The geographic distribution of anthropogenic aerosol emissions has evolved continuously over the industrial era and is projected to continue to do so. Emissions have transitioned from a locus in western Europe and North America in the late 19th and early 20th century to southern and eastern Asia in the present day due to changing patterns of industrialization and air quality regulation (Hoesly et al., 2018). Future projections of aerosol emissions, while highly uncertain (Samset et al., 2019), contemplate new growth of emissions in regions like South America and East Africa as industrialization of these regions accelerates (Lund et al., 2019).

With this redistribution of aerosol emissions comes the potential redistribution of aerosols' impact on the hydrological cycle, which is known to be substantial. Aerosols have been shown to have strong in situ and remote hydrological-cycle impacts via their influence on the dynamics, thermodynamics, and microphysics that control precipitation (Boucher et al., 2013). Anthropogenic aerosols' (AAs) rapid spatiotemporal evolution over the 20th and early 21st century and their effect on the large-scale circulation have been identified as the dominant driver of an observed southward shift in the tropical Pacific rain belt (Allen et al., 2015) and the collapse and recent recovery in Sahel precipitation (Marvel et al., 2020). Observed weakening of the South Asian monsoon and East Asian monsoon, meanwhile, has been attributed to both large-scale and local-scale aerosol forcing (Bollasina et al., 2011, 2014; B. Dong et al., 2019; Li et al., 2016; Singh et al., 2019).

A range of studies have established that the precipitation response to aerosols is dependent on the spatial distribution of aerosol forcing. The ongoing evolution in the spatial distribution of global aerosol emissions has been associated with an evolution in the spatial pattern of the corresponding global precipitation response (Deser et al., 2020; Kang et al., 2021). A range of studies isolating the response to historical emissions in individual regions or latitude bands have identified common underlying features of this geographic dependence. Multi-model and single-model studies applying both idealized and historical regional aerosol perturbations generally find that higher-latitude aerosol sources (Europe, North America) produce a stronger global-mean precipitation response than lower-latitude sources (southern or eastern Asia) when normalized by radiative forcing or atmospheric concentration (Kasoar et al., 2018; Liu et al., 2018; Shindell et al., 2012; Westervelt et al., 2018). However, some studies show strongly differing spatial distributions of precipitation response (Ishizaki et al., 2013; Liu et al., 2018; Shindell et al., 2012), while others argue that the spatial patterns of response are similar regardless of source region (Kasoar et al., 2018).

Identifying the specific role of emissions location in the climate response to aerosol, however, remains difficult based on the existing literature. Previous studies use actual or scaled historical emissions, which are unequal across regions, latitude bands, and/or time periods. Responses can be normalized by emissions, concentrations, or forcing to approximate the role of source location. For example, Shindell et al. (2012) proposed and estimated a set of “regional precipitation potentials” based on their analysis of the response to historical aerosol emissions in separate latitude bands, which quantify the precipitation response per unit of radiative forcing for aerosol emissions in a given latitude band. However, these approaches assume linearity in the response to different amounts of aerosol. Additionally, existing studies tend to sample the effects of a relatively small subset of regions (generally confined to southern and eastern Asia, North America, and Europe). Given the high uncertainty and potential for growth in aerosol emissions within individual regions outside of this subset (Lund et al., 2019), understanding the relative importance of aerosol emissions from a larger range of regions may be potentially beneficial for near-term climate prediction as well as for fundamental understanding of the climate system response to heterogeneous forcing.

In Persad and Caldeira (2018), we designed a set of simulations in a coupled atmosphere–slab ocean global climate model (GCM) to evaluate the global-scale temperature response to identical aerosols (equal to year 2000 Chinese sulfate, black carbon, and organic carbon emissions) emitted from eight different regions and found a 14-fold range in the global-mean temperature response due solely to differences in emissions source location. By fixing the amount of aerosol based on a historical reference but varying the source location, the simulations isolate the role of the geographic location of emissions in setting the climate response to aerosols within a quasi-realistic framework. This strategy is analogous to Green's function approaches, in which a climate model is perturbed with an identical anomaly in several different locations one by one, which have been used to evaluate radiative feedbacks (Y. Dong et al., 2019; Zhou et al., 2017). In Persad and Caldeira (2018), the anomaly – i.e., fixed emissions of aerosol – is structured to match the national boundaries along which the policy and technological shifts that determine aerosol trends typically occur (O'Neill et al., 2015; Rao et al., 2017; Riahi et al., 2017). The goal is to achieve an experimental setup that provides fundamental physical insight within a framework that is directly translatable to policy-relevant tools like emulators, emissions metrics, or social cost calculations that scale based on national emissions. This allows for an approach for understanding the importance of aerosols' geographic distribution to their climate response that is complementary to the unequal historical emissions or highly idealized forcing experiment designs that have previously been pursued.

In this study, the simulations from Persad and Caldeira (2018) are analyzed to understand the dependence of the hydrological-cycle response to aerosols on emissions source location. Section 2 describes the simulations and analysis techniques used to assess the influence of identical aerosol emissions from different regions on global precipitation. Section 3 presents results assessing impacts on both global-mean precipitation and the spatial distribution of precipitation and explores a potential theory for why certain source regions produce strongly localized precipitation responses and others do not. Section 4 places the findings in the context of existing understanding, and Sect. 5 summarizes and explores the implications of the results in the context of the continuing spatial redistribution of anthropogenic aerosol emissions.

Simulations designed to test the global climate response to identical aerosol emissions in different source regions are conducted in the National Center for Atmospheric Research Community Earth System Model version 1.2 (CESM1) using the Community Atmosphere Model version 5 (CAM5) coupled to the Community Land Model version 4 (CLM4) and a mixed-layer ocean (Hurrell et al., 2013). The three lognormal mode (MAM3) modal aerosol module is used in CAM5, which allows for interactive transport, growth, internal mixing, and removal of aerosol emissions by the internal physics of the model (Liu et al., 2012). This version of the model has been shown to produce minimal (< 10 %) biases in aerosol concentrations and aerosol radiative forcing compared to more complicated atmospheric chemistry models (Ghan et al., 2012; Liu et al., 2012). Simulations in CESM1 (CAM5) using historical aerosol emissions compare well with observations of spatial and temporal patterns of aerosol optical depth (AOD), though low biases in AOD are apparent over southern and eastern Asia. CAM5 allows for simulation of aerosol–cloud interactions within the model's two-moment microphysical parameterization (Liu et al., 2012). Microphysical aerosol–precipitation interactions are permitted within the stratiform cloud microphysics representation but are excluded from the cumulus cloud parameterization (Ghan et al., 2012). The fully coupled CESM1 consistently performs among the top 10 fifth-generation Coupled Model Intercomparison Project (CMIP5) models in simulation of historical temperature and precipitation trends and spatial patterns (Koutroulis et al., 2016).

The simulation suite consists of a control simulation and eight perturbation experiments. In the control simulation, year 2000 conditions are imposed for all external and internal forcers, with the exception of non-biomass-burning anthropogenic emissions of black carbon, organic carbon, sulfur dioxide, and sulfate. These are set to 1850 values using CAM5's standard historical emissions fields (Lamarque et al., 2010). In the eight perturbation experiments, these fields of anthropogenic aerosol emissions are modified only within the relevant region to impose additional total annual emissions of black carbon, organic carbon, and sulfate precursor equivalent to China's total year 2000 values (1.61, 4.03, and 22.4 Tg, respectively). This is achieved by scaling that region's year 2000 CAM5 standard historical emissions fields at each grid point by a fixed factor such that the total change in anthropogenic aerosol emissions between each of the eight perturbation experiments and the control is identical. The spatial distribution of the emissions perturbation thus follows the realistic year 2000 spatial pattern within a given region (Appendix Fig. A1). The eight perturbation regions chosen (Brazil, China, East Africa India, Indonesia, South Africa, the US, and western Europe) are selected to sample a range of past, present, and projected future major emissions source regions as well as a range of climate regimes (e.g., tropical, monsoonal, and extratropical in both hemispheres). A comparable set of simulations are run in atmosphere-only mode with sea surface temperatures (SSTs) and sea ice fixed. Further discussion of simulation characteristics and behavior can be found in Persad and Caldeira (2018).

The atmosphere-only simulations are used to calculate effective radiative forcing (ERF) and atmospheric absorption and to decompose the fast and slow precipitation responses. ERF is calculated as the change in global-mean top-of-atmosphere energy balance in each of the eight atmosphere-only perturbation experiments compared to the atmosphere-only control, following the standard fixed SST definition of ERF (Ramaswamy et al., 2018). The fast precipitation response is calculated as the precipitation response within the atmosphere-only simulation. Atmospheric absorption values shown are also calculated within the atmosphere-only simulation. The slow precipitation response is calculated as the residual of the precipitation response in the slab ocean simulations minus the fast precipitation response.

Model output is produced at a nominal 2∘ × 2∘ longitude–latitude resolution. Slab ocean simulations are run for 100 years, and atmosphere-only simulations are run for 60 years. The initial 40 (slab ocean simulations) or 20 (atmosphere-only simulations) years are treated as the transient response (based on analysis of trends in the top-of-atmosphere energy imbalance) and discarded. The simulation design used here, in which a signal from a given perturbation (e.g., regional aerosol emissions) is characterized by imposing that perturbation as the only modification to a control simulation and running the resulting simulation in repeating annual-cycle mode for an extended period, is a standard methodology used across the aerosol–climate interaction literature. Examples include simulations conducted as part of the Precipitation Driver and Response Model Intercomparison Project (PDRMIP) (Liu et al., 2018; Myhre et al., 2016; Samset et al., 2016) with idealized regional aerosol perturbations and within multi-model (Westervelt et al., 2017, 2018) and single-model experiment designs (Kasoar et al., 2018) simulating removal of present-day aerosol emissions in individual regions. In this experiment design, the perturbation signal is characterized as the difference between the long-term mean of the perturbation and control experiments after they have reached quasi-equilibrium, and the effects of internal variability are estimated using the interannual variability between individual years of the simulation.

One concern with this approach not addressed in prior studies is that persistent modes of internal variability may emerge within the equilibrium simulations and could be conflated with the perturbation signal. While atmosphere-only simulations cannot sustain long-term modes of internal variability, this concern may apply to the slab ocean coupled simulations used here. To address this concern, two approaches are applied in this study. First, statistical significance is estimated using either the last 60 years (slab ocean simulations) or the last 40 years (atmosphere-only) of the simulations as the sample, but effective sample size is adjusted to account for autocorrelation between simulation years following the methodology of Santer et al. (2000). The 95 % confidence level (i.e., 1.96σ) based on year-to-year variability in the difference between the control simulation and each perturbation experiment is provided for all global-mean values, and statistical significance for maps is estimated at the 95 % confidence level using a two-tailed t test, both using this adjusted effective sample size. Second, the slab ocean coupled experiments are repeated with slightly adjusted initial conditions (initial conditions drawn from a different year of the control simulation), allowing for a different trajectory of internal variability to emerge within the equilibrium simulation. Results from this second experiment set are provided in Appendix A and demonstrate that the central findings of the study are unlikely to be the result of persistent modes of internal variability emerging in either equilibrium simulation set but rather can robustly be assumed to result from perturbations of the regional aerosol emissions imposed.

These findings highlight that understanding the processes that control the relative contribution of fast versus slow precipitation response to the total precipitation response is important for constraining the magnitude and spatial distribution of precipitation response to regional aerosol forcing. In particular, understanding the processes controlling the fast precipitation response, namely atmospheric absorption, may provide an important constraint on the expected prevalence of localized precipitation responses to regional aerosol emissions. The dependence of the atmospheric absorption and strength of the fast precipitation response on aerosol location seen here aligns with results from highly idealized studies. Dagan et al. (2019) forced an aquaplanet atmospheric general circulation model (ICON, ICOsahedral Nonhydrostatic) with equivalent, radially symmetric absorbing aerosol optical depth plumes in the deep tropics versus mid-latitudes and found higher resulting atmospheric absorption in the deep tropics due to stronger cloud feedbacks. However, the aquaplanet formulation reduces the comparability of the resulting fast precipitation responses with those seen in this study. A follow-on study in the same atmosphere-only model with a realistic land surface found a stronger local reduction in fast precipitation response over land in response to a tropical scattering AOD plume than to a comparable higher-latitude plume, though the use of a purely scattering plume as opposed to the mixed scattering and absorbing aerosols used in this study again limits direct comparison (Dagan et al., 2021). Regardless, the results of this analysis indicate that, particularly for tropical aerosol sources, the resulting amount of atmospheric absorption – and, consequently, the strength of the fast precipitation response – can be a strong determinant of the overall precipitation response. The importance of the atmospheric absorption for the precipitation response is particularly notable, since scattering rather than absorption by the mixed aerosol emissions in this study dominates the temperature and overall radiative response (Fig. 2b and Persad and Caldeira, 2018). There are known model biases and limited observational constraints on atmospheric absorption, particularly at the regional scale (Samset et al., 2018). The importance to the hydrological response seen here, however, reinforces the need for improved observations and modeling of the processes that control atmospheric absorption.

The regions that manifest local fast versus slow precipitation responses in the simulations analyzed here also overlap with regions identified in existing studies, including those utilizing Precipitation Driver and Response Model Intercomparison Project simulations (PDRMIP; Myhre et al., 2016). Samset et al. (2018) evaluated the regions for which fast precipitation responses dominate slow precipitation responses for PDRMIP multi-model simulations of idealized global forcings, including 10 times present-day global black carbon emissions and 5 times present-day global sulfate emissions. Although the spatial pattern of imposed perturbation differs from this study (i.e., globally distributed versus regionally confined perturbations), they also find that the total precipitation response to both global black carbon (BC) and global sulfate is dominated by the fast response over parts of southern Asia and most of the African continent. High-latitude precipitation responses to these two forcers, meanwhile, are dominated by the slow precipitation response in the multi-model simulations (Samset et al., 2016), though individual models show conflicting results (Zhang et al., 2021). Similar PDRMIP multi-model simulations with regional idealized aerosol emissions over Asia and Europe (Liu et al., 2018), however, also showed a strong local fast precipitation response to Asian aerosols and almost no fast precipitation response to European aerosols. The appearance of a fast precipitation response in low-latitude continental regions in response to both localized and global-scale aerosol forcing and the absence of one at high latitudes thus appears to be a robust feature across models and aerosol perturbation setups.

The latitudinal dependence of the strength of the fast precipitation response and consequently the total local precipitation response to regional aerosol emissions may bear the signature of other processes that differentiate between the tropics and extratropics. For example, differences in local energy budget closure and the strength of horizontal energy and moisture gradients between the tropics and extratropics, associated with poleward strengthening of the Coriolis force, have been leveraged to explain the latitudinal dependence of fast precipitation responses to idealized aerosol forcing (Dagan et al., 2019, 2021). The relative role of local- versus large-scale precipitation processes in supplying precipitation to a region may also play a role in its susceptibility to in situ aerosol forcing. The tropics and extratropics differ strongly in the proportion of total precipitation that is supplied by convective versus large-scale precipitation (Fig. A7), though specific patterns are highly model dependent (Dai, 2006; Kyselý et al., 2016). To the (imperfect) extent to which these two model-derived flavors of precipitation correspond with precipitation that is controlled by local- versus large-scale processes in the real world (Norris et al., 2021), regions with climatological precipitation dominated by convective (i.e., local-scale) precipitation may be more susceptible to local aerosol forcing. Large-scale precipitation processes, meanwhile, may be relatively insensitive to localized forcing from regional aerosol emissions, since they are more strongly controlled by large-scale moisture and energy gradients (Wang et al., 2021).

The plausibility of this process is hinted at by the fact that regions that exhibit strong in situ precipitation responses to local aerosol emissions (Figs. 1, 4a) are also those whose climatological precipitation is overwhelmingly supplied by convective precipitation (Fig. A7). Conversely, the regions that show a negligible local response are those in which climatological precipitation is partly or primarily supplied by large-scale precipitation. Additionally, the local precipitation response is dominated by convective precipitation change in all cases (Fig. A8). However, it should be noted that CAM5, like many CMIP5 and CMIP6 generation models, includes aerosol microphysical effects on precipitation in its convective precipitation scheme but not its large-scale precipitation scheme (Hurrell et al., 2013). Thus, this signal may be largely dependent on the parameterization approach. Applying well-established energetic analysis approaches (Dagan et al., 2021; Liu et al., 2018; Ming et al., 2010; Zhang et al., 2021) to more realistic aerosol perturbations in combination with moisture-tracking algorithms (Mei et al., 2015) that allow for better characterization of precipitation sources could help further clarify the potential role of local- versus large-scale precipitation controls in determining the emergence of in situ precipitation responses to regional aerosols.

The greater capability of higher-latitude emissions sources at generating total global-mean precipitation change also appears to be a robust feature of the response to aerosols. Studies analyzing the global-mean precipitation response to removal of present-day aerosols from individual regions find that removal of European or North American emissions generates stronger global-mean precipitation per unit of radiative forcing than removal of southern or eastern Asian emissions (via fully coupled HadGEM3-GA4 simulations in Kasoar et al., 2018; via fully coupled GISS – Goddard Institute for Space Studies – E2, GFDL – Geophysical Fluid Dynamics Laboratory – CM3, and NCAR – National Center for Atmospheric Research – CESM1 simulations in Westervelt et al., 2018), in line with the findings here. This reinforces the latitudinal dependence in the climate response to heterogeneous regional forcing found in earlier studies (Shindell and Faluvegi, 2009; Shindell et al., 2012) and indicates that it continues to apply as the forcings become more regionalized.

Despite the promising alignment of this study's findings with prior single- and multi-model work, the single-model, slab ocean setup used here for computational tractability may create biases that encourage future multi-model coupled investigation of these questions. The regional aerosol response in the fully coupled model NCAR CESM1 is comparable with results from other contemporary coupled models, such as GFDL CM3, but the ITCZ response to aerosols is somewhat stronger (Westervelt et al., 2018). Ocean dynamical adjustments present in the fully coupled model but not in the slab ocean configuration used here can either damp or amplify the response to aerosols, depending on the spatial pattern of forcing (Kang et al., 2021). In particular, they may damp ITCZ shifts relative to those seen in a slab ocean model (Zhao and Suzuki, 2019). Nevertheless, slab ocean models can provide valuable insights into hydrological-cycle responses to anthropogenic forcings when computational efficiency is needed (Held and Soden, 2006; Ming and Ramaswamy, 2009). The strong dependence of the hydrological response to aerosols on regional distribution seen here and in other studies, across perturbation setups and models, highlights the importance of continued investment in developing a comprehensive and consensus theory of what drives this dependence.

This study identifies a strong dependence of the global-mean precipitation response and its spatial distribution on the geographic location of given aerosol emissions. Coupled atmosphere–slab ocean GCM simulations, in which an identical, quasi-realistic amount and composition of aerosol is separately placed into a range of emitting regions, are used to isolate the importance of source location in determining aerosols' hydrological effects. A 6-fold difference in global-mean precipitation response emerges, largely driven by a 14-fold difference in global-mean temperature response. This arises from a combination of differing strengths of global-mean radiative forcing generated by each region's emissions and diversity in the efficacy of that radiative forcing at generating cloud and ice albedo feedbacks (Persad and Caldeira, 2018). Major distinctions in the geographic distribution of the precipitation response, particularly the prevalence of local versus remote responses, also arise. Tropical regions, which also tend to have the greatest societal vulnerability to precipitation disruption, are the most susceptible to dynamically driven precipitation changes generated by both local and remote aerosol emissions, as has been highlighted elsewhere (e.g., Scannell et al., 2019; Westervelt et al., 2018; Zanis et al., 2020). However, these regions are also the least effective at generating remote precipitation responses. Conversely, mid-latitude emissions sources are highly effective at generating global-mean precipitation change, but these changes occur almost entirely outside of the source region. These results indicate that the time-evolving geographic distribution of global aerosol emissions may have substantial implications for the precipitation stability of vulnerable regions outside of key past or projected emissions hotspots. Building a comprehensive theory of what determines the relative strength of local versus remote hydrological impacts of changes in regional aerosol emissions will be important to understanding and predicting these trends.

The continuous spatial redistribution of aerosol emissions constitutes an ongoing source of uncertainty and variability in the global hydrological cycle (Deser et al., 2020), and near-term regional aerosol trends may be a major determinant of climate risk over the next several decades (Luo et al., 2020; Samset et al., 2019). Notably, the regions identified in this study to have the strongest global-mean and remote precipitation impacts (Europe and the US) have seen strongly declining aerosol emissions since the mid-20th century (Hoesly et al., 2018). Conversely, those with the weakest global-mean but the strongest local precipitation impacts (e.g., India, Indonesia, East Africa) are among those in which emissions could continue to increase substantially through the mid-21st century (Lund et al., 2019). This implies that the future spatial distribution of aerosol emissions may have a lower overall effectiveness at changing global-mean precipitation but may preferentially concentrate precipitation impacts into vulnerable regions. Known nonlinearity in the climate response to simultaneous regional aerosol variations (Herbert et al., 2021) limits the direct application of the regional dependence quantified here to estimate how ongoing spatial redistribution of aerosol emissions will affect global precipitation patterns. However, the strong dependence on source location seen in this study demonstrates that the geographic distribution of aerosol emissions must be accounted for when quantifying the human influence on global precipitation patterns.