Madden-Julian Oscillation-induced extreme rainfalls constrained by global warming mitigation

Floods expose nearly 300 million people annually, with large economic losses. Extreme rainfall-driven flooding is particularly destructive in low-lying, rapidly developing tropical Asia and Oceania, as evidenced by severe impacts during boreal winter 2020. IPCC AR6 reports likely increases in tropical extreme precipitation with regional inconsistency and uncertainties in mechanisms. The Paris Agreement aims to limit warming below 2 °C, yet it remains uncertain how mitigation—primarily driven by mid-latitude nations—will affect vulnerable tropical regions where the hydrological cycle is intensifying. The MJO, a dominant intra-seasonal driver of convection and rainfall variability over tropical Asia and northern Australia during boreal winter, modulates enhanced/suppressed convection regions and can trigger extremes locally and via teleconnections to extratropics. While many studies project future MJO activity changes using ESMs, it is unclear how MJO-induced extreme rainfall and societal exposure will change with warming and mitigation. CMIP5 models often underestimated MJO amplitude and propagation; several CMIP6 models show improvements, especially through better horizontal moisture advection over the Maritime Continent. This study leverages improved CMIP6 models to assess projected MJO-induced extreme rainfall changes and associated exposure under different warming scenarios, focusing on boreal winter when MJO is most active.

Prior work indicates increasing precipitation extremes with warming and regional heterogeneity (e.g., O’Gorman & Schneider; Donat et al.). CMIP5 models struggled with MJO amplitude and coherent eastward propagation, limiting confidence in projections. CMIP6 shows improved MJO simulation skill, especially over the Maritime Continent (Ahn et al.). Studies suggest MJO activity may intensify with warming due to increased moisture, instability, and convective heating, though uncertainty remains in structure/propagation and teleconnections (e.g., Maloney et al., Adames et al.). Teleconnected impacts to mid- and high-latitudes may strengthen or vary by phase. Evidence also highlights land–sea contrast and topographic barrier effects of the Maritime Continent on MJO propagation, influencing rainfall extremes.

• Models and periods: Historical and ScenarioMIP outputs from eight CMIP6 ESMs (BCC-CSM2-MR, CESM2-WACCM, EC-Earth3, EC-Earth3-Veg, GFDL-CM4, MIROC6, MRI-ESM2-0, NESM3), each r1i1p1f1, regridded to 1.25°×1.25°. Historical period: 1994–2014. Present: 2015–2035. Future: 2079–2099. Scenarios: SSP5-8.5 (high), SSP2-4.5 (medium), SSP1-2.6 (low).
• Study domain and sub-regions: Tropical Southeast Asia and northern Australia; sub-regions are Maritime Continent (Peninsular Malaysia, Indonesian Archipelago, Papua New Guinea), Philippines, Mainland Southeast Asia (Thailand, Cambodia, Laos, Vietnam, southern Myanmar), and northern Australia. Analyses focus on boreal winter (Nov–Apr).
• Reference datasets: NCEP–DOE Reanalysis 2 and NOAA Interpolated OLR for MJO structure (1979–2001), GPCC daily precipitation for evaluation (regridded to 1.25°). Land mask from NOAA high-resolution blended analysis aggregated to 1.25°; land-weighted averages applied.
• MJO characterization: Real-time Multivariate MJO (RMM) index derived via combined EOFs of OLR, U850, U250 following CLIVAR MJO WG protocol. For models and reanalysis, anomalies filtered (201-point Lanczos) to intra-seasonal band; first two modes define RMM1 and RMM2. MJO amplitude = sqrt(RMM1²+RMM2²). MJO-active days: amplitude > 1; phases (1–8) from arctangent(RMM2/RMM1). Trends of MJO occurrence (days) and amplitude computed over 2015–2099 for scenarios using linear regression.
• Precipitation metrics: Total rainfall = boreal winter accumulated precipitation annually. Extreme rainfall threshold per grid = 95th percentile of daily precipitation on wet days (>1 mm) during historical boreal winter. Extreme rainfall amount = multi-season average of precipitation exceeding threshold; frequency = fraction of boreal winter days exceeding threshold. MJO-induced rainfall/extremes assigned using a 6-day lag after MJO-active days to account for delayed effects; partitioned by MJO phase via composites.
• Bias handling and changes: Projected changes computed as multi-model median differences between future (2079–2099) and present (2015–2035), then added to historical levels to reduce systematic-forcing differences among models.
• Exposure assessment: Exposure defined as population and urban area within 1.25° grids where future MJO-induced extreme rainfall exceeds hazard thresholds (2-year and 5-year return levels). Return levels estimated by fitting GEV to 1964–2014 boreal winter extremes (maximum likelihood), spatially smoothing parameters (3×3 filter), then inverting GEV for 2- and 5-year levels. Population: GPW v4.11 (2015) interpolated to 1.25°. Urban area: 30 m global urban map (2015) aggregated to grid. Exposure evaluated using current (2015) population and urban extent (conservative, ignores future growth).

• Regional increases under SSP5-8.5: Multi-model mean boreal winter total rainfall increases by 103.3 mm (~11%) region-wide by 2079–2099 vs 2015–2035, largest in the Maritime Continent (+176.6 mm), with smaller increases in Mainland Southeast Asia (+36.8 mm), Philippines (+53.2 mm), and northern Australia (+75.2 mm). Much of the increase is MJO-associated; in some areas non-MJO rainfall fraction declines.
• Extreme rainfall amplification: Areal mean extreme rainfall increases by 103.9 mm (~60% of historical level), exceeding the total rainfall increase. Maritime Continent sees +245.9 mm; northern Australia +55.9 mm. On average, 84% of extreme rainfall increases are MJO-associated. MJO shares of total extreme increases: Maritime Continent 64%, Philippines 74%, Mainland Southeast Asia 68%, northern Australia 101%.
• Spatial concentration over land: Increases in extreme rainfall, especially MJO-induced, concentrate over land, notably Indonesia, Malaysia, Papua New Guinea, and northern Australia. Patterns of MJO-induced total and frequency changes mirror total changes.
• Temporal trends: Post-2015 extreme rainfall increases significantly (p<0.01). After 2015, extreme rainfall trend ~+18.5 mm/decade; MJO-induced extreme rainfall ~+15.6 mm/decade. Little change before ~2039; acceleration towards century end. MJO-induced increases: ~+11.6 mm/decade (2039–2059) and ~+31.0 mm/decade (2079–2099).
• Exposure under SSP5-8.5 (5-year return threshold): 96.68 million people and 9.72 million km² urban area exposed regionally. Maritime Continent dominates (93.85 million people; 9.40 million km² urban); Mainland Southeast Asia 2.76 million (0.23); northern Australia 0.06 million (0.09); Philippines 0.
• MJO phase asymmetry: Warming induces asymmetric increases in MJO occurrence and amplitude across phases, with notable increases in occurrence (nearly +2 days) for phases 2, 5, 6, and significant amplitude enhancement for all but phases 3 and 7. Extreme rainfall peaks shift forward in phase over the Maritime Continent (from phase 3 ~30 mm historically to phases 2–3 ~50 mm). Amplification strongest over land during phases 2–5; northern Australia increases linked to phase 5.
• Physical drivers: Land–sea contrast and topography (Maritime Continent barrier) interact with asymmetric MJO intensification to enhance convergence and vertical motion over land, boosting extremes. Increased extremes over NE Australia seas in phases 5–6 suggest a stronger MJO may overcome historical barrier effects.
• Mitigation effects: Under SSP2-4.5 and SSP1-2.6, projected MJO occurrence/amplitude increases are reduced or insignificant/negative, respectively. MJO-induced extreme rainfall changes are constrained: 29% (SSP2-4.5) and 11% (SSP1-2.6) vs 88% (SSP5-8.5). Phase-specific reductions: strong in Maritime Continent (especially phase 2), phase 4 for Philippines/Mainland SE Asia, phases 5–8 for northern Australia.
• Exposure reductions with mitigation: Under SSP2-4.5 (5-year threshold), exposed population and urban areas reduce by ~95% and 99% (to 4.85 million people and 0.12 million km²). Under SSP1-2.6, exposure is nearly eliminated across regions except Papua New Guinea (~0.71 million people; 0.04 million km²).

Findings link large projected increases in boreal winter extreme rainfall over tropical Asia and northern Australia predominantly to MJO intensification under high-emission warming. Asymmetric changes in MJO phase occurrence and amplitude (notably phases 2–5) shift the timing and enhance magnitude of extremes, with land–sea contrast and topographic effects focusing impacts over land. The results address the research question by quantifying the MJO’s contribution (84%) to extreme rainfall increases and by mapping associated exposure, identifying hotspots (western Maritime Continent, Java, Sumatra, Papua New Guinea, and around Darwin). Mitigation substantially constrains MJO-induced extremes and exposure, demonstrating policy relevance: SSP2-4.5 can avert ~95–99% of population/urban exposure relative to SSP5-8.5 under current distributions. Broader implications include potential changes in MJO teleconnections affecting extratropical extremes; phase-dependent changes imply regionally varying downstream impacts. Continued improvements in MJO representation and phase-asymmetric responses are critical for S2S prediction and climate risk management.

Using eight CMIP6 models with credible MJO representation, the study projects nearly 60% increases in boreal winter extreme rainfall across tropical Asia and northern Australia by late century under SSP5-8.5, with 84% of increases tied to MJO intensification. Increases are concentrated over land in the Maritime Continent and northern Australia and are modulated by asymmetric MJO phase changes (enhanced phases 2–5). Exposure analysis indicates that, with current population and urban distribution, 96.68 million people and 9.72 million km² of urban area are at risk from increased MJO-induced extremes under SSP5-8.5, but most exposure can be avoided under SSP2-4.5 and nearly all under SSP1-2.6. Future work should refine MJO and BSISO simulation, understand phase-dependent teleconnections, and integrate dynamic population/urban growth to better quantify evolving risk. Proactive adaptation and infrastructure design will be essential even under mitigation scenarios, particularly in Maritime Continent hotspots.

• Model biases: Despite selection of CMIP6 models with robust MJO propagation, some underestimate MJO amplitude and extreme precipitation. Multi-model medians underestimate extreme rainfall; models overestimate MJO-induced share of total rainfall by 20–30%, though MJO-induced extreme fractions are closer to observations.
• Threshold sensitivity and definitions: Extreme thresholds (95th percentile on wet days) and choice of 6-day lag for MJO effects introduce methodological uncertainty; results are robust across tested amplitude thresholds but residual sensitivity remains.
• Scenario and bias-correction approach: Future–present differencing reduces forcing inconsistencies but may not remove all structural biases.
• Exposure estimation: Uses static 2015 population and urban extent; true future exposure likely higher due to growth and urbanization patterns. Zero exposure in some regions under 5-year threshold partly reflects boreal winter focus (local dry season) rather than annual risk.
• Process uncertainties: Asymmetric MJO phase changes and teleconnections depend on SST warming patterns and model physics. BSISO representation is poor in the ensemble, limiting summer-season insights.
• Affiliation of phase-specific trends: Some phases show insignificant amplitude changes; inter-model spread exists.