Scientifically quantitative effectiveness assessment is a critical challenge for weather modification. To mitigate the limitation of traditional assessment methods, this study proposes an effectiveness assessment method for convective cloud precipitation suppression based on deep learning radar echo extrapolation. Firstly, a convective echo extrapolation dataset was constructed using 10 years (2013-2022) of SA-band radar observation data from the Beijing area. Secondly, through the training and comparative analysis of various mainstream deep learning models, including ConvLSTM, PredRNN++, MIM, SimVP, and TAU, PredRNN++ was selected as the foundational extrapolation model, and an observation-driven rolling forecast model was established based on it. By utilizing real-time observations to update model inputs, this strategy effectively mitigates the error accumulation associated with long-term extrapolation. This method was utilized to conduct a case study evaluation of a precipitation suppression operation on July 1, 2021. The results indicate that: (1) the composite radar reflectivity in the downstream (southeast side) of the analysis region showed significant weakening approximately 30 min after cloud seeding; (2) quantitative evaluation reveals significant precipitation suppression effects, with a reduction ratio ranging from 86.61% to 87.92% in total precipitation downstream of the operation stations.

Utilizing airborne measurements to investigate the response of microphysics to glaciogenic cloud seeding is instrumental in understanding and evaluating the seeding effect. In this study, the changes in microphysical characteristics of a supercooled liquid cloud induced by aircraft-based cloud seeding is analyzed. The data was collect in North China on Dec. 12, 2023, by using an S-band ground-based weather radar and airborne in-situ measurements. The results show that after cloud seeding, the cloud droplet number concentration and the supercooled liquid water content decreased by 48.5 % and 28.5 %, respectively, while the ice crystal number concentration increased from 0 to 100.8 L-1 on average, leading to a significant broadening of the ice particle spectrum. The ground-based weather radar observed distinct radar echo bands after the cloud seeding was completed. The enhanced radar reflectivity lasted for approximately 30 min before dissipation. A comparative analysis of five previous aircraft-based glaciogenic cloud seeding experiments in the same region revealed that the cloud seeding effect is sensitive to seeding-level temperature, cloud droplet number concentration, and supercooled liquid water content.

Hail forecasting using numerical models remains a challenge due to the uncertainties and deficiencies in microphysics schemes. In this study, we assessed the hail simulation performance of the two-moment Milbrandt-Yau (MY2) microphysics scheme within the Weather Research and Forecast (WRF) model by simulating three heavy rainfall events in Meiyu systems in which hail was rarely observed, rather than focusing on hail cases as done in previous research. Simulation results showed that the MY2 scheme produced noticeable hail in these rainstorms. Further analysis revealed that the overprediction of hail was caused by the imperfect graupel-to-hail conversion parameterization method adopted in the MY2 scheme. By incorporating the graupel spongy wet growth process, the modified scheme significantly mitigated the hail overforecasting. Moreover, the modified MY2 scheme kept the ability to simulate hail in real hail cases, demonstrating its ability to differentiate between heavy rainfall and hail events - a distinction the original scheme lacked. By comparing simulations of both rainstorms and hailstorms, it is concluded that the upward transport of large raindrops near the 0 degrees C level is critical for the graupel-to-hail conversion.

Warm rain prevails in clean marine clouds and is one of the most abundant types of precipitation in the world. However, warm rain is rarely observed in urban polluted atmospheric environments due to high number concentrations of aerosols generally suppress the occurrence of warm rain. Three aircraft observation of clouds were used during summer in Beijing area. Here, we documented warm rain processes similar to shallow marine clouds in the urban (Beijing) atmospheric environment after cleansing air pollution by extensive precipitation observed on 11 September 2019. Our results showed that, compared with aircraft observations of similar cumulus clouds under polluted conditions, large-scale precipitation can efficiently scavenge air pollution in Beijing and lead to marine-like clouds with warm rain. The average aerosol concentration (Na) near the cloud base in these clean cases was only 225 cm-3 with cloud droplet number concentration (Nd) of 36 cm-3, (maximum of 142 cm-3) which led to enhanced coalescence. The warm rain was initiated at 600 m above the cloud base. Our results highlight the significance of pollution scavenging by precipitation in warm rain production in urban environment.

Using airborne Ka-band probe radar (KPR) and cloud probe data obtained on 16 June 2021, the microphysical structures and particle characteristics in a stratiform cloud case with embedded convection were studied. At the mature stage, ice particles with sizes between 200 and 600 mu m, including plates and linears, were predominant in the stratiform cloud regions, and the maximum particle concentration was greater than 80 L- 1; however, particles larger than 600 mu m, such as plates, irregulars, and aggregates, were predominant at the edge of the stratiform cloud system, and the maximum particle concentration was less than 3.5 L- 1. The irregulars at the edge of the mature stratiform cloud region exhibited larger sizes compared to those in the inner region, while their occurrence frequency was significantly higher than that within the inner region of the mature stratiform cloud. In the embedded convection cloud region during the stratiform dissipation stage, particles ranging from 125 to 625 mu m were predominantly composed of plates, irregulars, and spheres (droplets). The particle concentration in this region was significantly higher compared to that observed in the stratiform cloud regions. We found that the average area of linear, spherical, plate, and dendritic particles in the stratiform cloud region was larger than that in the convective cloud region. The average area of graupel and aggregates in the embedded convection cloud region was larger than that in the stratiform cloud region. Moreover, we found that the larger the average area of graupel was, the lower the proportion of particles in the inner region of the mature stratiform cloud was.

The ice-phase microphysical characteristics of a stratiform cloud system over the Qilian Mountains in northwestern China on 15 September 2022 were analyzed via aircraft data. The stratiform cloud system developed under southwesterly flows at 500 hPa and was affected locally by topography. Synoptic features and aircraft observations revealed strengthened cloud development on the leeward slope. The ice particle habits and microphysical processes at heights of 6–8 km were investigated. The cloud system was characterized by extremely low supercooled liquid water content at temperatures between −4°C and −17°C. The ice particle concentrations ranged predominantly from 10 to 30 L−1, corresponding to ice water content ranging from 0.01 to 0.05 g m−3. Active ice aggregation was observed at temperatures colder than −10°C. The windward side of the cloud system exhibited weaker development and two distinct cloud layers. Intense orographic uplift on the leeward slope enhanced ice particle aggregation. The clouds on the leeside presented lower ice particle concentrations but larger sizes than those on the windward side. The influence of aggregation on the ice particle size distribution was reflected in two main aspects. One aspect was the bimodal spectra at −16°C, with the first peak at 125 µm and subpeak at 400–500 µm; the other was the broadened size spectra at −13°C due to significant aggregation of dendrites.

Several recent studies have reported complete cloud glaciation induced by airborne-based glaciogenic cloud seeding over plains. Since turbulence is an important factor controlling mixed-phase clouds, including ice initiation, snow growth, and cloud longevity, it is hypothesized that turbulence may have an impact on the seeding effect. To understand the role of turbulence in seeded clouds, idealized Weather Research and Forecasting (WRF) large eddy simulations over flat terrain were conducted for a shallow stratiform cloud in which complete glaciation was observed. The results show that the model can reasonably capture the magnitude and spatial distributions of radar echoes in seeded areas. Sensitivity tests suggest that, for this case, stronger turbulence enhanced the particle dispersion, the nucleation of silver iodide (AgI) particles, and the growth of ice crystals, which accelerated cloud glaciation, even though the condensation of droplets was also enhanced. The faster cloud glaciation intensified precipitation within a short time after seeding, while the liquid water was quickly consumed, leading to a decrease in precipitation rate in the further downwind areas. Such a transition from positive to negative seeding effect is more pronounced for seeding with a higher AgI release rate. This study provides strong evidence that turbulence plays a vital role in the physical chain of events associated with cloud seeding.

Riming is one of the key factors influencing the control of phase partitioning and precipitation formation in mixed-phase clouds. The riming growth rate of ice particles is strongly affected by the spatial distributions of liquid and ice particles. However, models often assume particles are homogeneously distributed in a given grid box when modelling riming, and few observational studies have quantified the effect of sub-grid heterogeneous particle distributions on riming. In this study, based on airborne in situ measurements made in isolated mixed-phase cumulus clouds, the impact of heterogeneous particle distribution on riming growth rate (Rrim) is quantified by defining an impact factor Frim, which is calculated using the observed mean Rrim divided by Rrim assuming a homogeneous particle distribution. The results show that Frim varies from ~0 to 2.8, demonstrating that the heterogeneous particle distribution has a strong nonlinear impact on riming. Additionally, it is found that Frim can be parameterized using the homogeneity of liquid-ice mixing and the range of normalized riming rate, which are related to the predictable condensed water content and ice particle concentration, respectively. Moreover, based on simulations driven by observed particle size distributions and assuming a static condition, it is suggested that riming has a significant feedback on Frim. These findings will be beneficial to improve the capability of models in simulating mixed-phased clouds and deepen our understanding of sub-grid scale cloud microphysics.

Warm-cloud hygroscopic seeding is widely used in precipitation enhancement, but the conditions under which seeding amplifies or suppresses rainfall remain unclear. Here, we use a two-dimensional slab-symmetric spectral bin microphysics model from Tel Aviv University to simulate a warm convective cloud that occurred over Hainan, China, on 11 May 2024, and design three sets of sensitivity experiments in which hygroscopic particles of different characteristic diameters are introduced under a fixed-mass injection constraint. We find that seeding with submicrometer particles (0.1–0.9 µm) systematically suppresses precipitation, with the strongest reduction for 0.1 µm particles. When super-micrometer particles (1–9 µm) are used, the precipitation response transitions from suppression to enhancement as particle size increases, and this transition occurs at about 2 µm. Seeding with ultra-giant particles (>10 µm) generally enhances rainfall and also advances its onset, with the enhancement strengthening up to ~60 µm before weakening for even larger particles. We further show that the transitional particle size at which the seeding effect changes sign decreases with increasing background aerosol loading, from maritime to polluted urban conditions. These results identify an environment-dependent critical particle size that governs the sign and efficiency of hygroscopic seeding in warm convective clouds.

This two-part study introduces a novel latent space data assimilation (LSDA) framework comprised of an autoencoder-observation to latent space, referred to as the AE-O2L network. This network allows observation-only analysis (LSDA-OOA) as demonstrated in Part I. The present work (Part II) extends AE-O2L to incorporate background fields into the data assimilation together with observations, referred to as observation and background assimilation (LSDA-OBA). As in Part I, the 2-m temperature (T2m) of a 1-km-grid numerical weather prediction (NWP) system over a complex surface in eastern China is used to train and test the AE-O2L-based LSDA-OBA framework. The result shows that assimilating backgrounds through the latent space improves LSDA performance. LSDA-OBA also outperforms the variational LSDA (LSDA-Var) method, especially when observations are sparse. By assimilating 40 real observations, LSDA-OBA achieves analyses of 933 test cases with an MAE of 0.72 K as verified against the seven data-withheld stations versus 0.76 K for LSDA-Var. Furthermore, LSDA-OBA runs two orders of magnitude faster than LSDA-Var. Sensitivity experiments show that the increment of each element of the latent vector corresponds to a mode perturbation in the NWP model space, and this relationship is roughly linear. We also demonstrate that the space spanned by these modes approximates the decoding space of the autoencoder. When performing an LSDA process, different modes are activated for different weather scenarios. Furthermore, the accumulated effect of the most active modes can approximate the final analysis with proper structures and intensity, which explains how LSDA works with such a small latent space.

Abstract. Hail forecasting using numerical models remains a challenge due to the uncertainties and deficiencies in microphysics schemes. In this study, we assessed the hail simulation performance of the 2-moment Milbrandt-Yau (MY2) microphysics scheme within the Weather Research and Forecast (WRF) model by simulating three heavy rainfall events in Meiyu systems in which hail was rarely observed, rather than focusing on hail cases as done in previous researches. Simulation results showed that MY2 scheme produced noticeable hail in these rainstorms. Further analysis revealed that the overprediction of hail was caused by the imperfect graupel-to-hail conversion parameterization method adopted in the MY2 scheme. By incorporating the graupel spongy wet growth process, the modified scheme significantly mitigated the hail overforecasting. Moreover, the modified MY2 scheme kept the ability to simulate hail in real hail cases, demonstrating its ability to differentiate between heavy rainfall and hail events—a distinction the original scheme lacked. By comparing simulations of both rainstorms and hailstorms, it is concluded that the upward transport of large raindrops near the 0 °C level is critical for the graupel-to-hail conversion.

Accurate short-term weather forecasting plays a vital role in disaster response, agriculture, and energy management, where timely and reliable predictions are essential for decision-making. Graph neural networks (GNNs), known for their ability to model complex spatial structures and relational data, have achieved remarkable success in meteorological forecasting by effectively capturing spatial dependencies among distributed weather stations. However, most existing GNN-based approaches rely on pairwise station connections, limiting their capacity to represent higher-order spatial interactions. Moreover, their dependence on supervised learning makes them vulnerable to spatial heterogeneity and temporal non-stationarity. This paper introduces a novel spatial–temporal pretraining framework, Hypergraph-enhanced Meteorological Pretraining (HyMePre), which combines hypergraph neural networks with self-supervised learning to model high-order spatial dependencies and improve generalization across diverse climate regimes. HyMePre employs a two-stage masking strategy, applying spatial and temporal masking separately, to learn disentangled representations from unlabeled meteorological time series. During forecasting, dynamic hypergraphs group stations based on meteorological similarity, explicitly capturing high-order dependencies. Extensive experiments on large-scale reanalysis datasets show that HyMePre outperforms conventional GNN models in predicting temperature, humidity, and wind speed. The integration of pretraining and hypergraph modeling enhances robustness to noisy data and improves generalization to unseen climate patterns, offering a scalable and effective solution for operational weather forecasting.

The growth processes and size distributions of ice particles in two stratiform clouds over the Qilian Mountains in northwestern China were investigated. A comparison of the moist Froude number between the two cases suggested stronger cross-barrier flow on 29 August than that on 16 August 2020, corresponding to relatively high precipitation amounts on 29 August. The stratiform cloud regions exhibited stable structures and dominated by vapor deposition. Convective regions contained more graupel, dendrites, and larger aggregates and grew primarily through riming and aggregation. Secondary ice production was responsible for the significant increase in ice particle concentrations in the convective region. Evidence of the Hallett-Mossop process was observed at temperatures between -3 degrees C and -8 degrees C and the estimated ice splinter production rate in the Hallett-Mossop region was positively related to the high ice particle concentration. Ice-ice collisional fragmentation might coexist at approximately -4 degrees C in the embedded convective region. The observed ice particle size distributions (PSDs) in stratiform regions were weakly dependent on temperatures ranging from -5 degrees C to -18 degrees C and exhibited bimodal distributions with breakpoints at 500-600 mu m. The observed PSDs in convective regions were also bimodal, but exhibited varied peaks due to the variation of dominant microphysical processes. The incomplete gamma fitting technique effectively characterized the PSDs from 300 to 2,000 mu m. The significant increase in N0 due to high ice particle concentrations was possibly attributed to secondary ice production processes within clouds, and the decrease in lambda corresponded to broadened ice particle spectra associated with aggregation.

Cloud seeding models are essential for understanding seeding mechanisms, yet their reliability remains insufficiently verified due to limited cases with confirmed seeding effects. On 19 March 2017, significant seeding signals were observed by multiple instruments following airborne cloud seeding over a stratiform cloud system with abundant supercooled water in northern China. This study performed an ensemble simulation of the case using two cloud microphysics schemes and three silver iodide (AgI) nucleation parameterizations, successfully replicating the vertical structure and evolution of the seeding‐induced cloud. The simulated seeding impact area, precipitation intensity, and changes in raindrop spectra closely aligned with observations. Results indicate that cloud seeding increased ice crystal amounts primarily through the deposition nucleation of AgI particles, activated the auto‐conversion of ice crystals to snow, enhanced snow deposition and riming processes, and ultimately increased surface precipitation through enhanced snow melting.

Ground-based cloud microphysical observations were continuously conducted at Mount (Mt.) Lu site from 2015 to 2020, which is located in the East Asian monsoon areas of eastern China. The statistical characteristics of cloud microphysics under different meteorological conditions were analyzed by using visibility, cloud droplet and raindrop size distribution data, and meteorological variables. The mean cloud droplet number concentration (Nc), liquid water content (LWC) and effective diameter (De) are 151 cm-3, 0.08 g m- 3 and 11.4 mu m, respectively. Nc and LWC observed at Mount Lu are lower than the observation results of cloud from other mountain stations all over the world. Broader cloud droplet spectra are correlated with higher temperatures and lower wind speeds. The clouds are classified into nonprecipitating, lightly-raining and raining clouds. Statistics show that lightlyraining and raining clouds account for 66% of all clouds. Nc and LWC of lightly-raining (drizzling) clouds are highest, while spectral width of raining clouds is largest. Spectral dispersion of raining clouds converges to the narrow range of 0.4-0.7. Notably, medium sized cloud droplets (11-40 mu m) promote the formation of large cloud droplets (40-50 mu m, pre-drizzle size) and drizzle drops which initiate the auto-conversion from condensation to collision-coalescence growth. This study relying on in-situ continuous cloud observation from mountain station are helpful to understand cloud microphysical structures in East Asian monsoon region, and fitting parameters also provide observational evidence for numerical models.

Convective initiation (CI) nowcasting is crucial for reducing loss of human life and property caused by severe convective weather. A novel deep learning method based on the U-Net model (named as CIUnet) was developed for forecasting CI during the warm season with eight interest fields of Himawari-8 Advanced Himawari Imager (AHI) and terrain height. The results showed that the CIUnet model produced probability forecasts of CI occurrence location and time with probability of detection (POD) at 93.3% +/- 0.3% and false alarm ratio (FAR) at 18.3% +/- 0.4% at a lead time of 30 min. Sensitivity and permutation importance experiments on the input fields of the CIUnet model revealed that the dif-ferences in brightness temperature for spectral channels were more critical for CI nowcasts than the original infrared chan-nel brightness temperatures. The brightness temperature difference between band 10 (7.3 mm) and band 13 (10.4 mm), which represents the cloud-top height relative to the lower troposphere, is identified as the most important input fields for CI nowcasting. The tri-spectral brightness temperature difference (TTD), which represents cloud-top glaciation, is ranked the second and it significantly reduced the FAR of the CI forecast. Using terrain heights as an extra input feature improved the POD, but slightly overestimated CI over complex terrain. In addition, a layer-wise relevance propagation (LRP) analyses was performed, and confirmed that the CIUnet model can effectively identify the crucial regions and features of the input fields for accurate CI prediction. Therefore, both permutation importance experiments and LPR analyses are useful for improving the CIUnet model and advancing the understanding of CI mechanisms.

The charge structure in thunderstorms may be strongly affected by different secondary ice production (SIP) processes, but has not been well understood. In this study, the impacts of three SIP mechanisms on microphysics and electrification in a squall line are investigated using model simulation, including the rime-splintering, ice-ice collisional breakup, and shattering of freezing drops. The parameterization of the three SIP mechanisms, a noninductive and an inductive charging parameterization are implemented in the spectral bin microphysics. The results show that with SIP processes included, the modeled radar reflectivity is more consistent with observation. It is found that both the mass and concentrations of graupel/hail are enhanced by SIP processes, while the diameter decreases. The mixing ratio of ice/snow decreases due to the rime-splintering, and increases in mixing ratio are due to the shattering of freezing drops. Particle charging is significantly affected by SIP, leading to a dipole structure of the total charge density, which includes a lower negative and an upper positive charge region. With both the noninductive and inductive charging considered, the charge carried by graupel/hail changes from negative to a bipolar structure, and the charge sign carried by ice/snow is inverted due to the SIP. The modeled lightning activity is enhanced by implementing all three SIP processes, while if only considering the rime-splintering process, the flash rate would be suppressed. The insights obtained from this study highlight the importance of considering different mechanisms of SIP in modeling the charge structure and lightning activity in thunderstorms. Electrification in thunderstorms is strongly related to microphysics, especially the ice-phase processes. However, the ice microphysics in deep convective clouds is very complicated, and the impacts of different ice-phase processes on cloud electrification are not well understood. One of the unresolved questions is how the charge structure in thunderstorms can be affected by various secondary ice production (SIP) mechanisms. Until now, only a small number of studies have investigated this issue, and most of them focused on a single SIP (rime-splintering process). However, recent studies have shown that other SIP mechanisms, especially the shattering of freezing drops and ice-ice collisional breakup, can greatly enhance ice generation in convective clouds. Therefore, the various mechanisms of SIP may potentially have strong impacts on the charge structure in thunderstorms. In this study, the parameterizations of three different mechanisms of SIP, as well as the noninductive and inductive charge parametrization are implemented in a spectral bin microphysics model scheme in WRF (Weather Research and Forecasting model). A squall line is modeled to investigate the impacts of the three mechanisms of SIP on charge structure and flash rate. The results highlight the importance of considering various mechanisms of SIP in modeling cloud electrification and lightning. The impacts of three secondary ice production (SIP) mechanisms on the electrification of a squall line is investigated using a model The SIP processes have strong impacts on the cloud microphysics, resulting in significant modification of charge structure The flash rate is suppressed by rime-splintering, while enhanced by ice-ice collisional breakup and shattering of freezing drops

Observation-validated cloud seeding simulation is valuable in assisting in evaluating seeding effect, but its sensitivity to microphysics schemes and cloud seeding parameterizations is rarely investigated. In this research three cloud seeding parameterizations (the Hsie, Demott and Xue parameterizations) are coupled with two microphysics schemes (the Thompson and Milbrandt schemes), to perform simulations of the rainfall suppression cloud seeding operation on a convective rainfall event occurred in North China on 1 July 2021, aiming to evaluate the seeding effect and investigate its sensitivity to microphysics schemes and seeding parameterizations. The differences of rainfall suppression effect between three seeding parameterizations are smaller than those between two microphysics schemes. The rainfall suppression ratios produced by the simulations configured with the Milbrandt scheme range from 16.3% to 24.7%, while those with the Thompson scheme range from 0.47% to 1.38%. The difference of the seeding effect between these two microphysics schemes arises from their different method in parameterizing the snow deposition growth process. In addition, in the seeding simulations with the Milbrandt scheme, the surface rainfall decrease region is followed by a rainfall increase region. This rainfall decrease-increase pattern is caused by the fact that the reduced graupel particles caused by cloud seeding falls to the ground earlier with its larger fall velocity, while the seeding-increased snow particle falls to the ground later due to its smaller fall velocity. This result suggests there is an ephemeral cloud seeding window for rainfall suppression operation.

Pulse hailstorms over southwestern China generally produce small hailstones with diameters <20 mm, causing nonnegligible damage to crops. This study examines the mechanisms behind the typical pulse hailstorms which frequently occur in plateau regions in southwestern China. Utilizing radar observations and the Cloud Model 1 (CM1) numerical model developed by George Bryan, the influence of environmental factors on pulse hailstorms and the associated small hailstones is investigated. The results indicate that: firstly, hailstones are primarily generated from frozen droplets, which further grow through collecting supercooled liquid water; secondly, a higher convective available potential energy (CAPE) and lower low-level relative humidity (RH) lead to the raising of convection initiation height; and lastly, higher CAPE values result in smaller hailstones on the surface due to the intensified hailstone melting process caused by higher melting layer and warmer temperatures within the warm layer (>0 °C) in this situation.

Robust water management is in intense demand in many water scarcity areas, such as arid and semi-arid regions in the world. As part of the regional water management strategy, rain enhancement is vital to replenish groundwater reservoirs, and the key challenge is how to assess its effectiveness. Some recent weather modification experiments attained cloud seeding effect through advanced in situ measurement coupled with accurate numerical simulation. However, there is still a lack of an objective and scientific approach to quantitatively evaluate the rain enhancement effect, especially for many non-randomized operational cloud seeding activities in China. In this study, we proposed a composite evaluation approach by analyzing two operational aircraft cloud seeding cases in stratus clouds in Shaanxi, China. By calculating the aircraft cloud seeding agent plumes, the target areas (as well as the control areas) of cloud seeding were dynamically and roughly determined. Physical properties, such as radar reflectivity and precipitation, were individually quantified in these areas. The cloud seeding effect was then evaluated by calculating the difference in parameter variation between target and control areas. This approach can be applied to qualitative analysis in a single aircraft cloud seeding operation and can also provide quantitative statistical results from multiple cloud seeding cases. We found that the average precipitation enhancement percentage of 18 operational aircraft cloud seeding cases is ~4.84%. Note that the homogeneity hypothesis of the seeding cloud, the error in the calculation of the target area, and the selection of control areas are the major uncertainties likely in the evaluation of the cloud seeding effect by this approach.