Changes in tropical width can have important consequences in sectors including ecosystems, agriculture, and health. Observations suggest tropical expansion over the past 30 years although studies have not agreed on the magnitude of this change. Climate model projections have also indicated an expansion and show similar uncertainty in its magnitude. This study utilizes an objective, longitudinally varying, tropopause break method to define the extent of the tropics at upper levels. The location of the tropopause break is associated with enhanced stratosphere–troposphere exchange and thus its structure influences the chemical composition of the stratosphere. The method shows regional variations in the width of the upper-level tropics in the past and future. Four modern reanalyses show significant contraction of the tropics over the eastern Pacific between 1981 and 2015, and slight but significant expansion in other regions. The east Pacific narrowing contributes to zonal mean narrowing, contradicting prior work, and is attributed to the use of monthly and zonal mean data in prior studies. Six global climate models perform well in representing the climatological location of the tropical boundary. Future projections show a spread in the width trend (from ~0.5° decade−1 of narrowing to ~0.4° decade−1 of widening), with a narrowing projected across the east Pacific and Northern Hemisphere Americas. This study illustrates that this objective tropopause break method that uses instantaneous data and does not require zonal averaging is appropriate for identifying upper-level tropical width trends and the break location is connected with local and regional changes in precipitation.
The width of the tropics is not fixed in space or time and is linked to changes in weather and climate, including precipitation (Fu et al. 2006; Hu and Fu 2007; Seager et al. 2007; Seidel et al. 2008). Changes in tropical width can have important societal and economic consequences in a variety of sectors including natural ecosystems, agriculture, and health. Observations suggest the tropics have expanded over the past 30 years due to increased greenhouse gas concentrations in the atmosphere (e.g., Seidel et al. 2008; Lucas et al. 2014). In conjunction with this tropical expansion, observations have shown that tropical plants and animals have migrated poleward in an attempt to track their preferred climatic conditions (Parmesan and Yohe 2003; Chen et al. 2011; VanDerWal et al. 2013). Vector-borne diseases have also spread poleward (e.g., into the United States) (Githeko et al. 2000; Kovats et al. 2001; Gatewood et al. 2009). The expansion of the tropics and its associated atmospheric circulations also impact atmospheric chemistry through changes in the global transport of pollutants that can have significant impacts to human health (Eyring et al. 2010).
An expansion of the tropics has been extensively documented in the literature in a zonal mean sense (e.g., Seidel et al. 2008; Lu et al. 2009; Davis and Rosenlof 2012). However, these previous studies have not agreed on the magnitude of the observed changes with estimates of tropical widening from ~0.25° (Archer and Caldeira 2008) to ~3° (Seidel and Randel 2007) latitude per decade. The range of widening estimates may be the result of differing methods to define the tropics, datasets, and/or time periods (e.g., Birner 2010) and may also reveal deficiencies in our understanding of the mechanisms controlling the width of the tropics. Climate model projections have also indicated a marginal to significant expansion of the tropics and show similar uncertainty (Lu et al. 2007; Davis and Rosenlof 2012; Davis and Birner 2017; Grise et al. 2018).
There is no unanimously accepted definition of the boundary between tropics and extratropics. Diagnostics of tropical width frequently use the zonal mean meridional mass streamfunction, which can identify the extent of the Hadley circulation (e.g., Frierson et al. 2007; Hu and Fu 2007; Lu et al. 2007; Johanson and Fu 2009). Studies have also used observations of cloud and precipitation features consistent with the locations of the Hadley circulation (e.g., Hu and Fu 2007; Johanson and Fu 2009; Quan et al. 2014; Mantsis et al. 2017) or measures based on jet stream locations (e.g., Archer and Caldeira 2008; Manney and Hegglin 2018). Hudson et al. (2006) used atmospheric ozone concentrations to identify the width of the tropics, but this approach has been recently shown to be challenging due to discontinuities in the ozone record (Davis et al. 2018).
Another distinguishing feature of the tropics is the height of the tropopause. The transition in tropopause altitude between the tropics and extratropics is often discontinuous and is referred to as the tropopause break. Prior studies have used the frequency of high tropical tropopause heights to measure the tropical width (e.g., Seidel and Randel 2007; Lu et al. 2009; Birner 2010; Lucas et al. 2012; Ao and Hajj 2013), but few studies have used a method based on the the tropopause break (Davis and Rosenlof 2012; Davis and Birner 2013, 2017). While the tropopause break has been shown to be uncorrelated with Hadley circulation variations (Davis and Birner 2017; Waugh et al. 2018), the location of the tropopause break and the associated upper-level boundary between tropical and extratropical air is associated with enhanced stratosphere–troposphere exchange and mixing, with implications as to the chemical composition of the stratosphere (Pan et al. 2004).
Each individual width metric has its own advantages and disadvantages but there are weaknesses that are common across previously used definitions. The primary weakness is that most metrics require zonal averaging and those that do not are presented in a zonal mean sense to facilitate comparisons. Hence, spatial variability of the tropical width cannot be assessed, aside from differences between the Northern and Southern Hemispheres. Lucas and Nguyen (2015) used radiosonde measurements to identify regional characteristics of tropical expansion, but this discrete approach does not cover the full tropics. A limited number of studies investigate regional trends but do not use metrics based on tropopause height (Chen et al. 2014; Grise et al. 2018; Staten et al. 2019). A second limitation is that previous studies use Eulerian monthly mean fields from observations or models to diagnose the tropical boundaries and measure changes in width over time. However, the impact of using monthly mean versus instantaneous (e.g., 0000 UTC) fields has not been documented. To improve upon these limitations, this study uses tropopause break metrics (i.e., defining the tropics as the region where the tropopause is elevated, as opposed to a definition based on the Hadley cell) to understand long-term trends and continental-scale variability in upper-level tropical width in the recent past and the twenty-first century.
2. Data and methods
Temperature data on model levels are utilized from reanalysis and climate model output to define the tropopause break and upper-level tropical width. Model level data are required as traditional pressure levels often do not have enough vertical resolution to reliably identify the tropopause and tropopause break. Instantaneous data at 0000 UTC (unless stated otherwise) as well as monthly mean output are used to define the tropopause break. Note that all available instantaneous data at synoptic times (0000, 0600, 1200, and 1800 UTC) were originally used to define the tropopause break in the reanalyses, but the analysis using a single synoptic time was found to be consistent with that using the larger dataset and less computationally demanding (both in terms of required disk space and processing; see also section 2c).
a. Reanalyses and observations
Data between 1981 and 2015 from four reanalyses are used in this study. The use of multiple reanalysis products enables uncertainty quantification in the location and trend of tropical width, as the reanalyses have differing spatial resolutions, assimilated data and assimilation methods, underlying model physics, and biases (see, e.g., Fujiwara et al. 2017). The reanalysis data used in this study are the National Centers for Environmental Prediction Climate Forecast System Reanalysis (CFSR), the European Centre for Medium-Range Weather Forecasts (ECMWF) Interim Reanalysis (ERA-Interim), the Japan Meteorological Agency (JMA) Japanese 55-year Reanalysis (JRA-55), and the National Aeronautics and Space Administration (NASA) Modern-Era Retrospective analysis for Research and Applications version 2 (MERRA-2). While CFSR is only available from 1979 to 2010, CFSR output is extended to the year 2015 in this study using analyses from the Climate Forecast System version 2 (CFSv2) model. Table 1 lists information about the model-level reanalysis data including the horizontal and vertical resolution.
To alleviate possible concerns regarding trends in the tropopause break being dependent on model-based reanalysis and climate models only, comparisons are also made with observations from the Integrated Global Radiosonde Archive (IGRA) version 2 (Durre et al. 2006; Durre and Yin 2008). The IGRA contains observations of pressure and temperature from global radiosonde and pilot balloon observations since 1905, but data between 1981 and 2015 will be used in this study. Only those locations with nearly continuous observations during the 35-yr study period are retained for analysis [see section 2 of Xian and Homeyer (2019) for more detail]. Monthly mean precipitation observations from the Global Precipitation Climatology Project version 2.3 (1979–2018) are also utilized to connect the tropopause break to surface weather and climate (Adler et al. 2003).
b. Climate model output
To determine historical and future trends in tropical width, global climate model output from eight members of phase 5 of the Coupled Model Intercomparison Project (CMIP5; Taylor et al. 2012) is used (see Table 2). The models chosen were based upon availability of instantaneous model-level temperature data at 6-h intervals. All model output is obtained for 0000 UTC, except IPSL-CM5A-LR for which 0000 UTC was not available and 0300 UTC was used instead. We utilize output from two CMIP5 simulation types: fully coupled ocean–atmosphere historical simulations (1950 to 2005) and future representative concentration pathway (RCP8.5) simulations (2006 to 2100). The RCP8.5 scenario was chosen as it represents an 8.5 W m−2 radiative forcing in 2100, the largest forcing available in the CMIP5 simulations. Horizontal resolution of the six models are similar (~2.5°), but the number of vertical model levels range from 24 to 80 (Table 2).
c. Tropopause break identification
There are two key identification methods in this study: one to locate the altitude of the tropopause and another to identify the location of the tropopause break. Methods to identify the altitude of the tropopause vary widely in the literature, but the conventional World Meteorological Organization (WMO 1957) definition is the most commonly used due to its reliability in demarcating dynamical and chemical transitions between the troposphere and stratosphere (e.g., Birner et al. 2002; Pan et al. 2004; Gettelman et al. 2011). The WMO definition identifies the tropopause using the temperature lapse rate as the lowest altitude at which the lapse rate decreases to 2°C km−1, provided that the average lapse rate from this level to any point within the next higher 2 km does not exceed 2°C km−1. This definition is applied to every grid column in the reanalyses, model output, and IGRA observations, following the mathematical approach outlined in Homeyer et al. (2010).
Tropical boundaries are identified in each hemisphere using two distinct methods: 1) an “objective” tropopause break method applied to instantaneous reanalysis and climate model output and 2) a tropopause altitude “frequency” method applied to IGRA observations, reanalyses, and climate model output (not shown). The objective method is equivalent to that outlined in Homeyer and Bowman (2013), for which we provide a short summary here and an illustration of this approach is provided in Fig. 1. First, frequency distributions of instantaneous global WMO tropopause altitudes are computed and the minimum tropopause altitude frequency between the tropical high-altitude mode and extratropical low-altitude mode is identified. Global contours of the minimum frequency tropopause altitude, which correspond to the sharp tropical–extratropical transition, are then drawn and stored as the instantaneous objective locations (i.e., latitudes) of the tropopause break at every longitude in each hemisphere for the given time (e.g., 0000 UTC). Note that the synoptic time used to diagnose break latitudes and trends was evaluated for each reanalysis and the results reported here were insensitive to this choice (not shown).
The frequency method for tropopause break identification uses the monthly frequency of tropopause altitudes (pressures) above and below 150 hPa. The sensitivity of the results to the 150-hPa threshold were tested and results are consistent in the pressure range from 125 to 175 hPa (the common width of the “well” in tropopause frequency between the high-altitude tropical mode and the low-altitude extratropical mode). Ultimately, to determine the tropopause break latitude in each hemisphere (and therefore the tropical width), a frequency threshold must be chosen. We use a 50% threshold to identify the tropopause break, which we find to provide results that are generally consistent with the objective method for the reanalyses (cf. Figs. 2 and 3).
d. Trends and significance
Long-term trends in the latitude of the tropopause break (or frequency of tropopause altitudes above/below a threshold) are computed using linear regression of monthly mean tropopause break latitudes at each longitude. Trends are determined in degrees (or percent) per decade and are deemed significant if they exceed the 3σ uncertainty of the measured slope, which is analogous to statistical significance at the 99% confidence level for a Gaussian distribution. All time series are deseasonalized prior to trend calculation using a low-pass filter that removes variability at time scales less than or equal to 1 year.
a. Tropopause break location climatology and trends
The tropopause break location climatology from the reanalysis using the objective method is shown in Fig. 2 (top panels) for the Northern Hemisphere (NH) and Southern Hemisphere (SH). The break location varies between approximately 30° and 38°N in the NH (mean of 34.1°N) and 26° and 34°S in the SH (mean of 32.2°S) with the SH break location being more zonal, consistent with prior work identifying the tropopause break latitude and sensitivities to it (Birner 2010). Overall there is good agreement in the longitudinal variability of the tropopause break latitude in each hemisphere but magnitudes vary between the reanalysis products, where CFSR is often the outlier (especially in the SH). CFSR is consistently 2°–3° equatorward of the other reanalyses in the SH and ~1° equatorward in the NH, with the largest differences over the ocean basins (where the radiosonde network is sparse). The tropopause break is farthest equatorward in the east Pacific (~210°–250°E) in both hemispheres and farthest poleward in the NH west Pacific (~160°E). The agreement between MERRA-2, JRA-55, and ERA-Interim is strong across the entire SH and the eastern NH. The highest agreement between all four reanalyses is over the East Asian coast, where the radiosonde network is extremely dense.
The frequency method (Fig. 3) produces an upper-level tropical width climatology that is almost identical to the objective method (Fig. 2) in the reanalysis. The equatorward break location in the east Pacific (and east Atlantic) in each hemisphere is evident, as well as the good agreement between three of the four reanalyses. CFSR is still equatorward of all the other reanalysis products in each hemisphere although the bias is reduced in the frequency method.
The historical trends by longitude from the reanalysis products are shown in the bottom panels of Figs. 2 and 3 for the objective and frequency methods, respectively. The two methods produce trends that are very similar in terms of the spatial pattern and the magnitude. In fact, correlations between the longitudinally resolved time series for each method (not shown) are near 1 throughout most of each hemisphere, with reductions to ~0.6 in the NH and ~0.85 in the SH within the eastern ocean basins. JRA-55, ERA-Interim, and MERRA-2 show consistency in the longitudinal trend in both hemispheres. Of particular interest is the significant historical narrowing of the upper-level tropical width in each hemisphere in the east Pacific of up to ~1° decade−1. ERA-Interim and MERRA-2 show the largest trends in the east Pacific, with JRA-55 weaker (up to ~0.5° decade−1) but still significant. In the NH outside of the east Pacific, small but significant widening has been observed in the west Pacific and across most of Asia. In the South Atlantic, narrowing trends have been evident since 1981 in three of the four reanalyses but there is much more disagreement in the North Atlantic with JRA-55 and MERRA-2 showing significant but opposite trends. As the tropopause break location varies seasonally, the trends also vary with the season although the overall pattern of widening and narrowing is consistent with that of the annual mean (see the appendix and Figs. A1–A4).
These tropopause break trends are consistent with regional trends in the location of the subtropical jet, including a narrowing in the east Pacific (Manney and Hegglin 2018). They are also in the same sense as the observed narrowing of the intertropical convergence zone (ITCZ) in the deep tropics (Wodzicki and Rapp 2016; Byrne et al. 2018). Although previous work has shown that the position of the ITCZ can play a role in setting the location of the Hadley cell edge in a warming climate (Kang and Lu 2012), it is important to note that the tropopause break location is not a direct measure of the Hadley cell edge.
As seen in the climatological location of the tropopause break and the historical trend, CFSR shows patterns that are inconsistent with the remaining reanalysis products. While the tropopause break is located farther equatorward in each hemisphere, CFSR also shows a large and significant widening across both hemispheres of up to ~1° decade−1. The east Pacific narrowing seen in the other reanalyses is not evident in CFSR in either hemisphere, but the diagnosed widening is weaker (and insignificant) there. The largest widening trends for CFSR are observed in in the NH Atlantic, where there is the greatest spread (and hence uncertainty) in the sign of the trend between all four reanalysis products.
A spatial comparison of the objective and frequency methods applied to reanalysis, as well as a frequency diagnosis of observations from radiosondes are shown in Fig. 4. The objectively defined break location matches well with the 50% frequency criteria as discussed in section 2c. The frequency method is also applied to the radiosonde data (Fig. 4, top left) and the resulting patterns are similar to those found in the reanalyses, except for CFSR, which is once again an outlier. For radiosonde observations and reanalyses, a significant reduction in frequency of low tropopause heights (or higher pressures) is observed across the extratropical continents in both hemispheres. The largest reductions (approximately 2% decade−1) are observed across China and East Asia, with a split feature evident over China in all of the reanalyses (and observations). This split feature may be indicative of seasonal changes in the location of the tropopause break there. As a reduced frequency of tropopause heights greater than 150 hPa means more high tropical tropopauses, this is consistent with upper-level tropical expansion across the continents where radiosonde data are prevalent, similar to the objective analysis in Fig. 2. This result and the consistency between radiosondes and reanalyses adds additional confidence to the tropopause break identified using the objective method and corresponding historical trend analysis.
The upper-level tropical narrowing in the east Pacific and Atlantic diagnosed from the reanalyses is evident in the few radiosonde locations in these regions. Namely, an increasing trend of approximately 1% decade−1 in low tropopause altitudes (pressures greater than 150 hPa) is seen in radiosonde data from Hawaii, Central America, and the Caribbean and some South Pacific islands, which are within broad regions of 1%–2% increases per decade in the reanalyses. It is worth noting that high-resolution temperature profiles from Global Navigation Satellite System–Radio Occultation (GNSS-RO) could, in time, be used to more uniformly and globally diagnose tropopause characteristics and long-term changes and better evaluate their representation in the reanalyses, but these data records are not yet sufficient (data are only available from early 2001 to the present).
The large narrowing trends in the reanalyses are in locations with few radiosonde observations (Fig. 2). The time series of the tropopause break latitude at two locations of large trends (210°E in the NH and 240°E in the SH) are shown in Fig. 5, along with the year at which high-resolution temperature profiles from the GNSS-RO retrieval method were first assimilated in each reanalysis system. It is evident that, in both locations (and others examined but not shown here), there are no sharp discontinuities associated with assimilation changes. Thus, the large trends in regions of fewer radiosonde observations are not simply due to the addition of high-resolution, globally distributed satellite data to the assimilation system and are likely real, resolved signals of upper-level tropical narrowing in the east Pacific.
b. Historical width of the tropics in CMIP5
The tropopause break climatology using the objective method with the CMIP5 historical output (1971–2005) is shown in Fig. 6. The model spread in the NH is slightly larger than that found in the reanalyses (approximately 26° to 37°N) due to the equatorward location of the break across North America and the Atlantic Ocean in five of the six models. The GFDL-CM3 model is closest to the reanalysis mean in the NH, and is the only model that lies on the poleward side of the reanalysis mean location across the east Pacific and Atlantic Oceans. In the SH, a similar pattern emerges, with five of the six models placing the tropopause break up to 4° equatorward of the reanalysis mean across nearly all longitudes. Again, the tropopause break in the GFDL-CM3 is poleward of the reanalysis mean, making the climatological upper-level tropical width larger in that model.
Historical trends in the CMIP5 models are shown in the bottom panels of Fig. 6 using the objective method for 1971–2005. This period was chosen to match the number of years from the reanalysis. When using the longer 1951–2005, the trends are smaller (less than 0.5° decade−1) and more closely clustered around zero (not shown). There are no consistent trends observed in the CMIP5 models in the 30-yr historical period, which is expected given the different realizations of natural variability in the historical simulations. The importance of natural variability in tropical width trends is discussed further in Grise et al. (2019). The trends are generally smaller and less significant than those in the reanalysis (Fig. 2), consistent with Davis and Birner (2017), with two exceptions. The IPSL-CM5A-LR model shows large (up to 1° decade−1) and significant narrowing trends across both hemispheres and NorESM1-M shows the opposite, especially in the SH. The east Pacific narrowing observed in the reanalysis and radiosonde data is less evident in the CMIP5 models, especially in the NH, suggesting that this may be a signal of natural or internal variability as opposed to external anthropogenic forcing. However, in the SH, four of the six models show some significant narrowing in the east Pacific, although not as large or zonally extensive as the reanalysis and observations and shifted to the west.
c. Future projections of tropical width
The future projections of the tropopause break location, utilizing the objective method, show many similarities with the historical CMIP5 simulations in terms of the longitudinal variability and intermodel spread (Fig. 7). The GFDL CM3 tropopause break latitude remains poleward of the remaining models in both hemispheres. In addition to being the most poleward in the climatology, the GFDL CM3 model has the largest future narrowing trend in each hemisphere (approximately 0.5° decade−1). CanESM2 has the largest widening trend in each hemisphere (up to 0.4° decade−1) with longitudinal variations that are almost the reflection of GFDL CM3 (i.e., they both have the largest trends in the east Pacific and Atlantic but of opposite signs). The smallest trends across all models occur between 30° and 90°E in the NH, suggesting little change to the width of the tropics across west and central Asia.
In contrast to the historical period (Fig. 6), there is a significant upper-level tropical narrowing projected in the NH east Pacific in four of the six models of approximately 0.2° decade−1. Three of the four models show a distinct peak in the region, similar to that found in the reanalysis (Fig. 2), with GFDL CM3 showing a more longitudinally extensive narrowing. There is no consensus for east Pacific narrowing in the future in the SH from these models.
It is evident from Fig. 7 that there is no clear, zonally consistent, forced response in the location of the tropopause break in the future. However, there are regions with model clustering including weak changes across eastern Asia, widening across Europe, and narrowing in the east Pacific and the Americas. The uncertainty in the SH is larger than the NH. This highlights the necessity to examine the changes beyond a zonal mean, as distinct (and zonally varying) signals can be hidden using such methods (e.g., Tables 3–5).
a. Method comparison
The tropopause break identification methods (frequency and objective) agreed well in the historical period, as described in sections 3a and 3b. However, the use of the objective method was necessary for the analysis of the future projections. An increase in tropopause height in the future has been extensively documented (e.g., Lorenz and DeWeaver 2007; Vallis et al. 2015). The nonstationarity of the tropopause altitude in the future has a major impact on the trends diagnosed using the frequency method when using a fixed threshold (150 hPa in this case). Due to the increase in height of the tropopause over time, the trend in the frequency of occurrences above 150 hPa is dominated by the overall troposphere expansion rather than changes in the break location.
The objective method does not suffer from these same stationarity issues but does have its own limitations. Instantaneous contour drawing can be challenging or fail in regions with large wave breaking events, where deformations or fractures in globe-circling tropopause contours prevent break identification. These failure modes are infrequent when the tropopause altitude representing the frequency minimum is diagnosed at every model analysis time, as done in this study. However, successful identification of the frequency minimum depends on the bin resolution used in the frequency distribution analysis, which must be consistent with the vertical grid spacing of each model (i.e., it requires careful adjustment by the analyst).
b. Zonal mean trends
The observed narrowing in this study contradicts prior studies (e.g., Stachnik and Schumacher 2011; Davis and Birner 2017) that show a widening trend in multiple reanalyses using a variety of metrics including the height of the tropopause. One potential reason for these discrepancies is the use of monthly mean data in prior studies rather than the instantaneous data that are used here. Figure 8 illustrates the differences in tropopause break location and trends when using monthly mean versus instantaneous data from ERA-Interim and JRA-55 as input for the objective method. While the differences are relatively small in the NH, the monthly mean tropopause break location in the SH is 2°–3° poleward of the instantaneous location. The trend in the SH also shifts substantially from a narrowing to a widening trend, although narrowing is still evident in the east Pacific. Monthly mean data calculated from daily data can be strongly impacted by extremes at the daily scale, leading to unrealistic trends that depend more on variability, which we have found to be significantly increasing over time in our analysis using the objective method (not shown), especially in the regions where the trends are largest. This is especially apparent in the SH where data are sparse.
An apples-to-apples comparison of the results presented here with prior studies requires calculating zonal mean tropical width trends, which are shown for the reanalyses (Table 3), historical models (Table 4), and future models (Table 5) calculated using zonal mean latitudes of the break locations from the objective method applied to instantaneous data. For ERA-Interim and JRA-55, zonal mean trends calculated from the objective analysis using monthly mean fields as input are also provided in Table 3. MERRA-2, ERA-Interim, and JRA-55 all show negative trends in the zonal mean width of the tropics between 1981 and 2015 based on instantaneous data, or a narrowing in the upper-level tropics between −0.1° and −0.52° decade−1. As expected from Figs. 2 and 3, CFSR shows a large widening trend of the zonal mean at 0.98° decade−1.
Changes in trends when using monthly mean fields as input the objective method are also consistent with the longitudinally varying comparison in Fig. 8. Namely, the sign of the trend in tropical width changes from narrowing in the instantaneous analysis to widening in the monthly mean analysis. Similar sensitivities to data input were found in a recent study that diagnosed tropical boundaries using potential temperature gradients along the dynamical tropopause—a potential vorticity isosurface (Maher et al. 2020). However, the diagnosed edge latitudes and trends for monthly and daily data used in Maher et al. (2020) showed the opposite sensitivity to that here (i.e., more poleward edge latitudes and widening using daily data as input and more equatorward edge latitudes and narrowing using monthly data as input). These differences highlight the sensitivity of tropical width diagnosis to both method design and data input.
The historical CMIP5 simulations show on average, zero trend in the zonal mean tropical width, with a large range in the diagnosed trends. Four of the six models show relatively small trends in the tropical width (less than 0.1° decade−1), with IPSL-CM5A-LR matching most closely with three of the reanalyses in narrowing, and NorESM1-M most closely resembling the large widening in CFSR. In the future, CMIP5 models simulate a model average zonal mean of slight upper-level tropical narrowing, again with large spread and no strong relationship between the historical and future trends.
c. Connection with precipitation
Work by Waugh et al. (2018) and Solomon et al. (2016) has highlighted the lack of correlation between tropical width metrics based on the tropopause and measures of the Hadley cell and precipitation minus evaporation. While we do not suggest here that the tropopause break is measuring the width of the Hadley cell, it is a measure of the tropical air mass at upper levels, separating mostly barotropic environments from mostly baroclinic ones. By using the objective method and creating a zonally varying tropopause break location, we can reexamine the relationship between the break and precipitation here.
The utility of the objective tropopause break analysis with respect to precipitation is illustrated first in Fig. 9, which shows probability distribution functions of precipitation rate (from GPCP monthly data) as a function of latitude. The mean tropopause break location approximately distinguishes the subtropical and midlatitude precipitation features in each hemisphere in the global Eulerian distribution. However, binning the precipitation in a tropopause break–relative sense makes this precipitation boundary more distinct, particularly in the SH where the precipitation distribution is less impacted by land. The tropopause break location clearly marks the poleward boundary of the dry subtropics, providing confidence that not only is the tropopause break location important for chemical exchange in the upper atmosphere, but it can be a useful diagnostic for the global distribution of precipitation. Similar results are also seen in the tropopause break–relative precipitation from the CMIP5 historical simulations (not shown).
Figure 10 addresses the covariability of the tropopause break and precipitation anomalies spatially using the zonally varying tropopause break (Figs. 10a–c) and the zonal mean tropopause break (Figs. 10d–f). Here we utilize the instantaneous tropopause break from the objective method averaged over each month (and zonally averaged for Figs. 10d–f), regressed against monthly mean precipitation from GPCP. The results do not change when using precipitation minus evaporation. Using the NH break location, there are only very small, and not widely significant, regression values between precipitation and tropopause break location (zonally varying or zonal mean) as suggested by other studies. However, regression values are larger in the SH in the zonally varying break location (Fig. 10b) in comparison to the zonal mean (Fig. 10e).
The largest and most significant covariability is seen when using the zonally varying tropical width rather than the hemispheric break locations (Fig. 10c), as the width is defined by the distance between the two breaks. Widening tropics in the Maritime Continent are significantly correlated (at the 95% significance level) with increased precipitation anomalies. Significant positive regression values are also observed across the Sahel region of Africa. In each of these locations, one standard deviation change in the tropical width can explain over 20% of the total precipitation standard deviation (not shown). The largest (but not significant) negative regression values where precipitation decreases in association with upper-level tropical widening, are observed in the Arabian Peninsula and across portions of the Amazon and South Atlantic convergence zone.
Utilizing the zonal mean break location (Fig. 10f), which is calculated using the instantaneous tropopause break locations from the objective method and creating monthly averages before taking the zonal mean, shows weaker regression coefficients and explains less of the total precipitation standard deviation (less than 15% globally). The positive coefficients in the Pacific warm pool and negative values across the central equatorial Pacific are reminiscent of the precipitation response to variability in El Niño–Southern Oscillation (ENSO) (e.g., Smith et al. 2012) and differ from the zonally varying metric (Fig. 10c). However, it is clear from Fig. 10f that if the precipitation is also zonally averaged, as in prior studies, the majority of these signals cancel, especially across the tropics, leading to the insignificant correlation between tropopause break location and precipitation in some other studies (e.g., Davis and Birner 2017). As the largest trends in the tropopause break location are observed in the east Pacific (Fig. 2), the variations in this region are dominating the zonal average break metrics and leading to an ENSO-like signal. In the zonally varying break metrics, we can quantify local responses to local changes in the width that the zonal mean metric cannot capture. The CMIP5 models produce very similar correlation patterns between tropopause break locations and precipitation anomalies in the historical period (not shown).
In this study we investigated trends in the width of the tropics using an objective tropopause break metric applied to 6-hourly instantaneous reanalysis and CMIP5 model output (from six models) for the past and future. This objective tropopause break metric distinguishes the boundary between the tropics and extratropics at upper levels and is a good measure of the extent of the dry subtropics. An additional benefit of this method is that it produces a zonally varying measure of the extent of the tropics, which is a significant advantage over previous methods. The key results from this study are summarized below.
Modern reanalyses show consistent longitudinal variability in the climatological tropopause break location in each hemisphere, with strong agreement in latitude in most locations except for the NH eastern Pacific and Atlantic Ocean basins.
Historical simulations from six CMIP5 models show more spread in the tropopause break location relative to reanalyses, but consistent longitudinal variations.
Reanalyses show significant contraction of the tropics over the eastern Pacific between 1981 and 2015, and slight but significant expansion in other regions.
For RCP8.5 projections, five out of six CMIP5 models show significant contraction of the tropics over the NH east Pacific and the Americas (especially North America) by 2100.
The tropical width as defined using the tropopause break metric covaries with precipitation anomalies most significantly in the Maritime Continent and the Sahel and produces more robust signals than the individual break edges (e.g., Northern and Southern Hemispheres).
Applying the objective metric to tropopause altitudes computed from monthly mean temperature fields or using a frequency based identification metric led to significant differences in the results for historical and future periods respectively.
Further analysis of seasonal (see the appendix), interannual, and regional variations in the tropical width, as well as regional connections with precipitation and changing chemical composition of the upper troposphere and lower stratosphere are necessary to further understand the implications and mechanisms causing the observed and simulated trends in upper-level tropical width. The objective tropopause break metric requires model-level instantaneous data and consequently the number of available models is currently limited. With the utility of this method we hope that more groups will make such model output available to the community to thoroughly assess the range of future possibilities. This new method and associated results has implications for regional ecosystem variations to changes in the extent of the tropics as well as upper troposphere–lower stratosphere composition via chemical exchange across the tropopause.
The authors thank Paul Staten, editor Darren Waugh, and two anonymous reviewers for feedback that improved this manuscript. This research was supported by a grant from the Research Council of the University of Oklahoma Norman campus. We acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP, and the climate modeling groups for producing and making model output available. 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.
Data availability statement. CMIP5 data were obtained through the Earth System Grid Federation (ESGF) data portals. MERRA-2 data (https://doi.org/10.5067/IUUF4WB9FT4W) are available at MDISC, managed by the NASA Goddard Earth Sciences (GES) Data and Information Services Center (DISC). The remainder of the data were accessed via the NCAR Research Data Archive (RDA), including ERA-Interim (https://doi.org/10.5065/D6CR5RD9), JRA-55 (https://doi.org/10.5065/D6HH6H41), CFSR (https://doi.org/10.5065/D69K487J), CFSv2 (https://doi.org/10.5065/D61C1TXF), and GPCP (https://doi.org/10.5065/D6SN07QX). Tropopause break locations (latitudes) for all reanalysis and model datasets are available by request from the authors.
The width of the tropics varies substantially between seasons, which is not reflected in the annual mean analysis contained in the main section of this manuscript. Figures A1–A4 show the climatological location and historical trend in the tropopause break location broken down by 3-month seasons (e.g., DJF, MAM, JJA, SON) using the objective method with reanalysis data. While overall, the seasonal trends show many similarities with the annual mean (Fig. 2), there are some important differences. In the NH, the narrowing trend in the Pacific is evident in each season, but the longitudinal location of the maximum narrowing shifts. For example, in the NH summer (Fig. A3), the narrowing is farther westward and closer to the central Pacific in comparison to the NH winter (Fig. A1). In the SH, trends are smallest in the winter (JJA; Fig. A3) and largest in the summer (DJF; Fig. A1), with maximum narrowing exceeding 2° decade−1 in the eastern Pacific.