Gas cannon operation weather influence evaluation method and system fused with AI algorithm

Abstract

The present application relates to the technical field of weather modification, and in particular to a gas cannon operation weather influence evaluation method and system fusing an AI algorithm. The method obtains weather radar data of an operation area and a control area, identifies a cloud three-dimensional structure and calculates a cloud physical parameter change amount by using a deep learning model, and then decomposes an artificial influence component by using a machine learning model. Finally, the artificial influence component is normalized and corrected and compared by using the control area data to generate a quantitative evaluation result. The present application realizes objective and quantitative evaluation of the gas cannon operation effect.

Description

Technical Field

[0001] This invention relates to the field of weather modification technology, and in particular to a method and system for assessing the meteorological impact of gas cannon operations by integrating AI algorithms. Background Technology

[0002] In the field of weather modification, the scientific evaluation of the effectiveness of gas cannon operations, as a physical intervention method for localized artificial rain enhancement or hail suppression, has always been a key focus of operational practice and research. Traditional evaluation methods mainly rely on statistical analysis of specific meteorological elements in the operational area before and after the operation, such as comparing changes in surface precipitation, radar echo intensity, or macroscopic cloud characteristics before and after the operation.

[0003] Conventional assessment procedures are typically based on weather radar observation data. Operators compare radar echo images of the target area before and after operations to qualitatively or semi-quantitatively determine changes in echo intensity, area, or height, thereby inferring the potential impact of the operations. Another common practice is to select a non-operational area with a similar weather background to the target area as a control, and use statistical tests to analyze the significance of differences in meteorological elements between the two areas to assess the effectiveness of the operations. These methods rely heavily on manual interpretation and experience, and their analysis often focuses on two-dimensional planar features or limited single-point parameters.

[0004] Current conventional practices have significant limitations. The assessment process relies heavily on human experience, and the analysis of radar echo data often remains at the level of image comparison and simple parameter statistics, lacking a quantitative characterization of the detailed three-dimensional structure of clouds and the continuous spatiotemporal evolution of their physical parameters. This makes it difficult to effectively separate natural weather evolution from changes caused by human intervention, and the assessment conclusions are easily influenced by subjective judgment and natural variability, resulting in insufficient objectivity and accuracy.

[0005] Furthermore, traditional methods typically assume that the control region is completely synchronized with the weather evolution of the target region and is only influenced by natural processes when constructing the control region. However, in real-world complex weather systems, even different regions under the same weather background may have inherent differences in the microphysical processes and evolution paths of their cloud systems. Conventional methods fail to quantify and correct for these differences in natural evolution between regions, directly attributing observed differences to human influences. This introduces systematic errors into the assessment results, and their reliability and universality need to be improved. Summary of the Invention

[0006] This invention provides a method and system for assessing the meteorological impact of gas cannon operations by integrating AI algorithms, which can solve the problems in the prior art.

[0007] A first aspect of this invention provides a method for assessing the meteorological impact of gas cannon operations by incorporating AI algorithms, comprising:

[0008] Acquire meteorological radar echo data of the target area before and after the gas cannon operation, and perform quality control processing on the meteorological radar echo data to obtain cloud precipitation echo data;

[0009] The cloud precipitation echo data is input into a cloud recognition deep learning model to obtain the three-dimensional structure features of clouds in the target area before and after the operation. Based on the three-dimensional structure features of clouds, the changes in cloud physical parameters in the target area before and after the operation are calculated. The changes in cloud physical parameters include changes in cloud top height, changes in liquid water content, and changes in cloud volume.

[0010] By performing spatiotemporal continuity analysis on multi-time radar echo data, the movement trajectory and evolution characteristics of clouds within the target area are determined. The movement trajectory and evolution characteristics are then input into a machine learning model for human impact identification to decompose the changes in cloud physical parameters observed after the operation to obtain the human impact component caused by the gas cannon operation.

[0011] Acquire meteorological radar echo data of a control area that is in the same weather system as the target area and has not carried out gas cannon operations; extract cloud evolution characteristics of the control area based on the meteorological radar echo data of the control area.

[0012] The artificial influence component is normalized and corrected using the evolution characteristics of the cloud body in the control area to obtain the normalized artificial influence index. The normalized artificial influence index is then compared with the preset effect evaluation standard to generate a quantitative evaluation result of the gas cannon operation effect.

[0013] The cloud precipitation echo data is input into a cloud recognition deep learning model to obtain the three-dimensional structure features of the cloud in the target area before and after the operation. Based on the three-dimensional structure features of the cloud, the changes in cloud physical parameters in the target area before and after the operation are calculated, including:

[0014] The cloud precipitation echo data is vertically layered to divide the target area into multiple height layers. The cloud precipitation echo data of each height layer is input into the cloud body recognition deep learning model. The model extracts the horizontal distribution features and vertical gradient features of the reflectivity factor for each height layer to obtain the layered cloud structure features.

[0015] Based on the aforementioned layered cloud structure characteristics, the phase distribution of cloud water particles at each altitude layer is identified to obtain phase distribution information;

[0016] The liquid water content at each altitude layer is calculated based on the layered cloud structure characteristics and the phase distribution information, and the three-dimensional cloud structure characteristics of the target area are obtained by vertical integration of the liquid water content at each altitude layer.

[0017] The three-dimensional structural features of the cloud before and after the operation are compared. The changes in cloud top height, liquid water content and cloud volume at each altitude level are calculated. The changes at each altitude level are then spatially integrated to obtain the changes in cloud physical parameters in the target area.

[0018] Based on the layered cloud structure characteristics and the phase distribution information, the liquid water content at each altitude layer is calculated, and the three-dimensional cloud structure characteristics of the target area are obtained by vertical integration of the liquid water content at each altitude layer, including:

[0019] Based on the spatial locations of the supercooled water region, ice crystal region, and mixed phase region in the phase distribution information, the corresponding phase weighting coefficients are determined for each height layer.

[0020] Based on the phase weighting coefficient, the reflectivity factor in the layered cloud structure features is phase-corrected to obtain the phase-corrected reflectivity factor.

[0021] The phase-corrected reflectivity factor is correlated with the particle spectrum distribution characteristics of each phase region in the phase distribution information to determine the phase transition relationship between the reflectivity factor and the liquid water content;

[0022] Based on the phase transition relationship, the phase-corrected reflectivity factor of each height layer is converted into the liquid water content of the corresponding height layer to obtain the liquid water content of each height layer;

[0023] The liquid water content of each altitude layer is integrated layer by layer in the vertical direction from the cloud base to the cloud top, and the cloud boundary range of each altitude layer is determined by combining the horizontal distribution characteristics in the layered cloud structure features, so as to obtain the three-dimensional cloud structure features of the target area.

[0024] By performing spatiotemporal continuity analysis on multi-time radar echo data, the movement trajectory and evolution characteristics of clouds within the target area are determined. These movement trajectories and evolution characteristics are then input into a machine learning model for identifying human impacts. This model decomposes the observed changes in cloud physical parameters after the operation to obtain the human impact components caused by the gas cannon operation, including:

[0025] Based on meteorological radar echo data of the target area at multiple consecutive times before and after the gas cannon operation, a multi-time cloud feature sequence was obtained;

[0026] The centroid position coordinates of the cloud body at each time point are calculated to obtain the centroid position coordinates of the cloud body at each time point in the multi-time cloud body feature sequence. The spatial displacement vector between the centroid position coordinates of the cloud body at adjacent time points is calculated to obtain the centroid displacement vector sequence.

[0027] The centroid displacement vector sequence is subjected to a continuity test, and the centroid displacement vectors that pass the continuity test are connected in chronological order to form the movement trajectory of the cloud within the target area;

[0028] Based on the reflectivity factor intensity and cloud morphology characteristics of the cloud at each time point in the multi-time cloud feature sequence, the temporal change rate of the cloud reflectivity factor and the deformation rate of the cloud morphology between adjacent time points are calculated. Combined with the movement speed in the movement trajectory, the evolution characteristics of the cloud within the target area are extracted.

[0029] Based on the movement trajectory, the evolution characteristics, and the actual observed changes in cloud physical parameters after the operation, the artificial influence component caused by the gas cannon operation is obtained using the artificial influence recognition machine learning model.

[0030] Based on the movement trajectory, the evolution characteristics, and the actual observed changes in cloud physical parameters after the operation, the artificial impact components caused by the gas cannon operation are obtained using the artificial impact identification machine learning model, including:

[0031] Based on the aforementioned evolutionary characteristics, the sequence of observed cloud physical parameters at each moment before the operation was extracted;

[0032] Time series analysis is performed on the observed data sequence to determine the growth or dissipation pattern of cloud physical parameters over time, and the time evolution function of cloud physical parameters is determined based on the growth or dissipation pattern.

[0033] The time evolution function is extrapolated from the pre-operation period to the post-operation period to generate expected values ​​of cloud physics parameters at each moment in the post-operation period under the condition that no gas cannon operation was carried out. The time series formed by the expected values ​​is used as the natural evolution baseline.

[0034] The actual observed values ​​of cloud physical parameters at each time point within the spatial range and time window after the operation are obtained. The actual observed values, the natural evolution baseline, the movement trajectory, and the evolution characteristics are input into the artificial influence recognition machine learning model. The deviation at each time point is determined by the artificial influence recognition machine learning model.

[0035] The deviation at each time point is integrated over time, and spatial calculations are performed based on the cloud space range and the time integration results to obtain the human impact component caused by the gas cannon operation.

[0036] Acquire meteorological radar echo data of a control area that is in the same weather system as the target area but has not been subjected to gas cannon operations. Based on the meteorological radar echo data of the control area, extract cloud evolution characteristics of the control area, including:

[0037] Obtain the spatial extent boundary and weather system type characteristics of the weather system in which the target area is located, and determine a set of candidate areas that are in the same weather system as the target area based on the spatial extent boundary;

[0038] Based on each candidate region in the candidate region set, the similarity index between each candidate region and the target region on different element distributions is calculated. The candidate region set is then sorted according to the similarity index, and the candidate region with the highest similarity index is selected as the control region.

[0039] Meteorological radar echo data of the control area within the same time range as the target area before and after the operation are obtained. The meteorological radar echo data of the control area are then subjected to echo intensity threshold segmentation and three-dimensional spatial gridding to determine the cloud characteristics of the cloud body at each time point in the control area.

[0040] Based on the cloud characteristics of the control area at each time, the cloud physical structure parameters of the cloud in the control area at each time are calculated. The cloud physical structure parameters include cloud top height, liquid water content and cloud volume.

[0041] Calculate the time change rate of the cloud physical structure parameters between adjacent time points, and combine the cloud physical structure parameters at each time point with the time change rate to obtain the cloud evolution characteristic quantity of the control area.

[0042] The human influence component is normalized and corrected using the evolution characteristics of the cloud body in the control area to obtain a normalized human influence index. This normalized index is then compared with a preset effect evaluation standard to generate a quantitative evaluation result of the gas cannon operation effect, including:

[0043] The temporal rate of change of cloud physical structure parameters in the control area during the post-operation period is extracted from the cloud evolution characteristics of the control area, and the natural variation amplitude of cloud physical structure parameters in the control area during the post-operation period is calculated based on the temporal rate of change.

[0044] The ratio of the artificial influence component to the natural variation amplitude of the control area is calculated to obtain the normalized artificial influence index after eliminating the influence of regional background differences and the overall evolution of the weather system.

[0045] The normalized artificial influence index is compared with the normalized artificial influence index threshold corresponding to each effect level in the preset effect evaluation criteria. The effect level corresponding to the normalized artificial influence index is determined according to the numerical range in which the normalized artificial influence index is located.

[0046] Based on the normalized human impact index and the effect level corresponding to the normalized human impact index, a quantitative evaluation result of the gas cannon operation effect is generated.

[0047] A second aspect of this invention provides a meteorological impact assessment system for gas cannon operations that integrates AI algorithms, comprising:

[0048] The data acquisition unit is used to acquire meteorological radar echo data of the target area before and after the gas cannon operation, and to perform quality control processing on the meteorological radar echo data to obtain cloud precipitation echo data.

[0049] The cloud feature unit is used to input the cloud precipitation echo data into the cloud recognition deep learning model to obtain the three-dimensional structure features of the cloud in the target area before and after the operation, and to calculate the changes in cloud physical parameters in the target area before and after the operation based on the three-dimensional structure features of the cloud. The changes in cloud physical parameters include changes in cloud top height, changes in liquid water content, and changes in cloud volume.

[0050] The artificial influence unit is used to determine the movement trajectory and evolution characteristics of cloud bodies within the target area by performing spatiotemporal continuity analysis on multi-time radar echo data. The movement trajectory and evolution characteristics are then input into the artificial influence recognition machine learning model to decompose the changes in cloud physical parameters observed after the operation to obtain the artificial influence component caused by the gas cannon operation.

[0051] The reference area unit is used to acquire meteorological radar echo data of the reference area which is in the same weather system as the target area and has not carried out gas cannon operations. Based on the meteorological radar echo data of the reference area, the evolution characteristics of the cloud body in the reference area are extracted.

[0052] The effect evaluation unit is used to normalize and correct the artificial influence component by using the evolution characteristics of the cloud body in the control area to obtain the normalized artificial influence index. The normalized artificial influence index is then compared with the preset effect evaluation standard to generate a quantitative evaluation result of the gas cannon operation effect.

[0053] A third aspect of the present invention provides an electronic device, comprising:

[0054] processor;

[0055] Memory used to store processor-executable instructions;

[0056] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0057] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0058] This invention utilizes meteorological radar echo data combined with an artificial intelligence model to achieve an objective and quantitative evaluation of the effectiveness of gas-fired artillery weather modification operations. By extracting three-dimensional cloud structure features from radar data using a deep learning model, it can accurately calculate subtle changes in cloud physical parameters before and after the operation, effectively improving the automation and accuracy of feature extraction. This method overcomes the limitations of traditional methods that rely on manual interpretation and statistical comparison, providing high-resolution observational evidence of cloud microphysical processes for effectiveness evaluation.

[0059] This invention introduces spatiotemporal continuity analysis and machine learning recognition models to effectively separate the change components caused by human intervention from the background of natural weather evolution. By analyzing the movement trajectory and evolution characteristics of cloud bodies and combining them with radar observation data from a control area, the influence of the weather system's own evolution on the observation results is systematically removed, significantly reducing the uncertainty of the assessment results. This decomposition and correction process enhances the reliability of causal inference, making the assessment conclusions more convincing.

[0060] The normalized weather modification index generated by this invention provides a unified and quantifiable standard for measuring the effectiveness of weather modification operations. Corrected for the natural evolution characteristics of the control area, this index has better universality and comparability, facilitating direct comparison with preset evaluation thresholds and thus quickly generating clear quantitative evaluation conclusions. The entire process achieves full automation and intelligence from data preprocessing, feature extraction, impact separation to effect determination, significantly improving the efficiency and objectivity of the evaluation work and providing strong technical support for scientific decision-making and effect verification in weather modification operations. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating the meteorological impact assessment method for gas cannon operations based on an embodiment of the present invention, which incorporates AI algorithms.

[0062] Figure 2 This is a schematic diagram of the process for determining the artificial influence component according to an embodiment of the present invention. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0065] Figure 1 This is a flowchart illustrating the meteorological impact assessment method for gas cannon operations incorporating AI algorithms, as described in this embodiment of the invention. Figure 1 As shown, the meteorological impact assessment method for gas cannon operations that integrates AI algorithms includes:

[0066] Acquire meteorological radar echo data of the target area before and after the gas cannon operation, and perform quality control processing on the meteorological radar echo data to obtain cloud precipitation echo data;

[0067] The cloud precipitation echo data is input into a cloud recognition deep learning model to obtain the three-dimensional structure features of clouds in the target area before and after the operation. Based on the three-dimensional structure features of clouds, the changes in cloud physical parameters in the target area before and after the operation are calculated. The changes in cloud physical parameters include changes in cloud top height, changes in liquid water content, and changes in cloud volume.

[0068] By performing spatiotemporal continuity analysis on multi-time radar echo data, the movement trajectory and evolution characteristics of clouds within the target area are determined. The movement trajectory and evolution characteristics are then input into a machine learning model for human impact identification to decompose the changes in cloud physical parameters observed after the operation to obtain the human impact component caused by the gas cannon operation.

[0069] Acquire meteorological radar echo data of a control area that is in the same weather system as the target area and has not carried out gas cannon operations; extract cloud evolution characteristics of the control area based on the meteorological radar echo data of the control area.

[0070] The artificial influence component is normalized and corrected using the evolution characteristics of the cloud body in the control area to obtain the normalized artificial influence index. The normalized artificial influence index is then compared with the preset effect evaluation standard to generate a quantitative evaluation result of the gas cannon operation effect.

[0071] The cloud precipitation echo data is input into a cloud recognition deep learning model to obtain the three-dimensional structure features of the cloud in the target area before and after the operation. Based on the three-dimensional structure features of the cloud, the changes in cloud physical parameters in the target area before and after the operation are calculated, including:

[0072] The cloud precipitation echo data is vertically layered to divide the target area into multiple height layers. The cloud precipitation echo data of each height layer is input into the cloud body recognition deep learning model. The model extracts the horizontal distribution features and vertical gradient features of the reflectivity factor for each height layer to obtain the layered cloud structure features.

[0073] Based on the aforementioned layered cloud structure characteristics, the phase distribution of cloud water particles at each altitude layer is identified to obtain phase distribution information;

[0074] The liquid water content at each altitude layer is calculated based on the layered cloud structure characteristics and the phase distribution information, and the three-dimensional cloud structure characteristics of the target area are obtained by vertical integration of the liquid water content at each altitude layer.

[0075] The three-dimensional structural features of the cloud before and after the operation are compared. The changes in cloud top height, liquid water content and cloud volume at each altitude level are calculated. The changes at each altitude level are then spatially integrated to obtain the changes in cloud physical parameters in the target area.

[0076] After acquiring cloud precipitation echo data before and after the gas cannon operation, the echo data needs to be systematically layered to construct a complete three-dimensional cloud structure. Specifically, based on the vertical resolution characteristics of meteorological radar detection, the target area is vertically divided from the ground to the top of the troposphere at intervals of 500 to 1000 meters, forming multiple independent height-layer slices. Each height layer corresponds to a two-dimensional plane echo intensity distribution, which reflects the scattering characteristics of cloud water particles on radar waves at that altitude. Considering the differences in meteorological elements at different height layers, it is necessary to ensure that there is an appropriate overlap area between each layer during the division. The overlap height is usually set to 100 to 200 meters to facilitate subsequent vertical continuity analysis.

[0077] Cloud precipitation echo data extracted from each altitude layer are input into a deep learning model for cloud recognition. This model employs a convolutional neural network architecture, enabling it to automatically learn spatial feature patterns in the echo data. For each altitude layer, the model first extracts the horizontal distribution features of the reflectivity factor, including the spatial continuity of echo intensity, the location and extent of the echo core region, and gradient changes at the echo edges. The extraction of horizontal distribution features is achieved through multi-scale convolutional kernels; small-scale kernels capture local detail features, while large-scale kernels capture the overall distribution pattern. Simultaneously, the model performs differential calculations on the reflectivity factors of adjacent altitude layers to obtain vertical gradient features. The vertical gradient features reflect the trend of cloud water particle concentration with altitude; a positive gradient indicates that cloud water content increases with altitude, while a negative gradient indicates that cloud water content decreases with altitude. By fusing and encoding the horizontal distribution features and vertical gradient features, a comprehensive layered cloud structure feature vector is formed.

[0078] Based on the feature vectors of layered cloud structures, the phase distribution of cloud water particles at each altitude level is further identified. Phase identification is based on the differences in the scattering characteristics of radar echoes by particles of different phases. Liquid water droplets typically produce a strong reflectivity factor and a gentle vertical gradient change, while ice crystal particles have a relatively weak reflectivity factor and a drastic vertical gradient change. The mixed phase region exhibits a moderate reflectivity factor and a complex vertical gradient distribution. By setting a combination of reflectivity factor thresholds and vertical gradient thresholds, each altitude level is divided into liquid, ice, and mixed phase regions. The liquid region is mainly distributed in the lower part of the cloud, in areas with temperatures above zero degrees Celsius; the ice region is mainly located in the upper part of the cloud, in high-altitude areas with temperatures below minus ten degrees Celsius; the mixed phase region is located near the zero-degree layer and exhibits characteristics of both liquid and solid particles. The phase distribution information is stored in the form of a three-dimensional spatial matrix, with each element of the matrix identifying the dominant phase type at its corresponding spatial location.

[0079] After obtaining the phase distribution information, the liquid water content at each altitude level is calculated based on the layered cloud structure characteristics. The calculation of liquid water content uses an empirical formula that establishes a quantitative link between reflectivity factor and liquid water content. For altitude levels identified as liquid zones, the liquid water content inversion algorithm is directly applied; for mixed phase zones, the inversion results need to be corrected based on the ratio of liquid water to ice crystals; for ice phase zones, the liquid water content is considered zero. After calculating the liquid water content at each altitude level, an integration operation is performed vertically. Vertical integration is achieved by summing the products of the liquid water content at each altitude level and the corresponding layer thickness, yielding the total liquid water volume within a unit area column. This value characterizes the degree of water vapor accumulation in the cloud. Combining the spatial distribution of reflectivity factor, liquid water content distribution, and phase distribution at each altitude level, a complete three-dimensional cloud structure feature is constructed. The three-dimensional structure feature not only includes the geometric morphology information of the cloud but also encompasses the microphysical structure information within the cloud.

[0080] To quantify the impact of gas cannon operations on clouds, a systematic comparison of the three-dimensional structural characteristics of clouds before and after the operation is necessary. The comparative analysis begins with the cloud top height at each altitude level. Cloud top height is defined as the highest point within that altitude level where the echo intensity exceeds a set threshold. By extracting the cloud top height values ​​for each altitude level before and after the operation, the difference between the two values ​​is calculated to obtain the change in cloud top height. Positive values ​​indicate cloud top elevation, while negative values ​​indicate cloud top elevation. Next, the change in liquid water content at each altitude level is calculated, i.e., the liquid water content after the operation minus the liquid water content before the operation. An increase in liquid water content indicates enhanced water vapor condensation at that altitude level, while a decrease indicates that precipitation consumption or evaporation is dominant. Finally, the change in cloud volume is assessed. Cloud volume is calculated by counting the number of grid points within each altitude level where the echo intensity exceeds the threshold and multiplying this number by the volume of a single grid cell. The difference in cloud volume before and after the operation reflects the evolution of the overall cloud size.

[0081] After obtaining the changes in cloud top height, liquid water content, and cloud volume at various altitude levels, spatial comprehensive processing is performed to obtain the overall changes in cloud physical parameters of the target area. This spatial comprehensive processing is implemented in two steps: First, each parameter is normalized by dividing the change in cloud top height, the change in liquid water content, and the change in cloud volume by the baseline value of the cloud top height and volume of each layer before the operation. This yields the relative change rate (dimensionless) of each parameter, eliminating the incomparability caused by different dimensions. Second, a weighted average is calculated from the normalized dimensionless relative change rates. The weighting coefficients are determined based on the contribution rate of each altitude level to the overall precipitation. Generally, lower-level clouds contribute more to surface precipitation and are assigned higher weights; higher-level clouds mainly affect cloud development potential and are assigned moderate weights. Ultimately, the changes in cloud physical parameters in the target area are constituted by three components: the relative change rate of cloud top height, the relative change rate of liquid water content, and the relative change rate of cloud volume. The weighted average of each component can be output as a scalar indicator for overall evaluation, or the spatial distribution of the three components at each height level can be output as a vectorized description. These changes in cloud physical parameters serve as the foundational data for subsequent human influence identification and effect evaluation; their accuracy directly impacts the reliability of the final evaluation conclusion. The entire processing flow ensures a complete conversion from raw echo data to structured changes in cloud physical parameters, providing solid data support for the quantitative evaluation of the gas cannon's operational effects.

[0082] Based on the layered cloud structure characteristics and the phase distribution information, the liquid water content at each altitude layer is calculated, and the three-dimensional cloud structure characteristics of the target area are obtained by vertical integration of the liquid water content at each altitude layer, including:

[0083] Based on the spatial locations of the supercooled water region, ice crystal region, and mixed phase region in the phase distribution information, the corresponding phase weighting coefficients are determined for each height layer.

[0084] Based on the phase weighting coefficient, the reflectivity factor in the layered cloud structure features is phase-corrected to obtain the phase-corrected reflectivity factor.

[0085] The phase-corrected reflectivity factor is correlated with the particle spectrum distribution characteristics of each phase region in the phase distribution information to determine the phase transition relationship between the reflectivity factor and the liquid water content;

[0086] Based on the phase transition relationship, the phase-corrected reflectivity factor of each height layer is converted into the liquid water content of the corresponding height layer to obtain the liquid water content of each height layer;

[0087] The liquid water content of each altitude layer is integrated layer by layer in the vertical direction from the cloud base to the cloud top, and the cloud boundary range of each altitude layer is determined by combining the horizontal distribution characteristics in the layered cloud structure features, so as to obtain the three-dimensional cloud structure features of the target area.

[0088] After obtaining the layered cloud structure characteristics and phase distribution information of the target area before and after the gas cannon operation, it is necessary to convert this basic information into quantifiable cloud physical parameters. For cloud data observed by meteorological radar, there are significant differences in the phase characteristics of water condensate at different altitudes. Directly using a unified reflectivity factor conversion relationship will introduce a large error. Therefore, it is necessary to establish a liquid water content calculation method that takes into account phase differences.

[0089] When calculating the liquid water content at various altitudes, it is first necessary to address the influence of phase distribution information on the reflectivity factor. Phase distribution information includes the spatial locations and vertical distribution characteristics of the supercooled water region, ice crystal region, and mixed phase region. These different phase regions exhibit fundamentally different microphysical characteristics of condensate particles. In the pure supercooled water region, water droplets remain liquid, and their scattering characteristics are similar to those of room-temperature liquid water. In the ice crystal region, the dielectric constant of ice crystal particles is significantly lower than that of liquid water, resulting in a higher water content for the same reflectivity factor. The mixed phase region contains both liquid and solid particles, and its scattering characteristics fall between the two.

[0090] To address this phase difference, a corresponding phase weighting coefficient is determined for each altitude layer. Specifically, for altitude layers identified as pure supercooled water regions, the phase weighting coefficient is set to 1.0, indicating that the conversion is performed entirely according to the scattering characteristics of liquid water. For altitude layers in pure ice crystal regions, the phase weighting coefficient is determined based on the ratio of the dielectric constants of ice crystals to liquid water, typically ranging from 0.15 to 0.20. For mixed-phase regions, the phase weighting coefficient is calculated based on the mass ratio of liquid particles to solid particles in that altitude layer. This ratio can be estimated using the correlation coefficient and differential reflectivity factor in the dual-polarization radar parameters. When the correlation coefficient is greater than 0.98, it indicates that the altitude layer is dominated by liquid water, and the phase weighting coefficient tends towards 1.0. When the correlation coefficient is less than 0.90 and the differential reflectivity factor is close to 0, it indicates that the altitude layer is dominated by ice crystals, and the phase weighting coefficient tends towards a value between 0.15 and 0.20.

[0091] After obtaining the phase weight coefficients for each altitude layer, phase correction is applied to the reflectivity factor in the layered cloud structure characteristics. The specific implementation of phase correction is as follows: the original reflectivity factor is multiplied by the phase weight coefficient of the corresponding altitude layer, thereby converting the reflectivity factors of different phases to reference values ​​under the equivalent liquid water scattering characteristics. For example, if the original reflectivity factor of a certain altitude layer is 35 dBZ, and this layer is located in the ice crystal region with a phase weight coefficient of 0.18, then the phase-corrected reflectivity factor needs to be converted to obtain the equivalent liquid water reflectivity factor. In practice, the reflectivity factor is first converted from logarithmic to linear form, multiplied by the phase weight coefficient, and then converted back to logarithmic form. This ensures that the phase-corrected reflectivity factor accurately reflects the scattering intensity corresponding to the equivalent liquid water content of that altitude layer.

[0092] After phase correction, a quantitative conversion relationship between reflectivity factor and liquid water content needs to be established. Traditional ZM relationships use a unified empirical formula, but the particle spectrum distribution in different phase regions of actual clouds varies significantly. By correlating the phase-corrected reflectivity factor with the particle spectrum distribution characteristics of each phase region in the phase distribution information, specific conversion relationships can be established for different phase regions. For the supercooled water region, the particle spectrum distribution exhibits a single-peak characteristic, with the peak position corresponding to particle diameters typically between 100 and 300 micrometers. In this case, the conversion relationship exponent parameter is close to 1.4. For the mixed phase region, the particle spectrum exhibits a bimodal or multimodal distribution, with small particle peaks corresponding to droplets and large particle peaks corresponding to snowflakes or graupel. In this case, the conversion relationship exponent parameter needs to be adjusted according to the relative intensity of the two peaks, with a value ranging from 1.2 to 1.6. By establishing this phase-specific conversion relationship, the actual water content characteristics at different altitudes can be reflected more accurately.

[0093] Based on the established phase-state conversion relationship, the phase-corrected reflectivity factor of each altitude layer is converted into the liquid water content of the corresponding altitude layer. During the conversion process, each horizontal grid point of each altitude layer is calculated independently. The converted liquid water content is expressed in grams per cubic meter (g / m³), typically ranging from 0.01 to 3.0 g / m³. For strong convective clouds, the liquid water content in the cloud core region exceeds 3.0 g / m³, while the liquid water content in the cloud edge region is less than 0.1 g / m³. To ensure the physical validity of the calculation results, a threshold check is performed. When the calculated liquid water content of a certain grid point exceeds the saturated water content under the temperature conditions of that altitude layer, it is corrected to the saturated water content value.

[0094] After obtaining the liquid water content at each altitude layer, the horizontal two-dimensional distribution of the liquid water content at each altitude layer is correlated with the corresponding altitude coordinates to form a three-dimensional spatial distribution matrix of liquid water content in the horizontal and vertical directions. This matrix records the liquid water content value at each three-dimensional grid point within the target area, constituting one of the core components of the cloud's three-dimensional structural features. Based on this, vertical integration is further performed to obtain the vertically integrated liquid water content, which serves as another component of the cloud's three-dimensional structural features. The vertical integration starts from the cloud base height and accumulates layer by layer upwards to the cloud top height. The cloud base height and cloud top height are determined by the vertical profile of the echo intensity in the layered cloud structural features. When the echo intensity of a certain altitude layer first exceeds 15 dBZ, it is identified as the cloud base; when the echo intensity drops below 15 dBZ and three consecutive layers are below this threshold, it is identified as the cloud top. During integration calculation, the distance between adjacent altitude layers is determined based on the radar scanning elevation angle and range library resolution, with typical values ​​ranging from 250 to 500 meters. For the k-th layer, its contribution to the integral result is the liquid water content of that layer multiplied by the vertical thickness of that layer. The vertical integral liquid water content is obtained by summing the contributions of all layers, in kilograms per square meter.

[0095] While obtaining the three-dimensional spatial distribution matrix of liquid water content and the vertically integrated liquid water content, it is also necessary to combine the horizontal distribution characteristics in the layered cloud structure features to determine the cloud boundary range of each height layer, and establish the three-dimensional spatial contour of the cloud layer by layer as the third component parameter of the cloud three-dimensional structural features. The horizontal distribution features reflect the horizontal extension range of the cloud at each height layer, and the cloud boundary of different height layers may be inconsistent. By setting an echo intensity threshold for each height layer, grid points exceeding the threshold are identified as the effective cloud area of ​​that layer, and areas below the threshold are considered as outside the cloud boundary. After this processing, the vertical integration process only accumulates all effective cloud layers at each horizontal position, avoiding the inclusion of noise echoes from outside the cloud in the calculation. The three-dimensional spatial distribution matrix of liquid water content, the three-dimensional spatial contour of the cloud, and the vertically integrated liquid water content together constitute the three-dimensional structural features of the cloud in the target area. Among them, the three-dimensional spatial distribution matrix and the three-dimensional spatial contour reflect the spatial distribution of the microphysical structure and geometry inside the cloud, while the vertically integrated liquid water content provides a holistic quantitative description of the water content of the cloud.

[0096] Figure 2 This is a schematic diagram illustrating the process of determining the artificial influence component according to an embodiment of the present invention. Figure 2 As shown, by performing spatiotemporal continuity analysis on multi-time radar echo data, the movement trajectory and evolution characteristics of clouds within the target area are determined. These movement trajectories and evolution characteristics are then input into a machine learning model for identifying artificial influences. This model decomposes the observed changes in cloud physical parameters after the operation to obtain the artificial influence components caused by the gas cannon operation, including:

[0097] Based on meteorological radar echo data of the target area at multiple consecutive times before and after the gas cannon operation, a multi-time cloud feature sequence was obtained;

[0098] The centroid position coordinates of the cloud body at each time point are calculated to obtain the centroid position coordinates of the cloud body at each time point in the multi-time cloud body feature sequence. The spatial displacement vector between the centroid position coordinates of the cloud body at adjacent time points is calculated to obtain the centroid displacement vector sequence.

[0099] The centroid displacement vector sequence is subjected to a continuity test, and the centroid displacement vectors that pass the continuity test are connected in chronological order to form the movement trajectory of the cloud within the target area;

[0100] Based on the reflectivity factor intensity and cloud morphology characteristics of the cloud at each time point in the multi-time cloud feature sequence, the temporal change rate of the cloud reflectivity factor and the deformation rate of the cloud morphology between adjacent time points are calculated. Combined with the movement speed in the movement trajectory, the evolution characteristics of the cloud within the target area are extracted.

[0101] Based on the movement trajectory, the evolution characteristics, and the actual observed changes in cloud physical parameters after the operation, the artificial influence component caused by the gas cannon operation is obtained using the artificial influence recognition machine learning model.

[0102] After acquiring meteorological radar echo data from multiple consecutive moments before and after the gas cannon operation in the target area, time series extraction is required. Specifically, a time window from 30 minutes before the operation to 60 minutes after the operation is selected, and radar volume scan data is extracted at 6-minute intervals, forming an echo dataset containing 15 time points. For the three-dimensional radar echo data of each time point, a cloud recognition deep learning model is applied to process the data, extracting key parameters such as the spatial distribution characteristics of clouds, the vertical profile of reflectivity factors, cloud top height, and cloud base height at each moment, constructing a multi-time-series cloud feature sequence. This sequence contains complete information on the evolution of clouds over time, providing a data foundation for subsequent trajectory tracking and impact decomposition.

[0103] For each moment in the multi-time cloud feature sequence, the three-dimensional spatial centroid position of the cloud is calculated. The centroid calculation adopts a reflectivity factor weighted method, that is, the reflectivity factor of each grid point is used as the weight, and the longitude, latitude, and altitude coordinates of that grid point are weighted and averaged. Horizontally, the target area is divided into 1 km × 1 km grids, and vertically, a 500 m interval is used for the layering. For the cloud at time i, its centroid position coordinates include longitude... ,latitude and height Three components. After calculation, the centroid position coordinates of adjacent time points are interpolated to obtain the centroid displacement vector. The centroid displacement vector from time i to time i+1 includes a horizontal displacement component and a vertical displacement component. The horizontal displacement component reflects the horizontal movement direction and velocity of the cloud, while the vertical displacement component reflects the cloud's development or dissipation trend. All adjacent centroid displacement vectors are arranged in chronological order to form a centroid displacement vector sequence.

[0104] The continuity test of the centroid displacement vector sequence aims to eliminate discontinuities caused by cloud splitting, merging, or newly formed cloud bodies. The continuity test criteria include: the spatial distance between adjacent centroid displacements does not exceed 15 kilometers, the directional angle between adjacent displacement vectors does not exceed 45 degrees, and the change in vertical displacement does not exceed 2 kilometers. Centroid displacement vectors that meet these criteria are considered to belong to the continuous evolution process of the same cloud body. The centroid displacement vectors that pass the continuity test are connected end-to-end in chronological order to form the cloud's movement trajectory within the target area. This trajectory is expressed as a three-dimensional spatial curve, recording the complete movement path of the cloud's centroid from before to after the operation. During trajectory generation, for a few moments that fail the continuity test, linear interpolation between the preceding and following moments is used to supplement the trajectory, ensuring its temporal integrity.

[0105] Based on multi-time cloud feature sequences, the evolutionary characteristics of clouds are further extracted, mainly including two categories: intensity evolution characteristics and morphological evolution characteristics. Intensity evolution characteristics are characterized by calculating the time-varying rate of change of the cloud reflectivity factor between adjacent time points. Specifically, the time derivatives of the average reflectivity factor, maximum reflectivity factor, and vertical integral value of the reflectivity factor are calculated for each time point. Morphological evolution characteristics are described by the deformation rate of the cloud shape, including the rate of change of the cloud's horizontal area, the rate of change of its vertical thickness, and the changes in its geometric shape. The changes in cloud geometry are assessed using an ellipse fitting method, fitting the horizontal projection of the cloud at each time point into an ellipse and calculating the time-varying rates of change of the ellipse's major axis, minor axis, and eccentricity. Combining the ratio of the spatial distance between the centroid positions of adjacent time points in the trajectory to the time interval, the cloud's movement velocity is obtained. The movement velocity is decomposed into a component along the average airflow direction of the weather system and a component perpendicular to the average airflow direction. The former reflects the translation of the cloud with the large-scale environmental airflow, while the latter indicates the local disturbance caused by the gas cannon operation.

[0106] The extracted movement trajectory features, evolution features, and post-operation changes in cloud physical parameters are used as inputs to a human impact identification machine learning model. This model is constructed using a gradient boosting decision tree algorithm, and its training phase utilizes a large number of historical gas cannon operation cases and corresponding meteorological observation data. The model's input feature vector includes 48 dimensions: 12 dimensions of movement trajectory-related features (including average movement speed, movement direction, trajectory curvature, etc.), 18 dimensions of intensity evolution features (including reflectivity factor change rate, echo top height change rate, vertical integral liquid water content change rate, etc.), and 18 dimensions of morphological evolution features (including cloud area change rate, shape parameter change rate, etc.). The model output is an estimate of the human impact component, representing the proportion of the total changes in cloud physical parameters observed after the operation that can be attributed to the gas cannon operation.

[0107] During model inference, the input features are first standardized, and then ensembled predictions are performed using 200 decision trees. Each decision tree is trained based on different feature subsets and sample subsets, and the final human impact component is output through a weighted average. To improve the reliability of the estimation, the model also outputs a prediction confidence interval, the width of which reflects the uncertainty of the estimation result. When the input features show that the cloud movement trajectory closely matches the operation point location, the evolution features show a significant turning point near the operation time, and this turning point pattern is similar to historical effective operation cases, the model gives a larger human impact component and a narrower confidence interval. Conversely, if the cloud movement trajectory is far from the operation point, or the evolution features do not show any abnormal changes related to the operation, the human impact component is smaller or even close to zero. In this way, a reasonable decomposition of the changes in cloud physical parameters observed after the operation is achieved, effectively distinguishing between the natural evolution part and the human impact part, and providing a quantitative indicator of human impact for subsequent effect evaluation.

[0108] Based on the movement trajectory, the evolution characteristics, and the actual observed changes in cloud physical parameters after the operation, the artificial impact components caused by the gas cannon operation are obtained using the artificial impact identification machine learning model, including:

[0109] Based on the aforementioned evolutionary characteristics, the sequence of observed cloud physical parameters at each moment before the operation was extracted;

[0110] Time series analysis is performed on the observed data sequence to determine the growth or dissipation pattern of cloud physical parameters over time, and the time evolution function of cloud physical parameters is determined based on the growth or dissipation pattern.

[0111] The time evolution function is extrapolated from the pre-operation period to the post-operation period to generate expected values ​​of cloud physics parameters at each moment in the post-operation period under the condition that no gas cannon operation was carried out. The time series formed by the expected values ​​is used as the natural evolution baseline.

[0112] The actual observed values ​​of cloud physical parameters at each time point within the spatial range and time window after the operation are obtained. The actual observed values, the natural evolution baseline, the movement trajectory, and the evolution characteristics are input into the artificial influence recognition machine learning model. The deviation at each time point is determined by the artificial influence recognition machine learning model.

[0113] The deviation at each time point is integrated over time, and spatial calculations are performed based on the cloud space range and the time integration results to obtain the human impact component caused by the gas cannon operation.

[0114] After acquiring the cloud movement trajectory and evolution characteristics, it is necessary to accurately decompose the changes in cloud physical parameters observed after the operation to distinguish between the natural evolution part and the part caused by the human operation. This process first extracts the observed value sequence of cloud physical parameters at each moment from the radar echo data in the period before the operation. These cloud physical parameters include radar reflectivity factor, cloud top height, cloud base height, echo intensity, and echo volume. During the extraction process, the parameter values ​​at each scanning moment are arranged into a continuous sequence according to time order. Data from 30 to 60 minutes before the operation is usually selected, and the time resolution is kept consistent with the radar scanning cycle, generally 5 to 6 minutes. The length of the observed value sequence directly affects the accuracy of subsequent time series analysis. If the sequence is too short, the evolution pattern will not be obvious, while if the sequence is too long, it will contain cloud characteristics at different development stages.

[0115] When performing time series analysis on observed data sequences, a trend decomposition method is used to decompose the sequence into a trend term, a periodic term, and a random term. The trend term reflects the overall direction of change of cloud physical parameters and is extracted using the moving average method or polynomial fitting method. In practical applications, a first-order difference operation is performed on the radar reflectivity factor sequence. If the difference values ​​at three consecutive time points are all positive and gradually increase, it is determined to be a growth mode; if the difference values ​​are all negative and gradually increase in absolute value, it is determined to be a dissipation mode; if the difference values ​​fluctuate near zero and the fluctuation amplitude is less than a set threshold, it is determined to be a stable maintenance mode. For growth modes, exponential or logarithmic functions are used for fitting. The choice of function form depends on the growth rate characteristics of the observed data sequence. When the growth rate is decreasing, the logarithmic function is preferred, and when the growth rate is increasing, the exponential function is chosen. For dissipation modes, a negative exponential decay function is usually used for fitting. During the fitting process, the least squares method is used to solve for the function parameters to ensure that the sum of squared residuals between the fitted curve and the observation points is minimized.

[0116] After determining the time evolution function of cloud physical parameters, this function is extrapolated from the pre-operation period to the post-operation period. The starting point of the extrapolation is set as the start time of the gas cannon operation, and the extrapolation duration is determined based on the expected duration of the operation's impact, generally 30 to 90 minutes after the operation. In the extrapolation calculation, each time point after the operation is substituted into the time evolution function to calculate the expected values ​​of cloud physical parameters at each time point under conditions where no gas cannon operation was carried out. These expected values ​​are arranged in chronological order to form a natural evolution baseline, which represents the evolution trajectory of the cloud body under natural conditions. To ensure the rationality of the extrapolation results, physical constraints must be imposed on the extrapolation process. For example, the extrapolated value of the radar reflectivity factor cannot exceed the theoretical maximum value of this type of cloud body under current weather conditions, and the extrapolated value of the cloud top height cannot exceed the equilibrium height corresponding to the current ambient temperature stratification.

[0117] Simultaneously, radar scanning was used to acquire actual observations of the cloud's spatial range and cloud physical parameters at various times within the operational period after the operation. The spatial range was determined centered on the operational point, based on the theoretical influence radius of the gas-fired cannon, generally ranging from 5 to 15 kilometers horizontally and from the cloud base to the cloud top vertically. The time window was kept consistent with the time length of the natural evolution baseline. Extraction of actual observations required spatial matching to ensure that the observations extracted at each moment corresponded to the same cloud region. This was achieved through cloud identification and tracking algorithms, guaranteeing the continuity of spatial location.

[0118] The model for identifying human impacts is fed with actual observations, natural evolution baselines, movement trajectories, and evolutionary features. This model employs an ensemble learning framework, combining random forest and gradient boosting decision tree algorithms to handle multi-dimensional input features and output deviations at each time step. The input feature vector includes the difference between the current observed and expected values, the deviation from the previous time step, the cloud movement velocity vector, the cloud area change rate, and the echo intensity gradient. During model training, a sample set is constructed using historical operational and non-operational data. Positive samples represent cases where the operation actually had an impact, while negative samples represent cases where no operation was performed or the operation was ineffective. Cross-validation is implemented during training, dividing the dataset into training and validation sets in a 7:3 ratio. Hyperparameters such as tree depth, minimum number of samples per leaf node, and learning rate are adjusted to minimize the root mean square error (RMSE) on the validation set.

[0119] The deviation output by the model represents the degree of deviation of the actual observed values ​​at each time point from the natural evolution baseline. Positive deviation indicates that cloud physical parameters are enhanced due to human intervention, while negative deviation indicates a weakening. The deviation at each time point is integrated over time, and the trapezoidal integral method is used to calculate the cumulative effect of the deviation over the entire post-operation period. The formula for calculating the time integral is to multiply the arithmetic mean of the deviations at two adjacent time points by the time interval, and then sum the results over all time periods. The integral result reflects the overall strength of human influence in the time dimension.

[0120] Based on time integration, spatial calculations are performed according to the cloud's spatial extent, dividing the area affected by the operation into several grid cells. Each cell has a horizontal scale of 1 km x 1 km and a vertical scale of 500 m. For each grid cell, the weighting coefficient of the operation's impact is determined based on its distance from the operation point and its azimuth, combined with the cloud's movement trajectory. Grid cells closer to the operation point and located downwind of the operation point have greater weights. The time integration result is multiplied by the volume and weighting coefficient of each grid cell, and then summed over all grid cells to obtain the artificial impact component, which characterizes the total artificial impact on the entire cloud system. The physical meaning of this component is a comprehensive reflection of the spatial and temporal increments in cloud physical parameters caused by the gas cannon operation, and it can be used to quantitatively assess the operation's promoting or inhibiting effect on cloud development.

[0121] Acquire meteorological radar echo data of a control area that is in the same weather system as the target area but has not been subjected to gas cannon operations. Based on the meteorological radar echo data of the control area, extract cloud evolution characteristics of the control area, including:

[0122] Obtain the spatial extent boundary and weather system type characteristics of the weather system in which the target area is located, and determine a set of candidate areas that are in the same weather system as the target area based on the spatial extent boundary;

[0123] Based on each candidate region in the candidate region set, the similarity index between each candidate region and the target region on different element distributions is calculated. The candidate region set is then sorted according to the similarity index, and the candidate region with the highest similarity index is selected as the control region.

[0124] Meteorological radar echo data of the control area within the same time range as the target area before and after the operation are obtained. The meteorological radar echo data of the control area are then subjected to echo intensity threshold segmentation and three-dimensional spatial gridding to determine the cloud characteristics of the cloud body at each time point in the control area.

[0125] Based on the cloud characteristics of the control area at each time, the cloud physical structure parameters of the cloud in the control area at each time are calculated. The cloud physical structure parameters include cloud top height, liquid water content and cloud volume.

[0126] Calculate the time change rate of the cloud physical structure parameters between adjacent time points, and combine the cloud physical structure parameters at each time point with the time change rate to obtain the cloud evolution characteristic quantity of the control area.

[0127] When evaluating the effectiveness of gas cannon operations, it is necessary to introduce a control area that is under the same weather system as the target area but has not undergone operations for comparison. By comparing and analyzing the differences in cloud evolution between the two areas within the same time period, the influence of the weather system's own evolution can be effectively eliminated, thereby more accurately identifying the actual effects of the gas cannon operations.

[0128] The spatial boundaries of the weather system encompassing the target area are determined. Based on meteorological satellite cloud images and weather analysis maps, a comprehensive assessment is conducted. For frontal systems, the location and extent of the frontal zone are determined by identifying the zone of maximum surface temperature gradient and cloud system boundaries. For low-pressure systems, the location of the low-pressure center and its influence range are delineated based on isobar distribution characteristics. For typhoon systems, the spatial scale of typhoon influence is determined based on the typhoon eye location and spiral cloud band distribution. Simultaneously, weather system type characteristics are extracted, including system movement direction, speed, intensity level, and main influence altitude. Based on the determined spatial boundaries of the weather system, a set of candidate regions is delineated within the system's influence area. Candidate regions must meet the basic conditions of being under the same weather system as the target area and not having undergone any weather modification operations during the operation period.

[0129] For each candidate region in the candidate region set, similarity indices between it and the target region in the distribution of multiple meteorological elements are calculated. First, altitude similarity is calculated by obtaining topographic parameters such as average altitude and topographic relief of the candidate and target regions using a digital elevation model, and a normalized distance metric is used to assess the similarity of topographic conditions. Second, airflow condition similarity is calculated by extracting parameters such as average wind speed, wind direction, and vertical velocity of the candidate and target regions during the operation period using wind field data output from numerical weather prediction models, and a vector similarity method is used to assess the consistency of atmospheric dynamic conditions. Third, water vapor condition similarity is calculated by obtaining water vapor parameters such as whole-layer precipitable water and vertical profiles of relative humidity for the two regions based on radiosonde or reanalysis data, and a correlation coefficient method is used to assess the similarity of the water vapor environment. Finally, cloud initial state similarity is calculated by extracting cloud top height, cloud base height, echo intensity distribution, cloud volume, and other cloud characteristic parameters of the two regions before the operation, and an eigenvector distance method is used to assess the closeness of the initial cloud states.

[0130] The aforementioned similarity indicators are weighted and synthesized, with the weighting coefficients dynamically adjusted based on different weather system types and operational objectives. For convective precipitation operations, atmospheric dynamic conditions and water vapor conditions have higher weights; for stratiform cloud precipitation operations, the initial cloud state has a relatively higher weight. The candidate region set is then sorted in descending order based on the comprehensive similarity index, and the candidate region with the highest similarity index is selected as the control region. When the highest similarity score falls below a set threshold, the search range of candidate regions is expanded or the similarity evaluation criteria are adjusted to ensure sufficient comparability of the control regions.

[0131] Meteorological radar echo data of the control area were acquired within the same time range as that of the target area before and after the operation. Assuming the observation period before the operation in the target area was t_1 to t_2, the operation time was t_2, and the observation period after the operation was t_2 to t_3, all radar volume scan data of the control area within the same time range t_1 to t_3 were acquired. The acquired meteorological radar echo data of the control area were then segmented by echo intensity threshold. An echo intensity threshold of 15 dBZ was set, and echoes below this threshold were considered noise or non-precipitation echoes and discarded, retaining only valid cloud precipitation echo signals.

[0132] The effective echo data is processed into a three-dimensional spatial grid. Horizontally, an equally spaced Cartesian grid is used with a grid spacing of 1 km. Vertically, equal-height layers are used with a height layer interval of 500 meters, ranging from ground level to 20 km. A three-dimensional spatial interpolation algorithm is used to transform the echo data from the radar spherical coordinate system to the Cartesian grid coordinate system. Kriging interpolation or inverse range-weighted interpolation is employed to reduce interpolation errors while maintaining spatial resolution. During the gridding process, each grid point is assigned radar observation parameters such as echo intensity, radial velocity, and spectral width.

[0133] Based on gridded echo data, a connected component labeling algorithm is used to identify independent cloud bodies at different times within the control area. A set of interconnected grid points in three-dimensional space with echo intensities exceeding a threshold is defined as a cloud body. Cloud body segmentation is performed using a six-neighbor or twenty-six-neighbor connectivity criterion. For each identified cloud body, its three-dimensional spatial extent is extracted, and geometric features such as maximum height, minimum height, horizontal coverage area, and cloud volume are calculated. The echo intensity distribution within the cloud body is statistically analyzed, and intensity features such as average echo intensity, maximum echo intensity, and vertical echo intensity profile are calculated. For convective clouds with significant vertical development, the location and height of their strong echo core region are identified, and the intensity value and spatial scale of the echo core are extracted.

[0134] Based on the extracted cloud features, the cloud physical structure parameters of the control area at each time point were calculated. According to the empirical relationship between echo intensity and rainfall intensity, the echo intensity data was converted into precipitation intensity distribution, and the total cloud precipitation was obtained by integration. Using the vertical profile characteristics of the echo intensity, the height distribution of ice and liquid phase particles within the cloud was identified, and the liquid water content and ice crystal content of the cloud were estimated. By analyzing the echo top height and bright band characteristics, the maturity of the cloud and the precipitation type were determined. For stratiform clouds, the average cloud thickness and cloud water path were calculated; for convective clouds, the convection intensity index and vertical development speed were calculated.

[0135] Calculate the rate of change of cloud physical structure parameters between adjacent time points to reflect the dynamic characteristics of cloud evolution. For cloud height parameters, calculate the rate of change of cloud top height. ,in This represents the change in cloud top height between adjacent time points. This represents the time interval. Positive values ​​indicate that the cloud is moving upwards, while negative values ​​indicate that the cloud is dissipating or sinking.

[0136] The human influence component is normalized and corrected using the evolution characteristics of the cloud body in the control area to obtain a normalized human influence index. This normalized index is then compared with a preset effect evaluation standard to generate a quantitative evaluation result of the gas cannon operation effect, including:

[0137] The temporal rate of change of cloud physical structure parameters in the control area during the post-operation period is extracted from the cloud evolution characteristics of the control area, and the natural variation amplitude of cloud physical structure parameters in the control area during the post-operation period is calculated based on the temporal rate of change.

[0138] The ratio of the artificial influence component to the natural variation amplitude of the control area is calculated to obtain the normalized artificial influence index after eliminating the influence of regional background differences and the overall evolution of the weather system.

[0139] The normalized artificial influence index is compared with the normalized artificial influence index threshold corresponding to each effect level in the preset effect evaluation criteria. The effect level corresponding to the normalized artificial influence index is determined according to the numerical range in which the normalized artificial influence index is located.

[0140] Based on the normalized human impact index and the effect level corresponding to the normalized human impact index, a quantitative evaluation result of the gas cannon operation effect is generated.

[0141] After obtaining the artificial weather modification component, it is necessary to compare and analyze the cloud evolution characteristics with those of the control area to eliminate the influence of background differences between different regions and the overall evolution trend of the weather system, thereby obtaining a normalized evaluation index that truly reflects the effectiveness of the gas cannon operation. The selection of the control area needs to meet the conditions of being under the control of the same weather system as the target area, having similar geographical environmental characteristics, but without any artificial weather modification operations. Typically, an area within 20 to 50 kilometers upwind or laterally of the target area is selected as the control area.

[0142] When extracting the temporal change rate of cloud physical structure parameters in the post-operation period from the cloud evolution characteristics of the control area, it is necessary to pay attention to the temporal evolution characteristics of several key cloud physical parameters. These parameters include cloud top height, cloud base height, cloud vertical thickness, cloud horizontal scale, vertical integral value of radar reflectivity factor, echo top height, maximum reflectivity factor intensity, and reflectivity factor centroid height. For each cloud physical structure parameter, a time series analysis method is used to calculate its change rate in the post-operation period. Specifically, radar observation data of the control area cloud body from 0 to 60 minutes after the operation are selected. Each volume scan cycle is typically 6 minutes, which yields approximately 10 consecutive observation samples. For the cloud top height parameter, assuming that the cloud top height observed in the first time interval after the operation is 9.2 km and the cloud top height observed in the tenth time interval is 8.7 km, the total change in cloud top height in the 60 minutes after the operation is -0.5 km, corresponding to a temporal change rate of approximately 8.3 meters per minute. A similar method is used to calculate the temporal change rate of other cloud physical structure parameters.

[0143] When calculating the natural variation of cloud physical structure parameters in the control area based on the time-varying rate of change, the corresponding natural variation amplitude is obtained by multiplying the time-varying rate of change of each parameter by the length of the observation period. Considering that different cloud physical parameters have different dimensions and numerical ranges, it is necessary to standardize the natural variation amplitude of each parameter. The relative variation amplitude is used as the standardization index, that is, the absolute change of the parameter is divided by its initial value. Taking the vertical integral value of radar reflectivity factor as an example, assuming that the initial value of this parameter after the operation is 450 kg / m², and it decreases to 405 kg / m² after 60 minutes of natural evolution, the absolute change is -45 kg / m², and the relative variation amplitude is -10%. This relative variation amplitude reflects the natural development trend of the cloud body under the background of the overall evolution of the weather system without human intervention.

[0144] To achieve normalization correction, the ratio of the artificial impact component to the natural variation amplitude of the control area must be calculated, ensuring that both use the same parameter system and calculation period. The artificial impact component is obtained through decomposition using an artificial impact identification machine learning model, representing the contribution of gas cannon operations to the changes in cloud physical parameters of the target area. Assuming that for the vertical integral value of the radar reflectivity factor, the artificial impact component is -68 kg / m², while the natural variation amplitude of this parameter in the control area during the same period is -45 kg / m², then the normalized artificial impact index is the ratio of the two, with a value of 1.51. This index indicates that, after deducting the influence of natural weather system evolution, gas cannon operations amplified the variation amplitude of this parameter in the target area by 51% compared to natural evolution. When the normalized artificial impact index is close to 1, it indicates that the impact of the operation is comparable to the degree of natural evolution; when the index is significantly greater than 1, it indicates that the operation produced a significant additional impact; when the index is less than 1, it indicates that the impact of the operation is weaker than the natural evolution trend or that the operation did not produce the expected effect.

[0145] To obtain a comprehensive Normalized Artificial Impact Index (NAI), it is necessary to weight and fuse the NAIs of multiple cloud physics parameters. Weight coefficients are assigned based on the importance of different parameters to the precipitation process. For example, parameters directly related to precipitation, such as the vertical integral of the radar reflectivity factor, echo top height, and cloud vertical thickness, are given higher weights, while morphological parameters such as cloud horizontal scale are given lower weights. Assuming five core parameters are selected for fusion, with weight coefficients of 0.30, 0.25, 0.20, 0.15, and 0.10, the corresponding NAIs are 1.51, 1.32, 1.18, 1.45, and 1.08, respectively. The comprehensive NAI, calculated through weighted summation, is approximately 1.33.

[0146] The preset effect evaluation criteria need to be established based on statistical analysis of a large number of historical operation cases, classifying the effect of gas cannon operations into multiple levels, such as significantly enhanced, noticeably enhanced, moderately enhanced, no significant impact, and suppressive effect. Each effect level corresponds to a specific normalized human impact index (NHI) threshold range. For example, when the NHI is greater than 1.5, it is judged as a significantly enhanced effect; when the index is between 1.2 and 1.5, it is judged as a noticeably enhanced effect; when the index is between 1.0 and 1.2, it is judged as a moderately enhanced effect; when the index is between 0.8 and 1.0, it is judged as no significant impact; and when the index is less than 0.8, it is judged as having a suppressive effect. The setting of these thresholds needs to comprehensively consider the differences in operation effects under different regions, seasons, and weather system types, and can be appropriately adjusted according to regional characteristics.

[0147] When determining the corresponding effect level based on the numerical range of the Normalized Anthropogenic Impact Index (NARI), a step-by-step comparison and judgment are required. For the calculated NARI of 1.33, the first step is to determine if it is greater than 1.5; the result is no, therefore the significantly enhanced level is excluded. The next step is to determine if it falls within the range of 1.2 to 1.5; the result is yes, therefore the effect level of this gas cannon operation is determined to be significantly enhanced. This level determination indicates that, after deducting the influence of natural weather system evolution and regional background differences, the gas cannon operation did indeed significantly promote the cloud precipitation process in the target area, with an enhancement of approximately 33%.

[0148] When generating quantitative evaluation results of the gas cannon operation effect based on the normalized human impact index and its corresponding effect level, a structured evaluation report containing multi-dimensional information is required.

[0149] A second aspect of this invention provides a meteorological impact assessment system for gas cannon operations that integrates AI algorithms, comprising:

[0150] The data acquisition unit is used to acquire meteorological radar echo data of the target area before and after the gas cannon operation, and to perform quality control processing on the meteorological radar echo data to obtain cloud precipitation echo data.

[0151] The cloud feature unit is used to input the cloud precipitation echo data into the cloud recognition deep learning model to obtain the three-dimensional structure features of the cloud in the target area before and after the operation, and to calculate the changes in cloud physical parameters in the target area before and after the operation based on the three-dimensional structure features of the cloud. The changes in cloud physical parameters include changes in cloud top height, changes in liquid water content, and changes in cloud volume.

[0152] The artificial influence unit is used to determine the movement trajectory and evolution characteristics of cloud bodies within the target area by performing spatiotemporal continuity analysis on multi-time radar echo data. The movement trajectory and evolution characteristics are then input into the artificial influence recognition machine learning model to decompose the changes in cloud physical parameters observed after the operation to obtain the artificial influence component caused by the gas cannon operation.

[0153] The reference area unit is used to acquire meteorological radar echo data of the reference area which is in the same weather system as the target area and has not carried out gas cannon operations. Based on the meteorological radar echo data of the reference area, the evolution characteristics of the cloud body in the reference area are extracted.

[0154] The effect evaluation unit is used to normalize and correct the artificial influence component by using the evolution characteristics of the cloud body in the control area to obtain the normalized artificial influence index. The normalized artificial influence index is then compared with the preset effect evaluation standard to generate a quantitative evaluation result of the gas cannon operation effect.

[0155] A third aspect of the present invention provides an electronic device, comprising:

[0156] processor;

[0157] Memory used to store processor-executable instructions;

[0158] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0159] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0160] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0161] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing the meteorological impact of gas cannon operations by integrating AI algorithms, characterized in that, include: Acquire meteorological radar echo data of the target area before and after the gas cannon operation, and perform quality control processing on the meteorological radar echo data to obtain cloud precipitation echo data; The cloud precipitation echo data is input into a cloud recognition deep learning model to obtain the three-dimensional structure features of clouds in the target area before and after the operation. Based on the three-dimensional structure features of clouds, the changes in cloud physical parameters in the target area before and after the operation are calculated. The changes in cloud physical parameters include changes in cloud top height, changes in liquid water content, and changes in cloud volume. By performing spatiotemporal continuity analysis on multi-time radar echo data, the movement trajectory and evolution characteristics of clouds within the target area are determined. The movement trajectory and evolution characteristics are then input into a machine learning model for human impact identification to decompose the changes in cloud physical parameters observed after the operation to obtain the human impact component caused by the gas cannon operation. Acquire meteorological radar echo data of a control area that is in the same weather system as the target area and has not carried out gas cannon operations; extract cloud evolution characteristics of the control area based on the meteorological radar echo data of the control area. The artificial influence component is normalized and corrected using the evolution characteristics of the cloud body in the control area to obtain the normalized artificial influence index. The normalized artificial influence index is then compared with the preset effect evaluation standard to generate a quantitative evaluation result of the gas cannon operation effect.

2. The method according to claim 1, characterized in that, The cloud precipitation echo data is input into a cloud recognition deep learning model to obtain the three-dimensional structure features of the cloud in the target area before and after the operation. Based on the three-dimensional structure features of the cloud, the changes in cloud physical parameters in the target area before and after the operation are calculated, including: The cloud precipitation echo data is vertically layered to divide the target area into multiple height layers. The cloud precipitation echo data of each height layer is input into the cloud body recognition deep learning model. The model extracts the horizontal distribution features and vertical gradient features of the reflectivity factor for each height layer to obtain the layered cloud structure features. Based on the aforementioned layered cloud structure characteristics, the phase distribution of cloud water particles at each altitude layer is identified to obtain phase distribution information; The liquid water content at each altitude layer is calculated based on the layered cloud structure characteristics and the phase distribution information, and the three-dimensional cloud structure characteristics of the target area are obtained by vertical integration of the liquid water content at each altitude layer. The three-dimensional structural features of the cloud before and after the operation are compared. The changes in cloud top height, liquid water content and cloud volume at each altitude level are calculated. The changes at each altitude level are then spatially integrated to obtain the changes in cloud physical parameters in the target area.

3. The method according to claim 2, characterized in that, Based on the layered cloud structure characteristics and the phase distribution information, the liquid water content at each altitude layer is calculated, and the three-dimensional cloud structure characteristics of the target area are obtained by vertical integration of the liquid water content at each altitude layer, including: Based on the spatial locations of the supercooled water region, ice crystal region, and mixed phase region in the phase distribution information, the corresponding phase weighting coefficients are determined for each height layer. Based on the phase weighting coefficient, the reflectivity factor in the layered cloud structure features is phase-corrected to obtain the phase-corrected reflectivity factor. The phase-corrected reflectivity factor is correlated with the particle spectrum distribution characteristics of each phase region in the phase distribution information to determine the phase transition relationship between the reflectivity factor and the liquid water content; Based on the phase transition relationship, the phase-corrected reflectivity factor of each height layer is converted into the liquid water content of the corresponding height layer to obtain the liquid water content of each height layer; The liquid water content of each altitude layer is integrated layer by layer in the vertical direction from the cloud base to the cloud top, and the cloud boundary range of each altitude layer is determined by combining the horizontal distribution characteristics in the layered cloud structure features, so as to obtain the three-dimensional cloud structure features of the target area.

4. The method according to claim 1, characterized in that, By performing spatiotemporal continuity analysis on multi-time radar echo data, the movement trajectory and evolution characteristics of clouds within the target area are determined. These movement trajectories and evolution characteristics are then input into a machine learning model for identifying human impacts. This model decomposes the observed changes in cloud physical parameters after the operation to obtain the human impact components caused by the gas cannon operation, including: Based on meteorological radar echo data of the target area at multiple consecutive times before and after the gas cannon operation, a multi-time cloud feature sequence was obtained; The centroid position coordinates of the cloud body at each time point are calculated to obtain the centroid position coordinates of the cloud body at each time point in the multi-time cloud body feature sequence. The spatial displacement vector between the centroid position coordinates of the cloud body at adjacent time points is calculated to obtain the centroid displacement vector sequence. The centroid displacement vector sequence is subjected to a continuity test, and the centroid displacement vectors that pass the continuity test are connected in chronological order to form the movement trajectory of the cloud within the target area; Based on the reflectivity factor intensity and cloud morphology characteristics of the cloud at each time point in the multi-time cloud feature sequence, the temporal change rate of the cloud reflectivity factor and the deformation rate of the cloud morphology between adjacent time points are calculated. Combined with the movement speed in the movement trajectory, the evolution characteristics of the cloud within the target area are extracted. Based on the movement trajectory, the evolution characteristics, and the actual observed changes in cloud physical parameters after the operation, the artificial influence component caused by the gas cannon operation is obtained using the artificial influence recognition machine learning model.

5. The method according to claim 4, characterized in that, Based on the movement trajectory, the evolution characteristics, and the actual observed changes in cloud physical parameters after the operation, the artificial impact components caused by the gas cannon operation are obtained using the artificial impact identification machine learning model, including: Based on the aforementioned evolutionary characteristics, the sequence of observed cloud physical parameters at each moment before the operation was extracted; Time series analysis is performed on the observed data sequence to determine the growth or dissipation pattern of cloud physical parameters over time, and the time evolution function of cloud physical parameters is determined based on the growth or dissipation pattern. The time evolution function is extrapolated from the pre-operation period to the post-operation period to generate expected values ​​of cloud physics parameters at each moment in the post-operation period under the condition that no gas cannon operation was carried out. The time series formed by the expected values ​​is used as the natural evolution baseline. The actual observed values ​​of cloud physical parameters at each time point within the spatial range and time window after the operation are obtained. The actual observed values, the natural evolution baseline, the movement trajectory, and the evolution characteristics are input into the artificial influence recognition machine learning model. The deviation at each time point is determined by the artificial influence recognition machine learning model. The deviation at each time point is integrated over time, and spatial calculations are performed based on the cloud space range and the time integration results to obtain the human impact component caused by the gas cannon operation.

6. The method according to claim 1, characterized in that, Acquire meteorological radar echo data of a control area that is in the same weather system as the target area but has not been subjected to gas cannon operations. Based on the meteorological radar echo data of the control area, extract cloud evolution characteristics of the control area, including: Obtain the spatial extent boundary and weather system type characteristics of the weather system in which the target area is located, and determine a set of candidate areas that are in the same weather system as the target area based on the spatial extent boundary; Based on each candidate region in the candidate region set, the similarity index between each candidate region and the target region on different element distributions is calculated. The candidate region set is then sorted according to the similarity index, and the candidate region with the highest similarity index is selected as the control region. Meteorological radar echo data of the control area within the same time range as the target area before and after the operation are obtained. The meteorological radar echo data of the control area are then subjected to echo intensity threshold segmentation and three-dimensional spatial gridding to determine the cloud characteristics of the cloud body at each time point in the control area. Based on the cloud characteristics of the control area at each time, the cloud physical structure parameters of the cloud in the control area at each time are calculated. The cloud physical structure parameters include cloud top height, liquid water content and cloud volume. Calculate the time change rate of the cloud physical structure parameters between adjacent time points, and combine the cloud physical structure parameters at each time point with the time change rate to obtain the cloud evolution characteristic quantity of the control area.

7. The method according to claim 1, characterized in that, The human influence component is normalized and corrected using the evolution characteristics of the cloud body in the control area to obtain a normalized human influence index. This index is then compared with a preset effect evaluation standard to generate a quantitative evaluation result of the gas cannon operation effect, including: The temporal change rate of cloud physical structure parameters in the control area during the post-operation period is extracted from the cloud evolution characteristics of the control area, and the natural variation amplitude of cloud physical structure parameters in the control area during the post-operation period is calculated based on the temporal change rate. The ratio of the artificial influence component to the natural variation amplitude of the control area is calculated to obtain the normalized artificial influence index after eliminating the influence of regional background differences and the overall evolution of the weather system. The normalized artificial influence index is compared with the normalized artificial influence index threshold corresponding to each effect level in the preset effect evaluation criteria. The effect level corresponding to the normalized artificial influence index is determined according to the numerical range in which the normalized artificial influence index is located. Based on the normalized human impact index and the effect level corresponding to the normalized human impact index, a quantitative evaluation result of the gas cannon operation effect is generated.

8. A meteorological impact assessment system for gas cannon operations integrating AI algorithms, used to implement the method as described in any one of claims 1-7, characterized in that, include: The data acquisition unit is used to acquire meteorological radar echo data of the target area before and after the gas cannon operation, and to perform quality control processing on the meteorological radar echo data to obtain cloud precipitation echo data. The cloud feature unit is used to input the cloud precipitation echo data into the cloud recognition deep learning model to obtain the three-dimensional structure features of the cloud in the target area before and after the operation, and to calculate the changes in cloud physical parameters in the target area before and after the operation based on the three-dimensional structure features of the cloud. The changes in cloud physical parameters include changes in cloud top height, changes in liquid water content, and changes in cloud volume. The artificial influence unit is used to determine the movement trajectory and evolution characteristics of cloud bodies within the target area by performing spatiotemporal continuity analysis on multi-time radar echo data. The movement trajectory and evolution characteristics are then input into the artificial influence recognition machine learning model to decompose the changes in cloud physical parameters observed after the operation to obtain the artificial influence component caused by the gas cannon operation. The reference area unit is used to acquire meteorological radar echo data of the reference area which is in the same weather system as the target area and has not carried out gas cannon operations. Based on the meteorological radar echo data of the reference area, the evolution characteristics of the cloud body in the reference area are extracted. The effect evaluation unit is used to normalize and correct the artificial influence component by using the evolution characteristics of the cloud body in the control area to obtain the normalized artificial influence index. The normalized artificial influence index is then compared with the preset effect evaluation standard to generate a quantitative evaluation result of the gas cannon operation effect.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images