Realistic representation of mixed-phase clouds increases projected climate warming

The study addresses a key uncertainty in climate projections: the phase partitioning of mixed-phase clouds (MPCs), which contain both ice and supercooled liquid water. Cloud phase strongly influences Earth’s radiation budget via albedo and longwave effects, and MPCs are prevalent in extratropical storm tracks and over high-latitude oceans. Many climate models assume homogeneously mixed ice and liquid within grid boxes, leading to overly efficient Wegener–Bergeron–Findeisen (WBF) conversion and too much ice at mixed-phase temperatures. This bias increases modeled shortwave cooling during warming (via excessive ice-to-liquid transition), thereby underestimating climate sensitivity. Observations also show MPCs are heterogeneous both horizontally and vertically, with supercooled liquid frequently concentrated at cloud tops, but prior global constraints used only interior (bulk) phase. The research question is whether incorporating distinct cloud-top and interior phase observations into model constraints improves present-day energy budget representation and alters projected warming and cloud feedbacks. The purpose is to use new CALIOP lidar retrievals separating cloud-top from interior supercooled liquid fraction (SLF) to constrain NorESM2 and quantify impacts on top-of-atmosphere radiation, cloud feedbacks, and 21st-century warming. The importance lies in reducing MPC-related uncertainty in climate sensitivity and correcting potential low bias in projected warming across models that overestimate ice in MPCs.

Prior work shows cloud phase exerts a strong control on radiative fluxes and that MPCs substantially affect the shortwave radiation budget, with active sensors (CloudSat/CALIPSO) providing more reliable phase detection than passive sensors. Many CMIP5/6 models overestimate ice at mixed-phase temperatures and exhibit persistent biases over the Southern Ocean and the Arctic. Studies document spatial heterogeneity of MPC phase and limitations of the WBF mechanism under realistic conditions, indicating reduced ice growth efficiency when ice and liquid are not co-located. Earlier observational constraints focused on bulk/in-cloud SLF, not capturing vertical inhomogeneity and the tendency for liquid-dominant tops. Observational and modeling studies have linked these MPC phase biases to underestimated climate sensitivity; however, a global constraint that separately targets cloud tops and interiors had not been implemented.

Observations: The team used CALIOP lidar (CALIPSO, v4 Level 2, 5 km horizontal resolution) to derive global SLF for cloud tops and interiors on isotherms from −40 to 0 °C (5 °C bins) over June 1, 2009–May 31, 2013. Cloud-top SLF was assigned at the highest layer where cloud optical depth (COD) ≥ 0.35 (filtering out optically thin cirrus). In-cloud (interior) SLF was computed from cloud layers with accumulated optical depth < 3. Each cloud layer’s temperature was interpolated from MERRA-2. The approach leverages CALIOP’s ability to penetrate opaque clouds on average by ~1.67 ± 0.49 km, enabling retrievals from distinct top and interior layers. Model and tuning: The Norwegian Earth System Model (NorESM2-MM; CAM6-based atmosphere with two-moment microphysics; 1.25° × 0.9375° atmosphere, 32 vertical levels) was used. First, atmosphere-only simulations were nudged to ERA-Interim circulation (2009–2013) to tune microphysical parameters to match observed SLF at both cloud tops and interiors globally and regionally. Three parameters were adjusted to increase SLF: (1) reduce WBF efficiency (WBF factor 0.5 vs 1.0 in control), (2) reduce ice-nucleating particle availability (INP factor 0.001 vs 1.0), and (3) modify convective detrainment phase partitioning by lowering the temperature at which all detrained condensate is liquid (from −5 °C in control to as low as −30 °C, depending on experiment) while keeping the all-ice threshold at −35 °C. Parameter choices differ when constraining both top and interior simultaneously versus interior only. Experimental design: After tuning, identical parameter changes were implemented in fully coupled NorESM2 simulations: pre-industrial control, historical, and SSP5-8.5 to 2100, for three cases: (i) Global SLF constraint (82°S–82°N), (ii) Northern extratropics (N-ET, ≥30°N), and (iii) Southern extratropics (S-ET, ≥30°S). Comparison to an unconstrained NorESM2 Control SSP5-8.5 submission served as baseline. Diagnostics: Top-of-atmosphere (TOA) net radiation biases were evaluated against CERES. Cloud feedbacks were computed using COSP2 outputs and the radiative kernel method (Zelinka et al.), partitioned into total, amount, altitude, and optical depth components (shortwave, longwave, net), focusing on 2041–2050. A transient “energy-sensitivity” metric was introduced: cumulative RESTOM (sum of global net shortwave and longwave residual TOA flux from 2015–2100) versus surface temperature change; higher temperature for a given RESTOM indicates greater climate sensitivity.

• CALIOP observations show MPC tops are more liquid than interiors globally and year-round. At −20 °C, cloud-top SLF exceeds interior SLF everywhere; extratropics show 50–60% more supercooled liquid at tops than interiors, with the smallest differences in the tropics. Seasonal analysis reveals persistent top–interior SLF gaps, largest between −30 and −10 °C.
• NorESM2 initially underestimates SLF at both tops and interiors (mean biases: −15.3% top, −12.8% interior) relative to CALIOP. After tuning for both top and interior, biases are reduced to +4.2% (top) and +0.3% (interior).
• Increasing SLF (more liquid) cools TOA net radiation in atmosphere-only, nudged runs by −1.8 W m−2 globally, with strong regional cooling in extratropics (N-ET ≈ −1.8 W m−2; S-ET ≈ −3.6 W m−2) and little change in the tropics. This reflects thicker, brighter MPCs reflecting more shortwave radiation.
• Constraining SLF reduces TOA net radiation bias versus CERES by −0.8 W m−2 globally, with up to ~8 W m−2 bias reduction over the Southern Ocean.
• Fully coupled SSP5-8.5 projections: For a cumulative TOA forcing of +100 W m−2 (RESTOM), global warming increases from 2.6 °C (Control) to 3.6 °C (Global constraint), i.e., +1.0 °C more warming. At +50 W m−2, warming increases from 1.4 °C to 2.4 °C. Regional constraints yield smaller global increases: N-ET and S-ET each warm 2.9 °C at +100 W m−2 (+0.3 °C vs Control).
• Hemispheric sensitivities: In N-ET, warming is strongly influenced by remote cloud-phase changes; at +100 W m−2, N-ET warms +6.2 °C (Global) vs +4.8 °C (N-ET only) vs +3.8 °C (Control). In S-ET, sensitivity is controlled by local constraints: Global and S-ET warm similarly by ~+1.5 °C at +100 W m−2, while Control and N-ET-only are lower.
• Cloud feedback decomposition (2041–2050 relative to PI): Global net cloud feedback increases from +0.16 to +0.26 W m−2 K−1 (+39%) in the Global constraint, mainly due to a less negative net optical depth (Tau, dominated by shortwave) and more positive altitude feedback. Over S-ET, total cloud feedback is less negative (−0.64 → −0.42 W m−2 K−1; +34%), driven by SW optical depth. Over N-ET, net cloud feedbacks are similar (0.17 vs 0.18), with more negative Tau offset by more positive altitude in the Global constraint.

The findings demonstrate that explicitly constraining both cloud-top and interior MPC phase corrects a pervasive low-SLF bias in NorESM2, improves the present-day energy budget (smaller TOA radiation bias, especially over the Southern Ocean), and increases transient climate sensitivity. By elevating the base-state SLF, the model avoids an exaggerated future shortwave cooling from ice-to-liquid transitions, yielding more positive (or less negative) cloud feedbacks—primarily via reduced shortwave optical depth feedback and a modestly enhanced altitude feedback. The top–interior SLF contrast is a robust global feature, implying that parameterizations assuming homogeneous mixing and efficient WBF are unrealistic. Hemispheric analyses reveal differing roles of local versus remote MPC changes: Northern extratropical sensitivity is strongly affected by remote (tropical and southern extratropical) cloud-phase teleconnections, whereas southern extratropics respond mainly to local constraints, likely reflecting slow oceanic adjustment. Overall, models that underestimate supercooled liquid in MPCs likely under-project warming; incorporating cloud-top phase constraints is thus crucial for realistic climate sensitivity and projections.

This work introduces a global observational constraint that distinguishes cloud-top and interior phases in mixed-phase clouds and applies it to NorESM2. The approach corrects present-day MPC phase biases, reduces TOA radiation biases (notably over the Southern Ocean), and increases projected 21st-century warming by about +1 °C for a given cumulative forcing under SSP5-8.5. The enhanced sensitivity arises from less negative cloud optical depth feedback and slightly more positive altitude feedback. The results highlight the importance of representing vertical phase inhomogeneity (liquid-dominant tops) and suggest that many CMIP-class models may underestimate future warming due to low base-state SLF. Future research should: (i) extend top–interior MPC constraints across multiple models, (ii) refine microphysical parameterizations (WBF efficiency, INPs, convective detrainment) to physically represent heterogeneity without ad hoc tuning, (iii) investigate teleconnections and timescales, especially the interaction with tropical clouds and Southern Ocean adjustment, and (iv) leverage additional active/passive satellite datasets and in situ observations to validate vertical cloud phase structure.

• Model-specific tuning: Constraints were implemented only in NorESM2; generalizability to other models remains to be tested.
• Nudged tuning phase: Atmosphere-only tuning prescribed ERA-Interim circulation, preventing circulation adjustments to microphysical changes; feedbacks in coupled mode may differ.
• Observational period: CALIOP-derived constraints rely on a 4-year period (2009–2013); while seasonality is small, longer records could better capture variability.
• Retrieval assumptions: CALIOP penetration depth and COD thresholds (top at COD ≥ 0.35; interior at accumulated COD < 3) and binary phase classification introduce uncertainties.
• Parameter degeneracy: WBF efficiency, INP factors, and convective detrainment temperatures can compensate each other; multiple parameter sets may match observations.
• Cloud feedback estimation windows: COSP2 output limits cloud feedback analysis to specific decades (e.g., 2041–2050), not continuous time.
• External teleconnections and ocean adjustment: The interpretation of remote influences, particularly on N-ET sensitivity and slow Southern Ocean responses, could be sensitive to model ocean dynamics and timescales.