly increased the accumulated precipitation at a range of 18%-23%. Therefore, despite the uncertainty introduced by different model settings, Fani underwent an intensification effect from warming trend (which is mostly warming) accompanied by a weakening effect from anthropogenic aerosols (but less pronounced than warming). These results echo the finding of (Sobel et al 2019) regarding TC potential intensity but differ in that climate warming intensifies a TC more so than aerosols reduce it, at least in the BOB. Reasons of the apparent discrepancy with those of (Sobel et al 2019) are manifold. In the regional model (WRF-Chem), SST is ingested from the GFS forcing data and therefore does not undergo the atmosphere feedback as in the fully coupled global model. The physic schemes, gaseous chemistry and aerosol mechanism all differ between the global and regional models. Moreover, the different resolutions and aerosol components can lead to somewhat different simulation results.

In-cyclone mechanisms
The processes by which the different experiments produced different amounts of precipitation from Fani are discussed as follows. The removal of aerosol interactions did not produce as much of a change in the potential temperature as the other detrend simulations did (figure 5 left column). Most of the detrend simulations produced two stabilization zones, one in the lower troposphere since most of the atmospheric warming took place in the lower levels (not shown), and the other in the upper troposphere, corresponding to the reduced convective precipitation (cf figure 3(d)) and an associated reduction in latent heat release (not shown). Likewise, the double-warming runs without aerosol interactions destabilize the lower troposphere while warming the upper troposphere through enhanced deep convection. Meanwhile, the effect of anthropogenic aerosols is not consistent, at least between April 27 and May 1; this echoes the fluctuated precipitation difference in figure 3(d).
By plotting the vertical profile of the concentration of raindrops ('qnrain' in WRF-Chem) within the 12 • domain centered on Fani, the simulations without engaging the aerosol and its direct and indirect effect (experiments named with RAERO) produced a consistent reduction in the concentration of raindrops, particularly in the middle troposphere (figure 5 right column). This reduction in 'qnrain' echoes the previous simulations of (May et al 2011) and (Hassim et al 2016), who found decreased number concentrations of smaller rain droplets in the core of convective storms during high aerosol conditions and the resultant decline in the deposition growth of cloud ice crystals. Such an effect of anthropogenic aerosols is different from that of sea-salt aerosols, which can strengthen a tropical cyclone through enhancing the condensation process (Jiang et al 2019). (Note that all the detrend simulations with the aerosol module turned off produced the same result in 'qnrain' , hence the single profile of RAERO.) The detrend experiments with an active aerosol interaction suggest that the atmospheric warming trend mitigates the weakening effect of anthropogenic aerosols on the concentration of raindrops bỹ 50%, while the SST trend (or the combination of air and SST trends) completely offset that weakening effect, hence delineating a 'tug of war' in their respective effects on Fani.
We should note that the selected domain size for evaluation may influence the intensity of precipitation since more intense rainfall occurs closer to Fani's center. By checking the vertical profile of 'qnrain' within 6 • domain centered on Fani (not shown), despite more intense rainfall as expected, the variance of 'qnrain' profile is consistent with 12 • x 12 • domain box centered on Fani. These analyses suggest that anthropogenic aerosols affected Cyclone Fani primarily by modulating the raindrop concentration, while climate trends modulated Fani mainly through changing the atmospheric instability.

Concluding remarks
Through the present WRF-Chem simulations, the removal of warming trends from the troposphere and SST appears to reduce Fani's intensity and decrease the cyclone precipitation, while the removal of aerosol interactions in the simulation initially strengthened the tropical cyclone, but later weakened it. However, when aerosols and climate trends were both removed from the simulations, the intensification effect of regional/BOB aerosols and climate warming on Fani (DALL_RAERO) was 'mitigated' by over 49% during April 28-May 2, suggesting certain nonlinear dynamics involved in the aerosol interactions with the increased heat in the air and ocean. An assessment could be made, based on the cyclone precipitation ratio averaged from April 28 to May 2 (figure 3(d)), that the removal of post-1990 regional climate warming alone (while keeping the aerosol effects; DALL) could have increased Fani's precipitation by 51%, from which aerosols would deduct up to 13%. This leaves the net intensification of Fani due to the BOB climate warming and aerosol combined effects to be about 38%. This result provides a rudimentary, yet quantitative estimate for what a premonsoon TC like Fani might have been like had it developed in the pre-1990 atmospheric and oceanic conditions and without the presence of anthropogenic aerosols.
Caution should be exercised when interpreting these statements since the simulations presented here did have drawbacks. First, the cloud resolving (4 km) simulations were not done in a large-ensemble fashion. The challenge lies in that setting up the interactive aerosol module of high-resolution WRF-Chem for large ensembles that involve engaging various physics schemes and different resolutions in the detrend experiments is extremely difficult, for the model would become unstable. As a result, model sensitivity and uncertainty have not been fully evaluated (such as the effect of the ensemble choices on Fani tracks). Second, the regional climate warming effect was estimated in terms of the 'change' in the air and SST, while the aerosols effect was examined in a 'with or without' manner, so the two effects were not analyzed symmetrically. Third, one should expect that the simulation results may fluctuate by changing the period of climate trends and/or using a different model. Finally, one should consider the longterm effect of the persistent increase in anthropogenic aerosols on the intensification of the pre-monsoon trough of the BOB, a trend that has modified the pre-monsoon tropical cyclone activity including frequency and intensity (Wang et al 2013).
Amid the model uncertainty, the simulations conducted and presented here call for future efforts with more sophisticated models in understanding how the rapidly increasing anthropogenic aerosols in the BOB interact with the pronounced warming in producing the compound effects on high-impact tropical cyclones like Fani.