Moderate Indian Ocean Dipole dominates spring fire weather conditions in southern Australia

Patterns of the Indian Ocean Dipole (IOD) exhibit strong diversity, ranging from being dominated by the western tropical Indian ocean (WTIO) to by the eastern tropical Indian ocean. How the different types of the IOD variability patterns affect Australian fires differently is unknown, nor is it certain how the impacts may change under greenhouse warming. Here, we find that the moderate IOD, dominated by WTIO sea surface temperature (SST) variability, plays a primary role in affecting southern Australian fire weather conditions during austral spring. During a positive moderate IOD, broad-scaled warm SST anomalies in WTIO force an atmospheric stationary Rossby wave with a high-pressure anomaly over southern Australia. This elevated pressure and associated anomalous atmospheric conditions provide suitable fire weather with hot, dry, and windy conditions, raising fire risks in southern Australia. Such impact is distinctively different from that strong IOD-induced. As predicted by climate models, decreased moderate IOD variability in the future will result in weakened Australian fire weather responses.


Introduction
Australian fires are well-known for their frequency, destruction, and blow to the economy, such as the 1983 'Ash Wednesday' , the 2009 'Black Saturday' , and the 2019/2020 'Black Summer' bushfires (Cai et al 2009b, Borchers Arriagada et al 2020, Filkov et al 2020, Wang and Cai 2020, Ward et al 2020).The occurrence of a wildfire is linked to four critical factors that ignition (lightning or human activities), fuel abundance and continuity, fuel dryness, and suitable fire weather conditions with hot, dry, and windy (Harris and Lucas 2019).In Australia, rich dry temperate forests (mostly eucalypts) act as sufficient fuels, and then fires mainly occur depending on fire weather conditions and ignitions.The main climate variabilities, including El Niño-Southern Oscillation (ENSO), Indian Ocean Dipole (IOD), and the Southern Annular Mode (SAM), could influence fire weather in different parts of the Australian continent and at various periods of the year (Harris and Lucas 2019, Abram et al 2021), resulting in fire crisis and even extreme events.
For tropical climate modes, ENSO, as the most significant one, dominates interannual climate variability and has established a strong correlation with Australian fire weather (e.g.Harris et al 2014, Liu et al 2023).ENSO is typically connected to hot and dry conditions over northeastern Australia (Harris et al 2014, Dowdy 2017, Liu et al 2023), raising dangerous fire risks in austral spring and summer.During ENSO events, particularly central Pacific (CP) El Niño events, reinforced atmospheric convection over tropical Pacific warm sea surface temperature (SST) anomalies (SSTAs) promote subsidence over eastern Australia, suppressing rainfall and enhancing drought conditions (Allan et al 2019).However, the ENSO does not have a direct dynamical pathway to influence southern Australia; instead, the IOD is associated with it (Cai et al 2011, Wang andCai 2020) and could potentially modulate fire weather conditions there.The Dipole Mode Index (DMI) is commonly used to characterize and assess IOD as well as its impacts (Harris and Lucas 2019, Wang and Cai 2020, Abram et al 2021), which is represented as the difference between SSTAs in the western (10 • S-10 • N, 50 • E-70 • E) and eastern (10 • S-0 • , 90 • E-110 • E) tropical Indian Ocean (Saji et al 1999).However, the definition of DMI implicitly assumes no diversity in the IOD anomaly pattern.Therefore, whether the IOD impact differs with IOD anomaly patterns is unknown.
In reality, positive IOD (pIOD) events display different flavors, ranging from being dominated by the western tropical Indian Ocean (WTIO) to by the eastern tropical Indian Ocean (ETIO).One such manifestation shows two pIOD regimes, that are depicted by two separate indices defined as the moderate pIOD dominated by western warm SSTAs and strong pIOD with strong eastern cool SSTAs, respectively (Cai et al 2021).During a moderate pIOD, the dominant forcing of western prevailing warm SSTAs is the Ekman pumping induced by Rossby waves propagating westward.Anomalous easterly induced oceanic downwelling Rossby waves and their reflection at the western boundary downwelling Kelvin waves arise vertical Ekman pumping, warming the WTIO (Du et al 2009, Cai et al 2021, Yang et al 2021a).For a strong pIOD, however, preceding CP El Niño provides a significant necessary but not sufficient condition for the occurrence (Yang et al 2021a).A CP El Niño during the austral summer and autumn favors easterly wind anomalies over the equatorial eastern Indian Ocean, shallowing thermocline here and driving upwelling near Sumatra-Java coast, which forms positive Bjerknes feedback along with the initial cold SSTAs and then contributes to the development of a strong pIOD.
The large difference in IOD dynamics performs differences in relative strength both of SSTAs and tropospheric convective anomalies in the western and eastern centers.Thus, IOD events dominated by the WTIO and ETIO could induce vastly different atmospheric teleconnection patterns, possibly with different climate impacts.In this study, we identify the dominant role of moderate IOD on southern Australian fire weather conditions, and explore the associated atmospheric teleconnections utilizing reanalysis datasets.To further substantiate our findings, we use pacemaker experiments and composite analysis to provide robust validation of the observed impacts.Since greenhouse warming modifies climate variabilities, we further examine how the impacts will change in the future.

Reanalysis data and climate variability index
Monthly mean reanalysis data we used in this study come from the fifth generation European Center for Medium Range Weather Forecasts reanalysis (ERA5) with a 0.25 • horizontal resolution and 37 vertical levels in the atmosphere components (Hersbach et al 2020).We regrid the above datasets to a common 1 • × 1 • latitude-longitude grid before diagnosing.Anomalies are constructed with reference to the mean of the period 1979-2022, and then detrended.Considering that ENSO and IOD both participate in modulating Australian climate variability (Harris and Lucas 2019), we employ partial regression (partial correlation) to explore the 'pure' IOD impacts on (connections with) Australian fire weather conditions (Yang et al 2021b), and the significance is obtained by the t-test.ENSO signal is represented by the Niño3.4index that spatially averaged SSTAs in the region 170 • W-120 • W, 5 • S-5 • N and temporally 5 month running mean (Rayner et al 2003), which is downloaded at https://psl.noaa.gov/gcos_wgsp/Timeseries/Nino34.The DMI we used is downloaded at https://psl.noaa.gov/gcos_wgsp/Timeseries/DMI/.

Fire proxy data
Monthly fire activity proxies, including the fire weather index (FWI), Fire Danger Index (FDI), fire carbon emissions, and burned area, are employed here.The FWI is a dimensionless numerical rating of the potential frontal fire intensity, and is commonly used as a reliable indicator of fire risk (Dowdy et al 2010, Liu et al 2023).It is derived from two fire behavior indices: the initial spread index and build-up index, which represent the rate of forward fire spread and the fuel available, respectively.The initial spread index is derived from the wind speed and fine fuel moisture code, quantifying the speed at which a fire can spread in a given weather condition.The buildup index is derived from the duff moisture code and drought code, measuring the total amount of fuel available for combustion by a moving flame front (Simpson et al 2014).This combination of these two factors assesses the potential fire risk and intensity comprehensively.The FWI is calculated based on four daily atmospheric variables, including 2 m air temperature, 10 m wind, relative humidity and total precipitation (Vitolo et al 2020).In addition, the FDI is a widely used tool in Australia for forecasting fire danger and describing the influence of surface meteorological conditions as well as landscape dryness on wildfires (Dowdy 2017).It is also a dimensionless index.Both daily FWI and FDI can be downloaded online and converted into monthly mean datasets for the period 1979-2022 (10.24381/cds.0 × 10 89 c522).
The burned area is one of the most crucial fire characteristics, influencing carbon cycle and global climate through emissions of greenhouse gas and particulate matter (Chen et al 2011).Here, we used the burned area data in 2001-2021 from Collection 6 Moderate Resolution Imaging Spectroradiometer (MODIS C6) at a 0.25 • × 0.25 • resolution (Giglio et al 2013; https://modis-fire.umd.edu/ba.html).We also used the fire carbon emissions data come from the Global Fire Emissions Database version 4 that includes the contribution of small fires (GFED4s) with a resolution of 0.25 • × 0.25 • for the period 1997-2022 (van der Werf et al 2017; www.geo.vu.nl/ ∼gwerf/GFED/GFED4/).All these data are removed linear trends before analysis to focus on the interannual variability.

Rossby wave source (RWS) and wave activity flux
The stationary Rossby wave dynamics suggests that anomalous tropical convective and associated diabatic heating induce divergence anomalies in the upper troposphere, exciting RWS and stationary Rossby wave trains, which teleconnects tropical forcing to extratropical regions (Hoskins and Karoly 1981).Following the methodology of Sardeshmukh and Hoskins (1988), the RWS can be derived from the barotropic vorticity equation at 200 hPa: Additionally, the wave activity flux (WAF) is used to investigate the Rossby wave propagation.Zonal and meridional components of the WAF on the pressure coordinates are calculated using where U = (u, v) is the basic flow, with u and v representing zonal (eastward) and meridional (northward) wind velocity, respectively; ψ is the quasi-geostrophic stream function.Perturbations are denoted by primes and the subscripts x and y are derivative in the zonal and meridional direction, respectively (Takaya and Nakamura 1997).

CMIP6 models and Bootstrap test
To gauge how the impacts on Australian fire weather conditions will change under greenhouse warming, we take monthly SST, 200 hPa geopotential height (Z200), surface air temperature (SAT), and precipitation flux outputs from 45 CMIP6 models (Eyring et al 2016) over the 1900-2099 period (table S1).These models are forced under historical anthropogenic and natural forcings until 2014, and thereafter under future greenhouse gases forcing of the Shared Socioeconomic Pathways (SSP) 5-8.5 (SSP5-8.5)emission scenario.Monthly anomalies of all variables are obtained with reference to the monthly climatology of 1900-1999 and quadratically detrended (Geng et al 2023).
We use the Bootstrap test (Austin and Tu 2004) to assess whether the multimodel mean result is statistically significant.Specifically, the given 45 models are resampled randomly to construct 10 000 realizations of a multimodel mean value for the 1900-1999 and 2000-2099 periods; in this process, any model can be selected again.We calculate the standard deviation (s.d.) of the 10 000 realizations of the mean value for the two periods.If the multimodel mean value difference between these two periods is greater than the sum of the two s.d.values, then the difference is considered statistically significant above the 95% confidence level.

The dominant role of moderate IOD influences southern Australian fire weather
Given that an IOD usually peaks in austral spring with a seasonal dependence of its atmospheric teleconnections, we choose September-October-November (SON) period for analysis.The time series of Australian FWI is obtained by spatially-averaged FWI in the region <110 • E-155 • E, 45 • S-10 • S> where ocean masked.The peak of Australian FWI also occurs in austral spring (red bar in figure S1(a)), with an enhanced variance on interannual timescales (figure 1(a)) and a peak at 5-6 yr (significant at the 95% confidence level) (figure S1(b)).The Australian FWI is strongly correlated with tropical SST variabilities, including ENSO and IOD (r = 0.55 and r = 0.54, respectively, p < 0.001; figure S2).Considering that IOD events tend to co-occur with ENSO events, the result presented here cannot distinctly attribute the impacts to IOD alone.More importantly, while ENSO is the major forcing of Australian fire weather, the primary climate drivers may vary regionally.The isolated results without ENSO show that the DMI influence is much weaker and confined to the southeastern edge of Australia (figure 1(b)).The decrese of the correlations in northeastern Australia is due to the removal of the ENSO-related fire weather there (figure 1(c)).
Using the DMI determined from the observed poles assumes that the IOD possesses little pattern diversity, which cannot identify the role of each individual pole and the associated impacts., respectively (figure 1(g)).A higher value of α indicates a more distinct separation between these two types of the IOD.The WTIO and ETIO SST time series show strong correlations with the moderate IOD (M-index) and strong IOD (Sindex), at 0.78 and 0.90, respectively.The grid-point correlation coefficients obtained using the M-and S-indices (figures 1(h) and (i)) largely resemble the WTIO-and ETIO-FWI relationships but with more pronounced differences, which better reflects the substantial correlations between southern Australian fire weather and WTIO-dominated moderate IOD.
To ensure that Australian fire activities are not biased by using the FWI-only, we examine other fire products, including the FDI, burned area, and fire carbon emissions (Dong et S1(a)).For the relationships with the two types of the IOD during SON, the FDI-related results are almost identical to that in the FWI, with significant and high correlations between WTIO-dominated moderate IOD and the FDI covering southeastern Australia (figure S4).Besides, both the burned area and fire carbon emissions are more correlated with moderate IOD than strong IOD in central Australia (figures S5 and S6).Thus, the multiple fire products support and confirm the robustness of the finding that WTIO-dominated moderate IOD plays a primary role in southern Australian fire weather.

Mechanism whereby moderate IOD SST influence on Australian FWI
To establish a climate fingerprint of moderate IOD dominated by WTIO on Australian FWI, we utilize reanalysis datasets and pacemaker experiments to diagnose atmospheric teleconnections and associated fire weather conditions.According to the Rossby wave theory (e.We further examine the WAF anomalies, and it shows that a wave train propagating eastward and poleward starts from this efficient RWS (vectors in figure 2(b)), featured by a center of high-pressure anomaly over southern Australia (shading in figure 2(b)).Careful observation of the WAF vectors shows some evidence of flux convergence to southern Australia because of the wave reflecting surface in the south of Australia (figure S8, white regions along about 40 • S).Near the mean jet edge there, the meridional gradient of the absolute vorticity is enhanced and acts as a waveguide to trap and reflect the stationary wave train (Li et al 2015a(Li et al , 2015b)).
Compared with moderate IOD cases (figure 2, left panel), the teleconnection associated with ETIOdominated strong IOD (figure 2, right panel) shows a different Rossby wave pattern extending from the subtropical Indian Ocean (30 • S, 70 • E) to higher latitudes of the Southern Hemisphere.Differences in the relative strength of convective anomalies in the two SSTA centers result in different wave train patterns.Although convections in the WTIO have covarying anomalies with that in the ETIO for SON (Cai et al 2011), the location of subtropical RWS anomaly is balanced by the relative magnitude of the two convective centers (McIntosh and Hendon 2018).Specifically, eastern reduced convective anomaly center is stronger and is shifted westward than that WTIO-induced, causing vortex stretching and vorticy advection anomalies (i.e. two linear terms of RWS) away from southwestern Australia.Combined together with the atmosphere background state in this season (Li et al 2015a(Li et al , 2015b)), RWS anomalies in the two cases (figure S7) explain the initiation and evolution of their associated Rossby waves.The difference in wave trains induced by WTIO and ETIO can also be obtained by the M-and S-indices (figure S9).
To provide further evidence for our proposed teleconnection process, we utilize pacemaker experiments using a coupled climate model, in which the tropical Indian Ocean SSTAs are nudged to the observed values whereas ocean and atmosphere are fully coupled elsewhere.This experiment isolates the impacts of the tropical Indian Ocean, while representing the role of air-sea interaction.The results show that the moderate IOD dominated by WTIO indeed excites a wave train propagating eastward and poleward, characterized by positive 200 hPa geopotential height anomalies over southern Australia (figure 2(c)).However, such geopotential height anomalies are not observed under the effect of strong IOD dominated by ETIO, with its associated wave train confined to the southern Indian Ocean sector (figure 2(f)).The consistence in moderate IOD generated wave train between the observational results and pacemaker experiments confirms the robustness of our ENSO-removed finding.
Next we evaluate the responses of fire weather conditions.The associated Australian fire weather responses during moderate IOD events show conducive conditions for fire outbreak.The atmospheric circulation in extratropics exhibits an equivalent barotropic response, with a high sea level pressure (SLP) anomaly across southern Australia (figure 3 Besides, the heat increases the vapour pressure deficit and encourages the plant to dry out (Nicholls 2004, Williams et al 2019, 2020) that in turn magnifies warm air temperatures across the boundary layer (Miralles et al 2014, Yin et al 2014, Abram et al 2021).Consequently, this land-atmosphere feedback provides favourable fire weather conditions and raises fire risks in southern Australia during SON.These results hold during moderate IOD events using the M-index (figure S10).However, the anomalies associated with the ETIO are distinct from those with the WTIO and display very limited, not statistically significant atmospheric responses over Australia (not shown).In responses to the WTIO SSTA warming, pacemaker experiment results also show a high SLP anomaly across southern Australia, anomalous SAT warming and decreased rainfall, consistent with the observational results (figures 3(d)-(f)).
To further distinguish and clarify the role played by WTIO-dominated moderate IOD, we use composite analysis in observations.During moderate pIOD events without ENSO, as in 1992ENSO, as in , 1996ENSO, as in , and 2012 (figure S11) (figure S11), the composite SSTAs display broadscaled warm anomalies in the western and central   and S10).Therefore, considering the nonlinear coefficient α between PC1 and PC2 of EOF over the tropical Indian Ocean, we use the M-index to diagnose the moderate IOD impacts on future Australian fire weather.
Multimodel mean result indicates a decline in Mindex s.d. in the future period but without assessment on each model (figure S13).Compared with a value of α of 0.4262 in reanalysis, 15 models among all 45 models produce a value greater than 50% of that (i.e.α ⩾ 0.2131; see 'Alpha' and 'Model type' in table S1), which shows a reasonable ability to reproduce the nonlinear relationship of the first two PCs and separate moderate IOD (figures 4(a) and (b)).A total of 13 out of the 15 selected models (87%) simulate decreased M-index variability, and the projected reduction in multimodel mean is 20%, statistically significant above the 95% confidence level based on the Bootstrap test (figure 4(c)).As the teleconnections generated by the moderate pIOD (WTIO) dominate spring fire weather conditions in southern Australia in reanalysis and pacemaker experiment results, we investigate the wave train and local atmospheric responses in climate models.Z200, SAT, and precipitation flux responses to M-index during the present-day and future periods display significant differences over southern and eastern Australia, suggesting a weakened wave train and a decrease in conditions favourable for Australian fire in the future (figures 4(d)-(i)).The decreased M-index variability is consistent with the result obtained from a combination of 15 CMIP5 models and 5 CMIP6 models in Cai et al (2021).
The moderate IOD variability is impacted by mean state atmosphere circulation changes.Under greenhouse warming, the lower troposphere warms faster than the surface (figure S14), indicating an increased static stability of the troposphere and reduced surface zonal wind feedback (Zheng et al 2010(Zheng et al , 2013)).There is a linear relationship between zonal wind stress and Ekman pumping, that is, westward wind stress is associated with a positive Ekman pumping (downwelling) south of the equator, and the associated reflected downwelling waves leads to the warming.The required wind variability results in decreased SST warming and thus decreased M-index variability.

Conclusions
Tropical SST variability is one of the dominant drivers in Australian fire weather conditions.Such a connection is established in tropical Indian Ocean commonly using the DMI, and could fail to distinguish impacts from various IOD patterns.Our study reveals that during austral spring, fire weather in southern Australia is primarily influenced by WTIOdominated moderate IOD and is relatively little related to ETIO-dominated strong IOD or ENSO.This is supported by three independent methods including observational regressions, composite analysis, and pacemaker experiments.For a moderate pIOD, dominated by warm WTIO SSTAs, the associated teleconnection is via an anomalous atmospheric Rossby wave train, with a high-pressure anomaly over southern Australia, which affects fire weather conditions there.The enhanced pressure anomaly results in hot and dry conditions, together with the enhanced easterly wind anomalies, creating weather conditions that raise fire risks.During a strong pIOD, dominated by cool ETIO SSTAs, the associated wave train forms prevailing weak low pressure anomalies across Australia, which make little contributions to fires.Under greenhouse warming, climate models project a decreased variability in WTIO SST and moderate IOD, thus pointing to a reduced Australian fire weather response to tropical Indian Ocean.
where ζ a denotes the vertical component of absolute vorticity; D is divergence and F is the frictional term.The horizontal velocity field v can be split into rotational v ψ and divergent v χ components, where ∇ • v χ = D. Then the barotropic vorticity equation can be rewritten as S = −ζ a D − v χ • ∇ζ a , where S denotes the RWS.
To distinguish the impact of the individual poles, time series in the WTIO (10 • S-10 • N, 50 • -70 • E) and the ETIO (10 • S-0 • , 90 • E-110 • E) are considered alongside the overall tropical Indian Ocean.ETIO SST time series is nearly identical (r = 0.87) to the DMI, while a weak correlation (r = 0.39) exists in WTIO time series and the DMI (figure 1(d)).The grid-point correlation coefficients suggest the dominant influence of the WTIO SST variability on the southern Australian FWI.Specifically, southern Australia, especially southeastern Australia, exhibits high correlations between the WTIO and FWI (figure 1(e)).This correlation is similar to but larger than the DMI-FWI relationship (figure 1(b)).By contrast, the ETIO has low correlations with the FWI throughout Australia, except for the eastern edge (figure 1(f)), although the correlation coefficient between ETIO and DMI is much higher than that with WTIO.Previous studies used the first two principal components (PCs) of an empirical orthogonal function (EOF) analysis in a tropical Indian Ocean region (5 • S-5 • N, 40 • E-100 • E) to separate moderate IOD and strong IOD (figure S3), dominated by WTIO and ETIO, respectively (Cai et al 2021).These two PCs have a strong nonlinear relationship (coefficient as α), and the combination of them can construct indices of moderate and strong pIOD, described by the M-index ( (PC1 − PC2) / √ 2 ) and the S-index ( (PC1 + PC2) / √ 2 ) g. Alexander et al 2002, Li et al 2021), anomalous tropical SST warming triggers vertical tropospheric deep convection, which modulates Walker and Hadley circulations and subsequently excites anomalous RWS over the subtropical upper-level (Mo and Higgins 1998, Garreaud and Battisti 1999, Ding and Steig 2013).These wave sources further generate atmospheric stationary Rossby waves propagating poleward and eastward, characterized by alternating positive and negative geopotential height anomalies (Simpkins et al 2014, 2016, Li et al 2015a, 2015b).Here, corresponding to the broad-scaled warm SSTAs in WTIO (figures 2(a) and S3(c)), a west-east dipole of 200 hPa divergence anomalies, with west divergent and east convergent flows (figure 2(a), vectors, tropical regions), appears over the tropical Indian Ocean.It can be understood from the outgoing longwave radiation (OLR) anomalies (figure 2(a), contours) that the negative OLR indicates upwelling flow in the midtroposphere (implying upper troposphere divergence) and opposite sign for positive OLR.These tropical convections perturb the meridional Hadley circulation, creating subtropical meridional divergent wind anomalies in the upper troposphere (figure 2(a), vectors, 10 • S-25 • S).Subsequently, the related vorticity forcings induce pronounced positive RWS anomaly near the southwestern edge of Australia (figure S7(a)), initiating the development of Rossby waves.
(a), shading).Compared to climatological wind (figure 3(a), pink vectors), enhanced easterly wind anomalies (figure 3(a), green vectors) along with anomalous warming SAT (figure 3(b)) and decreased rainfall (figure 3(c)) lead to soil moisture depletions that promote drying (Kriwoken 1996, Cai et al 2009a, Nolan et al 2016), contributing to increased fuels and serving for the development and spread of destructive fires in southern Australia.Under the control of a high SLP anomaly, descending flow and dry air prevent cloud formation, contributing to rainfall reduction and increasing incoming solar shortwave radiation that is conducive to SAT warming (Yin et al 2014, Abram et al 2021).

Figure 3 .
Figure 3. Responses of Australian fire weather conditions in reanalysis and pacemaker experiment.Partial regressions of (a) SLP (shading; Pa) with 10 m wind (green vectors; m s −1 ), (b) SAT ( • C) and (c) precipitation (mm) onto normalized WTIO SST time series for the 1979-2022 SON period.(d), (e) Same as (a), (b), but for the pacemaker experiment.(f) Same as (c), but for precipitation rate (10 −8 m s −1 ) in the pacemaker experiment.The pink vectors in (a), (d) denote climatological 10 m wind (m s −1 ).Stippling indicates the 95% confidence level based on the t-test.

3. 3 .
Weakened moderate IOD impacts on Australian fire weather in the futureWe assess the impact of greenhouse warming on moderate IOD changes by taking outputs from 45 CMIP6 coupled global climate models forced with historical forcing until 2014 and with SSP5-8.5 emission scenario from 2015 onward to 2099.Due to the range of the DMI observed poles differs vastly from one model to another, the definition of the WTIO SST time series is difficult to standardize.In contrast, the M-index accounts for diversity in patterns and thus facilitates assessment of the response of the WTIO SST variability as simulated by each individual model(Cai et al 2021).Additionally, the responses of atmospheric teleconnections and fire weather conditions to the M-index and WTIO SST are consistent (figures S9

Figure 4 .
Figure 4. Decreased in moderate IOD variability and reduced impacts on Australian fire weather under greenhouse warming.(a) A total of 15 models {i.e.models in (c)} produce a nonlinear relationship between PC1 and PC2 with α ⩾ 0.2131; that is, 50% of the observed value (0.4262), and are thus selected for further analysis.A quadratic fit using PC1 and PC2 time series from the aggregate of the 15 models produces a value of α of 0.27.(b) Same as (a), but for the discarded 30 models {i.e.models in figure S13 but not in (c)} with α < 0.2131.(c) A total of 13 out of the 15 models (87%) generate a decrease in M-index variability from the present-day (1900-1999; blue bars) to the future climate (2000-2099; red bars), with the exception of 2 models generating an increase (indicated by gray shading).The multimodel mean decrease of 20% is statistically significant above the 95% confidence level based on the Bootstrap test.(d)-(f) Partial regressions of (d) Z200 (m), (e) SAT ( • C), and (f) precipitation flux (10 −6 kg m −2 s −1 ) onto normalized M-index for the present-day.(g)-(i) Same as (d)-(f), but for the future climate.Stippling indicates the 95% confidence level based on the t-test.
Yang et al 2020) are nudged to the Extended Reconstruction of SST version 3b (ERSST-V3b) observations (Smith et al 2008), with the rest of the model's coupled climate system to evolve freely.
and sea ice components with a resolution of nominal 1 • (Hurrell et al 2013).Each member of this ensemble starts with slightly different initial conditions.The SSTAs of the tropical Indian Ocean (TIOexperiment) (15 • S-15 • N, Africa coast-174 E, the linearly tapering buffer zone is 20 • S-20 • N, Africa coast-174 E) (D.