Muted extratropical low cloud seasonal cycle is closely linked to underestimated climate sensitivity in models

Accurate projection of future climate is urgently needed to inform mitigation and adaptation policies, yet global climate models (GCMs) exhibit large uncertainty in how Earth’s climate responds to anthropogenic forcing. Equilibrium climate sensitivity (ECS)—the global mean near-surface air temperature increase following a doubling of CO2—varies widely across models. In CMIP6, ECS spans roughly 1.8–5.6 °C, exceeding the CMIP5 spread (2.1–4.7 °C). A key contributor to this spread is uncertainty in shortwave cloud radiative feedbacks associated with low clouds, particularly over extratropical oceans. This study restricts analyses to ocean grid cells to avoid land complexities and defines low cloud fraction (LCF) as cloud fraction below 700 hPa using a maximum overlap assumption. The authors investigate whether present-day seasonal variations of extratropical LCF provide an emergent constraint on model ECS by linking seasonal and trend-based LCF–temperature relationships, and assess model fidelity against satellite observations.

Prior work has identified low clouds—especially in the extratropics—as a major source of uncertainty in climate feedbacks and ECS. Studies have connected mixed-phase cloud physics, cloud fraction, and aerosol–cloud interactions to variations in cloud radiative feedbacks and model sensitivity. Emergent constraint frameworks have been proposed to relate observable present-day processes to ECS, and previous metrics have focused on tropical low-cloud regimes, subtropical marine LCF, and regime-averaged low-cloud feedbacks. CMIP6 models often show higher ECS than CMIP5, with evidence pointing to altered representations of extratropical low-cloud processes. Observational datasets like CloudSat/CALIPSO and CERES have been used to evaluate vertical cloud structure and radiative effects, and reanalyses (e.g., ERA5) to characterize large-scale controls. However, passive cloud datasets have limitations in multilayer cloud scenes typical of extratropical cyclones, motivating reliance on active sensors for vertical cloud profiling.

• Domain and definitions: Analyses focus on ocean grid points over extratropics defined as 30–60°N and 30–60°S. Low cloud fraction (LCF) is derived from vertical cloud fractions below 700 hPa under a maximum overlap assumption. Surface temperature (TS) refers to skin temperature (sea surface temperature over ocean).
• Time scales and dLCF/dTS metrics: Seasonal dLCF/dTS is computed at each grid cell via linear regression of climatological monthly LCF against TS using the 12-month seasonal cycle from 1980–2014 CMIP6 historical simulations; then averaged over extratropical oceans (30–60° in both hemispheres). Short-term trend dLCF/dTS (1980–2014) is computed by linear regression of annual mean LCF against TS at each grid cell and averaged over the same region. Long-term trend dLCF/dTS is computed as the ratio of spatially averaged changes in LCF and TS between 2061–2095 (SSP585) and 1980–2014 (historical) over extratropical oceans.
• Model ensembles: Correlations between ECS and the seasonal dLCF/dTS are computed across 26 CMIP6 models. Long-term trend analyses are based on 18 models with available data. Additional comparisons with CMIP5 models assess generality of the metric.
• Observational datasets: Seasonal dLCF/dTS is derived from CloudSat/CALIPSO (2B-GEOPROF-LIDAR) vertical cloud profiles for 2006–2011 using the same LCF definition, and TS from NOAA OISSTv2. TOA shortwave cloud radiative effect (SWCRE) is assessed using CERES EBAF Ed4.1. Large-scale fields (e.g., vertical velocity) are taken from ERA5 reanalysis. Uncertainty in observed seasonal dLCF/dTS is quantified from interannual variability (mean ≈ −1.65%/K, SD 0.27%/K; climatological estimate ≈ −1.71%/K).
• Cloud regime analysis: For detailed process attribution, three representative CMIP6 GCMs are analyzed with daily vertical cloud fractions: CESM2 and HadGEM3-GC31-LL (high ECS), MPI-ESM1-2-LR (low ECS). An EOF analysis of daily vertical cloud fraction anomalies over 45–55°S is performed. The latitude band is partitioned into 12 equal-area subregions (~30° longitude × 10° latitude). Daily profiles on 19 pressure levels (1000–1 hPa) for 1980–2014 (12,775 days) are concatenated and decomposed. Leading modes correspond to high-, mid-, and low-top cloud regimes. Reconstructed contributions from these modes are used to interpret seasonal and interannual LCF variability and trends, and their coupling to mid-tropospheric vertical velocity and EIS. Composite LCF as a function of 700 hPa vertical velocity and EIS is constructed from climatological seasonal cycles to diagnose controlling factors, and changes in the frequency of occurrence under future climate (2060–2094 SSP585) are contrasted with historical (1980–2014).

• Strong emergent relationship: Across 26 CMIP6 models, ECS is strongly negatively correlated with the seasonal extratropical dLCF/dTS (r ≈ −0.82). Models with larger seasonal reductions of LCF as TS increases (more negative dLCF/dTS) have higher ECS.
• Long-term feedback linkage: The extratropical low-cloud fraction feedback associated with long-term warming (2061–2095 vs 1980–2014) is highly negatively correlated with ECS across 18 models (r ≈ −0.81). Seasonal and long-term dLCF/dTS are themselves strongly correlated, indicating shared controlling processes.
• Observational constraint: CloudSat/CALIPSO indicates an observed seasonal extratropical dLCF/dTS ≈ −1.71%/K (mean from annual estimates ≈ −1.65%/K, SD 0.27%/K), closer to high-ECS model behavior and inconsistent with weak or positive slopes in many low-ECS models. This suggests low ECS values in such models may be underestimated.
• Seasonal LCF structure: High-ECS models and observations show pronounced extratropical LCF seasonality with winter maxima and summer minima. Low-ECS models show muted seasonality and often summer maxima, opposite to observations.
• Short-term trends (1980–2014): Despite similar TS trends across groups, high-ECS models show significant decreases in extratropical LCF and increases in TOA SWCRE, consistent with CERES; low-ECS models show little to no LCF or SWCRE trend. The short-term dLCF/dTS correlates with long-term dLCF/dTS (cor=0.80), with seasonal dLCF/dTS (cor=0.82), and with ECS (r = −0.69, p < 0.001).
• Cloud-regime processes: In high-ECS models (CESM2, HadGEM3-LL), LCF variability is dominated by mid-top or combined mid- and low-top regimes strongly coupled to mid-tropospheric ascending motion (storm-track activity), yielding strong winter LCF and reductions with weakened ascent in summer and in a warming climate. In the low-ECS model (MPI-ESM1-2-LR), a low-top regime dominates, tied mainly to lower-tropospheric stability (EIS), producing summer LCF peaks and weak trends; low clouds are decoupled from vertically extended clouds and vertical motion changes.
• Controlling factors diagnostics: High-ECS models and observations show strong LCF dependence on 700 hPa vertical velocity; low-ECS models show stronger dependence on EIS, especially under ascent. Future projections show increased occurrence of strong descent associated with reduced LCF in high-ECS models, while in low-ECS models compensating changes in EIS and descent yield little net LCF change.
• Ancillary: Annual mean extratropical LCF also correlates with ECS (r = −0.56, p = 0.0024) but is a weaker constraint than seasonal dLCF/dTS. CMIP5 alone shows no significant correlation between ECS and seasonal dLCF/dTS, but combining CMIP5+CMIP6 yields a significant relationship (r ≈ −0.56), highlighting CMIP6 changes in extratropical low-cloud feedback representation.

The study demonstrates that the seasonal sensitivity of extratropical low-cloud fraction to surface temperature is a powerful emergent constraint on model climate sensitivity. Models that realistically capture the strong winter–summer reduction in extratropical LCF—consistent with CloudSat/CALIPSO—also simulate stronger positive extratropical low-cloud feedbacks and higher ECS. Mechanistically, high-ECS models link LCF to storm-track dynamics and mid-tropospheric ascent; reductions in ascent from winter to summer and under greenhouse warming lead to reduced LCF and increased shortwave absorption, amplifying warming. Low-ECS models, dominated by a low-top regime tied to EIS and decoupled from vertical motion, lack this strong positive feedback, yielding muted LCF seasonality and trends that conflict with observations. The findings indicate that differences in how models couple extratropical low clouds to large-scale dynamics are central to ECS spread. The emergent metric aligns with short-term observed-like trends and supports the view that very low ECS in models with muted extratropical LCF seasonality is likely underestimated. The results also contextualize broader circulation changes (e.g., Hadley expansion, jet shifts) as potential drivers of ascent weakening, though model responses and cloud-regime sensitivities vary.

This work proposes and validates an emergent constraint on equilibrium climate sensitivity based on the seasonal slope of extratropical low-cloud fraction versus surface temperature. The metric captures the linkage between present-day seasonal cloud variability and long-term low-cloud feedbacks and correlates strongly with ECS. Satellite observations support the stronger seasonal LCF reductions seen in high-ECS models, implying that models with weak or positive seasonal dLCF/dTS likely underestimate ECS. Process analysis shows that high-ECS models couple low clouds to storm-track ascent via vertically extended cloud regimes, while low-ECS models emphasize low-top regimes tied to stability and are decoupled from ascent, explaining divergent feedbacks. Future research should (1) disentangle the roles of Hadley Cell expansion, storm-track variability, and jet shifts in modulating extratropical ascent and cloud regimes; (2) quantify contributions from cloud phase and optical-depth feedbacks alongside fraction changes; (3) extend emergent constraints across model generations and regions; and (4) improve representation of extratropical cyclone cloud structures to reduce ECS uncertainty.

• Cloud regime attribution uses only three CMIP6 models with available daily vertical cloud profiles, limiting generality of regime conclusions across all models.
• CloudSat/CALIPSO observations cover 2006–2011 due to mission constraints; while seasonal signals are robust, the short period limits trend assessment.
• ERA5 vertical velocity fields lack direct observational constraints, complicating evaluation of ascent trends.
• Passive satellite low-cloud products are not directly comparable to active-sensor and model-derived LCF in multilayer scenes typical of extratropical cyclones.
• Internal climate variability over 1980–2014 can affect short-term dLCF/dTS estimates and their correlation with ECS.
• ECS is influenced by multiple global feedbacks beyond extratropical low-cloud fraction; conclusions pertain primarily to this feedback component.