Unraveling ice multiplication in winter orographic clouds via in-situ observations, remote sensing and modeling

Mixed-phase clouds (MPCs) strongly influence radiative forcing and precipitation but are highly heterogeneous at scales below 100 m, affecting processes such as the Wegener-Bergeron-Findeisen mechanism and glaciation efficiency. Ice crystal number concentration (ICNC) is a key parameter; above -20 °C, scarce ice-nucleating particles (INPs) cannot explain observed ICNCs, implying an important role for secondary ice production (SIP). Models neglecting SIP underestimate ICNCs and misrepresent cloud radiative and microphysical evolution. SIP mechanisms include Hallett-Mossop rime splintering, ice-ice collisional breakup (BR), droplet shattering upon freezing (DS), and sublimational breakup (SUBBR). HM is limited to -8 to -3 °C and may be overestimated, while BR, DS, and SUBBR can operate at colder subzero temperatures. A major challenge is detecting SIP in global MPCs and constraining its mechanisms and intensity. Ground-based remote sensing, especially Doppler spectrograms from vertically pointing radars, often shows spectral multimodality in the dendritic growth layer (DGL, -20 to -10 °C), but interpretation is ambiguous without modeling support. This study leverages an NWP model with advanced SIP and a forward radar simulator to interpret zenith W-band radar observations of a winter orographic snowfall event in Greece, aiming to link characteristic Doppler spectral signatures and higher-order moments to specific microphysical processes, particularly SIP.

Prior work has established the widespread importance of SIP across laboratory, field, remote sensing, and modeling studies. HM is commonly included in models but restricted to a narrow temperature range and specific microphysical conditions. Radar polarimetry and multi-frequency observations have helped infer HM-related signatures (e.g., columnar crystals near -5 °C), yet SIP signatures at colder temperatures remain less explored. Remote-sensing studies reported Doppler spectral bimodalities and skewness changes within the DGL suggestive of new ice formation and interactions between fast and slow falling particles. Alternative SIP processes (BR, DS) may generate disk-like or small fragments and skew Doppler spectra toward slower-falling populations. Integrating models with forward radar simulators has been suggested as a robust path to evaluate SIP mechanisms inferred from radar. This study builds on these insights by combining a SIP-aware WRF configuration, radar simulation, and Doppler spectra analysis to identify BR-related fingerprints without polarimetric data.

- Field campaign and observations: The CALISHTO campaign (Mount Helmos, Greece; Oct 2021–Mar 2022) provided in-situ and remote sensing data at multiple elevations. A W-band FMCW vertically pointing radar (WProf, 94 GHz) at Vathia Lakka (VL, ~1850 m AMSL) collected full Doppler spectra and moments (equivalent reflectivity Ze, mean Doppler velocity, skewness) up to ~10 km AGL. W-band reflectivities were corrected for gaseous and cloud liquid attenuation using PAMTRA with atmospheric and liquid profiles from the SIP-aware WRF run. LWP was retrieved from an 89-GHz radiometer using a dedicated algorithm, with ~18% relative error for cloudy cases. - Modeling framework: WRF v4.0.1 with the Morrison double-moment microphysics (M09) was configured with three two-way nested domains (12/3/1 km), 97 vertical eta levels to 50 hPa, time steps 36/9/3 s, and 5-min output. Simulations covered Dec 17–19, 2021 (storm Carmel). A Cloud-Resolving Model Radar Simulator (CR-SIM) was configured to WProf specifications and driven by the nearest WRF grid cell to VL to generate forward-simulated radar observables. - Sensitivity simulations: Three runs isolated microphysical influences: 1) CONTROL: Default WRF primary ice production (PIP) with temperature-dependent schemes. 2) DEMOTT: PIP replaced by the aerosol-aware DeMott et al. (2010, DM10) INP parameterization using measured aerosol concentration >0.5 µm (0.30 cm−3). Bigg freezing for large raindrops retained. 3) ALLSIP: DEMOTT PIP plus SIP mechanisms: HM (Reisner), BR (Phillips et al.), DS (Phillips et al.), and SUBBR (Deshmukh et al.). HM thresholds rarely met due to colder temperatures. BR implemented with assumptions on rimed fraction (0.2 used after sensitivity tests), ice habit (planar or dendritic temperature-dependent formulation), and a characteristic size cutoff (efficiency limited to particles >100 µm). All BR fragments were added to cloud ice. DS activation required raindrops >50 µm, rarely met; SUBBR activated in subsaturated layers. - Evaluation strategy: Time–height Ze comparisons between WProf and CR-SIM outputs for all runs across three cloud periods: nimbostratus, external seeder–feeder, and a single-layer orographic cloud. Doppler spectra at selected times were examined for bimodality. Higher-order moment (skewness) time series were analyzed to identify persistent signatures in the DGL. Model microphysical tendencies (aggregation, deposition, riming, SIP production rates) were profiled to interpret radar signals and constrain mechanisms. LWP time series from radiometer were compared to model to assess liquid–ice partitioning.

- SIP necessity and model–observation agreement: Replacing default PIP with DM10 (DEMOTT) lowered simulated Ze, especially below -20 °C, but still underestimated observed Ze at warmer subzero temperatures. Activating SIP (ALLSIP) increased Ze across altitudes, with >10 dBZ enhancement at warmer subzero temperatures in the nimbostratus period, improving agreement with WProf. - Dominant SIP mechanism: BR dominated SIP during the two ice seeding periods, with localized DS mainly in the third period and localized SUBBR in subsaturated layers. HM remained inactive due to temperature and threshold constraints. BR production rates exceeded 10^4 particles L−1 s−1, with efficiency peaking in the DGL (around −15 °C) and increasing by about a factor of 10 when cloud tops were warmer than about −25 °C or within the feeder region at T > −15 °C. - Aggregation–BR synergy: Aggregation contours consistently enveloped BR regions, indicating collisions among dendrites/aggregates inside the DGL drive BR. Although aggregation is an ICNC sink, BR compensated and often increased ICNC. - Microphysical restructuring and partitioning: ALLSIP shifted elevated ICNCs from colder layers to warmer subzero temperatures, reduced vertical LWC availability, and lowered LWP. In the nimbostratus case, ALLSIP mean LWP was 77 g m−2 (observed 70 g m−2), roughly 40% (80%) lower than CONTROL (DEMOTT). In the seeder–feeder period, ALLSIP reduced LWP by 10% (35%) relative to CONTROL (DEMOTT) but still exceeded the observed mean of 45 g m−2. - Doppler spectral fingerprints: Persistent bimodal Doppler spectra emerged within the DGL in both the nimbostratus and seeder–feeder periods. Secondary slow-falling modes appeared at T > −17 °C, with reflectivities near −0.4 dBZ (nimbostratus) and −1.0 dBZ (seeder–feeder) and broad signatures indicative of cloud ice rather than supercooled droplets. The faster-falling mode aligned with aggregates; overlaps of particle trajectories and downdrafts could influence observed mean velocities. - Skewness as SIP indicator: Positive Doppler spectral skewness was repeatedly observed within −20 to −10 °C, coincident with spectral bimodality and model-predicted BR and aggregation. This consistent positive skewness indicates spectra skewed toward slower-falling particles after SIP onset, suggesting a robust, non-polarimetric fingerprint of BR-driven SIP within the DGL. - Seeder–feeder details: In the feeder cloud, Ze saturated due to non-Rayleigh scattering by large particles (sizes comparable to 3.2 mm wavelength); model snow size distributions showed particles >1 mm below 1 km. Near the surface, BR yielded ~2×10^2 small fragments L−1 s−1 while aggregation acted as a sink, resulting overall in elevated cloud ice concentrations (up to two orders of magnitude higher than DEMOTT near the surface) and improved Ze and LWP agreement in ALLSIP. - Dry-layer mismatch: At higher altitudes (−16 to −25 °C), observed Ze reductions (likely from a dry/subsaturated layer and sublimation) were not replicated by any configuration. ALLSIP predicted SUBBR up to ~10 L−1 s−1 in this layer and shifted size distributions toward smaller particles, partially reducing Ze biases.

The study addresses the challenge of detecting and attributing SIP in mixed-phase orographic clouds by combining SIP-aware NWP modeling and forward radar simulation with vertically pointing W-band observations. Results show that ice-ice collisional breakup (BR) within the DGL is the dominant SIP mechanism during ice seeding events, amplifying ICNCs by 1–2 orders of magnitude, enhancing precipitation, and increasing reflectivity near the surface. The observed spectral bimodality and persistent positive skewness within −20 to −10 °C are consistent with model-predicted BR and aggregation, providing a physically grounded, qualitative fingerprint for SIP in the absence of polarimetric data. In seeder–feeder configurations, BR-generated small ice in the feeder cloud grows by deposition and riming, improving agreement with observed Ze and LWP. While a dry layer between seeder and feeder clouds challenged all model setups, it did not alter the main conclusion that BR-driven SIP strongly shapes hydrometeor populations and radar observables in the DGL. These findings demonstrate that model–observation synergy can identify microphysical mechanisms from standard radar moments and spectra, enabling broader exploitation of existing radar archives to constrain SIP in models and reduce MPC-related biases.

This work demonstrates that secondary ice production via ice-ice collisional breakup decisively impacts winter orographic mixed-phase clouds, especially during ice seeding events and within the dendritic growth layer. SIP-aware WRF simulations coupled with a radar simulator reproduce key radar signatures observed by a W-band profiler, including enhanced reflectivity at warmer subzero temperatures, Doppler spectral bimodality, and persistent positive skewness aligned with BR and aggregation. The identified skewness-based fingerprint provides a practical pathway to infer SIP occurrences using widely available vertically pointing radar records, even without polarimetry. The approach offers a means to constrain microphysical processes in models, improve representation of ICNCs and hydrometeor size distributions, and ultimately reduce radiative and precipitation biases associated with MPCs. Future work should extend the analysis to long-term, multi-site datasets to establish statistical robustness, incorporate polarimetric and multi-frequency observations for quantitative retrievals, refine SIP parameterizations (e.g., BR efficiency, DS activation), and integrate advanced remote sensing techniques to map SIP intensity and mechanisms across diverse environments.

- Observational constraints: Lack of polarimetric radar prevented direct habit identification; potential vertical beam misalignment and advection/turbulence could affect Doppler spectra and mean Doppler velocity interpretations; non-Rayleigh effects led to Ze saturation in the feeder cloud. - Model–observation mismatches: Timing offsets for cloud events; discrepancies in winds and relative humidity with respect to ice; difficulty capturing dry/subsaturated layers between seeder and feeder clouds. - Microphysics parameterization uncertainties: DEMOTT INP scheme omits some INP types (e.g., biological), potentially impacting PIP; assumed rimed fraction (0.2) for BR; BR efficiency limited to particles >100 µm; ice habit not explicitly resolved; DS rarely activated due to raindrop size threshold (>50 µm) in M09; HM inactive under studied conditions. - Case specificity: Findings are based on a single storm event in complex orography; broader generalization requires longer-term and multi-case statistics. - Liquid partitioning biases: ALLSIP improved but did not fully match low observed LWP during the seeder–feeder period.