Abstract:To enhance visibility observation capability along highways and improve the accuracy of visibility detection based on video surveillance images, this paper builds upon prior research by focusing on the ordinal relationships between visibility levels. By transforming the traditional visibility level classification task into a series of ordered binary classifiers, it imposes consistency constraints on visibility level detection results. This approach leads to a novel visibility detection method based on ordinal consistency constraints, resulting in more stable model predictions. Two datasets collected from real highways—JS-FHVI (Jing-Shi Foggy Highway Visibility Images) and DG-FHVI (Da-Guang Foggy Highway Visibility Images)—are used for experimental design and validation. Images are randomly partitioned into training (70%), validation (10%), and test (20%) sets. Visibility is categorised into six levels: ≤50 m, 50-100 m, 100-200 m, 200-500 m, 500-1000 m, and ≥1000 m. Image annotations are derived from meteorological station data nearest to the camera locations, with verification and corrections by professional meteorologists. A model based on ordinal consistency constraints (OCC) is designed, and an ordinal consistency constraint loss function is introduced, enabling the model to prioritise samples with inconsistent ordinal predictions. Further ablation studies compare four ordinal consistency measurement methods, revealing that Intersection over Union (IoU) yields optimal results and is adopted as the consistency metric. Subsequently, five scaling functions—linear, exponential, exponential square root, logarithmic, and softmax scaling—are evaluated, with exponential square root scaling achieving the highest classification accuracy. Based on these findings, two weighting strategies for hard samples are designed and tested. The strategy of increasing weights for hard samples proves more effective. To verify the performance of the proposed method, AlexNet, VGG16, ResNet-18, ResNet-50, EfficientNetB1, original ordinal regression, and the ordinal regression method with Ordinal Consistency Constraint (OCC) are employed as backbone networks to detect the visibility level of each highway image. The results demonstrate that the detection accuracy of the OCC-based method surpasses all other approaches. On the JS-FHVI test set, the method’s accuracy achieves 93.75%, and on the DG-FHVI test set, it achieves 86.94%. Its effectiveness is further validated through ablation studies. Cross-testing is conducted using models trained on JS-FHVI, DG-FHVI, and a unified model trained on merged data. Results demonstrate that models trained on specific highway data perform better locally, while the unified model exhibits superior generalisation across scenarios. Extensive experiments confirm that the proposed OCC-based method significantly enhances visibility detection accuracy. The current work focuses on discrete-level visibility detection; future efforts may extend to continuous numerical estimation to improve practical precision. Additionally, given the dynamic evolution of fog, short-term visibility forecasting or trend prediction remains a critical direction for further research.

Abstract:Guizhou Province ranks the fifth in China in terms of expressway mileage, yet the deployment of traffic meteorological stations remains sparse. Frequent low-visibility weather events, influenced by mountainous terrain, pose severe threats to road safety. Deep learning-based image recognition methods for visibility level classification assist in rapid assessment for traffic management during foggy conditions. However, conventional neural networks (CNNs) suffer from limited classification accuracy due to insufficient global feature extraction. To address this issue, this study integrates both local and global feature representations by constructing a hybrid network model, cTrans-Net, which combines CNNs and Transformer architectures. Surveillance video images from typical mountainous expressway sections in Guizhou are selected for model training and testing to achieve precise five-level visibility classification (L0-L4) that impacts traffic safety. Experimental results demonstrate that cTrans-Net achieves an overall accuracy of 89.17%, with an area under the curve (AUC) of 0.9822, outperforming several mainstream deep learning models. The overall accuracy of the independent validation set remains 87.75%. In evaluating low-visibility conditions (<500 m), which significantly affect traffic, the model attains the highest recognition recall of 90.66% for the most frequently occurring L2 level (100-200 m). For the less frequent L0 (>500 m) and L4 (≤50 m) categories, the recognition recall reaches 92.31% and 90.45%, respectively, indicating the model’s robustness in handling imbalanced datasets. Feature visualisation reveals that cTrans-Net effectively focuses on key regions such as road markings and fog distribution, which are critical for visibility assessment. This study provides a technical solution for fog-related visibility recognition in intelligent transportation systems, offering practical value for real-world applications in mountainous expressway environments. The proposed cTrans-Net demonstrates strong adaptability to imbalanced data scenarios and exhibits superior performance in critical low-visibility conditions, making it a viable tool for enhancing traffic safety management under adverse weather conditions.

Abstract:To enhance the efficiency of real-time meteorological data access for local disaster prevention and mitigation authorities, the Jiangsu Meteorological Bureau spearheads the development of a specialised Decision-Meteorological Service Mobile Application (APP). The APP, driven by user needs, adopts a strategic approach to prioritise key decision-making factors while enhancing the user experience. The APP integrates multi-source meteorological data, expertise, and diverse application scenarios to provide comprehensive and scenario-based meteorological services. By deploying behaviour-driven and weather-integrated recommendation algorithms, it enables personalised service customisation for different user groups. Leveraging the “DeepSeek-R1-Distill-Qwen-32B” large language model and integrating speech recognition and dynamic dialogue state tracking technologies, the meteorological conversation robot “Su Xiaoce” is developed to support intelligent question-and-answer and function navigation to deliver conversational services. Given its precision-tailored and operationally effective service, the APP is fully implemented across key disaster prevention departments in Jiangsu Province. The system provides preemptive early warnings, real-time emergency alerts, and sustained operational support during disaster weather prevention and mitigation, offering a replicable and promotable technical path for the construction of decision meteorological service capabilities.

Abstract:Based on the daily precipitation observation data from 109 national stations in Shanxi from July to August 1990 to 2022, the ECMWF numerical precipitation data, and the EFI (Extreme Weather Forecast Index) products from the ECMWF ensemble prediction system from July to August 2022, the different percentile values (maximum, 99%, 95%) of historical extreme precipitation in Shanxi are calculated using the percentile method. The actual situation of extreme precipitation from July to August 2022 is analysed, and the forecast performance of two precipitation forecast products at different lead times (0-24 h, 24-48 h, 48-72 h) is evaluated. A new method for forecasting extreme precipitation is proposed by combining the EFI with the ECMWF high-resolution model. The results show that: (1) The historical extreme precipitation values in Shanxi in July are generally higher than those in August, and the possibility of abnormal extreme precipitation is greater. The extreme precipitation values in the north of Shanxi are lower than those in the central and southern regions, and the centre of the maximum extreme precipitation value has a significant relationship with the topographic distribution. (2) The precipitation in most areas of Shanxi exceeds the historical extreme precipitation in 2022, with occurrences in the central and southern regions in July, and in most areas in August. (3) Both the EFI and the ECMWF models have certain forecasting capabilities for extreme precipitation in July and August 2022. The forecasting ability of EFI for more extreme precipitation is superior to ECMWF’s. (4) The new extreme precipitation forecast based on the ECMWF model improves the forecasting effect on the 95th percentile values of western Shanxi, the 99th percentile values, and the maximum extreme precipitation of the central region in July, the 95th percentile values in central and southern Shanxi, and the 99th percentile values of the northwestern region in August. The forecasting effect for more extreme precipitation improves significantly.

Abstract:Numerical models serve as critical reference bases for weather forecasting. Effective application of numerical models for forecasting first requires understanding the models’ forecasting performance, which is derived from verification. This study researches the forecasting capabilities of different numerical models in terms of precipitation associated with the Northeast Cold Vortex in Jilin Province. Utilising hourly precipitation observation data from 1436 stations in Jilin Province during the Northeast Cold Vortex period from May to September 2021-2023, a comparison analysis is conducted between the precipitation forecasting products of six numerical models, namely, EC, CMA-MESO, CMA-GFS, CMA-TYM, CMA-SH9, and CMA-BJ. The aim is to reveal the characteristic differences in forecasting capability and deviation contributions between three-hour-interval precipitation forecasts in plain and hilly areas using quantitative assessment, graded assessment, and diurnal variation forecast comparison. The results indicate that: (1) The observed precipitation associated with the Northeast Cold Vortex and the precipitation forecasting capability of numerical models are closely related to topographic distribution. In the plain group, both the precipitation amount and the number of observation stations are smaller, while the precipitation intensity is stronger, and the mean relative error (MRE) and mean absolute error (MAE) of the stations are larger. In contrast, the hill group shows opposite trends: larger precipitation amount, more station numbers, weaker intensity, and smaller MRE and MAE. (2) EC demonstrates the best performance in forecasting diurnal variations of precipitation in both plain and hilly groups; however, it lacks the ability to predict the peak of precipitation intensities, with a notable overestimation of precipitation frequency occurring around midday. In contrast, CMA-MESO and CMA-SH9 are more effective in capturing the diurnal peak or trend characteristics of precipitation frequency and intensities. (3) The weak precipitation forecasts under various models are overly frequent, while the forecasting capabilities for moderate and heavy precipitation are influenced by topographical differences, where the performance in hilly areas surpasses that in plain areas. Additionally, CMA-MESO and CMA-TYM outperform global models, with global models exhibiting significant dry deviations. (4) The forecasting skill clock plots indicate that the excessive forecast of frequency by EC around midday is mainly due to overforecasting weak precipitation, while the significant underforecast of strong precipitation throughout the day results from missed occurrences. Furthermore, CMA-MESO exhibits stronger forecasting capabilities for heavy precipitation compared to EC, successfully predicting evening peak values and the occurrence of heavy precipitation stations. Its errors arise from location deviation in the forecasts.

Abstract:The study of winter snowfall holds great significance for Changji Prefecture, located in the northwestern arid region. Abnormal winter snowfall frequently leads to heavy snowfall, which results in snow disasters and avalanches. Furthermore, blizzards and strong winds often cause snowdrifts that block traffic and severely reduce visibility. Therefore, it is essential to conduct investigations in this area. A notable example occurred in December 2015, when Changji Prefecture experienced an extremely rare blizzard characterised by widespread snowfall and a high number of stations reporting heavy snow, with many stations breaking historical records. Therefore, systematic research on winter snowfall is necessary. This study focuses on the Changji area. Based on daily precipitation records from 10 national meteorological stations between 1982 and 2022, the characteristics of snowfall amount, snowfall intensity, and the number of snowfall days are analysed using methods such as linear regression, the Mann-Kendall test, and wavelet analysis. The results show that, based on the winter daily snowfall observation data from 10 national stations in Changji Prefecture from 1982 to 2022, this study analyses the climatological characteristics of winter snowfall amount, snowfall days, and snowfall intensity using methods such as linear trend estimation, Morlet wavelet analysis, Mann-Kendall mutation test, and ArcGIS inverse distance weighting interpolation. The results reveal several noteworthy findings: (1) Significant regional differences exist in winter snowfall changes. The snowfall amount shows a significant increasing trend in the plain area and the Beitashan region, with rates of 2.6 mm/10a and 2.8 mm/10a, respectively, while the trends in the Tianchi and Caijiahu regions are not significant. The dominant months contributing to the increase differ, with December being the primary contributor in the plain area and February in the Beitashan region. (2) The reduction in snowfall days across various regions is consistently attributed to a significant decrease in light snowfall days. Concurrently, the number of heavy snowfall days increases significantly in the plain and Tianchi areas, and the winter snowfall intensity enhances significantly, indicating a trend towards more intense snowfall events. (3) Topography is a key factor influencing snowfall distribution, with both snowfall amount and heavy snowfall days exhibiting a spatial pattern of high values in mountainous areas (Tianchi), medium values in oasis-plain areas, and low values in desert (Caijiahu) and arid mountainous (Beitashan) areas. This study reveals the characteristics of “increasing amount, decreasing days, and increasing intensity” of winter snowfall in Changji Prefecture against a warming and wetting background, providing references for regional water resource assessment and snow disaster prevention.

Abstract:On July 25, 2022, a widespread and extreme thunderstorm gale event suddenly struck Henan Province, with the maximum instantaneous wind force reaching Level 13 and the maximum instantaneous wind speed at 12 national meteorological stations exceeding the historical extreme values for the same period since their establishment. Based on multi-source data, including conventional observation data, FY-4 high-resolution satellite data, dual-polarisation radar data, and minute-level ground observation data, this study analyses the evolutionary characteristics and formation mechanisms of the extreme gale-force winds in this event. The results indicate: (1) The event occurred on the margin of the subtropical high under the background of an eastward-moving trough, which was jointly triggered by surface convergence lines and weak cold air. During the initial stage, strong radiative warming occurred at the surface in the afternoon, combined with the development of warm and moist advection in the lower atmosphere overlapped with the upper-level cold trough, forming a strongly thermally unstable stratification. The upper-dry and lower-moist structure was conducive to the occurrence of thunderstorm gales. During the development and maintenance stage, the eastward movement of the trough led to a significant increase in dynamic lifting, vertical wind shear, and carrying layer wind, which was conducive to the maintenance of gales. (2) This extreme gale event exhibited three distinct stages: In the initial stage, small bow echoes formed by the forward propagation of scattered convection in northwestern Henan, with localised extreme gales first occurring at the rear of these echoes and near supercell storms. During the development stage, the system gradually organised into linear convection in central and western Henan, with extreme gales concentrated in the regions near the apex of the bow echo embedded in the linear convection and the strong divergent regions on its rear. In the maintenance and weakening stage, the linear convection evolved into a larger-scale typical bow echo in central and eastern Henan, with extreme thunderstorm gales appearing sporadically near the locations with the maximum curvature of the bow echo and the hook echoes. Mesocyclones, γ-mesoscale vortices, low-level gale zones, deep radial convergence, and areas of low differential reflectivity factor (Z_DR) and low specific differential phase (K_DP) had certain indicative significance for the early warning of extreme thunderstorm gales. (3) Strong downdrafts, downward momentum transport, cold pool density currents, and topographic effects were the primary causes of this extreme thunderstorm gale event. In the initial stage, negative buoyancy was the main factor leading to strong downdrafts. During the development stage, the combined dynamic forcing of negative buoyancy, precipitation drag, and γ-mesoscale vortices dominated the formation of strong downdrafts, superimposed with the synergistic effect of downward momentum transport, cold pool density currents, and downdraft divergent airflow, leading to an increase in the intensity and expansion of the extreme gales. In the maintenance and weakening stage, downward momentum transport and cold pool effects were predominant. Additionally, the convergence lines caused by topography significantly promoted the organised development of the convective system, while the superposition of local trumpet-shaped and narrow-tube terrain effects further enhanced the extremeness of near-surface wind speeds.

Abstract:The implementation of weather modification operations involves the advance planning and deployment of equipment and personnel. If the direction of the weather system and the nature of the cloud formations can be accurately forecasted one week in advance, it plays a significant and meaningful role in the allocation of field operation resources. Supercooled water clouds are the primary targets for weather modification operations such as rain enhancement, rain suppression, and hail suppression. The formation of supercooled water depends on specific environmental conditions, including temperature, humidity, and vertical motion. To support operational forecasting for weather modification, it is necessary to accurately characterise supercooled water and its environmental fields at least one week in advance. Using the temperature and humidity forecast parameters from the CMA-GFS global model, the cloud-top temperatures for different cloud layers are calculated. The CIP (Crystal Icing Potential) algorithm is improved into a supercooled water content potential algorithm. By establishing a relational function between SLW (Supercooled Liquid Water) content and key parameters including temperature, relative humidity, and cloud-top temperature, this enhanced algorithm enables effective identification of SLW potential conducive to precipitation enhancement. A cold cloud seeding potential forecast product with a 168-hour forecast period is developed. The supercooled water potential algorithm is evaluated using both the binary classification method for icing events and the probability of detection (POD) method, incorporating 91 aircraft icing observations. Additionally, the cold-cloud seeding potential forecast results for spring 2024 are validated against 10 weather modification aircraft observations. The results show that the supercooled water potential algorithm effectively represents the likelihood of supercooled water occurrence. Validation of the supercooled water potential algorithm is conducted using 91 aircraft observation cases. When applying a 100% threshold, the icing detection rate reaches 54.5%. The icing detection rate is 97.7% and the no-icing detection rate is 66.0% when using a 15% threshold. The TSS score is 0.74 when the threshold is 25%. The cold cloud seeding potential forecast product is applied during the spring 2024 weather modification operations for rain enhancement. Out of 8 flight cases involving icing, the forecast accuracy is 87.5%, and both the 2 flight cases without icing are accurately predicted. The predictable forecast time ranges between 60 to 168 hours, and the potential reflects the intensity of icing and supercooled water, showing certain advantages over quantitative supercooled water forecasts. This product provides technical support for the process forecasting and operational outlook of weather modification activities up to one week in advance.

Abstract:To gain an in-depth understanding of the raindrop size distribution (DSD) and microphysical characteristics of precipitation in the Yellow River Delta Nature Reserve, and to better understand the regulatory effects of precipitation on vegetation, hydrology, and ecosystems, we analyse data from 2021 to 2024 collected by precipitation disdrometers and automatic weather stations at the two national basic meteorological stations closest to the northern and southern parts of the reserve. We examine the characteristics of DSDs across different seasons and cloud types (stratiform, convective, and mixed), as well as the relationships between the parameters of the Gamma distribution. The results show that: (1) DSDs exhibit notable spatial and typological differences. In the northern region, precipitation in spring and summer, as well as from stratiform and convective clouds, shows a bimodal distribution, while autumn, winter, and mixed clouds exhibit a unimodal distribution. In contrast, the southern region displays unimodal DSDs across all seasons and cloud types. In both regions, raindrop concentration is dominated by small and medium-sized drops. However, rainfall is primarily contributed by medium and large drops in summer and from convective clouds, whereas small and medium drops play a major role in winter and in stratiform and mixed clouds. (2) The normalised intercept parameter (lgN_w) shows a consistent seasonal order of autumn > summer > winter > spring in both regions, while the mass-weighted mean diameter (D_m) follows summer > spring > autumn > winter. For different cloud types, both parameters decrease in the order: convective > stratiform > mixed. Both lgN_w and D_m increase with rainfall intensity (R). The increase in R is mainly attributed to an increase in D_m (i.e., a broadening of the drop size spectrum) and, to a lesser extent, to an increase in lgN_w (i.e., a higher drop concentration). Convective precipitation in both regions exhibits transitional continental-maritime characteristics. (3) Seasonal variations in the shape (μ) and slope (λ) parameters differ between the two regions: in the north, the order is autumn > winter > spring > summer, while in the south, it is winter > autumn > summer > spring. For cloud types, the order is mixed > stratiform > convective clouds in both regions. Both μ and λ decrease with increasing R. (4) The classic Z-R relationship (Z=300R^1.4) performs well for autumn and winter precipitation in the northern region but systematically overestimates rainfall in northern spring and summer, as well as in all seasons in the southern region. To improve quantitative precipitation estimation (QPE) accuracy, we derive localised Z-R relationships: for the northern region, Z=331.4R^1.56 is recommended for spring, and Z=276.3R^1.48 for the spring-summer-autumn period; for the southern region, Z=413.5R^1.54 for spring and Z=339.8R^1.45 for the spring-summer-autumn period. These localised relationships significantly reduce estimation errors.

Abstract:Using the observation data of cloud condensation nuclei (CCN) under 0.1%, 0.2%, 0.3%, 04%, 0.6%, and 0.8% supersaturations and meteorological data in the Liupan Mountain area from September 2022 to August 2023, this study explores the variation characteristics of CCN in Liupan Mountain and their influencing factors. The results show that, affected by meteorological factors, pollution sources, supersaturation, etc.: (1) The average concentration of CCN in the Liupan Mountain area is 589 cm^-3. The number distribution of CCN particles under each supersaturation shows a unimodal pattern. With the increase in supersaturation, the peak particle size of the number of particles increases, and the peak particle size under 0.8% supersaturation is 3.5 to 4.0 μm. The monthly variation of CCN number concentration shows the highest in June and the lowest in February; the diurnal variation shows that the concentration increases in the afternoon and evening. (2) The CCN number concentration is relatively high when the snowfall in the winter half-year is less than 2.5 mm and the rainfall in the summer half-year is less than 0.5 mm; in non-precipitation weather, the CCN number concentration is relatively high when the relative humidity is 30% to 90%, the wind speed is less than 15 m/s, the wind direction is easterly in the winter half-year, and the wind direction is 135° to 180° in the summer half-year. The CCN number concentration mean spectrum in non-snowfall weather in the winter half-year is higher than that in snowfall weather and the summer half-year. (3) When the rain intensity is 1.1 mm·h^-1 and 6.7 mm/h, the CCN number concentration reduction rates are 10.16 cm^-3·h^-1 and 20.11 cm^-3·h^-1, respectively; by fitting the CCN activation spectra, it is found that the parameter C of the CCN activation spectrum is significantly large in most periods (greater than 1000), and the fitting coefficient k is high (about 0.7 or more), indicating that the area often has obvious continental characteristics.

Abstract:Human climatic comfort has a significant impact on architectural design, human health, outdoor activities, and so on. Based on the daily average temperature, relative humidity, and wind speed of 1866 meteorological stations throughout China during 1963-2022, the human climatic comfort index is calculated, and the spatial distribution and temporal variation characteristics of comfortable days over China are analysed. The results show that the human climatic comfort level in China is dominated by cold to comfortable levels. The annual comfortable days are 100 days in China during 1963-2022, and the number of comfortable days is the highest in May and June, with 16.7 and 16.4 days, respectively. In terms of spatial distribution, the southeastern region has more annual comfortable days, exceeding 80 days, while the Qinghai-Xizang Plateau has the fewest annual comfortable days, less than 50 days. Under the background of climate change, the increase rate of the annual comfortable days is 1.5 d/10a. It increases most obviously in spring and autumn, while it shows a slight decreasing trend in summer, which has the most comfortable days. The annual comfortable days increase by about 5.8 days in the last 30 years (1993-2022) compared with the first 30 years (1963-1992). The annual comfortable days are on the rise in most parts of China, while they decrease in the central and southern parts of North China, the northern part of East China, and the southern part of South China. On the regional scale, Southwest China has the most annual comfortable days, reaching 133.2 days, mainly concentrated in spring, summer, and autumn. South China follows with 115.4 days, mainly occurring in spring, autumn, and winter. The annual comfortable days in North China, East China, Central China, and Xinjiang Regions are close to 100 days. Northeast China has 83.7 days. Northwest China and Inner Mongolia Region are around 74 days. The Tibet Region has the fewest annual comfortable days, with 6.6 days, only occurring in summer. Northeast China, Xinjiang, and Inner Mongolia regions have more than 60 comfortable days in summer, significantly higher than other regions. On the whole, the average annual comfortable days in all regions increase. Inner Mongolia, Xinjiang, and Northwest China have relatively large increases, while South China, Central China, East China, North China, and Southwest China see smaller increases. Moreover, the number of comfortable days in summer in these regions decreases to varying degrees. In general, climate warming is conducive to improving climatic comfort in China.

Abstract:Against the backdrop of global warming, understanding the occurrence patterns of continuous rain during the summer harvest and planting period is crucial for avoiding the increasingly frequent risks of continuous rain disasters during the maturation and harvest stages of rapeseed/wheat, as well as the sowing and emergence stages of rice/dryland crops. Using data on meteorology, soil moisture, and crop production progress from 1981 to 2020, this study calculates the continuous rain intensity index based on the start and end times of the summer harvest and planting period and the criteria for determining continuous rain. It establishes the corresponding relationship between the continuous rain intensity index and the increment of 10 cm soil relative humidity and defines the impact levels of continuous rain intensity during the summer harvest and planting period. By employing spatial-temporal statistics and EOF (Empirical Orthogonal Function) methods, the study classifies different regions, decades, intensity levels, and stages to analyse the spatial-temporal distribution characteristics and occurrence patterns of continuous rain during the summer harvest and planting period in Jiangsu Province. The results show that, in terms of time, the number of continuous rain occurrences in southern, central, and northern Jiangsu is 0.38 times/year, 0.72 times/year, and 0.98 times/year, respectively. The high-incidence period is concentrated in 2010-2020, with the occurrence intensity showing an upward trend of 0.15-0.25 per decade. The periods of strong occurrence are concentrated in 2001-2020, and mild, strong, extra strong continuous rains mostly occur in late May-early June and late June. In terms of space, the frequency and intensity of continuous rain show a pattern of “more in the south and less in the north” and “lighter in the north and heavier in the central and southern regions.” The number of occurrences of continuous rain at all intensity levels follows the order: southern Jiangsu > central Jiangsu > northern Jiangsu. In terms of typical years, the intensity of continuous rain in the Lixiahe region of Jiangsu is the strongest in 2003, 2015, and 2020; the intensity is relatively strong in the north-central part of the area between the Yangtze and Huaihe Rivers in 2003, 2006, and 2012; and the intensity is relatively strong in the central part of southern Jiangsu in 2003, 2015, and 2020. In actual production, focus should be placed on the frequent and severe occurrence regions and periods of continuous rain during the summer harvest and planting period, so as to scientifically prevent farmland waterlogging, optimise the allocation of agricultural machinery and equipment, and rationally achieve risk transfer.

Abstract:To enhance the silver iodide (AgI) seeding capability of the catalytic operation aircraft, this paper proposes a method that utilises the high-temperature, high-speed exhaust flame of a turbojet engine for efficient AgI seeding agent combustion and seeding. Based on theoretical and simulation analysis, a systemic architecture integrating storage, delivery, combustion, and control is designed, and a prototype is developed. A specialised seeding agent formulation with 40% AgI content is developed to adapt to the exhaust flame environment. Ground tests are conducted to evaluate the seeding agent’s stable combustion within the exhaust flame, the particle size distribution of the generated aerosols, and their ice-nucleating ability in a cloud chamber. The results show that this technology generates sub-micron-sized (less than 0.5 μm) artificial ice nuclei. Both the ice nucleus concentration and the ice formation efficiency meet operational requirements and are comparable to the performance of conventional AgI flares. This technology significantly improves effective payload and seeding efficiency, offering a new technical approach for UAV (unmanned aerial vehicle)-based weather modification operations.