Extreme weather event and ecological risk prediction method using generative adversarial network

Abstract

The invention belongs to the technical field of ecological risk prediction, and relates to an extreme weather event and ecological risk prediction method using a generative adversarial network, through a meteorological generation branch, spatial-temporal characteristics of a meteorological field are synchronously extracted by using a 3D residual network and Transform and a future meteorological state is predicted, and an ecological generation branch is based on a meteorological prediction result and topographic data. An ecological risk map is dynamically generated through hole convolution U-Net, and mismatch of resolution and dimension caused by cross-model interpolation is avoided; a discriminator introduces a physical conservation verification module and an ecological association module, physical rationality loss and ecological logic consistency loss of a generator are jointly optimized in adversarial training, in addition, a dynamic feedback mechanism corrects and generates errors through real-time observation data, and spectrum normalization is utilized to constrain model parameters, so that the reliability of the model is improved. The accuracy of extreme event and ecological risk coupling prediction is further improved; therefore, seamless space-time coupling of the weather and the ecological field is realized, and the generated result has both physical conservation and ecological relevance.

Description

Technical Field

[0001] This application belongs to the technical field of ecological risk prediction. More specifically, it relates to a method for predicting extreme weather events and ecological risks using generative adversarial networks. Background Art

[0002] The intensification of global climate change has led to frequent extreme weather events (such as hurricanes, heatwaves, heavy rains), and the ecological risks (such as vegetation degradation, loss of species habitats) caused by them show complex spatio-temporal correlations. Traditional meteorological predictions and ecological impact assessments are often carried out separately: meteorological models focus on simulating physical processes, while ecological models rely on static thresholds or lag response analysis. This separation leads to two key defects: Lack of physical rationality: Meteorological predictions do not provide real-time feedback on the dynamic responses of ecosystems and may violate the constraints of energy conservation or atmospheric motion. Ecological response lag: Risk assessments rely on historical experience and cannot capture the non-linear coupling relationship between extreme events and ecological degradation.

[0003] The current mainstream method uses a cascade of numerical weather prediction models (NWP) and ecological statistical models: Meteorological prediction stage: Based on physical equations (such as the Navier-Stokes equation), simulate atmospheric motion and output a high-resolution meteorological field. Ecological assessment stage: Input the meteorological prediction results into an ecological model (such as a species distribution model) and calculate the risk probability through regression analysis.

[0004] However, there are the following problems with the above method. The meteorological model and the ecological model operate independently, and data transfer depends on spatio-temporal interpolation, resulting in mismatches in the spatio-temporal resolution and physical dimensions between the meteorological field and the ecological field (such as the spatial misalignment between the wind speed gradient and the vegetation stress index). Summary of the Invention

[0005] The present invention provides a method for predicting extreme weather events and ecological risks using generative adversarial networks, aiming to solve the technical problem that the meteorological model and the ecological model operate independently, and data transfer depends on spatio-temporal interpolation, resulting in mismatches in the spatio-temporal resolution and physical dimensions between the meteorological field and the ecological field.

[0006] The method for predicting extreme weather events and ecological risks using generative adversarial networks includes the following steps: Data collection: Collect original meteorological data, remote sensing data, geographical data, and ecological baseline data. Data processing: Align the collected original meteorological data, remote sensing data, geographical data, and ecological baseline data in space and time, and then fuse them to construct a three-dimensional feature cube. Meteorological dynamic feature calculation: Based on the constructed three-dimensional feature cube, calculate the quasi-geostrophic potential vorticity, extract the water vapor input flux, and obtain the physical diagnostic field; Ecological response modeling: Construct a vegetation stress index and generate a species habitat suitability map; Generative adversarial joint training: Construct a two-branch generative adversarial network, where the generator includes a meteorological generation branch and an ecological generation branch. The meteorological generation branch extracts spatial features based on a 3D residual network and captures temporal dependencies by combining Transformer to output the meteorological field at future times; the ecological generation branch generates a dynamic risk probability map using a dilated convolutional U-Net model based on the predicted meteorological field and terrain data; the discriminator is divided into a physical conservation verification module and an ecological correlation module. The physical conservation verification module is used to verify the rationality of the generated meteorological field; the ecological correlation module verifies the logical consistency of the ecological response by analyzing the statistical correlation between the vegetation index and meteorological variables; during the training process, the generator simultaneously optimizes the adversarial loss, physical conservation loss, and ecological correlation loss, and finally generates a result of seamless coupling of meteorological prediction and ecological risk; Dynamic feedback optimization: Use the gradient information of the discriminator to locate the key areas of the generation error, combine the real-time observation data of satellites and buoy stations to perform linear correction on the input layer of the generator, and then use spectral normalization to limit the parameter range of the generator.

[0007] In the present invention, by performing spatio-temporal alignment on multi-source data and fusing them into a three-dimensional feature cube, the spatio-temporal consistency of the input data is ensured; subsequently, the meteorological generation branch uses a 3D residual network and Transformer to synchronously extract the spatio-temporal features of the meteorological field and predict the future meteorological state, while the ecological generation branch dynamically generates an ecological risk map based on the meteorological prediction result and terrain data through a dilated convolutional U-Net, avoiding the resolution and dimension mismatch caused by cross-model interpolation; the discriminator introduces a physical conservation verification module and an ecological correlation module to jointly optimize the physical rationality loss and ecological logical consistency loss of the generator in the adversarial training, ensuring that the generated result conforms to both the atmospheric dynamics law and the ecological response mechanism; in addition, the dynamic feedback mechanism corrects the generation error through real-time observation data and uses spectral normalization to constrain the model parameters, further improving the accuracy of the coupled prediction of extreme events and ecological risks; therefore, the present invention realizes the seamless spatio-temporal coupling of the meteorological and ecological fields, the generated result has both physical conservation and ecological relevance, and can dynamically adapt to real-time observation data, providing a reliable technical support for the high-precision collaborative early warning of extreme weather and its ecological impacts.

[0008] Preferably, the data processing includes the following steps: Downscale the meteorological data space through cubic spline interpolation, and use bilinear interpolation to improve the temporal resolution of remote sensing data to the hourly level to ensure that all data are in a unified spatio-temporal coordinate system; then eliminate invalid areas through dynamic masking technology, and finally fuse data of different dimensions into a multi-dimensional spatio-temporal dataset to obtain standardized spatio-temporal grid data.

[0009] Preferably, the calculation of the quasi-geostrophic potential vorticity includes the following steps: Relative vorticity calculation: Calculate the spatial gradients of the zonal wind and the meridional wind to obtain the rate of change of the wind field on the horizontal plane, and then calculate the relative vorticity based on the second-order central difference method; Coriolis parameter calculation: Calculate the Coriolis parameter for each grid point using the standard formula for calculating the Coriolis parameter; Potential temperature gradient calculation: Calculate the potential temperature value based on the pressure and temperature, and then calculate the gradient of the potential temperature in the vertical direction using the central difference method; Quasi-geostrophic potential vorticity calculation: Combine the relative vorticity, the Coriolis parameter, and the potential temperature gradient to calculate the quasi-geostrophic potential vorticity for each grid point, and then normalize the calculated quasi-geostrophic potential vorticity.

[0010] Preferably, the specific steps for extracting the water vapor input flux are as follows: Vertical integration: According to the 37-layer pressure data provided by ERA5, use the trapezoidal integration method to calculate the contribution of each layer to the total water vapor flux; Statistical water vapor flux components: Integrate the product of the zonal wind speed and the specific humidity to obtain the zonal flux; Integrate the product of the meridional wind speed and the specific humidity to obtain the meridional flux; At each level, use the interval of the pressure layer to de-weight the contributions of each layer; Synthesize the water vapor flux: Calculate the magnitude of the total water vapor flux by calculating the sum of the squares of the zonal flux and the meridional flux, and calculate the direction angle of the water vapor transport flux by calculating the ratio of the meridional flux to the zonal flux.

[0011] Preferably, the construction of the vegetation stress index includes the following steps: Heat stress component calculation: For each pixel, calculate the difference between its daily maximum temperature and the tolerance threshold of the species, and use the Logistic function to determine the heat stress component; Moisture stress component calculation: Measure the severity of moisture stress based on the ratio of the calculated surface soil moisture to the critical moisture; where a terrain adjustment factor is introduced to adjust the moisture stress component when calculating the moisture stress component, and the terrain adjustment factor is calculated based on the terrain slope, and the larger the slope, the smaller the terrain adjustment factor; Comprehensive stress index calculation: Comprehensively consider the heat stress component and the moisture stress component, and combine the NDVI value of the vegetation to correct the stress to obtain the comprehensive stress index.

[0012] Preferably, the specific steps for generating the species habitat fitness map are as follows: Feature engineering: Screen environmental variables significantly correlated with species distribution through Moran's I index to obtain the selected features; MaxEnt model optimization: Based on the selected features and species distribution data, train the MaxEnt model and output the suitability scores for each habitat; Phenological regulation: Based on the vegetation index data, calculate the phenological phase angle of the species, calculate the phenological regulation coefficient according to the phenological phase angle, apply the phenological regulation coefficient to the suitability values output by the MaxEnt model to obtain the seasonally adjusted habitat suitability scores; Generate the species habitat suitability map according to the suitability values of each pixel.

[0013] Preferably, the physical conservation verification module is used to check whether the generated meteorological data follows the principle of physical conservation, and the specific steps are as follows: Mass conservation residual : Calculate the divergence of the fluid velocity field according to the conservation equation of fluid mechanics: ; In the formula: represents the air density, which is calculated by the ideal gas state equation; represents the three-dimensional vector of the wind speed field; represents the divergence of the fluid, which is calculated by the central difference method; Energy conservation residual Verification: ; In the formula: represents the kinetic energy density of the gas; represents the specific heat capacity at constant volume; represents the temperature field; represents the time derivative; Based on the above calculated mass conservation residual and energy conservation residual, compare with the threshold value to judge whether the generated meteorological data follows the principle of physical conservation.

[0014] Preferably, the steps for the ecological correlation module to verify the logical consistency of the ecological response by analyzing the statistical correlation between the vegetation index and meteorological variables are as follows: Normalized difference vegetation index: ; In the formula: represents the near-infrared band reflectance; represents the infrared band reflectance; represents the normalized difference vegetation index; Temperature anomaly Calculation: ; In the formula: represents the actual temperature at the current moment; represents the long-term average temperature; Mutual information calculation: Based on mutual information, calculate the correlation between NDVI and temperature anomaly: ; In the formula: represents the NDVI value and the temperature anomaly is the probability of their simultaneous occurrence; , are the marginal distributions of NDVI and temperature anomaly respectively; represents the number of bins of NDVI and temperature anomaly; Risk map matching degree: Measure the difference between the generated risk map and the real risk map through KL divergence: ; In the formula: represents the probability distribution of the generated ecological risk map; represents the probability distribution of the real ecological risk map; N represents the number of pixels of the risk map; represents the probability value of the i-th pixel of the generated ecological risk map; represents the probability value of the i-th pixel of the real ecological risk map; represents the risk map matching degree; When the difference between the calculated mutual information and the mutual information of the real data is less than the threshold, and the risk map matching degree is less than the threshold, it passes the verification of the ecological correlation module. If either of the above conditions is not met, the verification is not passed.

[0015] The beneficial effects of the present invention include: In the present invention, by performing spatio-temporal alignment on multi-source data and fusing it into a three-dimensional feature cube, the spatio-temporal consistency of the input data is ensured. Subsequently, the meteorological generation branch uses a 3D residual network and Transformer to synchronously extract the spatio-temporal features of the meteorological field and predict the future meteorological state. The ecological generation branch, based on the meteorological prediction results and terrain data, dynamically generates an ecological risk map through a dilated convolutional U-Net, avoiding the resolution and dimension mismatch caused by cross-model interpolation. The discriminator introduces a physical conservation verification module and an ecological correlation module to jointly optimize the physical rationality loss and ecological logical consistency loss of the generator during adversarial training, ensuring that the generated results conform to both the laws of atmospheric dynamics and the ecological response mechanism. In addition, the dynamic feedback mechanism corrects the generation error through real-time observation data and uses spectral normalization to constrain the model parameters, further improving the accuracy of the coupled prediction of extreme events and ecological risks. Therefore, the present invention realizes the seamless spatio-temporal coupling of the meteorological and ecological fields, and the generated results have both physical conservation and ecological relevance, and can dynamically adapt to real-time observation data, providing a reliable technical support for the high-precision collaborative early warning of extreme weather and its ecological impacts. Description of the Drawings

[0016] To more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0017] Figure 1 It is a block diagram of the overall steps provided by the embodiment of the present invention.

[0018] Figure 2 It is an exemplary structural diagram of the generative adversarial network provided by the embodiment of the present invention. Detailed Embodiments

[0019] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application more clearly understood, the following further details the present application in conjunction with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0020] See Figure 1 As shown, a method for optimizing the energy efficiency of a diving device based on environmental data prediction includes the following steps: A method for predicting extreme weather events and ecological risks using a generative adversarial network includes the following steps: Data collection: Collect original meteorological data, remote sensing data, geographical data, and ecological baseline data; Meteorological data: including temperature, precipitation, wind speed, humidity, etc., usually from meteorological stations or satellite observations; Remote sensing data: including vegetation indices (such as NDVI), land cover data, etc., usually obtained from satellite remote sensing images; Geographical data: including terrain height, land use type, slope, etc., usually from remote sensing data or Geographic Information System (GIS) data; Ecological baseline data: including species distribution maps, habitat suitability, etc.

[0021] Data processing: Align the collected original meteorological data, remote sensing data, geographical data, and ecological baseline data in space and time, and then perform fusion to construct a three-dimensional feature cube; The data processing includes the following steps: Spatial downscaling of meteorological data: Original meteorological data usually has a low resolution, such as 0.25°, and it needs to be downscaled to 1 km to more precisely simulate the impact of meteorology on the ecosystem. To perform spatial downscaling, a cubic spline interpolation method is used to smoothly adjust the data to make it more accurate. For example, let the original meteorological data be , its spatial resolution is , and the target spatial resolution is . The meteorological data with the target resolution is obtained through cubic spline interpolation : ; In the formula: represents the cubic polynomial coefficient; represents the spatial position; Temporal upscaling of remote sensing data: Remote sensing data has a low temporal resolution and needs to be upscaled to the hourly level to match the high-frequency meteorological data. Therefore, bilinear interpolation is used to interpolate the temporal resolution of remote sensing data. For example, set the original temporal resolution of remote sensing data to , and the target temporal resolution to . Through bilinear interpolation, calculate the remote sensing data corresponding to the target time point : ; In the formula: and represent the known time points; and represent the remote sensing data corresponding to the corresponding time points; Data Fusion and Construction of 3D Feature Cube: After data alignment, a 3D feature cube is constructed, where each dimension represents time, space (longitude and latitude), and different meteorological or ecological variables; all the aligned data are sorted by time and spatial location to form a multi-dimensional dataset. Assume is meteorological data, is remote sensing data, is geographical data, and a 3D cube is constructed: ; where: represents the 3D feature cube, t represents time; x and y represent spatial coordinates; Dynamic Masking: For invalid areas (such as oceans, cities, etc.), the dynamic masking technique is adopted to eliminate the influence of invalid areas. Through the land use / land cover map in the geographical information data and high-resolution remote sensing images, a mask map is generated, set to 0 in invalid areas and 1 in valid areas, and finally all data are multiplied by the mask map.

[0022] Then, based on the masked data, standardization processing is performed, and the standardized data are mapped onto a regular spatio-temporal grid to obtain the final spatio-temporal grid dataset .

[0023] Calculation of Meteorological Dynamic Characteristics: Based on the constructed 3D feature cube, the quasi-geostrophic potential vorticity is calculated, and the water vapor input flux is extracted to obtain the physical diagnostic field; The calculation of the quasi-geostrophic potential vorticity includes the following steps: Relative Vorticity Calculation: Calculate the spatial gradients of the zonal wind and meridional wind to obtain the rate of change of the wind field on the horizontal plane, and then calculate the relative vorticity based on the second-order central difference method; ; In the formula: and are the zonal and meridional wind speeds respectively; x and y are the meridional and zonal coordinates; Coriolis Parameter Calculation: Calculate the Coriolis parameter at each grid point using the standard formula for calculating the Coriolis parameter; Exemplarily: ; In the formula: represents the Coriolis parameter; represents latitude; represents the angular velocity of the Earth's rotation; Potential Temperature Gradient Calculation: Calculate the potential temperature value based on pressure and temperature, and then calculate the gradient of the potential temperature in the vertical direction using the central difference method; ; In the formula: Denote the vertical gradient of potential temperature; k represents the vertical level index, where ERA5 provides 37 levels of pressure data; Denote the potential temperature of the upper layer; Denote the potential temperature of the lower layer; Denote the pressure of the upper layer; Denote the pressure of the lower layer; Calculate the quasi-geostrophic potential vorticity: Combine the relative vorticity, Coriolis parameter, and potential temperature gradient to calculate the quasi-geostrophic potential vorticity at each grid point, and then normalize the calculated quasi-geostrophic potential vorticity; ; In the formula: Denote the quasi-geostrophic potential vorticity; Denote the acceleration due to gravity; Normalize based on the historical climatological mean and standard deviation of the calculated quasi-geostrophic potential vorticity.

[0024] The specific steps for extracting the water vapor input flux are as follows: Integrate from the surface pressure to the top of the atmosphere according to the 37 pressure levels of ERA5. First, calculate the flux components. Integrate according to the product of the zonal wind speed and specific humidity to obtain the zonal flux component: ; In the formula: Denote the zonal flux component; Denote the specific humidity; Denote the zonal wind speed; Denote the pressure difference; k represents the index of the pressure level; Then integrate according to the product of the meridional wind speed and specific humidity to obtain the meridional flux component: ; In the formula: Denote the meridional flux component; Denote the meridional wind speed; Use the trapezoidal integration method to calculate the contribution of each layer: ; In the formula: Denote the pressure of the upper layer; Denote the pressure of the lower layer; Based on the zonal and meridional flux components, calculate the magnitude and direction angle of the flux: ; In the formula: Denote the magnitude of the flux; ; In the formula: Denote the direction angle of the flux.

[0025] By calculating the sum of the squares of the zonal flux and the meridional flux, the magnitude of the total water vapor flux is obtained. By calculating the ratio of the meridional flux to the zonal flux, the direction angle of the water vapor transport flux is obtained.

[0026] Ecological response modeling: Construct a vegetation stress index and generate a species habitat suitability map; As a possible implementation of this embodiment, the construction of the vegetation stress index includes the following steps: Calculation of heat stress component: For each pixel, calculate the difference between its daily maximum temperature and the tolerance threshold of the species, and use the Logistic function to determine the heat stress component; ; In the formula: represents the daily maximum temperature; represents the high-temperature tolerance threshold of the species; represents the steepness coefficient of the Logistic curve; e represents the base of the natural logarithm; Calculation of water stress component: Based on calculating the ratio of the surface soil moisture to the critical moisture to measure the severity of water stress; where a terrain adjustment factor is introduced to adjust the water stress component when calculating the water stress component, and the terrain adjustment factor is calculated based on the terrain slope, and the larger the slope, the smaller the terrain adjustment factor; ; ; In the formula: represents the surface soil moisture; represents the critical soil moisture; represents the water stress component; represents the terrain adjustment factor; represents the terrain slope; Calculation of comprehensive stress index: Comprehensively consider the heat stress component and the water stress component, and combine the NDVI value of the vegetation to correct the stress to obtain the comprehensive stress index; ; In the formula: represents the normalized vegetation index of the current pixel; represents the historical maximum value.

[0027] As a possible implementation of this embodiment, the specific steps for generating the species habitat suitability map are as follows: Feature engineering: Screen the environmental variables significantly related to the species distribution through the Moran's I index to obtain the selected features; ; In the formula: represents the spatial weight matrix; and respectively represent the values of environmental variables at different spatial positions; represents the mean value of the environmental variable; After screening by the Moran's I index, the following environmental variables are retained for subsequent modeling: PV anomaly intensity, water vapor flux divergence, VSI, and DEM (terrain elevation); MaxEnt model optimization: Based on the selected features and species distribution data, train the MaxEnt model to output the suitability score for each habitat; The MaxEnt model is a commonly used species distribution modeling method, suitable for dealing with the relationship between physical distribution and environmental variables. Its goal is to train the model through known species distribution points and generate species suitability predictions based on the characteristics of environmental variables; The output of the MaxEnt model is the habitat suitability score for each pixel; Phenological regulation: The habitat suitability of a species is not only affected by environmental conditions but also closely related to phenology (seasonal changes). Therefore, it is necessary to adjust the suitability prediction results of the model according to seasonal changes; Seasonal adjustment: Based on the MODIS EVI (vegetation index) data, calculate the phenological phase angle of the species, which reflects the growth and reproduction cycles of the species in different seasons; Phenological regulation coefficient: Adjust the habitat suitability through the phenological regulation coefficient: ; In the formula: represents the original suitability value given by the MaxEnt model; represents the regulation coefficient calculated based on phenological data; A represents the amplitude coefficient; t represents the current time point index; represents the phenological phase angle; N represents the total number of time points; Based on this, we then generate a habitat suitability map according to the suitability value of each pixel.

[0028] See Figure 2As shown in the figure, the generation adversarial joint training is as follows: construct a dual-branch generation adversarial network, where the generator includes a meteorological generation branch and an ecological generation branch. The meteorological generation branch extracts spatial features based on a 3D residual network, captures temporal dependencies by combining with a Transformer, and outputs the meteorological field at future time steps. The ecological generation branch generates a dynamic risk probability map using a dilated convolutional U-Net model based on the predicted meteorological field and topographic data. The discriminator is divided into a physical conservation verification module and an ecological correlation module. The physical conservation verification module is used to verify the rationality of the generated meteorological field. The ecological correlation module verifies the logical consistency of the ecological response by analyzing the statistical correlation between the vegetation index and meteorological variables. During the training process, the generator simultaneously optimizes the adversarial loss, physical conservation loss, and ecological correlation loss, and finally generates a result of seamless coupling of meteorological prediction and ecological risk. Exemplarily, the meteorological generation branch is as follows: The network structure includes a 3D ResNet-50 backbone network, a Transformer spatio-temporal attention module, and a decoding prediction layer; The input of the 3D ResNet-50 backbone network and are concatenated along the channel dimension (total number of channels ); where represents the meteorological field at time step t (temperature, humidity, wind field, etc., ); represents the quasi-geostrophic potential vorticity; Each residual block contains a 3D convolution (kernel size 3×3×3, stride 1) + batch normalization + ReLU; Based on the output feature map processed by the residual block ; Transformer spatio-temporal attention module: Spatial serialization: Reshape into a sequence of , then use multi-head attention to calculate the spatio-temporal correlation weights, and enhance the features based on the calculated spatio-temporal correlation weights to obtain spatio-temporal enhanced features ; Decoding layer prediction: Use a 3D transposed convolution layer (kernel size 3×3×3, stride 1) to reduce the number of channels to , and output the meteorological generation result ; In this embodiment, by combining the 3D ResNet-50 backbone network and the Transformer spatio-temporal attention module, the network can simultaneously capture the spatial features and temporal correlations of meteorological data; 3D convolution can preserve the spatio-temporal structure of the input data, while the multi-head attention mechanism of the Transformer can effectively enhance the spatio-temporal dependence relationship; considering spatio-temporal information comprehensively, the predicted meteorological field is more in line with the actual situation and has higher accuracy.

[0029] The ecological generation branch network structure is as follows: Atrous convolution encoder: Average along the time dimension , to obtain , and concatenate with and input it into the atrous convolution layer (dilation rate = 2, 4, 8), and output multi-scale features ; U-Net decoder: Concatenate the features of each layer of the encoder with the upsampling results of the decoder through skip connections; compress the number of channels to 1 through 1×1 convolution, and obtain the ecological risk probability map through Sigmoid activation: ; In the formula: represents Sigmoid activation; W represents the convolution kernel parameter; represents the bias term; represents the concatenation result of the skip connection; In this embodiment, through the atrous convolution encoder, the network can extract ecological features at different scales, ensuring the comprehensiveness and accuracy of ecological risk assessment; atrous convolution can expand the receptive field and does not increase the computational cost, which is suitable for processing complex ecological data; the U-Net decoder combined with skip connections can retain more information, making the generated ecological risk map more detailed and accurate.

[0030] Discriminator: The physical conservation verification module is used to check whether the generated meteorological data follows the physical conservation principle, and the specific steps are as follows: Mass conservation residual : According to the conservation equation of fluid mechanics, calculate the divergence of the fluid velocity field: ; In the formula: represents the air density, which is calculated by the ideal gas state equation; represents the three-dimensional vector of the wind speed field; represents the divergence of the fluid, which is calculated by the central difference method; Energy conservation residual Verification: ; In the formula: represents the kinetic energy density of the gas; represents the specific heat capacity at constant volume; represents the temperature field; represents the time derivative; Based on the mass conservation residuals and energy conservation residuals calculated above, compare them with the threshold to determine whether the generated meteorological data follows the principle of physical conservation.

[0031] In this embodiment, through the calculation of mass conservation and energy conservation residuals, it is ensured that the generated meteorological field data follows the basic physical laws; effectively reducing non-physically reasonable prediction results and improving the credibility of the model.

[0032] As a possible implementation of this embodiment, the steps for the ecological correlation module to verify the logical consistency of the ecological response by analyzing the statistical correlation between the vegetation index and meteorological variables are as follows: Normalized Difference Vegetation Index: ; In the formula: represents the reflectance in the near-infrared band; represents the reflectance in the infrared band; represents the Normalized Difference Vegetation Index; Temperature anomaly Calculate: ; In the formula: represents the actual temperature at the current moment; represents the long-term average temperature; Mutual information calculation: Based on mutual information, calculate the correlation between NDVI and temperature anomaly: ; In the formula: represents the NDVI value and temperature anomaly the probability of co-occurrence; , are the marginal distributions of NDVI and temperature anomaly respectively; represents the number of bins of NDVI and temperature anomaly; Risk map matching degree: Measure the difference between the generated risk map and the true risk map through KL divergence: ; In the formula: represents the probability distribution of the generated ecological risk map; represents the probability distribution of the true ecological risk map; N represents the number of pixels of the risk map; represents the probability value of the i-th pixel of the generated ecological risk map; Represents the probability value of the i-th pixel of the true ecological risk map; Represents the risk map matching degree; When the difference between the calculated mutual information and the mutual information of the true data is less than the threshold, and the risk map matching degree is less than the threshold, it passes the verification of the ecological association module. If either of the above conditions is not met, the verification is not satisfied.

[0033] In this embodiment, by calculating the mutual information between the vegetation index and meteorological variables and the KL divergence of the risk map, it is ensured that the generated ecological risk map is consistent with the actual ecological response, improving the reliability of ecological risk assessment.

[0034] Among them, the loss function of the generator includes three parts, adversarial loss, physical conservation loss, and ecological association loss. The specific expressions are as follows: Adversarial loss: ; In the formula: Represents the output of the discriminator; Represents the output of the generator; Represents the distribution of the true data; x represents the true data sample; Represents the logarithm of the probability of the discriminator for the true sample; z represents the sample in the latent space; Represents the distribution of the latent space; Represents the mapping of the generator to the latent vector z; Physical conservation loss: ; In the formula: Represents the weight of the mass conservation loss; Represents the weight of the energy conservation loss; Ecological association loss: ; In the formula: Represents the weight of the ecological association loss.

[0035] In this embodiment, the adversarial loss function is used to train the generator to generate a realistic meteorological field and ecological risk map, while enabling the discriminator to correctly distinguish between true data and generated data; the physical conservation loss function is used to ensure that the generated meteorological field satisfies the physical constraints of mass conservation and energy conservation; the ecological association loss function is used to ensure that the generated dynamic risk map conforms to ecological principles, that is, the matching degree between the generated map and the true map.

[0036] In this embodiment, a meteorological field and an ecological risk map are generated through a forward propagation generator, and then the adversarial loss, physical conservation loss, and ecological correlation loss of the generator are calculated. The parameters of the generator and the discriminator are updated according to the losses, and the training is completed by repeatedly iterating until the losses converge.

[0037] Dynamic feedback optimization: Use the gradient information of the discriminator to locate the key areas of the generation error, linearly correct the input layer of the generator in combination with the real-time observation data of satellites and buoy stations, and then use spectral normalization to limit the parameter range of the generator.

[0038] In this embodiment, since the training process of the generator depends on the feedback information of the discriminator, the gradient information of the discriminator can help locate the areas where errors occur when the generator outputs the generated meteorological field and ecological risk map. Therefore, by calculating the extraction of the discriminator until the generator adjusts the input data structure or network structure; based on this, through the chain rule, calculate the gradient information of the loss function of the discriminator with respect to the input of the generator: ; In the formula: represents the latent input of the generator; represents the output of the generator; represents the gradient of the discriminator loss with respect to the output of the generator; represents the derivative of the generator with respect to the latent input; Through the gradient information provided by the discriminator, correction can be performed on the input layer of the generator. The correction target is to modify the latent space of the generator or its input features, thereby improving the accuracy of generating the meteorological field and ecological risk map; Among them, the real-time data provides the actual measurement of the meteorological field and ecological risk map, which can help locate the errors of the generator in space and time; the accuracy of the generator output is evaluated by calculating the difference between the predicted value and the observed value: ; In the formula: represents the real observation data provided by satellites or buoy stations; represents the predicted value output by the generator; represents the generation error; Combining the gradient information of the discriminator and the real-time observation data, correct the generated input: ; In the formula: represents the corrected input of the generator; represents the learning rate, which controls the correction effect of the discriminator gradient on the input; represents the learning rate, which controls the correction effect between the observation data and the generation data error; Through linear correction, the input of the generator is adjusted so that the generated meteorological field and ecological risk map are more consistent with real observational data, while satisfying physical conservation and ecological logic; Furthermore, in this embodiment, to avoid the parameters of the generator network from being too large or unstable, spectral normalization technology is used to constrain the weights of the generator. By restricting the maximum eigenvalue of each convolutional layer, the generator parameters are ensured to be within a stable range, thereby improving the stability of training.

[0039] In the present invention, by performing spatio-temporal alignment on multi-source data and fusing them into a three-dimensional feature cube, the spatio-temporal consistency of the input data is ensured; subsequently, the meteorological generation branch uses a 3D residual network and Transformer to synchronously extract the spatio-temporal features of the meteorological field and predict future meteorological states, while the ecological generation branch, based on the meteorological prediction results and terrain data, dynamically generates an ecological risk map through dilated convolutional U-Net, avoiding the resolution and dimension mismatch caused by cross-model interpolation; the discriminator introduces a physical conservation verification module and an ecological correlation module to jointly optimize the physical rationality loss and ecological logic consistency loss of the generator in adversarial training, ensuring that the generated results conform to both the laws of atmospheric dynamics and the ecological response mechanism; in addition, the dynamic feedback mechanism corrects the generation error through real-time observational data and uses spectral normalization to constrain the model parameters, further improving the accuracy of the coupled prediction of extreme events and ecological risks; therefore, the present invention realizes the seamless spatio-temporal coupling of the meteorological and ecological fields, the generated results have both physical conservation and ecological relevance, and can dynamically adapt to real-time observational data, providing reliable technical support for the high-precision collaborative early warning of extreme weather and its ecological impacts.

[0040] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. The extreme weather event and ecological risk prediction method using generative adversarial networks is characterized by: The following steps are involved: Data collection: collect original meteorological data, remote sensing data, geographic data and ecological baseline data; Data processing: align the collected original meteorological data, remote sensing data, geographic data and ecological baseline data in time and space, and then fuse them to construct a three-dimensional feature cube; Calculation of meteorological dynamic characteristics: Based on the constructed three-dimensional characteristic cube, the quasi-geostrophic potential vorticity is calculated, the water vapor input flux is extracted, and the physical diagnostic field is obtained; Ecological response modeling: construct vegetation stress index and generate species habitat suitability maps; Generative adversarial joint training: A dual-branch generative adversarial network is constructed, in which the generator includes a meteorological generation branch and an ecological generation branch. The meteorological generation branch extracts spatial features based on a 3D residual network, combines Transformer to capture temporal dependencies, and outputs the meteorological field at future moments; the ecological generation branch generates a dynamic risk probability map based on the predicted meteorological field and terrain data using a dilated convolutional U-Net model; the discriminator is divided into a physical conservation verification module and an ecological association module. The physical conservation verification module is used to verify the rationality of the generated meteorological field; the ecological association module verifies the logical consistency of the ecological response by analyzing the statistical correlation between vegetation indices and meteorological variables; during the training process, the generator simultaneously optimizes the adversarial loss, physical conservation loss, and ecological association loss, and finally generates a result of seamless coupling of meteorological forecasts and ecological risks; Dynamic feedback optimization: The gradient information of the discriminator is used to locate the key areas of generation error, and the real-time observation data from satellites and buoy stations are combined to perform linear correction on the generator input layer. Then, spectral normalization is used to limit the parameter range of the generator.

2. The method for predicting extreme weather events and ecological risks using a generative adversarial network according to claim 1, characterized in that: The data processing comprises the following steps: The meteorological data space is downscaled through cubic spline interpolation, and the temporal resolution of remote sensing data is increased to the hourly level using bilinear interpolation to ensure that all data are in a unified spatiotemporal coordinate system. Then, invalid areas are eliminated through dynamic masking technology, and finally data of different dimensions are fused into a multidimensional spatiotemporal dataset to obtain standardized spatiotemporal grid data.

3. The method for predicting extreme weather events and ecological risks using a generative adversarial network according to claim 1, characterized in that: The calculation of quasi-geostrophic potential vorticity comprises the following steps: Relative vorticity calculation: Calculate the spatial gradients of zonal wind and meridional wind to obtain the rate of change of the wind field on the horizontal plane, and then calculate the relative vorticity based on the second-order central difference method; Calculate Coriolis parameters: Calculate the Coriolis parameters of each grid point using the standard formula for calculating Coriolis parameters; Calculate potential temperature gradient: Calculate the potential temperature value based on air pressure and temperature, and then use the central difference method to calculate the gradient of the potential temperature in the vertical direction; Calculate the quasi-geostrophic potential vorticity: Combine the relative vorticity, Coriolis parameter and potential temperature gradient to calculate the quasi-geostrophic potential vorticity of each grid point, and then normalize the calculated quasi-geostrophic potential vorticity.

4. The method for predicting extreme weather events and ecological risks using a generative adversarial network according to claim 1, characterized in that: The specific steps of extracting the water vapor input flux are as follows: Vertical integration: Based on the 37-layer air pressure data provided by ERA5, the contribution of each layer to the total water vapor flux is calculated using the trapezoidal integration method; Statistical water vapor flux components: integrate the product of zonal wind speed and specific humidity to obtain zonal flux; integrate the product of meridional wind speed and specific humidity to obtain meridional flux; at each level, use the interval of the pressure layer to weight the contribution of each layer; Synthetic water vapor flux: By calculating the sum of the squares of the latitudinal flux and the meridional flux, the modulus of the total water vapor flux is obtained. By calculating the ratio of the meridional flux to the latitudinal flux, the direction angle of the water vapor transport flux is obtained.

5. The method for predicting extreme weather events and ecological risks using a generative adversarial network according to claim 1, characterized in that: The construction of the vegetation stress index comprises the following steps: Calculation of heat stress component: For each pixel, the difference between its daily maximum temperature and the species tolerance threshold is calculated, and the heat stress component is determined using the Logistic function; Calculation of water stress component: The severity of water stress is measured based on the ratio of surface soil moisture to critical moisture. When calculating the water stress component, a terrain adjustment factor is introduced to adjust the water stress component. The terrain adjustment factor is calculated based on the terrain slope. The greater the slope, the smaller the terrain adjustment factor. Calculation of comprehensive stress index: The heat stress component and water stress component are comprehensively considered, and the stress is corrected in combination with the NDVI value of the vegetation to obtain the comprehensive stress index.

6. The method for predicting extreme weather events and ecological risks using a generative adversarial network according to claim 1, characterized in that: The specific steps to generate a species habitat fitness map are as follows: Feature engineering: The environmental variables significantly related to species distribution were screened through Moran's I index to obtain the selected features; MaxEnt model optimization: Based on the selected features and species distribution data, the MaxEnt model is trained to output the suitability score of each habitat; Phenological adjustment: Based on the vegetation index data, the phenological phase angle of the species is calculated, and the phenological adjustment coefficient is calculated according to the phenological phase angle. The phenological adjustment coefficient is applied to the suitability value output by the MaxEnt model to obtain the seasonally adjusted habitat suitability score; Based on the suitability value of each pixel, a species habitat suitability map is generated.

7. The method for predicting extreme weather events and ecological risks using a generative adversarial network according to claim 1, characterized in that: The physical conservation verification module is used to check whether the generated meteorological data complies with the physical conservation principle. The specific steps are as follows: Mass conservation residual : According to the conservation equations of fluid mechanics, calculate the divergence of the fluid velocity field: ； Where: represents the air density, calculated by the ideal gas state equation; A three-dimensional vector representing the wind speed field; represents the divergence of the fluid, which is calculated by the central difference method; Energy Conservation Residual verify: ； Where: represents the kinetic energy density of the gas; represents the specific heat at constant volume; represents the temperature field; represents the time derivative; The mass conservation residual and energy conservation residual calculated above are compared with the threshold to determine whether the generated meteorological data follows the principle of physical conservation.

8. The method for predicting extreme weather events and ecological risks using a generative adversarial network according to claim 1, characterized in that: The steps of the ecological correlation module to verify the logical consistency of ecological responses by analyzing the statistical correlation between vegetation indices and meteorological variables are as follows: Normalized Difference Vegetation Index: ； Where: Represents the reflectivity in the near-infrared band; Indicates the reflectivity in the infrared band; represents the normalized difference vegetation index; Temperature anomaly calculate: ； Where: Indicates the actual temperature at the current moment; represents the long-term average temperature; Mutual information calculation: Based on mutual information, the correlation between NDVI and temperature anomaly is calculated: ； Where: Indicates NDVI value and temperature anomalies Probability of simultaneous occurrence; , They are the marginal distributions of NDVI and temperature anomalies; Indicates the number of bins for NDVI and temperature anomalies; Risk map matching: The difference between the generated risk map and the true risk map is measured by KL divergence: ； Where: represents the probability distribution of the generated ecological risk map; represents the probability distribution of the real ecological risk map; N represents the number of pixels in the risk map; Represents the probability value of the i-th pixel in the generated ecological risk map; Represents the probability value of the ith pixel in the real ecological risk map; represents the matching degree of the risk graph; When the difference between the calculated mutual information and the mutual information of the real data is less than the threshold, and the risk map matching degree is less than the threshold, the verification of the ecological association module is passed. If any of the above conditions is not met, the verification is not met.

Images