Abstract

Through the latter half of the twentieth century, meridional shifts in tropical precipitation have been associated with severe droughts. Although linked to a variety of causes, the origin of these shifts remains elusive. Here, it is shown that they are unlikely to arise from internal variability of the climate system alone, as simulated by coupled ocean–atmosphere climate models. Similar to previous work, the authors find that anthropogenic and volcanic aerosols are the dominant drivers of simulated twentieth-century tropical precipitation shifts. Models that include the cloud-albedo and lifetime aerosol indirect effects yield significantly larger shifts than models that lack aerosol indirect effects and also reproduce most of the southward tropical precipitation shift in the Pacific. However, all models significantly underestimate the magnitude of the observed shifts in the Atlantic sector, unless driven by observed SSTs. Mechanistically, tropical precipitation shifts are driven by interhemispheric sea surface temperature variations, which are associated with hemispherically asymmetric changes in low-latitude surface pressure, winds, and low clouds, as well as the strength, location, and cross-equatorial energy transport of the Hadley cells. Models with a larger hemispheric aerosol radiative forcing gradient yield larger hemispheric temperature contrasts and, in turn, larger meridional precipitation shifts. The authors conclude that aerosols are likely the dominant driver of the observed southward tropical precipitation shift in the Pacific. Aerosols are also significant drivers of the Atlantic shifts, although one cannot rule out a contribution from natural variability to account for the magnitude of the observed shifts.

1. Introduction

Shifts in tropical precipitation, associated with north–south displacements of the intertropical convergence zone (ITCZ), have been observed going back to the last ice age (Peterson et al. 2000; Haug et al. 2001; Arbuszewski et al. 2013) and since the industrial revolution (Ridley et al. 2015) based on paleorecords, as well as through the latter half of the twentieth century based on rain gauges (e.g., Folland et al. 1986). These shifts are closely related to sea surface temperature (SST) variations (Folland et al. 1986; Giannini et al. 2003; Hoerling et al. 2006; Zhang and Delworth 2006; Schneider et al. 2014), particularly interhemispheric variations, whereby relative warming (cooling) of the Northern Hemisphere (NH) is associated with a northward (southward) shift in tropical precipitation (Broccoli et al. 2006; Allen and Sherwood 2011; Ming and Ramaswamy 2011; Hwang et al. 2013; Haywood et al. 2013). Studies have attributed these precipitation shifts and the interhemispheric SST variations to natural climate variability, including the Atlantic multidecadal oscillation (AMO)/Atlantic meridional overturning circulation (AMOC) (Zhang and Delworth 2006) and volcanic aerosols (Haywood et al. 2013), as well as to external forcings, including anthropogenic sulfate aerosols (Williams et al. 2001; Rotstayn and Lohmann 2002; Biasutti and Giannini 2006; Chang et al. 2011; Ackerley et al. 2011; Hwang et al. 2013). Extratropical thermal forcings—such as changes in high-latitude sea ice—can also displace the latitude of the ITCZ (Chiang and Bitz 2005; Broccoli et al. 2006; Kang et al. 2008; Frierson and Hwang 2012; Chiang and Friedman 2012).

Over the latter half of the twentieth century, from ~1950 to ~1985, tropical precipitation shifted southward, driving the Sahelian and Amazonian droughts (Folland et al. 1986; Hwang et al. 2013). Since ~1985, there has been a weaker northward recovery. Attribution studies suggest that these shifts are partially anthropogenically forced, primarily by sulfate aerosols (Rotstayn and Lohmann 2002; Hwang et al. 2013). The increase in sulfate emissions up to the mid-1980s preferentially cooled the NH, reducing the interhemispheric SST gradient, leading to a southward shift in the mean meridional circulation (MMC) and the ITCZ. The (weaker) northward recovery since ~1985 is consistent with a concurrent decrease in NH sulfate emissions (Lamarque et al. 2010) and preferential warming of the NH. Climate models, however, fail to capture the full magnitude of the observed tropical precipitation shifts—particularly the southward shift associated with the Sahelian and Amazonian droughts (Hwang et al. 2013).

A possible reason for poor simulation of the southward shift is related to underestimation of a contribution from natural variability, such as changes in the AMO/AMOC (Zhang and Delworth 2006). Thompson et al. (2010) and Dima and Lohmann (2010) suggest that an abrupt change in the AMOC, in response to a rapid freshening known as the “Great Salinity Anomaly,” is primarily responsible for the interhemispheric SST shift (relative NH cooling) around 1970. Similarly, Terray (2012) argues that internal ocean variability is primarily responsible for this abrupt interhemispheric SST shift. Friedman et al. (2013) show that CMIP5 models do not capture this shift, leading them to suggest that it is likely due to internal variability. They also, however, find that some individual model realizations produce a realistic interhemispheric SST shift at this time. It therefore remains an open question as to whether anthropogenic or natural factors—or a combination of both—are the dominant driver of late twentieth-century tropical precipitation shifts.

In this paper, we examine tropical precipitation shifts since 1950 using several observational datasets, archived experiments from phase 5 of the Coupled Model Intercomparison Project (CMIP5) (Taylor et al. 2012), and idealized simulations with the Community Atmosphere Model, version 3 (CAM3) (Collins et al. 2004) and version 5 (CAM5) (Neale et al. 2012). Similar to previous work, we find that anthropogenic and volcanic aerosols are the dominant drivers of simulated twentieth-century tropical precipitation shifts. Models that include both aerosol indirect effects (cloud-albedo and lifetime effects) yield significantly larger shifts than models that lack aerosol indirect effects (AIE) and also reproduce most of the observed southward tropical precipitation shift in the Pacific. However, all models significantly underestimate the magnitude of the observed shifts in the Atlantic sector, unless driven by observed SSTs (and hence, the real-world variation of interhemispheric SSTs). Models with a larger hemispheric aerosol radiative forcing gradient (i.e., those with aerosol indirect effects) excite a dynamical response involving tropical ocean–atmosphere coupling, which drives both low-latitude Northern Hemisphere cooling and low-latitude Southern Hemisphere warming. These models therefore yield larger hemispheric temperature contrasts and, in turn, larger meridional precipitation shifts that better resemble observations. This paper helps to explain why there is a current lack of model consensus on the role aerosols play in driving recent tropical precipitation shifts and identifies the reason why some models simulate a larger fraction of the observed shifts.

This paper is organized as follows. Section 2 discusses the datasets and methods used, including the observed precipitation data and the model experiments. Results are presented in section 3, including the relationship between the interhemispheric aerosol radiative forcing gradient and the magnitude of the simulated tropical precipitation shift. Conclusions and a discussion are presented in section 4.

2. Data and methods

a. Data

Monthly precipitation anomalies at 5° × 5° resolution come from the Global Historical Climatology Network (GHCN) (Peterson and Vose 1997), a database of precipitation from land surface stations that have been subjected to quality control. To minimize trend errors, only those grid boxes with at least 75% valid (nonmissing) years over the entire time period (1950–2012) and over each of the time periods (1950–85 and 1985–2012) are used. A valid annual mean requires three of four valid seasons, defined as January–March, April–June, July–September, and October–December. A valid season requires at least one valid month. Additional precipitation datasets include the University of Delaware dataset (UDEL) (Willmott and Matsuura 1995), the University of East Anglia Climate Research Unit dataset (CRU) (Harris et al. 2014), and the National Oceanic and Atmospheric Administration’s Precipitation Reconstruction over Land (PREC/L) (Chen et al. 2002). Each use GHCN stations, in addition to other sources of gauge data, and use interpolation methods to construct estimates over all land areas. Because these alternate datasets are based on GHCN stations, and include spatially interpolated values over relatively large areas, we primarily (but not exclusively) focus on the GHCN precipitation dataset. We also use the Global Precipitation Climatology Project (GPCP) (Adler et al. 2003), a 2.5° × 2.5° resolution monthly precipitation product based on precipitation estimates from low-orbit satellite microwave data, geosynchronous-orbit satellite infrared data, and surface rain gauge observations, to estimate tropical precipitation shifts over both ocean and land from 1985 to 2012.

Temperature observations come from the 5° × 5° HadCRUT4 dataset, a joint effort between the Met Office Hadley Centre and the Climatic Research Unit at the University of East Anglia (Morice et al. 2012). HadCRUT4 consists of monthly temperature anomalies from a blend of a land surface air temperature dataset (CRUTEM4) and sea surface temperature dataset (HadSST3). We also use temperature and other data, such as cloud cover and surface wind speeds, from several reanalyses including the Twentieth Century Reanalysis (20CR) (Compo et al. 2011), the NCEP–NCAR reanalysis (R1) (Kalnay et al. 1996), and the European Centre for Medium-Range Weather Forecasts twentieth-century pilot reanalysis (ERA-20C) (Poli et al. 2013).

Table 1 lists the CMIP5 models used in this study. Experiments include twentieth-century all forcing (ALL), greenhouse-gas-only forcing (GHG), natural forcings only (NAT), anthropogenic-aerosol-only forcing (AA), preindustrial control (PIC), and all forcings with observed sea surface temperatures (AMIP). The first five experiments are coupled ocean–atmosphere experiments; AMIP experiments are atmosphere-only experiments driven with observed SSTs. To extend the CMIP5 historical simulations beyond their ending year of 2005, we use the historical extended experiments or representative concentration pathway 4.5 (RCP4.5). Since few models archived historical extended experiments (only 10 are used here, as indicated in Table 1), RCP4.5 is primarily used to extend the historical simulations through 2012. Although only four AMIP models extend as far back as 1950 (most begin in 1979), we also use a larger set of 29 AMIP models to quantify tropical precipitation shifts from 1985 to 2012. Similar results for 1985–2012 are obtained with both AMIP subsets.

Table 1.

CMIP5 models and number of simulations used for each experiment, including ALL, GHG, NAT, AA, and AMIP. PIC refers to the number of years available for a model’s preindustrial control integration. Asterisks indicate that the twentieth-century experiment was extended through 2012 using the corresponding historical extended experiment. (Expansions of acronyms are available at http://www.ametsoc.org/PubsAcronymList.)

CMIP5 models and number of simulations used for each experiment, including ALL, GHG, NAT, AA, and AMIP. PIC refers to the number of years available for a model’s preindustrial control integration. Asterisks indicate that the twentieth-century experiment was extended through 2012 using the corresponding historical extended experiment. (Expansions of acronyms are available at http://www.ametsoc.org/PubsAcronymList.)
CMIP5 models and number of simulations used for each experiment, including ALL, GHG, NAT, AA, and AMIP. PIC refers to the number of years available for a model’s preindustrial control integration. Asterisks indicate that the twentieth-century experiment was extended through 2012 using the corresponding historical extended experiment. (Expansions of acronyms are available at http://www.ametsoc.org/PubsAcronymList.)

Model precipitation data are spatially interpolated to the GHCN grid boxes, and missing GHCN data are used to screen the model precipitation data (i.e., model precipitation is restricted to the grid boxes with valid GHCN data). Thus, our tropical precipitation shifts are over land. Nonetheless, models yield similar tropical precipitation shifts when both land and ocean are used (section 3a).

Interhemispheric radiative forcing estimates from the Monitoring Atmospheric Composition and Climate (MACC) aerosol reanalysis (Bellouin et al. 2013) are also used. The reanalysis assimilates total aerosol optical depth from the Moderate Resolution Imaging Spectroradiometer (MODIS) to correct for model (the Integrated Forecast System) departures from the observed aerosols. MACC reanalysis includes shortwave direct and first indirect radiative forcing estimates (60°S–60°N) of anthropogenic aerosols from 2003 to 2010. Direct forcing is computed for anthropogenic aerosols as the difference between an atmosphere containing all aerosols and an atmosphere containing natural aerosols only. The aerosol direct radiative forcing is scaled from cloud-free to all-sky conditions by multiplying the cloud-free direct radiative forcing (at each grid box) by the cloud-free fraction simulated by the model (this assumes no direct radiative forcing under cloudy-sky conditions). The all-sky direct radiative forcing is then corrected, based on the Hadley Centre climate model, to address differences between present-day natural aerosols and preindustrial aerosols as the reference state for the forcing. Aerosol indirect forcing is calculated as the change in cloud albedo exerted by a change in cloud droplet number concentration due to anthropogenic aerosols (Quaas et al. 2008) and relies on a statistical analysis of satellite retrievals and model-simulated cloud cover. There has been debate over the validity over this approach—specifically the use of aerosol optical depth as a proxy for cloud condensation nuclei number concentrations—with some suggesting it leads to an underestimation of aerosol indirect radiative forcing (Andreae 2009) and others suggesting it leads to an overestimation (Grandey and Stier 2010).

b. Methods

Meridional precipitation shifts are based on the trend of the intertropical precipitation index, defined as the difference between NH tropical (NHT; 0°–20°N) and Southern Hemisphere (SH) tropical (SHT; 0°–20°S) annual precipitation anomalies (NHT − SHT PRECT). Precipitation shifts are estimated over the Atlantic (75°W–30°E) and Pacific (30°E–75°W) sectors and are denoted as NHT − SHT Atlantic PRECT and NHT − SHT Pacific PRECT, respectively. We note that our definition of the “Pacific” sector includes the Indian Ocean; thus, our Pacific sector refers to everything except the Atlantic. We focus on a simple metric to capture the large-scale response and to minimize complications in isolating the exact latitude of the tropical rainfall maximum (which are partially due to the double rainfall maximum in most climate models). Results presented here are robust to changing the latitudinal boundaries of this index (i.e., using 0°–10°N and 0°–10°S yields similar conclusions).

Interhemispheric temperature trends are denoted as NH − SH TS for the entire globe (90°S–90°N) and NHT − SHT TS for the tropics (20°S–20°N) only. Interhemispheric temperature trends over the Atlantic and Pacific sectors follow the same nomenclature as that for the regional precipitation shifts but are calculated over ocean only (i.e., NH − SH Atlantic TS is based on SSTs from 90°S–90°N, 75°W–30°E, and NHT − SHT Atlantic TS is based on SSTs from 20°S–20°N, 75°W–30°E). Similar definitions apply to other climate variables (e.g., interhemispheric sea level pressure trends). All analyses (including precipitation, temperature, etc.) are based on monthly anomalies using the 1961–90 base period.

PIC experiments are used to generate a total of 21 444 30-yr (overlapping) segments for the distribution of unforced intertropical precipitation trends. A total of the number of PIC years minus 30 yr plus one 30-yr overlapping segments is used for each model, resulting in 21 444 realizations in total. The 95% confidence interval is calculated as the range in which 95% of the samples fall. Nonoverlapping 30-yr segments yielded similar PIC distributions.

Trends are estimated by taking a least squares trend of the 5-yr smoothed annual mean time series. Trend significance is based on a standard Student’s t test, with the effects of autocorrelation accounted for by reducing the sample size according to (1 − ρ1)(1 + ρ1)−1, where ρ1 is the lag-1 autocorrelation (Wilks 2006). The Wilcoxon–Mann–Whitney rank-sum test (Wilks 2006) is also used to evaluate the significance of the forced trends relative to the unforced, preindustrial control trends. This allows comparison of the CMIP5 all- and miscellaneous-forcing trends to the corresponding trends from the PIC experiments. Similar results are obtained using a standard Student’s t test for the difference between two independent samples and the pooled variance (Wilks 2006). Significance of correlations is estimated assuming a Student’s t distribution given by t = r(1 − r2/n − 2)−0.5, where n is the number of years (or model realizations) and r is the correlation.

Aerosol radiative forcing (RF) is obtained by differencing the net top-of-the-atmosphere (TOA) shortwave (SW) radiative flux between the CMIP5 sstClim and sstClimAerosol experiments (Ghan 2013). The sstClim experiment is integrated with climatological SSTs and sea ice from the preindustrial period. The sstClimAerosol experiment is analogous but uses anthropogenic aerosols based on the year 2000. This is referred to as the “effective radiative forcing,” and it includes rapid adjustments, like changes in atmospheric lapse rates and clouds (indirect and semidirect effects), in response to aerosols. When not available in the CMIP5 archive, we include estimates of the aerosol RF (using the same procedure) from the Atmospheric Chemistry and Climate Model Intercomparison Project (Shindell et al. 2013). We note that only shortwave radiative fluxes are used because the longwave (LW) contribution to aerosol RF is much smaller than the SW contribution. We obtain similar results using aerosol RF calculations based on SW TOA fluxes only or both SW and LW TOA fluxes. Moreover, only TOA SW RF is provided by the MACC aerosol reanalysis, which we compare the models to (model RF calculations are also restricted to 60°S–60°N, as in MACC reanalysis).

Atlantic multidecadal variability (AMV) is calculated as the area-weighted average North Atlantic (0°–70°N, 75°W–30°E) SST anomaly, and the long-term trend is removed (except for PIC). The tropical component of AMV (TAMV) is estimated similarly but based on tropical North Atlantic (0°–20°N, 75°W–30°E) SSTs. The Atlantic meridional mode (AMM) is calculated as the second principal component of monthly SST anomalies from 21°S–32°N, 74°W–15°E. Data are first detrended and a linear fit to the cold tongue index subtracted from each spatial point (Chiang and Vimont 2004). AMV/AMM variability is removed by linearly regressing the monthly AMV/AMM time series onto the appropriate field (e.g., precipitation). This yields a regression coefficient for each grid point, which is multiplied by the AMV or AMM time series and then subtracted from the field. The corresponding intertropical precipitation trend is then estimated, with AMV or AMM variability removed. In addition to natural ocean-circulation-driven changes associated with the AMO/AMOC, Booth et al. (2012) show that aerosols can externally force AMV, although the relative magnitudes of natural and forced variability remain uncertain (Zhang et al. 2013). Similar to Martin et al. (2014), we use AMV as an overarching term that represents contributions to SST variability from the AMO, AMM, and potential external forcing.

Tropical precipitation shift sensitivity to the interhemispheric temperature gradient is calculated by regressing the time series of the tropical precipitation shift onto the hemispheric temperature contrast. The corresponding regression coefficient is used for the sensitivity. This provides a measure of the magnitude of the precipitation shift associated with +1 K of relative warming in the NH. Sensitivities are based on 1950–2012 in observations, CMIP5 ALL, and AMIP. CMIP5 PIC sensitivities are obtained using all of the available PIC years (Table 1).

Northward cross-equatorial moist static energy (MSE) flux is estimated by integrating the equation for the atmospheric energy budget (assuming steady state):

 
formula

where QS is the net shortwave radiation at the top of the atmosphere, QL is the outgoing longwave radiation, QO is the net downward surface flux (including latent and sensible fluxes; Frierson and Hwang 2012), a is the radius of Earth, ϕ is latitude, and λ is longitude. CMIP5 northward ocean heat flux (OHF) data are interpolated to a regular grid using the Earth System Modeling Framework (ESMF) software, and trends in the northward cross-equatorial (2.5°S–2.5°N) OHF are also calculated. However, only approximately one-third of the CMIP5 models are archived northward OHF.

The Hadley cell is quantified in terms of the MMC by calculating the northward mass flux above 500 hPa. The maximum MMC is found at this level in both hemispheres. Zonal displacements of the MMC are based on the trend of the latitude of the maximum MMC. In the NH, the MMC is positive (clockwise), so a positive trend represents strengthening of the NH Hadley cell. In the SH, the MMC is negative (counterclockwise), so a negative trend represents strengthening of the SH Hadley cell.

Several CAM3 SST sensitivity experiments are performed to isolate the effects of Atlantic SSTs on tropical precipitation shifts. In these experiments, the Atlantic is defined as 70°S–70°N, 75°W–30°E. A control experiment is performed using global observational SSTs from 1940 to 2012. Several experiments are performed by keeping SSTs in specific regions (e.g., Atlantic sector) held fixed at their 1940 values. Taking the difference with the control experiment yields the climate response due to the temporal evolution of SSTs in that region. This assumes that the response is linear to the SST evolution in different regions. A total of three ensemble members are performed for each experiment. Different ensemble members are generated by applying a random perturbation to the surface temperature field (in 1940). Thus, these experiments allow 10 years of spinup—or, more specifically, for the atmosphere to diverge from the initial state in 1940. CAM3 SST experiments are driven with the Hadley Centre Sea Ice and Sea Surface Temperature dataset (HadISST) (Rayner et al. 2003).

Idealized simulations using CAM5, coupled to a slab ocean model (SOM), are also performed to evaluate the effects of the interhemispheric aerosol RF gradient on the interhemispheric temperature and tropical precipitation shifts. The monthly TOA SW anthropogenic aerosol radiative forcing from the Hadley Centre Global Environmental Model, version 2–Atmosphere [HadGEM2-A; the atmospheric component of the coupled earth system model (HadGEM2-ES)], is inserted into the CAM5 radiation module. TOA SW aerosol RF from HadGEM2-A is used because it exhibits a relatively large interhemispheric gradient, particularly in the Atlantic sector (NH − SH Atlantic RF from HadGEM2-A is −2.3 W m−2, compared to CAM5 at −1.0 W m−2). At every time step and grid box, CAM5’s default surface solar radiation is modified according to HadGEM2-A’s TOA SW aerosol RF (as in Allen and Sherwood 2011). The response is then estimated by differencing the experiment with a corresponding integration that lacks HadGEM2-A’s aerosol RF. Similar CAM5 experiments are performed using MACC TOA SW direct effect due to mineral dust (which also exhibits a relatively large interhemispheric gradient in the Atlantic sector at −2.85 W m−2). CAM5 SOM experiments are integrated for 60 yr, the last 40 yr of which are analyzed. We note that our CAM5 SOM results may be sensitive to the assumed depth of the ocean mixed layer (Donohoe et al. 2014).

3. Results

a. Observed and modeled tropical precipitation shifts

Observed precipitation trends from GHCN and CRU show that much of the tropical NH landmasses experienced a decrease in precipitation from 1950 to 1985, including northern Africa, the Caribbean, and Indonesia (Fig. 1). At the same time, the tropical SH landmasses experienced an increase in precipitation, most notably central South America. This pattern, however, largely reverses from 1985 to 2012. This is consistent with meridional shifts in the intertropical convergence zone, as quantified by the difference between NH tropical (0°–20°N) and SH tropical (0°–20°S) annual mean precipitation anomalies (NHT − SHT PRECT). GHCN shows a significant southward shift of −57.3 mm decade−1 from 1950 to 1985, followed by a significant northward recovery of 25.9 mm decade−1 from 1985 to 2012 (Table 2). Similarly, CRU shows a significant southward shift of −40.2 mm decade−1 from 1950 to 1985 and a nonsignificant northward recovery of 8.7 mm decade−1 from 1985 to 2012. Similar meridional displacements—particularly the southward shift from 1950 to 1985—also exist in alternate precipitation datasets over all land areas, including UDEL and PREC/L. The range of observed trends is from −34.8 to −57.2 mm decade−1 for 1950–85 and from 4.6 to 25.9 mm decade−1 for 1985–2012 (Table 2). Observations generally show similar trends—including a southward shift from 1950 to 1985 and a northward recovery from 1985 to 2012—in both the Atlantic (75°W–30°E) and Pacific (30°E–75°W) sectors. Reanalyses yield similar shifts, particularly during 1950–85.

Fig. 1.

Observed annual mean 1950–85 and 1985–2012 tropical precipitation trends (mm decade−1) based on (a),(b) GHCN and (c),(d) CRU.

Fig. 1.

Observed annual mean 1950–85 and 1985–2012 tropical precipitation trends (mm decade−1) based on (a),(b) GHCN and (c),(d) CRU.

Table 2.

Observed and model-mean intertropical precipitation trends (mm decade−1). Trends are based on the 5-yr smoothed annual average time series for 1950–85 (P1) and 1985–2012 (P2). Also included are the corresponding trends based on the three models—HadGEM2-ES, MIROC-ESM, and MIROC5—that yield the largest hemispheric aerosol radiative forcing gradient in the Atlantic (denoted as ALL BOTH AIE′). AMIP P2 corresponds to a larger set (29 models) of AMIP models (since most begin in 1979). Models and reanalyses have been subsampled according to GHCN observations, and thus represent land-based tropical precipitation shifts. Corresponding shifts over both land and ocean are also included in parentheses (GPCP land and ocean 1985–2012 shifts are also included). Trend significance is denoted by boldface values for ≥90%, a single asterisk for ≥95%, and double asterisks for ≥99%.

Observed and model-mean intertropical precipitation trends (mm decade−1). Trends are based on the 5-yr smoothed annual average time series for 1950–85 (P1) and 1985–2012 (P2). Also included are the corresponding trends based on the three models—HadGEM2-ES, MIROC-ESM, and MIROC5—that yield the largest hemispheric aerosol radiative forcing gradient in the Atlantic (denoted as ALL BOTH AIE′). AMIP P2 corresponds to a larger set (29 models) of AMIP models (since most begin in 1979). Models and reanalyses have been subsampled according to GHCN observations, and thus represent land-based tropical precipitation shifts. Corresponding shifts over both land and ocean are also included in parentheses (GPCP land and ocean 1985–2012 shifts are also included). Trend significance is denoted by boldface values for ≥90%, a single asterisk for ≥95%, and double asterisks for ≥99%.
Observed and model-mean intertropical precipitation trends (mm decade−1). Trends are based on the 5-yr smoothed annual average time series for 1950–85 (P1) and 1985–2012 (P2). Also included are the corresponding trends based on the three models—HadGEM2-ES, MIROC-ESM, and MIROC5—that yield the largest hemispheric aerosol radiative forcing gradient in the Atlantic (denoted as ALL BOTH AIE′). AMIP P2 corresponds to a larger set (29 models) of AMIP models (since most begin in 1979). Models and reanalyses have been subsampled according to GHCN observations, and thus represent land-based tropical precipitation shifts. Corresponding shifts over both land and ocean are also included in parentheses (GPCP land and ocean 1985–2012 shifts are also included). Trend significance is denoted by boldface values for ≥90%, a single asterisk for ≥95%, and double asterisks for ≥99%.

Figures 2a,b show the intertropical precipitation trend probability density functions based on observations and several CMIP5 experiments (Table 1)—including PIC—subsampled to the GHCN data. The distribution of unforced 30-yr trends is obtained from the CMIP5 preindustrial control simulations (section 2b). The 1950–85 GHCN trend is larger in magnitude than all 21 444 PIC realizations; the 1985–2012 GHCN trend is larger than 99.3% of the PIC realizations. Both GHCN trends therefore fall outside the 95% confidence interval based on unforced preindustrial control simulations.

Fig. 2.

CMIP5 (a) 1950–85 and (b) 1985–2012 intertropical precipitation trend probability density functions. Models that include both aerosol indirect effects (ALL BOTH AIE) and models that lack aerosol indirect effects (ALL NO AIE) are indicated, as are miscellaneous forcing experiments with GHG, AA, and NAT. AMIP runs include all forcings and are driven with observed SSTs. Thin gold lines in (b) represent a larger set (29 models) of AMIP experiments. Thick solid lines in (a),(b) indicate the experiment is significantly different than the PIC at the 95% confidence level, based on the Wilcoxon–Mann–Whitney rank-sum test (dashed lines indicate otherwise). Model precipitation has been subsampled according to GHCN observations. Similar results are obtained without subsampling (over all land areas, as with CRU, UDEL, and PREC/L).

Fig. 2.

CMIP5 (a) 1950–85 and (b) 1985–2012 intertropical precipitation trend probability density functions. Models that include both aerosol indirect effects (ALL BOTH AIE) and models that lack aerosol indirect effects (ALL NO AIE) are indicated, as are miscellaneous forcing experiments with GHG, AA, and NAT. AMIP runs include all forcings and are driven with observed SSTs. Thin gold lines in (b) represent a larger set (29 models) of AMIP experiments. Thick solid lines in (a),(b) indicate the experiment is significantly different than the PIC at the 95% confidence level, based on the Wilcoxon–Mann–Whitney rank-sum test (dashed lines indicate otherwise). Model precipitation has been subsampled according to GHCN observations. Similar results are obtained without subsampling (over all land areas, as with CRU, UDEL, and PREC/L).

Similar results are obtained if we restrict the PIC distribution to those models (top 50%) that yield the largest 95% confidence interval of tropical precipitation shifts. This model subset, therefore, simulates larger intertropical precipitation variability. The 1950–85 GHCN trend is larger in magnitude than all of these 10 017 PIC realizations, and the 1985–2012 GHCN trend is larger than 98.7% (significant at the 95% confidence interval, assuming a two-tailed test). We also obtain similar conclusions if we restrict the PIC distribution to the five models identified by Martin et al. (2014), which better simulate the teleconnection between North Atlantic SST and Sahel rainfall. Based on individual models’ PIC simulations, the GHCN southward (northward) shift is significant at the 95% confidence level in 100% (93%) of the models. Similar conclusions for the 1950–85 time period are obtained over all land areas, where both the UDEL and CRU southward shifts are significant at the 95% confidence level in 100% of the models; the PREC/L shift is significant at the 95% confidence level in all but one model. Although climate models may underestimate the full scale of natural variability, these results imply the observed shifts—particularly the southward shift—are unlikely to arise solely from internal variability of the climate system.

Figures 3a–c show the 1950–2012 5-yr smoothed time series of the intertropical precipitation index using GHCN observations and CMIP5 experiments by region. The CMIP5 twentieth-century all forcing (ALL) ensemble mean reproduces both shifts but smaller than observed. All forcing models without aerosol indirect effects (ALL NO AIE) yield weak ensemble-mean shifts of −1.9 and 2.8 mm decade−1 (Table 2), respectively, whereas models that include aerosol indirect effects (ALL AIE) yield larger shifts. In particular, models that include both the cloud-albedo and the cloud-lifetime effect (ALL BOTH AIE) yield ensemble-mean shifts of −14.6 and 5.7 mm decade−1, respectively (Table 2). Both shifts are significant at the 95% confidence level based on a standard Student’s t test and a Wilcoxon–Mann–Whitney rank-sum test (Wilks 2006). Similar results exist for the tropical precipitation shift over both land and ocean (Table 2), where ALL BOTH AIE yields statistically significant trends of −13.4 and 4.7 mm decade−1, respectively. As will be discussed below (section 3e), the subset of models that yield the largest hemispheric aerosol radiative forcing gradient (ALL BOTH AIE′) simulate the largest tropical precipitation shifts and interhemispheric temperature trends (Figs. 3d–f).

Fig. 3.

CMIP5 1950–2012 5-yr smoothed ensemble-mean annual average (a)–(c) intertropical precipitation and (d)–(f) interhemispheric temperature time series by region. Linear trends are included as solid lines for 1950–85 and 1985–2012. Also included is the subset of models that yield the largest hemispheric aerosol radiative forcing gradient (ALL BOTH AIE′); gray shading represents the standard deviation of these models. The timings of the three major volcanic eruptions are denoted by asterisks. Model precipitation has been subsampled according to GHCN observations.

Fig. 3.

CMIP5 1950–2012 5-yr smoothed ensemble-mean annual average (a)–(c) intertropical precipitation and (d)–(f) interhemispheric temperature time series by region. Linear trends are included as solid lines for 1950–85 and 1985–2012. Also included is the subset of models that yield the largest hemispheric aerosol radiative forcing gradient (ALL BOTH AIE′); gray shading represents the standard deviation of these models. The timings of the three major volcanic eruptions are denoted by asterisks. Model precipitation has been subsampled according to GHCN observations.

Among the various external forcings, the southward shift is primarily driven by anthropogenic aerosols, which yield a significant ensemble-mean trend of −12.1 mm decade−1 (Table 2). The bulk of the northward shift is driven by natural forcings at 6.8 mm decade−1, also significant at the 99% confidence level. Greenhouse gases yield tropical precipitation shifts in the opposite direction as observed. Although different models comprise the ALL and miscellaneous forcing experiments, similar conclusions are obtained using a consistent subset of models (i.e., the 10 AA models). When driven by the observed evolution of sea surface temperatures (AMIP), models yield much larger tropical precipitation shifts at −39.2 mm decade−1 from 1950 to 1985 and 26.5 mm decade−1 (17.3 mm decade−1 based on the larger set) for 1985–2012 (Table 2), in better agreement to those observed. This highlights the importance of the real-world evolution of sea surface temperatures.

Based on the Wilcoxon–Mann–Whitney test, AMIP intertropical precipitation trends for both 1950–85 and 1985–2012 are inconsistent with the null hypothesis. This indicates that AMIP intertropical precipitation trends are significantly different than unforced, preindustrial control trends at the 95% confidence level (Figs. 2a,b). Using the same procedure, we find that ALL AIE, ALL BOTH AIE, NAT, and AA 1950–85 intertropical precipitation trends are also significantly different than the corresponding PIC trends at the 95% confidence level. The 1950–85 CMIP5 GHG trends are also significantly different than the corresponding PIC trends at the 95% confidence level. However, the GHG trends are more positive than the PIC trends, in disagreement with the negative intertropical precipitation trend from observations. In addition to the previously mentioned AMIP trends, ALL AIE, ALL BOTH AIE, and NAT yield 1985–2012 intertropical precipitation trends that are significantly different than the corresponding preindustrial control trends. For both time periods, there are no significant differences between ALL NO AIE and PIC trends.

In the Atlantic sector, model underestimation of the observed intertropical precipitation and interhemispheric temperature trends is exacerbated, particularly for 1950–85 (Figs. 3b,e). Table 2 shows that observations yield significant 1950–85 southward shifts from −48.0 to −61.4 mm decade−1. The corresponding 95% confidence interval based on PIC is from −22.8 to 23.4 mm decade−1, so all observed southward shifts are significant relative to model-simulated unforced, natural variability. ALL NO AIE models yield the weakest simulated trends at −3.2 mm decade−1, and ALL BOTH AIE models (particularly ALL BOTH AIE′) yield the largest simulated trends at −6.6 mm decade−1. Both model subsets, however, are not significant at the 95% confidence level and are much smaller than the observed trends. Similar to the global tropical precipitation shift, the southward shift in the Atlantic sector is primarily driven by anthropogenic aerosols, which yield a significant ensemble-mean trend of −7.3 mm decade−1 (Table 2). The bulk of the northward shift is also driven by anthropogenic aerosols at 8.7 mm decade−1, significant at the 99% confidence level. Natural forcings qualitatively reproduce both observed Atlantic sector shifts, and GHGs yield the opposite. AMIP simulations again better resemble observations, yielding shifts of −54.0 and 47.0 mm decade−1 (43.0 mm decade−1 based on the larger set), respectively.

Observations also suggest that tropical precipitation shifted southward in the Pacific sector (30°E–75°W) from 1950 to 1985 from −18.3 to −46.4 mm decade−1. However, only GHCN (and ERA-20C and 20CR over land and ocean) yields a statistically significant shift. The corresponding 95% confidence interval based on PIC is from −26.5 to 28.0 mm decade−1, so the GHCN and UDEL southward shifts are significant relative to model-simulated unforced, natural variability. Relative to the Atlantic, models yield larger tropical precipitation shifts and interhemispheric temperature contrasts in the Pacific that better resemble observations, particularly the model subsets that include aerosol indirect effects (especially ALL BOTH AIE′) (Figs. 3c,f). ALL BOTH AIE, for example, yields a significant ensemble-mean southward shift of −23.7 mm decade−1 (Table 2). Most of this southward Pacific shift is also driven by anthropogenic aerosols at −16.9 mm decade−1. Natural forcings qualitatively reproduce both observed Pacific sector shifts and account for most of the 1985–2012 simulated northward shift at 9.3 mm decade−1. GHGs yield the opposite shift during 1950–85 and a weak northward shift during the second time period.

b. HadGEM2-ES results

Of the 44 CMIP5 coupled models considered, HadGEM2-ES yields the largest ensemble-mean intertropical precipitation trends. Subsampled to the GHCN data, the 1950–85 ensemble-mean tropical precipitation shift (over all four realizations) is −32.9 mm decade−1, and the subsequent northward recovery is 22.2 mm decade−1. Individual ensemble members yield consistent, relatively large trends, particularly for the 1950–85 time period at −20.4, −46.4, −23.5, and −41.2 mm decade−1. Individual ensemble member trends for the 1985–2012 time period are less consistent at −2.4, 27.0, 9.0, and 55.1 mm decade−1.Ensemble-mean trends of HadGEM2-ES are much larger than the CMIP5 ALL ensemble-mean trends and nearly as large as those simulated by the prescribed SST (AMIP) experiments. Thus, HadGEM2-ES yields ensemble-mean tropical precipitation shifts in better agreement to those observed. Because the ensemble mean is based on four realizations with the same external forcing, and each realization yields consistent relatively large trends, this suggests that they are primarily due to external forcing, particularly the 1950–85 southward shift. Internal variability, however, does contribute to the spread across realizations, particularly for 1985–2012.

In addition to the four historical all forcing HadGEM2-ES experiments, four GHG, four NAT, and three constant anthropogenic aerosol experiments exist (Table 1). The ensemble-mean HadGEM2-ES GHG intertropical precipitation trends are 6.0 mm decade−1 from 1950 to 1985 and −17.9 mm decade−1 from 1985 to 2012 (opposite those observed). The corresponding HadGEM2-ES NAT trends are −18.4 and 10.7 mm decade−1, respectively, and the anthropogenic aerosol (obtained by taking the difference of the historical and constant anthropogenic aerosol simulations) trends are −49.7 and 18.4 mm decade−1, respectively. Thus, the majority of the tropical precipitation shift in HadGEM2-ES is forced by both anthropogenic and volcanic aerosols, in agreement with the CMIP5 ensemble-mean results (section 3d discusses how aerosols have contributed to the shifts). The only difference is that the ensemble-mean HadGEM2-ES ALL intertropical precipitation trends are nearly as large as those observed, indicating that recent meridional shifts in tropical precipitation have a larger externally forced contribution (by anthropogenic and volcanic aerosols) than simulated by most models.

Similar results exist for the tropical precipitation shift in the Atlantic and Pacific sectors. In the Atlantic, HadGEM2-ES ALL yields ensemble-mean trends of −29.5 mm decade−1 from 1950 to 1985 and 35.1 mm decade−1 from 1985 to 2012. HadGEM2-ES AA yields corresponding Atlantic shifts of −65.4 and 24.8 mm decade−1 for 1950–85 and 1985–2012, respectively, and HadGEM2-ES NAT yields −25.7 and 22.6 mm decade−1. In the Pacific, HadGEM2-ES ALL yields tropical precipitation shifts of −35.3 mm decade−1 from 1950 to 1985 and 6.8 mm decade−1 from 1985 to 2012. HadGEM2-ES AA yields corresponding Pacific shifts of −30.7 and 11.6 mm decade−1 for 1950–85 and 1985–2012, respectively, and HadGEM2-ES NAT yields −12.6 and −3.7 mm decade−1.

As will be discussed below, consistent with larger tropical precipitation shifts in HadGEM2-ES, HadGEM2-ES also simulates relatively large interhemispheric temperature trends. Ensemble-mean interhemispheric temperature trends from HadGEM2-ES are −0.07 and 0.16 K decade−1 for 1950–85 and 1985–2012, respectively, which are larger than the CMIP5 ensemble mean (Table 3) and nearly as large as observed. We also note that HadGEM2-ES yields one of the largest hemispheric contrasts in aerosol radiative forcing, particularly in the Atlantic sector (section 3e).

Table 3.

Observed and model-mean interhemispheric temperature trends (K decade−1). Trends are based on the 5-yr smoothed annual average time series for 1950–85 (P1) and 1985–2012 (P2). Global interhemispheric temperature trends are calculated over both land and ocean; Atlantic and Pacific sector trends are calculated over ocean only. Also included are the corresponding trends based on the three models—HadGEM2-ES, MIROC-ESM, and MIROC5—that yield the largest hemispheric aerosol radiative forcing gradient in the Atlantic (denoted as ALL BOTH AIE′). Trend significance is denoted by boldface values for ≥90%, a single asterisk for ≥95%, and double asterisks for ≥99%.

Observed and model-mean interhemispheric temperature trends (K decade−1). Trends are based on the 5-yr smoothed annual average time series for 1950–85 (P1) and 1985–2012 (P2). Global interhemispheric temperature trends are calculated over both land and ocean; Atlantic and Pacific sector trends are calculated over ocean only. Also included are the corresponding trends based on the three models—HadGEM2-ES, MIROC-ESM, and MIROC5—that yield the largest hemispheric aerosol radiative forcing gradient in the Atlantic (denoted as ALL BOTH AIE′). Trend significance is denoted by boldface values for ≥90%, a single asterisk for ≥95%, and double asterisks for ≥99%.
Observed and model-mean interhemispheric temperature trends (K decade−1). Trends are based on the 5-yr smoothed annual average time series for 1950–85 (P1) and 1985–2012 (P2). Global interhemispheric temperature trends are calculated over both land and ocean; Atlantic and Pacific sector trends are calculated over ocean only. Also included are the corresponding trends based on the three models—HadGEM2-ES, MIROC-ESM, and MIROC5—that yield the largest hemispheric aerosol radiative forcing gradient in the Atlantic (denoted as ALL BOTH AIE′). Trend significance is denoted by boldface values for ≥90%, a single asterisk for ≥95%, and double asterisks for ≥99%.

c. Mechanisms driving tropical precipitation shifts

Meridional shifts in tropical precipitation are closely related to interhemispheric SST variations, such that relative warming (cooling) of the NH is associated with a northward (southward) shift in tropical precipitation (Broccoli et al. 2006; Allen and Sherwood 2011; Ming and Ramaswamy 2011; Hwang et al. 2013; Haywood et al. 2013). Regressing the 1950–2012 observed monthly intertropical precipitation time series onto SSTs yields hemispherically asymmetric SST changes, including warming in the NH and cooling in the SH; models yield a similar relationship (not shown). Figure 4 shows significant correlations between the interhemispheric temperature trend and tropical precipitation shift in both AMIP and CMIP5 ALL experiments. This relationship is particularly strong in the Atlantic sector, where the correlation between the tropical precipitation shift and the tropical interhemispheric temperature trend is 0.81 across all model realizations (and both time periods) (Fig. 4f).

Fig. 4.

Interhemispheric temperature trend vs intertropical precipitation trend scatterplots. Tropical precipitation shift vs (a),(b) interhemispheric temperature trend and (c),(d) tropical interhemispheric temperature trend for (left) AMIP and (right) CMIP5 ALL experiments. (e),(f) As in (c),(d), but for the Atlantic sector. Open symbols represent individual model realizations; black (blue) squares represent the mean over all model realizations (model ensemble means) and include the 2σ uncertainty. Small (large) symbols represent trends for 1950–85 (1985–2012). Also included in each panel is the correlation coefficient across all model realizations (r) and across all model ensemble means (r′). CMIP5 ALL model symbols are stratified according to aerosol indirect effects, including those with both (square); the cloud-albedo aerosol indirect effect only (triangles); and no aerosol indirect effects (crosses).

Fig. 4.

Interhemispheric temperature trend vs intertropical precipitation trend scatterplots. Tropical precipitation shift vs (a),(b) interhemispheric temperature trend and (c),(d) tropical interhemispheric temperature trend for (left) AMIP and (right) CMIP5 ALL experiments. (e),(f) As in (c),(d), but for the Atlantic sector. Open symbols represent individual model realizations; black (blue) squares represent the mean over all model realizations (model ensemble means) and include the 2σ uncertainty. Small (large) symbols represent trends for 1950–85 (1985–2012). Also included in each panel is the correlation coefficient across all model realizations (r) and across all model ensemble means (r′). CMIP5 ALL model symbols are stratified according to aerosol indirect effects, including those with both (square); the cloud-albedo aerosol indirect effect only (triangles); and no aerosol indirect effects (crosses).

Several CAM3 (Collins et al. 2004) SST sensitivity experiments were performed to isolate the effects of Atlantic SSTs on tropical precipitation shifts. A control experiment was performed using global observational SSTs from 1940–2012 (this experiment is denoted SST). Several experiments were performed by keeping SSTs in specific regions held fixed at their 1940 values. Taking the difference with the control experiment yields the climate response due to the temporal evolution of SSTs in that region. For example, to get the North Atlantic (0°–70°N, 75°W–30°E) SST climate response, an experiment was performed keeping North Atlantic SSTs held fixed to their 1940 values but allowing SSTs everywhere else to vary in accord with observations. This experiment is then differenced with the control experiment, which yields the late twentieth-century climate response to North Atlantic SSTs (denoted NA). Additional experiments include North and South Atlantic (70°S–70°N, 75°W–30°E) SSTs (NSA), tropical North Atlantic (0–30°N, 75°W–30°E) SSTs (NTA), and tropical North and South Atlantic (30°N–30°S, 75°W–30°E) SSTs (NSTA). A total of three ensemble members were performed for each experiment.

Similar to the CMIP5 AMIP results (Table 2), CAM3 simulations forced with observed SSTs yield relatively large tropical precipitation shifts (Fig. 5). In the Atlantic sector, the ensemble-mean shift is −68.9 mm decade−1 from 1950 to 1985 and 58.0 mm decade−1 from 1985 to 2012 (these are actually larger than observed, particularly for 1985–2012). North Atlantic SSTs alone reproduce most of CAM3’s simulated tropical Atlantic precipitation shifts; experiments with both North and South Atlantic SSTs better reproduce the shifts—particularly from 1950 to 1985. This is due to warming of the SH Atlantic Ocean throughout both time periods, which strengthens (weakens) the negative (positive) interhemispheric temperature trend from 1950 to 1985 (1985–2012), resulting in larger (smaller) tropical precipitation shifts. CAM3 shows ~75%–80% of the tropical Atlantic precipitation shifts are reproduced with Atlantic SSTs only (the remaining difference may be related to SSTs poleward of 70°N and 70°S). Although the interhemispheric temperature trends are weaker in tropical North and South Atlantic CAM3 experiments, the magnitudes of the tropical precipitation shifts are similar, showing the importance of low-latitude (30°S–30°N) Atlantic SSTs. Although these experiments do not identify the cause of the SST variations, they show that simulation of the interhemispheric Atlantic temperature gradient is necessary to reproduce the magnitude of the observed tropical Atlantic precipitation shifts.

Fig. 5.

CAM3 intertropical precipitation trends in the Atlantic sector. Boxes show the median trend (center line) and the interquartile range (length of box) for prescribed global SST, NA, NSA, NTA, and NSTA. Intertropical precipitation trends are based on subsampling to the GHCN data. Blue (red) rectangles represent 1950–85 (1985–2012).

Fig. 5.

CAM3 intertropical precipitation trends in the Atlantic sector. Boxes show the median trend (center line) and the interquartile range (length of box) for prescribed global SST, NA, NSA, NTA, and NSTA. Intertropical precipitation trends are based on subsampling to the GHCN data. Blue (red) rectangles represent 1950–85 (1985–2012).

Similar to prior results, Figs. 6 and 7 show that the precipitation shift and the interhemispheric temperature contrast are closely related to changes in the strength and latitude of the Hadley cells and the associated northward cross-equatorial MSE transport (e.g., Hwang et al. 2013; Schneider et al. 2014). In both AMIP (Fig. 6b) and CMIP5 ALL (Fig. 7b), relative cooling of the NH and a southward shift in tropical precipitation are associated with an increase in northward cross-equatorial MSE transport (and vice versa). Since the climatological cross-equatorial MSE transport is southward, this change represents a weaker flux of energy by the atmosphere from the Northern to the Southern Hemisphere. To decrease southward cross-equatorial MSE transport, the Hadley circulation adjusts with a stronger NH cell (Figs. 6e and 7g), a weaker SH cell (Figs. 6f and 7h), and a southward shift of both cells (Figs. 6c,d and 7e,f). Figure 8, which shows 1950–85 ensemble-mean MMC trends in AMIP and CMIP5 ALL BOTH AIE, better illustrates the adjustment of the Hadley circulation. Positive MMC trends act to weaken the SH Hadley cell but strengthen the NH Hadley cell. The southward shift of both cells is consistent with the largest positive MMC trends near the equatorial flanks of both cells. Since the tropical rain belt is collocated with the ascending branch of the Hadley cell, this results in a southward shift in tropical precipitation.

Fig. 6.

CMIP5 AMIP northward cross-equatorial MSE transport vs (a) tropical interhemispheric surface temperature and (b) intertropical precipitation. Intertropical precipitation vs (c) NH MMC latitude, (d) SH MMC latitude, (e) NH MMC strength, and (f) SH MMC strength. Open symbols represent individual model realizations; black (blue) squares represent the mean over all model realizations (model ensemble means) and include the 2σ uncertainty. Small (large) symbols represent trends for 1950–85 (1985–2012). Also included in each panel is the correlation coefficient across all model realizations (r) and across all model ensemble means (r′).

Fig. 6.

CMIP5 AMIP northward cross-equatorial MSE transport vs (a) tropical interhemispheric surface temperature and (b) intertropical precipitation. Intertropical precipitation vs (c) NH MMC latitude, (d) SH MMC latitude, (e) NH MMC strength, and (f) SH MMC strength. Open symbols represent individual model realizations; black (blue) squares represent the mean over all model realizations (model ensemble means) and include the 2σ uncertainty. Small (large) symbols represent trends for 1950–85 (1985–2012). Also included in each panel is the correlation coefficient across all model realizations (r) and across all model ensemble means (r′).

Fig. 7.

As in Fig. 6, but for CMIP5 ALL experiments. Also included is the northward cross-equatorial OHF vs (c) tropical interhemispheric surface temperature and (d) intertropical precipitation. Model symbols are stratified according to aerosol indirect effects, including those with both (square); the cloud-albedo aerosol indirect effect only (triangles); and no aerosol indirect effects (crosses).

Fig. 7.

As in Fig. 6, but for CMIP5 ALL experiments. Also included is the northward cross-equatorial OHF vs (c) tropical interhemispheric surface temperature and (d) intertropical precipitation. Model symbols are stratified according to aerosol indirect effects, including those with both (square); the cloud-albedo aerosol indirect effect only (triangles); and no aerosol indirect effects (crosses).

Fig. 8.

Ensemble-mean 1950–85 MMC trends for (a) AMIP and (b) CMIP5 ALL BOTH AIE. Trends are shown as color shading (1010 kg s−1 century−1) and the corresponding climatology as black contours. Climatological contour interval is 2.5 × 1010 kg s−1, with negative values dashed. The plus symbols represent the locations where more than 75% of the model realizations agree on the sign of the trend.

Fig. 8.

Ensemble-mean 1950–85 MMC trends for (a) AMIP and (b) CMIP5 ALL BOTH AIE. Trends are shown as color shading (1010 kg s−1 century−1) and the corresponding climatology as black contours. Climatological contour interval is 2.5 × 1010 kg s−1, with negative values dashed. The plus symbols represent the locations where more than 75% of the model realizations agree on the sign of the trend.

Although few CMIP5 models archived ocean heat flux, Figs. 7c,d show that relative cooling of the NH and a southward tropical precipitation shift are associated with an increase in northward cross-equatorial OHF (and vice versa). For example, the 1950–85 ensemble-mean increase in northward cross-equatorial OHF is 0.0192 ± 0.0124 PW decade−1. Since the climatological cross-equatorial OHF is northward, the 1950–85 change represents a stronger flux of heat by the ocean from the Southern to the Northern Hemisphere (a weaker but not statistically significant flux from 1985 to 2012 also exists at −0.0079 ± 0.0249 PW decade−1). An increase in northward cross-equatorial ocean heat flux would promote warming of the NH and thus act to weaken the interhemispheric temperature contrast and the southward tropical precipitation shift (similarly for the 1985–2012 northward tropical precipitation shift). Thus, we find that the ocean heat flux response in models acts to weaken the late twentieth-century tropical precipitation shifts, particularly from 1950 to 1985. Most of the ocean heat flux response occurs outside the Atlantic, as the northward cross-equatorial Atlantic OHF trend is smaller than the global trend at 0.0064 ± 0.0042 PW decade−1 from 1950 to 1985 (and −0.0060 ± 0.0138 PW decade−1 from 1985 to 2012). This suggests that model underestimation of the Atlantic tropical precipitation shift is not caused by an excessive northward cross-equatorial Atlantic OHF.

Interhemispheric temperature contrasts, as well as tropical precipitation shifts, are also related to variations in the low-latitude hemispheric contrast of surface wind speeds, surface pressure, and low clouds (Fig. 9). Relative cooling (warming) of the NH tropics is associated with a relative increase (decrease) in NH tropical surface pressure, surface wind speeds, and low cloud cover (vice versa for the SH). In the Atlantic, for example, the correlation between the low-latitude interhemispheric temperature (NHT − SHT Atlantic TS) trend and the low-latitude interhemispheric sea level pressure (NHT − SHT Atlantic SLP) trend is −0.91 across all CMIP5 ALL realizations (and both time periods) (Fig. 9a). Similar but somewhat weaker negative correlations exist between the NHT − SHT Atlantic TS trend and the NHT − SHT Atlantic wind speed (WS) and low cloud (CLOW) trend (Figs. 9c and 9e, respectively).

Fig. 9.

Tropical interhemispheric surface temperature trend in the Atlantic sector vs the corresponding trend of (a) SLP, (c) surface WS, and (e) CLOW. (b),(d),(f) The corresponding scatterplots based on tropical data over all longitudes. Model symbols are stratified according to aerosol indirect effects, including those with both (square); the cloud-albedo aerosol indirect effect only (triangles); and no aerosol indirect effects (crosses).

Fig. 9.

Tropical interhemispheric surface temperature trend in the Atlantic sector vs the corresponding trend of (a) SLP, (c) surface WS, and (e) CLOW. (b),(d),(f) The corresponding scatterplots based on tropical data over all longitudes. Model symbols are stratified according to aerosol indirect effects, including those with both (square); the cloud-albedo aerosol indirect effect only (triangles); and no aerosol indirect effects (crosses).

d. The effect of individual climate forcings on tropical precipitation shifts

This framework relating the tropical precipitation shift to the interhemispheric temperature contrast (and the aforementioned thermodynamical/dynamical responses associated with the MMC) allows us to better understand how individual climate forcings affect tropical precipitation shifts. Although GHGs are globally well mixed, continents are primarily located in the NH, which are more easily warmed or cooled compared to the ocean with its high heat capacity. Thus, the twentieth-century increase in GHGs has preferentially warmed the NH (Table 3), yielding a northward tropical precipitation shift (particularly from 1950 to 1985). The 1963 eruption of Mount Agung is associated with a northward shift, whereas the 1982 eruption of El Chichón is associated with a southward shift (Figs. 3a–c). Mount Agung, located at 8°S, resulted in a larger increase in SH stratospheric aerosol (Sato et al. 1993), which preferentially cooled the SH (not shown). In contrast, El Chichón, located at 17°N, resulted in a larger increase in NH stratospheric aerosols, which preferentially cooled the NH. In both cases, tropical precipitation shifted away from the hemisphere that cooled. The tropical precipitation response to the (much larger) eruption of Mount Pinatubo (located at 15°N) in 1991 is less clear, which is consistent with a more hemispherically uniform dispersal of its aerosols (Sato et al. 1993).

Similarly, anthropogenic aerosols are primarily emitted in the NH. A large increase in NH sulfate emissions occurred from 1950 to 1985, followed by a weaker decrease (Lamarque et al. 2010). Because sulfate aerosols reflect solar radiation, brighten clouds, and extended their lifetime, they preferentially decreased 1950–85 TOA net SW radiation in the NH relative to the SH. The CMIP5 AA ensemble-mean NH − SH net TOA SW trend is −0.22 ± 0.06 W m−2 decade−1. This cooled the NH relative to the SH (Table 3), resulting in a negative interhemispheric temperature trend (Friedman et al. 2013; Chiang et al. 2013) and a southward shift in tropical precipitation (Table 2). In the Atlantic sector, the opposite occurred from 1985 to 2012, consistent with the decrease in anthropogenic aerosol emissions from developed countries bordering the North Atlantic (e.g., the United States and European countries). The corresponding CMIP5 AA ensemble-mean NH − SH Atlantic net TOA SW trend is 0.30 ± 0.18 W m−2 decade−1. We also note that much of the simulated variation in the interhemispheric temperature trend (both global and Atlantic sector) is tightly coupled to the hemispheric contrast in TOA net SW radiation (r ≥ 0.80 in the CMIP5 AA and r ≥ 0.78 in CMIP5 ALL).

The relationship between interhemispheric temperature and intertropical precipitation also helps us to understand model underestimation of the observed shifts and the larger simulated shifts in the prescribed SST experiments. HadCRUT4 yields NH − SH TS trends of −0.07 and 0.17 K decade−1 for 1950–85 and 1985–2012, respectively. The corresponding AMIP ensemble-mean trends are similar at −0.09 ± 0.04 and 0.16 ± 0.03 K decade−1 (Table 3). The ALL NO AIE ensemble mean underestimates these trends, particularly from 1950 to 1985, where a 0.04 ± 0.02 K decade−1 warming trend is simulated. ALL BOTH AIE also underestimates, but to a lesser extent at −0.04 ± 0.04 and 0.13 ± 0.03 K decade−1, respectively. Similar conclusions apply for the Atlantic sector, where model underestimation of the interhemispheric temperature and intertropical precipitation trends are exacerbated. HadCRUT4 yields NH − SH Atlantic TS trends of −0.22 and 0.33 K decade−1 for 1950–85 and 1985–2012. ALL NO AIE yields negligible trends of 0.01 and 0.01 K decade−1, respectively. Consistent with larger tropical Atlantic precipitation shift from ALL BOTH AIE, the corresponding NH − SH Atlantic TS trends are also larger than ALL NO AIE but less than observed. Underestimation of the interhemispheric temperature trend is consistent with the corresponding underestimation of the tropical precipitation shifts.

Most of the tropical precipitation shift difference between ALL BOTH AIE and ALL NO AIE is in the Pacific sector (30°E–75°W) (Table 2). ALL BOTH AIE yields a 1950–85 southward shift of −23.7 mm decade−1, which is significantly larger (at the 99% confidence level) than the corresponding ALL NO AIE shift of −2.2 mm decade−1. This is again consistent with the corresponding interhemispheric temperature trends in the Pacific (Table 3); here, ALL NO AIE simulates relative NH warming of 0.04 K decade−1, whereas ALL BOTH AIE simulates relative NH cooling of −0.04 K decade−1. This is consistent with observations, with HadCRUT4 yielding relative NH Pacific cooling of −0.06 K decade−1. Moreover, the magnitude of the cooling simulated by ALL BOTH AIE is similar to most of the observations/reanalyses. Thus, ALL BOTH AIE (and especially ALL BOTH AIE′) better reproduces the observed hemisphere temperature contrasts in the Pacific and, in turn, better simulates the magnitude of the observed southward tropical Pacific precipitation shift.

e. Interhemispheric aerosol radiative forcing

Figure 10 shows scatterplots between the interhemispheric (NH − SH) anthropogenic aerosol RF over the ocean and the 1950–85 tropical precipitation shift for each region. Models with a larger hemispheric aerosol RF gradient yield larger 1950–85 precipitation shifts. The corresponding correlation coefficient is 0.78 for the Atlantic sector, 0.79 for the Pacific sector, and 0.71 for the globe. Consistent with the reduction in anthropogenic aerosol emissions from developed countries bordering the North Atlantic from 1985 to 2012 (Lamarque et al. 2010), a similar—but opposite—relationship exists between 1985–2012 tropical Atlantic precipitation shift and the Atlantic aerosol RF gradient (r = −0.76). Similar correlations exist when both SW and LW fluxes are used to calculate the aerosol RF gradient; the corresponding correlation coefficient is 0.74 for the Atlantic sector, 0.89 for the Pacific sector, and 0.85 for the globe.

Fig. 10.

Interhemispheric aerosol radiative forcing vs the model-mean 1950–85 tropical precipitation shift for the (a) global, (b) Atlantic, and (c) Pacific sectors. Aerosol radiative forcing is calculated over the ocean. Only those models that archived the appropriate experiments are included. Models are stratified according to aerosol indirect effects, including those with no aerosol indirect effects (crosses); first indirect effect only (triangles); and both indirect effects (squares). Also included is the correlation coefficient r, all of which are significant at the 99% confidence level based on a standard Student’s t test. Solid black squares show results based on the interhemispheric aerosol radiative forcing from the MACC reanalysis and the average observed 1950–85 tropical precipitation shift (using GHCN, CRU, UDEL, and PREC/L).

Fig. 10.

Interhemispheric aerosol radiative forcing vs the model-mean 1950–85 tropical precipitation shift for the (a) global, (b) Atlantic, and (c) Pacific sectors. Aerosol radiative forcing is calculated over the ocean. Only those models that archived the appropriate experiments are included. Models are stratified according to aerosol indirect effects, including those with no aerosol indirect effects (crosses); first indirect effect only (triangles); and both indirect effects (squares). Also included is the correlation coefficient r, all of which are significant at the 99% confidence level based on a standard Student’s t test. Solid black squares show results based on the interhemispheric aerosol radiative forcing from the MACC reanalysis and the average observed 1950–85 tropical precipitation shift (using GHCN, CRU, UDEL, and PREC/L).

The MACC aerosol reanalysis, which includes estimates of the shortwave direct and first indirect radiative forcing by anthropogenic aerosols, yields hemispheric aerosol RF gradients much smaller than those estimated by most models. MACC hemispheric RF gradient is −0.52 W m−2 for the globe, −0.16 W m−2 for the Atlantic sector, and −0.64 W m−2 for the Pacific sector. Although MACC provides one strand of evidence, and is itself based on a model, this suggests that CMIP5 models may exaggerate the hemispheric asymmetry in anthropogenic aerosol forcing.

Although global aerosol emissions significantly increased from 1950 to 1985 (Lamarque et al. 2010), the increase is less than that from preindustrial to present day (which the aerosol RF quantifies the radiative effects of). Moreover, the hemispheric contrast in aerosol RF is not necessarily the same as that from 1950 to 1985. Because the actual radiative forcing from 1950 to 1985 is not archived, we approximate this by scaling each model’s aerosol radiative forcing (obtained using the aforementioned procedure) to the log of the ratio of the total change in aerosol optical depth (AOD) from 1950 to 1985 relative to 1850–2000 (for those models that archived AOD). To illustrate using a simplified example (averaging over all available models), the ratio of the total change in AOD from 1950 to 1985 relative to 1850–2000 is 0.70 ± 0.07 in the NH and 0.78 ± 0.09 in the SH (uncertainty estimates are 2σ). Thus, the multimodel average hemispheric contrast in aerosol RF is reduced by a factor of log(0.70/0.78) (~10%). The corresponding reduction is larger in the Atlantic sector (25%), but smaller in the Pacific sector (4%). The models with a relatively large hemispheric aerosol RF gradient persist after implementing this AOD scaling, and our conclusions relating the hemispheric aerosol RF gradient to the magnitude of the tropical precipitation shift, the interhemispheric temperature trend, etc. remain robust.

This is consistent with CAM5, where the actual (1950–85) hemispheric aerosol RF gradient is similar to the approximated value. CAM5’s radiative forcing from 1950 to 1985 was calculated by performing two additional simulations driven by anthropogenic aerosol emissions in the years 1985 and 1950. The 1950–85 NH − SH aerosol RF contrast in CAM5 is −1.10 W m−2, compared to the scaled value of −0.96 W m−2. Similarly, the hemispheric RF gradient in the Atlantic sector is −0.10 W m−2 versus the scaled value of −0.20 W m−2. Similar effective radiative forcing calculations were also performed with HadGEM2-ES, allowing direct quantification of this model’s radiative forcing from 1950 to 1985. As expected, relative to CAM5, HadGEM2-ES yields much larger 1950–85 hemispheric aerosol RF gradients, which are nearly as large as those based on present-day versus preindustrial conditions. This is consistent with the large tropical precipitation shift simulated by HadGEM2-ES and the importance of the hemispheric aerosol RF gradient. We note that all CMIP5 models use the same emission inventory (Lamarque et al. 2010). This implies that the large range in the hemispheric aerosol RF gradient (from −0.13 W m−2 in MPI-ESM-LR to −2.55 W m−2 in MRI-CGCM3 and CAM5) is due to other factors, including transport and removal processes, as well as the representation of aerosol radiative effects. Since ALL BOTH AIE models (squares) have larger hemispheric aerosol RF gradients than ALL NO AIE models (crosses), the large range is related to aerosol indirect effects.

We note that similar relationships exist between the magnitude of the tropical precipitation shift and the 1950–85 trend in NH − SH TOA net SW radiation (based on over 40 ALL models, r = 0.69 in the Atlantic sector and 0.50 globally). TOA net SW radiation, however, is affected by changes in ozone, volcanic aerosols, and GHG-related climate feedbacks, in addition to aerosols. Restricting this analysis to the 10 AA models yields similar correlations (r = 0.77 in the Atlantic sector and 0.61 globally). Moreover, if we compare the tropical precipitation shift to the 1950–85 trend in NH − SH cloud-top effective droplet radius (reff, a measure of aerosol indirect effects), we find similar correlations (r = 0.71 in the Atlantic sector and 0.62 globally). As expected, models that exhibit the largest hemispheric contrast in aerosol RF also exhibit the largest 1950–85 trends in NH − SH TOA net SW radiation and NH − SH reff.

Models with a larger hemispheric aerosol RF gradient also yield larger 1950–85 interhemispheric temperature trends, in addition to larger hemispheric contrasts in low-latitude SLP, surface WS, and CLOW that are more like those observed. This is particularly the case in the Atlantic sector (Fig. 11), where the correlation between the hemispheric aerosol RF gradient and the 1950–85 NHT − SHT TS trend is 0.73. Similarly, the corresponding Atlantic sector correlations are −0.66 for NHT − SHT SLP, −0.76 for NHT − SHT WS, and −0.64 for NHT − SHT CLOW. Reanalyses and AMIP also show similar responses.

Fig. 11.

Interhemispheric aerosol radiative forcing vs 1950–85 interhemispheric trends in the tropical Atlantic for (a) surface temperature, (b) tropical precipitation shift, (c) SLP, (d) surface WS, (e) CLOW, and (f) equatorial MSE (calculated over all longitudes). Each panel includes the correlation coefficient r, all of which are significant at the 95% confidence level based on a standard Student’s t test. Also included are the corresponding 1950–85 trends based on 20CR (sold black line), ERA-20C (dashed black line), and the AMIP ensemble mean (solid gold line).

Fig. 11.

Interhemispheric aerosol radiative forcing vs 1950–85 interhemispheric trends in the tropical Atlantic for (a) surface temperature, (b) tropical precipitation shift, (c) SLP, (d) surface WS, (e) CLOW, and (f) equatorial MSE (calculated over all longitudes). Each panel includes the correlation coefficient r, all of which are significant at the 95% confidence level based on a standard Student’s t test. Also included are the corresponding 1950–85 trends based on 20CR (sold black line), ERA-20C (dashed black line), and the AMIP ensemble mean (solid gold line).

These results suggest that the interhemispheric aerosol RF gradient influences meridional displacements in tropical precipitation by directly modulating SSTs through changes in solar radiation but also by exciting a dynamical response involving tropical ocean–atmosphere coupling. Cooling of North Atlantic SSTs—where most of the (negative) aerosol forcing is located—yields a surface high pressure, which strengthens the northeast trade winds. Stronger winds further cool the SSTs, which in turn favors an increase in low cloud cover (stratocumulus), leading to additional (radiative) cooling of the SSTs. In contrast, relatively small SH aerosol RF has a small cooling effect on the underlying SSTs. However, the dynamical response induced in the NH leads to a reduction in low-latitude SH surface pressure and weaker southeast trade winds that act to warm the SH low-latitude SSTs (through reduced wind–evaporative cooling). This favors a decrease in low cloud cover, which further promotes (radiative) warming of SH low-latitude SSTs. A similar mechanism has been proposed to explain low-latitude interhemispheric temperature variations in the Atlantic (Dommenget and Latif 2000; Tanimoto and Xie 2002; Chiang and Vimont 2004; Evan et al. 2013; Bellomo et al. 2015).

Figures 12a–d shows the 1950–85 multimodel mean trend in several quantities related to tropical precipitation shifts, using the three models that yield the largest gradient in hemispheric aerosol RF in the Atlantic (HadGEM2-ES, MIROC-ESM, and MIROC5; denoted as ALL BOTH AIE′ in Tables 2 and 3). These models exhibit hemispherically asymmetric trends in TS, SLP, WS, and CLOW, consistent with large tropical precipitation shifts. For the NH, this includes a decrease in TS and an increase in SLP, WS, and CLOW—particularly in the Atlantic sector. These trends are reversed in the Southern Hemisphere. Tables 2 and 3 show that this model subset yields relatively large interhemispheric temperature trends (e.g., −0.12 and 0.11 K decade−1 in the Atlantic sector) and, in turn, relatively large tropical precipitation shifts (e.g., −21.4 and 24.2 mm decade−1 in the Atlantic sector).

Fig. 12.

1950–85 ensemble-mean trends based on (left) the subset of CMIP5 ALL models with the largest Atlantic aerosol RF gradient and (right) HadGEM2-ES anthropogenic aerosol experiment. (a),(e) Surface temperature, (b),(f) SLP, (c),(g) surface WS, and (d),(h) CLOW. Models with the largest Atlantic aerosol RF gradient include HadGEM2-ES, MIROC-ESM, and MIROC5. Dotted areas represent significance at the 95% confidence level.

Fig. 12.

1950–85 ensemble-mean trends based on (left) the subset of CMIP5 ALL models with the largest Atlantic aerosol RF gradient and (right) HadGEM2-ES anthropogenic aerosol experiment. (a),(e) Surface temperature, (b),(f) SLP, (c),(g) surface WS, and (d),(h) CLOW. Models with the largest Atlantic aerosol RF gradient include HadGEM2-ES, MIROC-ESM, and MIROC5. Dotted areas represent significance at the 95% confidence level.

The role of anthropogenic aerosols in driving these changes is further supported by HadGEM2-ES, the lone AA experiment archived by these three models (Figs. 12e–h). The response shows similar hemispherically asymmetric trends, including significant cooling of the NH, in addition to warming of the SH—particularly in the tropical Atlantic—consistent with the aforementioned dynamical mechanism. These responses are supported by idealized CAM5 experiments coupled to a SOM and forced with HadGEM2’s anthropogenic aerosol RF (Figs. 13a–f). Hemispherically asymmetric responses exist in low-latitude TS, SLP, WS, and CLOW (and tropical precipitation; not shown). Similar CAM5 results are obtained if HadGEM2’s aerosol forcing is restricted to the Atlantic (not shown). Moreover, the magnitude of the response increases, including warming of the tropical South Atlantic if the Atlantic forcing is increased (multiplied by 2).

Fig. 13.

CAM5 SOM response to (left) HadGEM2 anthropogenic aerosol radiative forcing and (right) MACC reanalysis mineral dust TOA SW direct effect. (a),(e) Surface temperature, (b),(f) SLP, (c),(g) surface WS, and (d),(h) CLOW. Dotted areas represent significance at the 95% confidence level.

Fig. 13.

CAM5 SOM response to (left) HadGEM2 anthropogenic aerosol radiative forcing and (right) MACC reanalysis mineral dust TOA SW direct effect. (a),(e) Surface temperature, (b),(f) SLP, (c),(g) surface WS, and (d),(h) CLOW. Dotted areas represent significance at the 95% confidence level.

Although we find no significant relationship between dust aerosols and tropical precipitation shifts, CMIP5 models poorly simulate African dust (Evan et al. 2014), which has been shown to be important for tropical Atlantic precipitation shifts (Evan et al. 2011). Since African dust primarily affects the tropical North Atlantic, its possible importance is consistent with the framework presented here and the role of the hemispheric aerosol RF gradient. Based on the MACC aerosol reanalysis, the NH − SH contrast in TOA SW direct effect due to mineral dust is −1.99 W m−2; the corresponding contrast in the Atlantic sector is −2.85 W m−2. Figures 13e–h show the CAM5 SOM response to MACC’s mineral dust TOA SW direct effect. In addition to significant cooling of the tropical North Atlantic, significant warming of the tropical South Atlantic occurs. Moreover, a low-latitude hemispheric contrast between SLP, WS, and CLOW exists (including tropical precipitation; not shown). This response is consistent with the mechanism outlined in this paper and further supports the role of the hemispheric contrast in aerosol radiative forcing in driving tropical precipitation shifts.

f. Natural variability and model sensitivity

The AMO and AMM have been associated with tropical precipitation shifts in the Atlantic sector (Chiang and Vimont 2004; Knight et al. 2006; Zhang and Delworth 2006). Both modes of SST variability are considered to be internally generated, although a recent analysis suggests that anthropogenic aerosols have significantly contributed to twentieth-century AMO evolution (Booth et al. 2012). Discrepancies between the anthropogenic aerosol simulation used by Booth et al. (2012) and other aspects of observed Atlantic multidecadal variability have been noted (Zhang et al. 2013). Similar to Martin et al. (2014), we use the term AMV as an overarching term that represents contributions to SST variability from both natural factors (AMO/AMOC) and potential external forcing.

The AMO and AMM both exhibit multidecadal variability that has contributed to cooling of the NH Atlantic from 1950 to 1985, and warming since. Moreover, as mentioned in the introduction, the AMO/AMOC have been implicated in driving the interhemispheric SST shift (relative NH cooling) around 1970. Both interhemispheric temperature and intertropical precipitation exhibit larger multidecadal variability in the Atlantic relative to the Pacific (Fig. 3). Removal of AMV and AMM variability from the observed and modeled precipitation fields (section 2b) yields intertropical precipitation trends (as well as the interhemispheric temperature trends) that are significantly reduced (Fig. 14). In particular, removal of AMV has the largest impact—the southward tropical precipitation shift decreases by ~50% in observations and ~80% in AMIP. The northward shift is eliminated or reversed. These results show that most of the tropical precipitation shifts are related to AMV and suggests that coupled models may underestimate the shifts as a result of deficient AMV simulation.

Fig. 14.

Intertropical precipitation box-plot trends for the Atlantic sector (left). Boxes show the median trend (center line) and the interquartile range (length of box) for CMIP5 all forcing (ALL) and all forcing with prescribed SST (AMIP) experiments. Also included are the corresponding trends after removal of AMV (center) and the AMM (right). Blue (red) symbols or boxes represent the 1950–85 (1985–2012) time period. Symbols represent precipitation observations, including GHCN, CRU, and UDEL. CMIP5 intertropical precipitation trends are shown based on subsampling to the GHCN data.

Fig. 14.

Intertropical precipitation box-plot trends for the Atlantic sector (left). Boxes show the median trend (center line) and the interquartile range (length of box) for CMIP5 all forcing (ALL) and all forcing with prescribed SST (AMIP) experiments. Also included are the corresponding trends after removal of AMV (center) and the AMM (right). Blue (red) symbols or boxes represent the 1950–85 (1985–2012) time period. Symbols represent precipitation observations, including GHCN, CRU, and UDEL. CMIP5 intertropical precipitation trends are shown based on subsampling to the GHCN data.

Figures 15a,b show that the HadCRUT4 observed AMV trends, at −0.16 and 0.18 K decade−1, respectively, fall outside the 99% confidence interval based on unforced, preindustrial control simulations. This suggests recent AMV is unlikely to arise solely from internal variability of the climate system, as simulated by state-of-the-art climate models. Similarly, if we compare the observed AMV trends to the PIC distribution of each model, both observed trends are 95% significant in 95.2% of the models (all but two). All three subsets of CMIP5 ALL models—particularly ALL AIE and ALL BOTH AIE—are significantly different from the distribution of unforced, preindustrial control AMV trends for both time periods. The ALL AIE ensemble mean accounts for more than 80% of the multidecadal variance in AMV and reproduces ~65% of the magnitude of recent AMV trends (the ensemble-mean ALL AIE AMV trends are −0.11 and 0.14 K decade−1, respectively). Similar to Booth et al. (2012), but based on several models, we find that the bulk of recent AMV is mostly forced by anthropogenic and volcanic aerosols. The ensemble-mean CMIP5 AA AMV trends are −0.05 K decade−1 from 1950 to 1985 and 0.04 K decade−1 from 1985 to 2012, both significant at the 95% confidence level. The corresponding CMIP5 NAT AMV trends are −0.03 and 0.08 K decade−1, respectively (the latter time period is significant at the 90% confidence level). CMIP5 GHG AMV trends are weaker and not statistically significant, at −0.02 and 0.00 K decade−1, respectively. The important role of aerosols is also demonstrated by significantly larger AMV trends in ALL AIE and BOTH AIE relative to ALL NO AIE. Similar conclusions apply for the TAMV, where ALL BOTH AIE models reproduce nearly 100% of the magnitude of recent TAMV trends (not shown). Thus, underestimation of recent AMV (particularly by models with aerosol indirect effects) does not appear to contribute to underestimation of tropical Atlantic precipitation shifts.

Fig. 15.

CMIP5 (a) 1950–85 and (b) 1985–2012 AMV trend probability density functions. As in Fig. 2, models are subdivided based on their representation of aerosol indirect effects. The distribution of AMV trends from PIC runs is in solid black. The HadCRUT4 observed AMV trend is shown as a dashed vertical line. Tropical (c) Atlantic and (d) Pacific precipitation shift sensitivity to tropical (20°S–20°N) and basinwide (90°S–90°N) hemispheric temperature contrasts. Box plots show the median (center line), interquartile range (box length), and max range (whiskers) for CMIP5 ALL, PIC, and AMIP experiments. The HadGEM2 models—which yield some of the largest Atlantic sensitivities—are indicated by crosses. Left (right) box plot in each pair shows the sensitivity with model precipitation subsampled according to GHCN observations (over all land and ocean grid boxes). Observed sensitivities are based on GHCN, CRU, and UDEL precipitation combined with 20CR (green) or HadCRUT4 (blue) temperature. Note that the y axis in (d) spans twice the range as in (c).

Fig. 15.

CMIP5 (a) 1950–85 and (b) 1985–2012 AMV trend probability density functions. As in Fig. 2, models are subdivided based on their representation of aerosol indirect effects. The distribution of AMV trends from PIC runs is in solid black. The HadCRUT4 observed AMV trend is shown as a dashed vertical line. Tropical (c) Atlantic and (d) Pacific precipitation shift sensitivity to tropical (20°S–20°N) and basinwide (90°S–90°N) hemispheric temperature contrasts. Box plots show the median (center line), interquartile range (box length), and max range (whiskers) for CMIP5 ALL, PIC, and AMIP experiments. The HadGEM2 models—which yield some of the largest Atlantic sensitivities—are indicated by crosses. Left (right) box plot in each pair shows the sensitivity with model precipitation subsampled according to GHCN observations (over all land and ocean grid boxes). Observed sensitivities are based on GHCN, CRU, and UDEL precipitation combined with 20CR (green) or HadCRUT4 (blue) temperature. Note that the y axis in (d) spans twice the range as in (c).

Although models reproduce the bulk of recent AMV, in order to simulate the resulting effects on the position of the tropical rain belt, models also need to accurately capture the sensitivity of tropical precipitation shifts to the interhemispheric temperature variations (which AMV has contributed to). Based on observations, the Atlantic tropical precipitation shift sensitivity (NHT − SHT Atlantic PRECT) to the tropical Atlantic hemispheric temperature contrasts (NHT − SHT Atlantic TS) ranges from 133 to 396 mm K−1 (Fig. 15c). Models, including ALL and PIC, are on the low end of this range, with median values of 145 and 134 mm K−1, respectively, and interquartile ranges of 103–187 and 110–189 mm K−1, respectively. Models also underestimate the corresponding shift sensitivity to the basinwide Atlantic hemispheric temperature contrast (NH − SH Atlantic TS). Observations yield sensitivities ranging from 101 to 236 mm K−1. CMIP5 ALL and PIC yield median shifts less than 100 mm K−1, at 78 and 69 mm K−1, respectively, and interquartile ranges of 58–110 and 48–86 mm K−1, respectively. AMIP models are on the high end of the observed sensitivities, particularly based on the basinwide Atlantic hemispheric temperature contrasts (median sensitivity is 222 mm K−1). Although the simulated sensitivities increase when the shifts are based on precipitation over both ocean and land (right bar in each pair), these results suggest that coupled models underestimate the Atlantic precipitation shift sensitivity to the interhemispheric temperature contrasts. We note that the observed sensitivities are robust to the time period chosen; the GHCN–HadCRUT4 low-latitude Atlantic shift sensitivity from 1950 to 2012 (1900–2012) is 396 (363) mm K−1. The corresponding basinwide Atlantic shift sensitivities are 151 and 180 mm K−1, respectively.

Interestingly, the HadGEM2 family of models, including HadGEM2-ES (which simulates the largest tropical precipitation shifts), yields some of the largest sensitivities in the Atlantic sector. These results are qualitatively similar to Martin et al. (2014), who found most CMIP5 models underestimate the teleconnection between North Atlantic SSTs and Sahel rainfall. However, Martin et al. (2014) identify five models that better simulate this teleconnection, and two of these models are HadGEM2 models. Although the corresponding Pacific shift sensitivity exhibits a large range in observations, models generally better reproduce the shift sensitivity to both low-latitude and basinwide Pacific hemispheric temperature contrasts, particularly when compared to AMIP sensitivities (Fig. 15d).

4. Discussion and conclusions

We have shown that late twentieth-century meridional tropical precipitation shifts, particularly the southward shift from 1950 to ~1985, are unlikely to arise solely from internal variability of the climate system, as simulated by state-of-the-art coupled ocean–atmosphere climate models (Fig. 2a). Similar conclusions are obtained if we restrict the unforced, preindustrial control (PIC) distribution to those models (top 50%) that yield the largest 95% confidence interval of tropical precipitation shifts (this model subset, therefore, simulates larger intertropical precipitation variability). Nonetheless, climate models may underestimate the full scale of natural variability. Several studies, for example, suggest that an abrupt change in the AMO/AMOC is primarily responsible for the interhemispheric SST shift (and, implicitly, the southward tropical precipitation shift, particularly in the Atlantic) around 1970 (Thompson et al. 2010; Dima and Lohmann 2010; Terray 2012). Although CMIP5 models—particularly those with aerosol indirect effects—capture most of the late-twentieth-century AMV (which is related to the AMO/AMOC) (Figs. 15a,b), they do not, in general, capture this abrupt interhemispheric Atlantic SST shift (Friedman et al. 2013) (Fig. 3e). Although this is also true for the abrupt southward tropical Atlantic precipitation shift (Fig. 3b), some ALL BOTH AIE′ realizations do yield an abrupt shift in the late 1960s (gray shading).

Externally forced models—particularly those that include both aerosol indirect effects—qualitatively reproduce both shifts, with anthropogenic and volcanic aerosols being the dominant drivers (Table 2). Moreover, models that include aerosol indirect effects yield significantly larger shifts than models that lack aerosol indirect effects. Since the prior generation of climate models (CMIP3) lacked aerosol indirect effects, our study implicitly shows a larger role of anthropogenic aerosols in driving late twentieth-century shifts in the tropical rain belt. This conclusion is similar to Hwang et al. (2013); however, their analysis is based on a very different approach using a global energetic framework.

In the Atlantic sector, CMIP5 AA experiments show that the bulk of the simulated southward and northward shift is driven by anthropogenic aerosols. However, models significantly underestimate the magnitude of both the hemispheric temperature contrasts and the tropical precipitation shifts, unless driven by the real-world evolution of sea surface temperatures. ALL BOTH AIE and ALL BOTH AIE′ once again yield the largest trends, but they too underestimate the observed shift. This includes the abrupt southward shift in the early 1970s (Figs. 3b,e), which many have suggested is due to internal variability associated with the AMOC. Although CMIP5 models—particularly ALL BOTH AIE—reproduce the bulk of recent AMV, we cannot rule out natural factors in driving part of the tropical Atlantic precipitation shifts.

Another possible factor contributing to model underestimation of the tropical Atlantic precipitation shifts involves underestimation of the sensitivity of the shift to the Atlantic interhemispheric temperature gradient—particularly in the extratropics (Fig. 15c). This result is consistent with Seo et al. (2014) and the importance of the extratropics in driving tropical precipitation shifts via cloud radiative feedbacks. Thus, most CMIP5 models may underestimate this feedback—particularly in the Atlantic. Future work will investigate this in detail.

In the Pacific sector (which is defined as everything except the Atlantic sector), CMIP5 AA experiments show that the bulk of the simulated southward tropical precipitation shift is due to anthropogenic aerosols. Models that include both the cloud-albedo and lifetime aerosol indirect effects (ALL BOTH AIE) reproduce most of the observed relative cooling of the NH Pacific and the southward tropical precipitation shift. The subset of models that yield the largest hemispheric aerosol RF gradient (ALL BOTH AIE′) yields larger relative cooling of the NH Pacific and a larger southward shift, in better agreement with the observations (Tables 2 and 3). Moreover, the contribution from natural variability—including the abrupt southward shift in the early 1970s—is smaller in the Pacific (Figs. 3c,f). Coupled models also generally better reproduce the observed and AMIP shift sensitivities in the Pacific sector (Fig. 15d). Thus, we conclude that anthropogenic aerosols are the dominant driver of the observed southward tropical precipitation shift in the Pacific sector.

Mechanistically, tropical precipitation shifts are driven by interhemispheric sea surface temperature variations, which are associated with hemispherically asymmetric changes in the strength, location, and cross-equatorial energy transport of the Hadley cells. Models with a larger hemispheric aerosol radiative forcing gradient (i.e., models with indirect effects) yield larger hemispheric temperature contrasts and, in turn, larger meridional precipitation shifts (Figs. 10 and 11). HadGEM2-ES AA (Figs. 12e–h) and idealized CAM5 SOM simulations (Fig. 13) show that aerosols can drive not only cooling in the Northern Hemisphere but also warming in the Southern Hemisphere. This hemispheric temperature contrast is consistent with direct modulation of SSTs through changes in solar radiation and by a dynamical response involving tropical ocean–atmosphere coupling associated with hemispheric contrasts in low-latitude surface pressure and winds as well as low clouds. This response scales with the interhemispheric aerosol RF gradient (Fig. 11) and exists in ALL BOTH AIE′ (Figs. 12a–d). This result helps to explain why there is a current lack of model consensus on the role aerosols play in driving recent tropical precipitation shifts and identifies the reason why some models simulate a larger fraction of the observed shifts.

Aerosol radiative forcing remains highly uncertain, with IPCC estimates ranging from −1.9 to −0.1 W m−2 (Myhre et al. 2013). Although one interpretation of our results is that many models may underestimate the hemispheric contrast in anthropogenic aerosol RF, the MACC aerosol reanalysis, which includes estimates of the shortwave direct and first indirect radiative forcing by anthropogenic aerosols, yields hemispheric aerosol RF gradients much smaller than those estimated by most models. Although the MACC reanalysis provides one strand of evidence, and is itself based on a model, this suggests that CMIP5 models may exaggerate the hemispheric asymmetry in anthropogenic aerosol forcing. If so, models may overestimate the anthropogenic contribution to twentieth-century tropical precipitation shifts.

All future emission scenarios show that anthropogenic aerosol emissions decrease substantially, reaching preindustrial values by 2100 (Moss et al. 2010). Consistent with the results discussed here, Rotstayn et al. (2015) and Allen (2015) find larger increases in the interhemispheric (north minus south) asymmetry of both tropical sea surface temperatures and precipitation in models that include a more sophisticated treatment of aerosols (i.e., indirect effects). Moreover, Allen (2015) finds a larger northward shift in the Pacific and a weaker shift in the Atlantic partially due to weakening of the AMOC. Thus, anthropogenic aerosols are likely the dominant drivers of tropical precipitation shifts—particularly in the Pacific—during both the twentieth and twenty-first centuries.

Acknowledgments

Robert J. Allen was supported by NASA Grant NNX13AC06G. Ben Booth was supported by the Joint UK DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). We acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups for producing and making available their model output. For CMIP, the U.S. Department of Energy’s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals.

REFERENCES

REFERENCES
Ackerley
,
D.
,
B. B. B.
Booth
,
S. H. E.
Knight
,
E. J.
Highwood
,
D. J.
Frame
,
M. R.
Allen
, and
D. P.
Rowell
,
2011
:
Sensitivity of twentieth-century Sahel rainfall to sulfate aerosol and CO2 forcing
.
J. Climate
,
24
,
4999
5014
, doi:.
Adler
,
R. F.
, and Coauthors
,
2003
:
The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present)
.
J. Hydrometeor.
,
4
,
1147
1167
, doi:.
Allen
,
R. J.
,
2015
:
A 21st century northward tropical precipitation shift caused by future anthropogenic aerosol reductions
.
J. Geophys. Res. Atmos.
, doi:, in press.
Allen
,
R. J.
, and
S. C.
Sherwood
,
2011
:
The impact of natural versus anthropogenic aerosols on atmospheric circulation in the Community Atmosphere Model
.
Climate Dyn.
,
36
,
1959
1978
, doi:.
Andreae
,
M. O.
,
2009
:
Correlation between cloud condensation nuclei concentration and aerosol optical thickness in remote and polluted regions
.
Atmos. Chem. Phys.
,
9
,
543
556
, doi:.
Arbuszewski
,
J. A.
,
P. B.
deMenocal
,
C.
Cléroux
,
L.
Bradtmiller
, and
A.
Mix
,
2013
:
Meridional shifts of the Atlantic intertropical convergence zone since the last glacial maximum
.
Nat. Geosci.
,
6
,
959
962
, doi:.
Bellomo
,
K.
,
A. C.
Clement
,
T.
Mauritsen
,
G.
Radel
, and
B.
Stevens
,
2015
:
The influence of cloud feedbacks on equatorial Atlantic variability.
J. Climate
,
28
,
2725
2744
, doi:.
Bellouin
,
N.
,
J.
Quaas
,
J. J.
Morcrettes
, and
O.
Boucher
,
2013
:
Estimates of aerosol radiative forcing from the MACC re-analysis
.
Atmos. Chem. Phys.
,
13
,
2045
2062
, doi:.
Biasutti
,
M.
, and
A.
Giannini
,
2006
:
Robust Sahel drying in response to late 20th century forcings
.
Geophys. Res. Lett.
,
33
,
L11706
, doi:.
Booth
,
B. B. B.
,
N. J.
Dunstone
,
P. R.
Halloran
,
T.
Andrews
, and
N.
Bellouin
,
2012
:
Aerosols implicated as a prime driver of twentieth-century North Atlantic climate variability
.
Nature
,
484
,
228
232
, doi:.
Broccoli
,
A. J.
,
K. A.
Dahl
, and
R. J.
Stouffer
,
2006
:
Response of the ITCZ to Northern Hemisphere cooling
.
Geophys. Res. Lett.
,
33
, doi:.
Chang
,
C. Y.
,
J. C. H.
Chiang
,
M. F.
Wehner
,
A. R.
Friedman
, and
R.
Ruedy
,
2011
:
Sulfate aerosol control of tropical Atlantic climate over the twentieth century
.
J. Climate
,
24
,
2540
2555
, doi:.
Chen
,
M.
,
P.
Xie
,
J. E.
Janowiak
, and
P. A.
Arkin
,
2002
:
Global land precipitation: A 50-yr monthly analysis based on gauge observations
.
J. Hydrometeor.
,
3
,
249
266
, doi:.
Chiang
,
J. C. H.
, and
D. J.
Vimont
,
2004
:
Analogous meridional modes of atmosphere–ocean variability in the tropical Pacific and tropical Atlantic
.
J. Climate
,
17
,
4143
4158
, doi:.
Chiang
,
J. C. H.
, and
C. M.
Bitz
,
2005
:
Influence of high latitude ice cover on the marine intertropical convergence zone
.
Climate Dyn.
,
25
,
477
496
, doi:.
Chiang
,
J. C. H.
, and
A. R.
Friedman
,
2012
:
Extratropical cooling, interhemispheric thermal gradients, and tropical climate change
.
Annu. Rev. Earth Planet. Sci.
,
40
,
383
412
, doi:.
Chiang
,
J. C. H.
,
C. Y.
Chang
, and
M. F.
Wehner
,
2013
:
Long-term behavior of the Atlantic interhemispheric SST gradient in the CMIP5 historical simulations
.
J. Climate
,
26
,
8628
8640
, doi:.
Collins
,
W. D.
, and Coauthors
,
2004
: Description of the NCAR Community Atmosphere Model (CAM 3.0). NCAR Tech. Note NCAR/TN-464+STR, 214 pp. [Available online at http://www.cesm.ucar.edu/models/atm-cam/docs/description/description.pdf.]
Compo
,
G. P.
, and Coauthors
,
2011
:
The Twentieth Century Reanalysis Project
.
Quart. J. Roy. Meteor. Soc.
,
137
,
1
28
, doi:.
Dima
,
M.
, and
G.
Lohmann
,
2010
:
Evidence for two distinct models of large-scale ocean circulation changes over the last century
.
J. Climate
,
23
,
5
16
, doi:.
Dommenget
,
D.
, and
M.
Latif
,
2000
:
Interannual to decadal variability in the tropical Atlantic
.
J. Climate
,
13
,
777
792
, doi:.
Donohoe
,
A.
,
D. M. W.
Frierson
, and
D. S.
Battisti
,
2014
:
The effect of ocean mixed layer depth on climate in slab ocean aquaplanet experiments
.
Climate Dyn.
,
43
,
1041
1055
, doi:.
Evan
,
A. T.
,
G. R.
Foltz
,
D.
Zhang
, and
D. J.
Vimont
,
2011
:
Influence of African dust on ocean–atmosphere variability in the tropical Atlantic
.
Nat. Geosci.
,
4
,
762
765
, doi:.
Evan
,
A. T.
,
R. J.
Allen
,
R.
Bennartz
, and
D. J.
Vimont
,
2013
:
The modification of sea surface temperature anomaly linear damping time scales by stratocumulus clouds
.
J. Climate
,
26
,
3619
3630
, doi:.
Evan
,
A. T.
,
C.
Flamant
,
S.
Fiedler
, and
O.
Doherty
,
2014
:
An analysis of aeolian dust in climate models
.
Geophys. Res. Lett.
,
41
,
5996
6001
, doi:.
Folland
,
C. K.
,
T. N.
Palmer
, and
D. E.
Parker
,
1986
:
Sahel rainfall and worldwide sea temperatures, 1901–85
.
Nature
,
320
,
602
607
, doi:.
Friedman
,
A. R.
,
Y. T.
Hwang
,
J. C. H.
Chiang
, and
D. M. W.
Frierson
,
2013
:
Interhemispheric temperature asymmetry over the twentieth century and in future projections
.
J. Climate
,
26
,
5419
5433
, doi:.
Frierson
,
D. M. W.
, and
Y.-T.
Hwang
,
2012
:
Extratropical influence on ITCZ shifts in slab ocean simulations of global warming
.
J. Climate
,
25
,
720
733
, doi:.
Ghan
,
S. J.
,
2013
:
Technical note: Estimating aerosol effects on cloud radiative forcing
.
Atmos. Chem. Phys.
,
13
,
9971
9974
, doi:.
Giannini
,
A.
,
R.
Saravanan
, and
P.
Chang
,
2003
:
Oceanic forcing of Sahel rainfall on interannual to interdecadal time scales
.
Science
,
302
,
1027
1030
, doi:.
Grandey
,
B. S.
, and
P.
Stier
,
2010
:
A critical look at spatial scale choices in satellite-based aerosol indirect effect studies
.
Atmos. Chem. Phys.
,
10
,
11 459
11 470
, doi:.
Harris
,
I.
,
P.
Jones
,
T.
Osborn
, and
D.
Lister
,
2014
:
Updated high-resolution grids of monthly climatic observations—The CRU TS3.10 dataset
.
Int. J. Climatol.
,
34
,
623
642
, doi:.
Haug
,
G. H.
,
K. A.
Hughen
,
D. M.
Sigman
,
L. C.
Peterson
, and
U.
Rohl
,
2001
:
Southward migration of the intertropical convergence zone through the Holocene
.
Science
,
293
,
1304
1308
, doi:.
Haywood
,
J. M.
,
A.
Jones
,
N.
Bellouin
, and
D.
Stephenson
,
2013
:
Asymmetric forcing from stratospheric aerosols impacts Sahelian rainfall
.
Nat. Climate Change
,
3
,
660
665
, doi:.
Hoerling
,
M.
,
J.
Hurrell
,
J.
Eischeid
, and
A.
Phillips
,
2006
:
Detection and attribution of twentieth-century Northern and Southern African rainfall change
.
J. Climate
,
19
,
3989
4008
, doi:.
Hwang
,
Y.-T.
,
D. M. W.
Frierson
, and
S. M.
Kang
,
2013
:
Anthropogenic sulfate aerosol and the southward shift of tropical precipitation in the late 20th century
.
Geophys. Res. Lett.
,
40
,
2845
2850
, doi:.
Kalnay
,
E.
, and Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project
.
Bull. Amer. Meteor. Soc.
,
77
,
437
471
, doi:.
Kang
,
S. M.
,
I. M.
Held
,
D. M. W.
Frierson
, and
M.
Zhao
,
2008
:
The response of the ITCZ to extratropical thermal forcing: Idealized slab-ocean experiments with a GCM
.
J. Climate
,
21
,
3521
3532
, doi:.
Knight
,
J. R.
,
C. K.
Folland
, and
A. A.
Scaife
,
2006
:
Climate impacts of the Atlantic multidecadal oscillation
.
Geophys. Res. Lett.
,
33
, L17706, doi:.
Lamarque
,
J. F.
, and Coauthors
,
2010
:
Historical (1850–2000) gridded anthropogenic and biomass burning emissions of reactive gases and aerosols: Methodology and application
.
Atmos. Chem. Phys.
,
10
,
7017
7039
, doi:.
Martin
,
E. R.
,
C.
Thorncroft
, and
B. B. B.
Booth
,
2014
:
The multidecadal Atlantic SST–Sahel rainfall teleconnection in CMIP5 simulations
.
J. Climate
,
27
,
784
806
, doi:.
Ming
,
Y.
, and
V.
Ramaswamy
,
2011
:
A model investigation of aerosol-induced changes in tropical circulation
.
J. Climate
,
24
,
5125
5133
, doi:.
Morice
,
C. P.
,
J. J.
Kennedy
,
N. A.
Rayner
, and
P. D.
Jones
,
2012
:
Quantifying uncertainties in global and regional temperature change using an ensemble of observational estimates: The HadCRUT4 data set
.
J. Geophys. Res.
,
117
,
D08101
, doi:.
Moss
,
R. H.
, and Coauthors
,
2010
:
The next generation of scenarios for climate change research and assessment
.
Nature
,
463
,
747
756
, doi:.
Myhre
,
G.
, and Coauthors
,
2013
: Anthropogenic and natural radiative forcing. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 659–740. [Available online at http://www.climatechange2013.org/images/report/WG1AR5_Chapter08_FINAL.pdf.]
Neale
,
R. B.
, and Coauthors
,
2012
: Description of the NCAR Community Atmosphere Model (CAM 5.0). NCAR Tech. Note NCAR/TN-486+STR, 274 pp. [Available online at http://www.cesm.ucar.edu/models/cesm1.0/cam/docs/description/cam5_desc.pdf.]
Peterson
,
L. C.
,
G. H.
Haug
,
K. A.
Hughen
, and
U.
Röhl
,
2000
:
Rapid changes in the hydrologic cycle of the tropical Atlantic during the last glacial
.
Science
,
290
,
1947
1951
, doi:.
Peterson
,
T. C.
, and
R. S.
Vose
,
1997
:
An overview of the global historical climatology network temperature database
.
Bull. Amer. Meteor. Soc.
,
78
,
2837
2849
, doi:.
Poli
,
P.
, and Coauthors
,
2013
: The data assimilation system and initial performance evaluation of the ECMWF pilot reanalysis of the 20th-century assimilating surface observations only (ERA-20C). ECMWF Tech. Rep., 59 pp. [Available online at http://old.ecmwf.int/publications/library/ecpublications/_pdf/era/era_report_series/RS_14.pdf.
Quaas
,
J.
,
O.
Boucher
,
N.
Bellouin
, and
S.
Kinne
,
2008
:
Satellite-based estimate of the direct and indirect aerosol climate forcing
.
J. Geophys. Res.
,
113
,
D05204
, doi:.
Rayner
,
N. A.
,
D. E.
Parker
,
E. B.
Horton
,
C. K.
Folland
,
L. V.
Alexander
,
D. P.
Rowell
,
E. C.
Kent
, and
A.
Kaplan
,
2003
:
Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century
.
J. Geophys. Res.
,
108
,
4407
, doi:.
Ridley
,
H. E.
, and Coauthors
,
2015
:
Aerosol forcing of the position of the intertropical convergence zone since AD 1550
.
Nat. Geosci.
,
8
,
195
200
, doi:.
Rotstayn
,
L. D.
, and
U.
Lohmann
,
2002
:
Tropical rainfall trends and the indirect aerosol effect
.
J. Climate
,
15
,
2103
2116
, doi:.
Rotstayn
,
L. D.
,
M. A.
Collier
, and
J.-J.
Luo
,
2015
:
Effects of declining aerosols on projections of zonally averaged tropical precipitation
.
Environ. Res. Lett.
,
10
,
044018
, doi:.
Sato
,
M.
,
J. E.
Hansen
,
M. P.
McCormick
, and
J. B.
Pollack
,
1993
:
Stratospheric aerosol optical depth, 1850–1990
.
J. Geophys. Res.
,
98
,
22 987
22 994
, doi:.
Schneider
,
T.
,
T.
Bischoff
, and
G. H.
Haug
,
2014
:
Migrations and dynamics of the intertropical convergence zone
.
Nature
,
513
,
45
53
, doi:.
Seo
,
J.
,
S. M.
Kang
, and
D. M. W.
Frierson
,
2014
: Sensitivity of intertropical convergence zone movement to the latitudinal position of thermal forcing. J. Climate, 27, 3035–3042, doi:.
Shindell
,
D. T.
, and Coauthors
,
2013
:
Radiative forcing in the ACCMIP historical and future climate simulations
.
Atmos. Chem. Phys.
,
13
,
2939
2974
, doi:.
Tanimoto
,
Y.
, and
S.-P.
Xie
,
2002
:
Inter-hemispheric decadal variations in SST, surface wind, heat flux and cloud cover over the Atlantic Ocean
.
J. Meteor. Soc. Japan
,
80
,
1199
1219
, doi:.
Taylor
,
K. E.
,
R. J.
Stouffer
, and
G. A.
Meehl
,
2012
:
An overview of CMIP5 and the experiment design
.
Bull. Amer. Meteor. Soc.
,
93
,
485
498
, doi:.
Terray
,
L.
,
2012
:
Evidence for multiple drivers of North Atlantic multi-decadal climate variability
.
Geophys. Res. Lett.
,
39
,
L19712
, doi:.
Thompson
,
D. W. J.
,
J. J.
Kennedy
, and
P. D.
Jones
,
2010
:
An abrupt drop in Northern Hemisphere sea surface temperatures around 1970
.
Nature
,
467
,
444
447
, doi:.
Wilks
,
D. S.
,
2006
: Statistical Methods in the Atmospheric Sciences. Academic Press, 627 pp.
Williams
,
K. D.
,
A.
Jones
,
D. L.
Roberts
,
C. A.
Senior
, and
M. J.
Woodage
,
2001
:
The response of the climate system to the indirect effects of anthropogenic sulfate aerosol
.
Climate Dyn.
,
17
,
845
856
, doi:.
Willmott
,
C. J.
, and
K.
Matsuura
,
1995
:
Smart interpolation of annually averaged air temperature in the United States
.
J. Appl. Meteor.
,
34
,
2577
2586
, doi:.
Zhang
,
R.
, and
T. L.
Delworth
,
2006
:
Impact of Atlantic multidecadal oscillations on India/Sahel rainfall and Atlantic hurricanes
.
Geophys. Res. Lett.
,
33
,
L17712
, doi:.
Zhang
,
R.
, and Coauthors
,
2013
:
Have aerosols caused the observed Atlantic multidecadal variability?
J. Atmos. Sci.
,
70
,
1135
1144
, doi:.