Cloud Seeding Simulations Demonstrate Reduced Surface Hail Kinetic Energy

High-resolution three-dimensional numerical simulations of a severe supercell thunderstorm indicate that operational cloud seeding significantly modifies surface hailfall characteristics and reduces hail kinetic energy, according to research conducted by atmospheric scientists Darko Savić, Vladan Vučković, and Dragana Vujović at the University of Belgrade. The study evaluated two established glaciogenic seeding methodologies using the Advanced Regional Prediction System model upgraded with a three-moment microphysics scheme and a two-moment aerosol module. Modeled outcomes revealed that introducing silver iodide aerosol particles into hail embryo formation zones altered microphysical conversion pathways, resulting in substantial reductions in both total ground-level kinetic energy and the specific land surface areas subjected to the most destructive hailfall intensities. Because kinetic energy flux serves as an established physical proxy for agricultural losses, these explicit simulations provide critical physical insights into how operational cloud seeding can mitigate storm damage while highlighting complex spatial redistributions at the surface.

Assessing the physical and operational efficacy of hail suppression has historically remained a formidable scientific challenge due to the intense non-linearity, rapid temporal evolution, and localized spatial scales of convective cloud systems. Operational hail mitigation programs across more than fifty countries have deployed glaciogenic agents such as silver iodide for decades under the beneficial competition hypothesis. This theoretical mechanism proposes that injecting an abundant population of artificial ice-nucleating particles into the hail embryo formation zone induces the freezing of supercooled liquid droplets and raindrops, producing a vastly elevated concentration of competing hail embryos. Consequently, these numerous competing hydrometeors divide the available supercooled liquid water, which limits individual stone growth and yields smaller hailstones that are more susceptible to complete or substantial melting before reaching the ground. Despite extensive operational adoption using rockets, aircraft, and ground generators, isolating a unambiguous physical signature of suppression from the natural variability of convective storms has long eluded observational verification.

To investigate these microphysical mechanisms under controlled conditions, the researchers employed version 5.3.4 of the Advanced Regional Prediction System, configured as a cloud-resolving model over a computational domain spanning 120 kilometers horizontally and 17.5 kilometers vertically. The spatial integration utilized a fine grid spacing of 500 meters horizontally and 250 meters vertically, operating with a large dynamical time step of 2.0 seconds and a split acoustic mode time step of 0.5 seconds across a three-hour storm lifecycle. Convection was initiated within a horizontally homogeneous environment using an idealized four-kelvin ellipsoidal thermal perturbation. Environmental conditions were prescribed from a well-documented composite radio-sounding based on observations from Fort Sill and Elmore City during the historic Del City supercell outbreak of May 20, 1977. This sounding was characterized by extreme thermodynamic instability, featuring a surface-based convective available potential energy of 3381 joules per kilogram, zero convective inhibition, a surface dew point temperature near 21 degrees Celsius, and intense vertical wind shear that fostered the rapid development of a right-moving supercell.

A critical technical enhancement of the numerical framework was the integration of a two-moment aerosol scheme with prognostic equations governing the mixing ratio and number concentration of aerosol particles in ambient air and within each hydrometeor category. Aerosol particles were represented by a continuous gamma distribution matching the physical and chemical properties of silver iodide, accounting for four distinct ice-nucleation modes: depositional, condensation-freezing, immersion, and contact freezing. The model incorporated all primary aerosol scavenging processes, including Brownian diffusion, phoresis, hydrodynamic capture, and electrostatic attraction under an electrification parameter set to a moderate cumulonimbus charge level. Tracking aerosol mass and number alongside six hydrometeor categories—liquid cloud droplets and raindrops, plus cloud ice, snow, graupel, and hail—permitted explicit monitoring of reagent injection, scavenging by hydrometeors, phase change initiation, and downstream redistribution through collision, accretion, and sedimentation.

The study directly contrasted two operational seeding protocols against an unseeded control simulation: the rocket-based methodology of the Republic Hydrometeorological Service of Serbia, designated RHSS, and the multi-parameter radar criteria proposed by Abshaev and colleagues, designated AB23. Under the RHSS protocol, which has operated in Serbia since 1967, seeding is triggered by specific radar reflectivity and isotherm criteria, confining injection between the zero and minus-ten-degree Celsius isotherms within horizontal radar reflectivity boundaries of 15 to 45 decibels relative to Z. In contrast, the AB23 protocol categorizes convective cells into four hierarchical seeding objects using more intricate criteria that integrate maximum reflectivity, vertical reflectivity heights, and integrated liquid water content above the freezing level. The AB23 seeding zone extends from 35 decibels relative to Z out to five kilometers beyond the radar echo boundary into cloud-free inflow air, delivering reagent cycles at distinct temporal intervals. In the model experiments, total injected silver iodide mass reached approximately 50 kilograms in RHSS and 70 kilograms in AB23.

Microphysical budget analyses across the simulation domain revealed immediate responses following aerosol introduction, primarily originating within liquid hydrometeors in the supercooled hail embryo formation zone. Injected silver iodide was rapidly scavenged by cloud droplets and raindrops, triggering sharp spikes in immersion and contact freezing. Immersion freezing of raindrops emerged as the predominant source term generating new hailstone numbers, while contact freezing contributed substantially to hail numbers despite adding negligible mass. The proliferation of newly formed ice crystals accelerated the conversion of cloud ice into snow and graupel through intense riming and accretion. Snow exhibited the most pronounced volumetric increase of any solid hydrometeor category, with elevated snow melting acting as a massive secondary source of raindrops. Because the newly generated raindrops were far more numerous, domain-averaged mean raindrop diameters decreased significantly, which subsequently diminished the efficiency of raindrop collection by existing hailstones and curtailed hail mass growth.

Surface precipitation results demonstrated that both operational seeding methodologies achieved meaningful reductions in cumulative hail mass and total ground-level kinetic energy compared to the unseeded control run. Total cumulative hailfall mass reaching the ground decreased by 26.4 percent under the RHSS methodology and by 14.0 percent under the AB23 methodology, while cumulative rainfall mass also underwent modest declines. Total surface hail kinetic energy integrated across the entire three-hour simulation dropped by 27.0 percent in the RHSS simulation and by 17.9 percent in the AB23 simulation relative to the unseeded baseline value of 6.36 times ten to the tenth joules. Furthermore, land surface areas exposed to moderate hail kinetic energy flux exceeding 100 joules per square meter decreased by 36.6 percent under RHSS and by 38.6 percent under AB23, underscoring that both operational seeding designs successfully mitigated intermediate surface impact footprints across the simulated terrain.

The suppression effect was most pronounced when evaluating the high-intensity impact areas that correlate with severe agricultural damage, defined by a cumulative kinetic energy flux exceeding 300 joules per square meter. In the unseeded control storm, this critical damage swath covered 12.5 square kilometers of surface area. Under the RHSS seeding methodology, the high-intensity impact footprint contracted by 66.0 percent, whereas the AB23 protocol produced an 88.0 percent reduction in this extreme exposure zone. Authors attributed the differential performance between the two operational schemes to their divergent aerosol delivery strategies. The direct injection into the hail embryo formation zone prescribed by RHSS produced rapid, aggressive droplet freezing and higher total scavenging rates, substantially suppressing domain-wide total kinetic energy. Conversely, the extended exterior seeding zone and staggered temporal cycles of AB23 provided a steadier, lower-concentration influx of ice nuclei, preventing intermittent operational windows during which hailstones could escape competition and grow into extreme size classes.

Despite substantial integrated reductions in kinetic energy and high-intensity damage footprints, local surface hail patterns displayed remarkable spatial complexity, shifting, and localized intensification. At specific ground locations within the main storm track, seeding caused the near-complete disappearance of large hailstones exceeding two to four centimeters in diameter, accompanied by marked drops in peak hailfall intensity. Paradoxically, neighboring peripheral areas that experienced no hailfall in the unseeded control run received moderate hail accumulations following seeding. Researchers explained that continuous latent heat release from induced freezing, combined with modified hydrometeor loading, altered the storm updraft velocity, structural tilt, and trajectory. These secondary dynamical perturbations displaced precipitation shafts and stretched hail concentration fields, demonstrating that point-by-point ground measurements can register apparent hail increases even when domain-wide damage metrics reflect significant mitigation.

Statistical evaluation using the non-parametric Wilcoxon signed-rank test confirmed that the observed differences in surface hail diameter and kinetic energy flux were statistically significant rather than numerical artifacts. Spatial neighborhood testing across three-by-three grid clusters identified significant microphysical shifts at confidence levels of p less than 0.05 and p less than 0.01 as early as 30 minutes into the simulation, roughly twelve minutes after the initiation of seeding. Domain-wide temporal distributions showed that both seeding methods significantly lowered median kinetic energy flux and compressed upper extreme percentiles during early storm stages between 30 and 60 minutes. The AB23 scheme exhibited greater temporal stability by maintaining significant separation from control values through the mid-stage between 90 and 120 minutes, whereas RHSS temporarily converged toward baseline. During final storm decay between 150 and 180 minutes, both schemes exhibited an upward rebound in median hail sizes, while domain-wide kinetic energy flux differences subsided to statistically non-significant levels.

The authors emphasized several fundamental physical and methodological limitations that frame the scope of their findings. The simulations focused on an idealized single-storm case initialized from a historic sounding over flat terrain without high-resolution surface orography or environmental cloud condensation nuclei variations. A substantial fraction of injected silver iodide—54 percent in RHSS and 65 percent in AB23—remained unscavenged after three hours, remaining suspended as submicron particles with potential implications for downwind inadvertent weather modification. Microphysical parameterizations for ice nucleation and aerosol scavenging introduce additional model uncertainties that require rigorous observational benchmarking. Consequently, the research team concluded that comprehensive multi-storm ensemble simulations, realistic three-dimensional meteorological initializations, and integrated field validation combining dual-polarization radar observations and surface hailpad networks will be essential before these numerical results can be definitively translated into operational agricultural protection policies.

Subject of Research: Numerical modeling of hailstorm cloud seeding and surface hail suppression efficacy

Article Title: Explicit numerical simulations of hailstorm seeding with a three–moment microphysics cloud–resolving model: a comparison of two operational methodologies for hail suppression

Article References: Savić, D., Vučković, V., & Vujović, D. (2026). Explicit numerical simulations of hailstorm seeding with a three–moment microphysics cloud–resolving model: a comparison of two operational methodologies for hail suppression. Atmospheric Chemistry and Physics, 26(21), 15785-15811. https://doi.org/10.5194/acp-26-15785-2026

Image Credits: AI Generated

DOI: 10.5194/acp-26-15785-2026

Keywords: cloud seeding, hail suppression, microphysics, numerical weather prediction, supercell thunderstorms, silver iodide, kinetic energy flux, atmospheric physics, Explicit, numerical, simulations, hailstorm