Poleward Shift of Northern Subtropics in Winter: Time of Emergence of Zonal Versus Regional Signals

The Northern Hemisphere mid‐latitudes will be exposed to hydroclimatic risk in next coming decades because the subtropical expansion. However, it is not clear when the anthropogenic signal will emerge from the internal climate variability. For this purpose, we investigate the time of emergence (ToE) of the hemispheric and regional shift of northern subtropical margins in the Max Planck Institute Grand Ensemble. For several indicators, the ToE of the poleward shift of Northern subtropical margin will not occur by the end of the 21st century, neither at regional nor at hemispheric scale. The exceptions are the Mediterranean/Middle East and, to a lesser degree, Western Pacific, where the ToE would occur earlier. According to our results, given the fundamental role played by internal variability, trends of Northern Hemisphere subtropical poleward shift that have been identified over last decades in reanalyses cannot be considered as robust signals of anthropogenic climate change.


Introduction
Almost the 40% of Earth's human population lives in subtropical areas that are prone to aridity or desertification in next coming decades. Hence, investigating the magnitude and the rate of Northern subtropical margin shift is fundamentally important for water availability and environmental and societal changes (Doxsey-Whitfield et al., 2015;Heffernan, 2016). Subtropical climate is characterized by lack of clouds, rainfall scarcity, transition from positive-to-negative surface moisture budget, subsidence, and high surface pressure, characteristics associated with the descending branch of the Hadley Circulation (HC). Several studies have leveraged on this link for developing indicators to investigate changes of HC width and poleward shift of its subtropical margins Nguyen et al., 2018;Schwendike et al., 2014;Staten et al., 2018).
Beyond that, the influence of natural/internal climate variability has been highlighted to be a source of discrepancy on zonal-mean HC trends in last decades Grise et al., 2018Grise et al., , 2019Kang et al., 2013;Staten et al., 2018). In fact, when accounting for different indicators to estimate HC edges, CMIP5 simulations show that most of recent trends in reanalyses do not exceed the range of natural variability in global climate models, because the length of the observational record over the satellite era (30-40 years) is simply too short to detect any anthropogenically forced signal Staten et al., 2018). In particular in the Northern Hemisphere, CMIP5 simulations indicate that the anthropogenic greenhouse gas (GHG) increase contributed to a very small poleward shift of zonal-mean HC edges, and such a small forced signal is only detectable using a large ensemble of model simulations. At regional scale, the forced responses to individual forcing (e.g., GHG increase, stratospheric ozone depletion, and anthropogenic aerosols) is difficult to quantify because the natural variability has a large impact , in particular because the GHG-induced HC poleward shift might be counteracted by the presence of aerosols. These aspects leave room for discussion about where the anthropogenic forcing, especially the CO 2 , will affect more the subtropical shift and HC expansion.
Additionally, it is unclear when the forced poleward shift of the HC will possibly emerge from the internal climate variability. A recent study on the Community Earth System Model Large Ensemble (CESM-LE), a global climate model composed of 40-ensemble members for the period 1920-2100 forced with Historical and RCP8.5 forcing, shows that two indicators used to identify HC edges (i.e., the zonal-mean mass streamfunction and the surface wind changes) indicate that the forced signal will not emerge from the noise during the 21st century in the Northern Hemisphere . However, as the time of emergence (ToE, i.e., the year when the signal emerges from the noise) is intrinsically related to some thresholds, results might vary among indicators, data sets, and regions.
Here, we quantify the relative role of the internal climate variability against the external forcing in the Max Planck Institute-Grand Ensemble (MPI-GE) (Bittner et al., 2016;Maher et al., 2019;Stevens, 2015) on the shift of Northern subtropical margin. We extend our investigation beyond the study by Grise et al. (2019), analyzing the ToE at hemispheric (zonal average) and at regional scale using six indicators for detecting subtropical margins, in order to identify the areas most sensitive to the anthropogenic forcing. The regional approach is an aspect never investigated before. Furthermore, we have considered the 1%CO 2 experiment, for an improved characterization of the ToE related to CO 2 -only forcing. We introduce the MPI-GE and the indicators used for the analysis in section 2, then we first analyze the latitudes, trends, and ToE of zonally averaged indicators (section 3.1), and afterward the regional characteristics of subtropical margins (section 3.2). We summarize and conclude the paper (section 4).

Data and Methods
The MPI-GE is currently the largest ensemble available among the climate models (Bittner et al., 2016;Giorgetta et al., 2013;Maher et al., 2019;Stevens, 2015). It consists of the ocean component, MPIOM , the ocean biogeochemistry model HAMOCC 5.2 (Ilyina et al., 2013), the atmospheric model ECHAM (Stevens et al., 2013) coupled to the land component, JSBACH, which includes land use transition with the standard fire module, and interactive vegetation . The atmospheric model resolution is 1.875°(T63L47) and the ocean model is on GR15L40 grid, featuring a horizontal resolution between 22 and 220 km and 40 vertical levels. For in-depth details refer to Maher et al. (2019).
In this study, we have investigated the evolution of subtropical margins in the Historical simulation  and in two other climate experiments, the RCP8.5 until the end of the 21st century and the 1%CO 2 (150 years). We use these three experiments to explore the sensitivity of our results to different forcing. The comparison between RCP8.5 and 1%CO 2 ensures that the evolution of the subtropical margins cannot be attributed to aerosols, other GHG gases (such as ozone) or land use changes in the subtropical regions.
We focus the analysis on boreal winter (December-to-February, DJF), when hydroclimate conditions of some densely populated areas (e.g., Mediterranean and California) at the Northern side of subtropics are particularly sensitive to storm track position (Mariotti et al., 2002;Seager, Neelin, et al., 2014), to HC expansion and/or poleward shift of subtropical margins (Seager et al., 2010) and hence more impacted by water scarcity. Furthermore, the Northern HC is strongest and well defined in DJF; therefore, it is possible to detect easily the regional position of its edges.
The latitude of the subtropical margins is computed by using six different indicators. Some of them are related to precipitation, which has relevant implications for society, and some of them have been already used in past studies as indicators of the position of the HC edges: • OLR (outgoing longwave radiation): poleward-most latitudes where OLR falls below 250 W/m 2 Chen & Bordoni, 2014;Davis & Rosenlof, 2012;Hu & Fu, 2007;Johanson & Fu, 2009). It is associated with the absence of clouds in the subtropics. • TP (total precipitation): latitudes of absolute precipitation minima in the subtropics (Chen & Bordoni, 2014;Hu & Zhou, 2010). It describes rainfall scarcity. • P-E (precipitation-evaporation): zero-crossing latitudes of P-E Davis & Birner, 2016;Davis & Rosenlof, 2012;Johanson & Fu, 2009;Lu et al., 2007;Staten et al., 2018), associated with the dry conditions and negative surface water balance in the subtropics. • MSLP (mean sea level pressure): latitudes of absolute maxima of the MSLP (i.e., high-pressure ridges in the subtropics) Hu & Zhou, 2010). • ω ϕ (vertical velocity associated with the meridional overturning circulation or, in other words, the vertical velocity associated with the meridional component of the divergent wind): latitudes of absolute maxima near the subsiding branches at 500 hPa (Schwendike et al., 2014(Schwendike et al., , 2015. It describes the subsidence in the subtropics. • ψ ϕ (streamfunction of the regional meridional overturning circulation): zero-crossing latitudes of ψ ϕ at 500 hPa (Nguyen et al., 2018;Schwendike et al., 2014), associated with the poleward limit of the HC in the subtropics.
Regional subtropical margins are computed for all indicators and at each longitude within the 5°-45°N range, as shown in Figure 1.
Recently, it has been emphasised that the regionality of the tropical atmospheric circulation depends also on the zonal component of the mean flow in some areas. This implies that the edge definition might be affected by the zonal flow too, as in the Pacific around 170°W (Raiter et al., 2020). The use of the indicators based on precipitation, surface pressure, and OLR and the degree of agreement they have with ω ϕ and ψ ϕ serves to avoid any ambiguity on results by using different definition of subtropical margins. In particular, we have noticed that most of the discrepancy between different indicators occurs over the Pacific and Atlantic ocean: These might be the areas where the mean atmospheric flow has the largest zonal component in DJF ( Figure 1). However, since we are more interested in the societal implication of subtropical margin shift on regional hydroclimate, we will analyze in detail only results over the continents.
The impact of the external forcing is described by the so-called forced response

10.1029/2020GL089325
Geophysical Research Letters evolution of the forced response (described by the ensemble mean), while IV is the variance associated to the overall variability of the ensemble members relative to the ensemble mean: Here, σ TV , σ FV , and σ IV denote the square root of TV, FV, IV, respectively, and represent the total uncertainty of the signal and its components (see the supporting information for the algebraic expressions).
The methods for estimating the signal-to-noise ratio S/N, which is needed for computing the ToE, differ among studies (Christensen et al., 2007;Deser et al., 2016;Giorgi & Bi, 2009;Hawkins & Sutton, 2012).
Here, the signal S j at time t j is given by the anomaly of the 31-year running mean ð · Þ j , centered at time t j , of the ensemble mean position F i f r of the subtropical margin with respect to its value in the reference period The noise N j is estimated as the standard deviation associated with the internal climate variability σ IV at each time step (t), considering 31-year-long time windows, centered at t j . The value of the ToE is computed as the year when S/N crosses the 0.5 or 1 threshold. In aforementioned studies Staten et al., 2019), the noise has been estimated using the internal climate variability in the reference period. In our case, we considered its possible variation with time. However, we found that this variation turns out to be negligible. We have estimated the S/N ratio and the corresponding ToE for each ensemble member. In this case, the signal is computed as the anomaly of the 31-year running mean with respect to the 1971-2000 mean and the noise N as the standard deviation in the reference period, 1971-2000. The best estimate of the ToE is the median of the members' ToEs. The 5th to 95th percentiles of the distribution are used to delimit its uncertainty range.
At regional scale, a slightly different method is applied. The regional signal is linearly approximated using its trend (computed using the whole duration of the experiment): S(λ, t) ≈ ∂F fr (λ, t)/∂t| Exp (t − t 0 ). The index Exp indicates the experiment (in turn Historical, 1%CO 2 , RCP8.5) that is used for computing the trend and t 0 the corresponding initial time. The noise σ IV (λ) | Exp is the standard deviation of the regional IV, which is a function of the longitude. The signal-to-noise ratio is then approximated as follows: This allows to estimate the time t needed to reach an established signal-to-noise ratio threshold, which is the ToE in terms of years after the initial date of each experiment.

Zonally Averaged Northern Subtropical Margins: Uncertainty in the Latitude, Trends, and ToE
The latitude of the Northern subtropical margins strongly depends on the adopted indicator ( Figure 1), with P-E and TP delimiting the most poleward and equatorward margins, respectively. For all indicators and in every experiment, most of the uncertainty in the latitude of the zonally averaged subtropical margins is due to the internal climate variability (ICV; Figure 2, left panels, diamonds). The unique exception is OLR, which exhibits a large sensitivity to the 1%CO 2 forcing (FV, stars), with the forced response accounting for 30% of the total variance (TV, dots). In fact, for all other indicators, the contribution of the forced response to the total variance is small or even negligible, ranging from a minimum 1% for P-E and MSLP in the Historical to a maximum 13% for MSLP in the 1%CO 2 . In spite of the large ensemble spread, for all indicators the ensemble mean of Northern subtropical margins exhibit a consistent poleward shift, which progressively increases with time in the 1%CO 2 and the RCP8.5 experiments (see Figure S1).
The magnitude of the trends of the ensemble mean (forced trends; Figure 2, right panels) depends on the indicator too, varying in the range from 0.04°per decade for TP in the 1%CO 2 experiment to 0.12°per decade for MSLP in the RCP8.5. Conversely, trends are small and uncertain in the Historical. Though the 1%CO 2 and RCP8.5 forced trends are always positive and statistical significant, individual members exhibit much uncertain trends (Figure 2, right panels) are sufficiently large to be statistically significant, while RCP8.5 trends are often not significant for several indicators (namely, TP, ω ϕ , and ψ ϕ ). The percentages annotated in the right panels of Figure 2 can be interpreted as probability of detecting a significant trend and are larger than 90% for most indicators only in the 1%CO 2 .
Summarizing, the poleward shift of Northern subtropics appears to be robust and significant across the different indicators, and it will increase as the climate warms. However, the forced shift emerges from the internal variability only if the signal is sufficiently large. The early signals in the Historical are too small for allowing their detection and attribution to the anthropogenic forcing. In the RCP8.5 signals are larger than in the Historical, but in any case for TP, ω ϕ and ψ ϕ the probability of detecting a significant trend is lower than 50%. Therefore, the probability of detecting a significant trend of the subtropical margins during the 21st century is low.
The comparatively large internal variability has important consequences on the ToE of the Northern subtropical margin's shift. Figure 3 shows the evolution of S/N (signal-to-noise ratio) for all indicators considering the time series built by joining the Historical and RCP8.5. If the threshold S/N = 1 is adopted for the ToE, only OLR provides a ToE within the 21st century (around 2050). If a much lower threshold S/N = 0.5 is considered, ToE is anticipated to 2011 for OLR and shifts to 2050, 2060, 2062, and 2084 for ω ϕ , P-E, MSLP, and TP, respectively, while for ψ ϕ it does not occur within the considered time range. Results are substantially equivalent if ToE is estimated separately for each member. The median values with the corresponding 5th to 95th uncertainty bars are shown in Figure 3b. Here we considered the threshold S/N = 1. Only OLR and P-E estimate the occurrence of ToE within this century at the 95th confidence level. Note that the adopted methodology actually prevents to detect ToE occurring beyond 2085.
Therefore, though the anthropogenic climate change will certainly cause the poleward shift of the Northern subtropical margin, it is unlikely that the signal will emerge from the natural variability within the 21st century (more precisely before the last decade of the 21st century) at hemispheric scale.

Regional Subtropical Margins: Uncertainty in the Latitudes, Trends, and ToE
The latitudes of regional subtropical margins in the Northern Hemisphere (Figure 1) are far from being zonally uniform. Regional features, which are usually hidden by taking the zonal average, exhibit large differences among indicators.
The OLR shows the smallest zonal variations, with northward deviations in areas free of clouds, such as the Mediterranean and the Western Pacific, and southward deviations in correspondence with cloudy areas above Indochina, North America and Eastern Pacific. The subtropical margins detected by ω ϕ and ψ ϕ , which are by definition two indicators directly related to the atmospheric circulation, have rather similar positions above the oceans and flat areas, but they differ in the presence of strong orography (e.g., Himalaya, Tibetan Plateau, Rocky Mountains, and Central America). The MSLP systematically deviates northward above continents because of the presence of the Siberian High and the high-pressure systems above North America in winter. There is a clear systematic deviation between the regional and zonal averaged positions of the precipitation minima (TP) and ω ϕ maxima. The former are always southward of the latter because lack of moisture in the atmosphere southward of the ω ϕ maxima is likely to be a key factor determining precipitation minima. The P-E locates the margins systematically north of the precipitation minima (TP) because it includes the effect of evaporation. This effect is large above oceans, particularly over their warm pools. The effect of the Mediterranean Sea is evident as well. These results stress the importance of different regional features on the positions of the subtropical margins provided by different indicators.
Also regional trends are not zonally uniform and differ among indicators (Figure 4). Differences are particularly large above Indochina and North America, where some indicators show equatorward trends instead. Considering a single indicator, its trend has a similar regional behavior in the three experiments, but its magnitude is largest in the RCP8.5 and smallest in the Historical. In spite of the spatially irregular behavior of the trends, some regions provide trends with a consistent positive sign with a similar magnitude: Mediterranean/Middle East and Western Pacific.  grey horizontal lines. For values of the trends above the 0.1, the signal will exceed the noise in one century. If a weaker criterion S/N = 0.5 is assumed, the trend the detection would occur in five decades. The threshold 0.25 denotes the level required for the signal exceeding the noise in four decades. In the Historical, trends are extremely small and their rate of change is everywhere far from the level that would allow its detection within the 21st century. However, the past rate of change is expected to accelerate in the future: RCP8.5 and 1%CO 2 present in fact larger values than the Historical. RCP8.5 and 1%CO 2 shows that the ToE of the poleward shift of Northern subtropics is not regionally uniform as well. For many indicators, the ToE will occur sooner over Mediterranean/Middle East and Western Pacific (see grey areas in Figure 5), while the large effect of internal climate variability prevents the ToE within the 21st century above the Eastern Pacific.
Regional processes, such as changes of available moisture provided by atmospheric advection, can explain the differences in trends among the indicators shown in Figure 4. The OLR presents large northward trends of subtropical margins over North America and the Mediterranean, where climate projections suggest increased aridity and cloudless conditions as consequence of reduced moisture mean transport (D'Agostino & Lionello, 2020;Seager et al., 2007). In particular, over North America, the northward trend of OLR disagrees with the southward trends provided by ψ ϕ and ω ϕ (direct indicators of the position of the HC descending branch) and by TP (the latitude of the precipitation minima). TP exhibits also a large southward trend of the subtropics over Eastern Asia, consistently with a future increased precipitation over China (Tian et al., 2015). The TP behavior is possibly explained by an increased northerly moisture transport (Ma et al., 2012), caused by the southward shift of the subtropical highs over the Chinese sea, a feature also shown by the MSLP indicator. Furthermore, the OLR presents a southward trend over the Eastern Pacific, where the low-stratus clouds are projected to decrease over the cool southward branch of the subtropical gyre (Su et al., 2014), while other indicators do not show any trend in this specific area. The disagreements discussed above shows the relevance of regional processes when comparing different indicators. Lack of consensus prevents from getting a robust conclusion for several regions and is the consequence of the complicated conditions in the subtropics, which cannot be attributed uniquely to the descending branch of the HC. The agreement among indicators over Mediterranean/Middle East suggests that these regions will be future hot spots, with a more evident poleward shift of the subtropics than in the rest of the Northern Hemisphere.

Summary and Conclusion
In the past decades several studies have investigated the poleward shift of the subtropical belt. These studies were mostly focused on the HC expansion and on the suitability of different indicators to detect its shift. Disagreements among studies have been attributed to differences among indicators, data sets, and, recently, to the role of internal climate variability.
In this study we have focused on the distinction between zonal average and regional poleward shift of Northern Hemisphere subtropical margins, using indicators associated with the characteristics of the subtropical belt, such as lack of clouds (OLR), rainfall scarcity (TP), transition from positive-to-negative surface moisture budget (P-E), subsidence, and high surface pressure (ω ϕ , MSLP). We have further included, as a reference, the meridional mass stream function (ψ ϕ ), computed by meridional divergent winds, which is the most common metric used for studies of the HC extent. The role of internal variability versus the forced response on regional subtropical shifts has been investigated in the MPI-GE. By design, the MPI-GE allows to disentangle the role of internal climate variability and forced response in the full phase space of both the ocean and the atmosphere states. Furthermore, in order to detect the sensitivity of our results to different concentration of GHG, we have used three different experiments from the MPI-GE: the Historical, the RCP8.5 and 1%CO 2 . We have considered the winter season when the subtropical shift strongly affects the water cycle of the Northern Hemisphere arid and semiarid areas, for which is extremely important estimating the regional ToE for triggering adaptation measures.
For all indicators and in all experiments, the uncertainty on the latitude of subtropical margins is predominately affected by the internal climate variability. If OLR is excluded, in the Historical experiment the forced response has a negligible role on the total variance of the subtropical margins (never exceeding 2%). The contribution of the forced response remains small also in the RCP8.5 (2006RCP8.5 ( -2099 and in the presence of nonlinear increment of CO 2 (1%CO 2 simulation). Similarly, uncertainty on trends are caused by large internal variability, with the effect of forced response never exceeding 15% of the total trend variance.
Different indicators produce different future rates of Northern subtropics' shift at hemispheric and regional scale. The disagreement among the regions where regional shifts are detected suggests that these indicators reflect changes produced by the anthropogenic climate change on regional processes. The forced response is more important for OLR than for all other indicators. OLR trends depend directly on thermodynamics, atmospheric moisture, and cloud distributions, and this explains its large sensitivity to anthropogenic climate change and its differences with respect to other indicators Waugh et al., 2018). Regional and local processes affect also the indicators based on precipitation, evaporation, MSLP, and vertical velocity fields, but their effect on OLR is so strong that its behavior should be cautiously considered to be representative of the shift of subtropical margins and of the HC expansion (Waugh et al., 2018).
Some previous studies attributed past observed poleward shift signals of the Northern subtropics to global warming Liu et al., 2012;Lu et al., 2007;Nguyen et al., 2013). However, due to the large influence of the internal variability, the forced response of the Northern subtropical margin shift in winter will not emerge at hemispheric scale before the end of the 21st, both in MPI-GE (as shown here) and in CESM-LE . Therefore, trends associated with a poleward shift of Northern subtropics observed in reanalyses over last decades, especially those based on the meridional mass stream function, cannot be considered a robust signal of anthropogenic forcing. On one hand, our results confirm that anthropogenic climate change will produce a poleward shift of the Northern Hemisphere subtropical margins but,

Geophysical Research Letters
on the other hand, suggest that a robust detection of an anthropogenically induced signal at hemispheric scale is unlikely within the 21st century. The latter conclusion is in contrast with the wide literature on reanalysis data and CMIP3 and CMIP5 models, as already pointed out in the analysis of the CESM-LE ensemble .
Disagreement among studies that have investigated past trends in observational data sets and reanalyses can be explained by their attempt to assess a signal that is too small to be detectable. Their results reflect mainly multidecadal fluctuations, as pointed out by  and not a sustained long-term climate trend. However, at regional scale, there are some areas particularly exposed to the subtropical northward shift: Mediterranean/Middle East and Western Pacific. A combination of factors, such as lack of precipitation, negative surface water budget, persisting subsidence and anticyclones characterize these areas as an hot spot of future climate change. In such areas, all indicators agree on a significant northward shift of subtropics, producing a signal that will be detectable within the end of the 21st century.

Data Availability Statement
The MPI-GE data that have been used in this study are available at esgf-data.dkrz.de/projects/mpi-ge/ and can be freely downloaded after having signed the data-use license.