Land space planning environment influence monitoring method and device

Abstract

The invention discloses a territorial space planning environment influence monitoring method and device. The method comprises the following steps: collecting and preprocessing multi-source environment monitoring data; performing spatio-temporal feature extraction on the preprocessed multi-source environment monitoring data by using a self-supervised spatio-temporal diagram mask transfer attention network to obtain initial spatio-temporal features; performing fusion processing on the initial spatial-temporal characteristics based on a multi-mode diffusion converter to obtain fusion characteristics; calculating an environmental impact evaluation index based on the fusion feature; and based on a preset visualization template and an early warning rule, generating a dynamic monitoring result including a space-time evolution diagram, a trend analysis diagram and early warning information according to the environmental influence evaluation index. Through deep learning and multi-modal data fusion technologies, the precision and efficiency of environmental impact monitoring are improved, and accurate evaluation and timely early warning of environmental impact in the land space planning implementation process are realized.

Description

Technical Field

[0001] The present application relates to the field of environmental monitoring technology, and in particular to a method and device for monitoring the environmental impact of land space planning. Background Art

[0002] With the acceleration of urbanization and the increase in the intensity of land space development and utilization, the importance of environmental impact monitoring in land space planning has become increasingly prominent. Environmental monitoring data has the characteristics of multi-source heterogeneity and strong temporal and spatial heterogeneity, and it is necessary to establish a systematic monitoring and evaluation system.

[0003] At present, the commonly used environmental monitoring methods mainly include change detection based on remote sensing images and parameter collection based on ground monitoring sites. These methods are relatively mature in processing single data sources, but they are still insufficient in multi-source data fusion.

[0004] Existing technologies usually use simple feature splicing or weighted average methods to fuse data and use traditional statistical models to perform environmental impact assessment. This method is difficult to fully explore the spatiotemporal correlation features between data and is prone to information loss when processing high-dimensional heterogeneous data.

[0005] In addition, the existing environmental monitoring systems generally have problems such as insufficient feature extraction and imperfect early warning mechanisms, making it difficult to accurately identify and promptly warn of environmental changes, and unable to meet the needs of real-time monitoring of environmental impacts in national land space planning. Summary of the invention

[0006] In view of this, the present application provides a method and device for monitoring the environmental impact of national land space planning, which solves the problems in the prior art of difficulty in effectively integrating multi-source heterogeneous data, insufficient feature extraction and imperfect early warning mechanism.

[0007] The present application provides a method for monitoring the environmental impact of land space planning, including: Collecting and preprocessing multi-source environmental monitoring data, wherein the multi-source environmental monitoring data includes at least two of satellite remote sensing images, meteorological station data, surface coverage data, and ecological environment monitoring data; The self-supervised spatiotemporal graph mask transfer attention network is used to extract spatiotemporal features from the preprocessed multi-source environmental monitoring data to obtain the initial spatiotemporal features. Based on a multimodal diffusion transformer, the initial spatiotemporal features are fused to obtain fused features; Calculating environmental impact assessment indicators based on the fusion features; Based on preset visualization templates and early warning rules, dynamic monitoring results including spatiotemporal evolution diagrams, trend analysis charts and early warning information are generated according to the environmental impact assessment indicators.

[0008] The present application also provides a national land space planning environmental impact monitoring device, including: A data collection and preprocessing module, used for collecting and preprocessing at least two types of multi-source environmental monitoring data including satellite remote sensing images, meteorological station data, surface coverage data and ecological environment monitoring data; The spatiotemporal feature extraction module is used to extract spatiotemporal features from the preprocessed multi-source environmental monitoring data using a self-supervised spatiotemporal graph mask transfer attention network to obtain initial spatiotemporal features; A feature fusion module, used for fusing the initial spatiotemporal features based on a multimodal diffusion transformer to obtain fused features; An assessment index calculation module, used to calculate environmental impact assessment indicators including land use change index, ecosystem damage degree and environmental quality comprehensive index based on the fusion features; The visual warning module is used to generate dynamic monitoring results including spatiotemporal evolution diagrams, trend analysis charts and warning information based on preset visual templates and warning rules according to the environmental impact assessment indicators.

[0009] This application has the following technical effects: Through the self-supervised spatiotemporal graph mask transfer attention network, the effective fusion of multi-source heterogeneous data is realized, and the accuracy of feature extraction is improved; the multimodal diffusion transformer is used for feature fusion processing to enhance the expression ability of environmental change characteristics; the early warning mechanism based on deep learning realizes the accurate identification and timely warning of environmental risks, and improves the monitoring effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solution of the embodiment of the present application, the following is a brief introduction to the drawings required for use in the embodiment: Figure 1 It is a flowchart of a method for monitoring the environmental impact of land space planning provided in an embodiment of the present application; Figure 2 It is a schematic diagram of the process of collecting and preprocessing multi-source environmental monitoring data in an embodiment of the present application; Figure 3 This is a specific flow chart of step S2 in the embodiment of the present application; Figure 4 This is a specific flow chart of step S3 in the embodiment of the present application; Figure 5 This is a specific flow chart of step S4 in the embodiment of the present application; Figure 6 This is a specific flow chart of step S5 in the embodiment of the present application; Figure 7 It is a schematic diagram of the land space planning environmental impact monitoring device in an embodiment of the present application. DETAILED DESCRIPTION

[0011] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0012] like Figure 1 As shown, the embodiment of the present application provides a method for monitoring the environmental impact of land space planning, including: S1: Collect and preprocess multi-source environmental monitoring data, wherein the multi-source environmental monitoring data includes at least two of satellite remote sensing images, meteorological station data, surface coverage data and ecological environment monitoring data; like Figure 2 As shown, the embodiment of the present application can be carried out in the multi-source environmental data collection stage through a pre-established data collection system, which has both data collection and data preprocessing functions (for example, data cleaning, format unification, spatial alignment and time synchronization, etc.). For satellite remote sensing image data, the embodiment of the present application uses high-resolution satellite remote sensing images with a spatial resolution better than 1 meter. Such remote sensing images are mainly derived from high-resolution series satellites or commercial remote sensing satellites. Its high spatiotemporal resolution characteristics enable it to clearly capture subtle changes in the surface, which helps to detect environmental changes in a timely manner. To ensure the continuity and timeliness of monitoring, the frequency of acquiring remote sensing images is set to 1-2 times per month.

[0013] The collection of meteorological station data mainly includes key meteorological parameters such as temperature, precipitation, and humidity. The embodiment of the present application preferably uses the observation data of the automatic weather station to achieve automatic collection around the clock. Taking into account the spatial variability of meteorological elements, a differentiated strategy is adopted in the layout of meteorological stations: one station is set up every 50-100 square kilometers in urban areas, and one station can be appropriately relaxed to 200-300 square kilometers in rural areas to ensure the spatial representativeness of data collection.

[0014] The collection of land cover data is mainly carried out through multi-temporal remote sensing images. The land cover information is obtained through the land object classification method, and the classification includes main types such as forest land, grassland, cultivated land, construction land, and water area. To ensure the accuracy of the land cover data, the embodiment of the present application also verifies and improves the classification results in combination with field survey data.

[0015] The collection of ecological environment monitoring data is more comprehensive, including monitoring data of air quality, water quality, soil and other environmental factors. Specifically, air quality monitoring mainly collects PM2.5, PM10, SO2, NO x Water quality monitoring includes indicators such as pH value, dissolved oxygen, total nitrogen, and total phosphorus of surface water bodies; soil monitoring focuses on key parameters such as heavy metal content and organic matter content. These data are usually obtained through fixed monitoring stations or portable monitoring equipment.

[0016] In the data preprocessing stage, the embodiment of the present application adopts a multi-stage processing strategy. First, the remote sensing image is geometrically corrected and radiometrically calibrated. The geometric correction uses the control point registration method to ensure that the spatial position accuracy is better than 1 pixel; the radiometric calibration converts the image grayscale value into the surface reflectivity through the atmospheric radiation transfer model. For meteorological and environmental monitoring data, outlier detection and missing value processing are mainly performed. Outlier detection uses the 3σ criterion to mark data that exceeds the normal range; missing values ​​are supplemented by time series interpolation methods to ensure data continuity.

[0017] In addition, the embodiment of the present application innovatively introduces a spatiotemporal data cube structure to organize monitoring data from different sources into a three-dimensional data volume with the same spatiotemporal resolution. In the spatial dimension, regular grid division is adopted, and the grid size is determined according to the size of the study area and the characteristics of the data, usually 100 meters × 100 meters; in the temporal dimension, an appropriate time interval is set according to the monitoring frequency, which can be days, weeks or months. Through this unified data organization method, the effective integration of multi-source heterogeneous data is achieved.

[0018] To ensure data quality, the present application also establishes a complete quality control system, including instrument calibration and operation and maintenance specifications for data collection, quality assessment indicators for preprocessing, and a cross-validation mechanism for processing results. These measures can significantly improve the reliability and accuracy of monitoring data.

[0019] Finally, in order to solve the problem of spatiotemporal alignment of different types of data, the embodiment of the present application adopts a fusion strategy based on spatial interpolation and time weighting. For point observation data with uneven spatial distribution (such as meteorological station data), the Kriging interpolation method is used to generate a continuous spatial distribution field; for data with inconsistent temporal resolution, unified processing is achieved through the time-weighted average method. This processing method not only maintains the characteristics of the original data, but also ensures the comparability and consistency between different data sources.

[0020] S2: Use the self-supervised spatiotemporal graph mask transfer attention network to extract spatiotemporal features from the preprocessed multi-source environmental monitoring data to obtain the initial spatiotemporal features; like Figure 3 As shown, S2 specifically includes: S2.1: Based on the pre-processed multi-source environmental monitoring data, the monitoring area is divided into space-time unit nodes, spatial edges are established based on geographical location proximity, and time edges are established based on time series continuity to construct a space-time relationship graph; The self-supervised spatiotemporal graph mask transfer attention network first needs to construct a spatiotemporal relationship graph. In the spatial dimension, the monitoring area is divided into regular grid cells, and each grid cell is used as a node in the graph. The initial feature vector of the node contains the remote sensing image depth features of the location (extracted by a pre-trained convolutional neural network) and environmental parameters. In terms of establishing spatial edges, an 8-neighborhood connection method is adopted, that is, each node is connected to its upper, lower, left, right and four diagonal directions. This connection method can effectively capture local spatial dependencies. In the time dimension, the corresponding nodes at different time points are connected through time edges to form a time series association. The weight of the time edge can be adjusted according to the distance of the time interval to reflect the time attenuation effect.

[0021] S2.2: For the nodes of the spatiotemporal relationship graph, 15%-30% of the node features are randomly masked, and the masked node features are reconstructed through the surrounding node information for self-supervised pre-training to obtain a pre-trained attention network model; In order to improve the model's ability to understand the characteristics of environmental changes, the embodiment of the present application designs a mask-based self-supervised learning strategy. Specifically, in each training iteration, 15%-30% of the nodes are randomly selected for feature masking. The setting of the masking ratio is based on experimental verification. Too low a masking ratio makes it difficult for the model to learn effective feature representations, while too high a masking ratio may result in excessive information loss. For masked nodes, their connection relationship in the graph is retained, but their feature vectors are set to special mask marks. The model needs to reconstruct the features of these masked nodes through the information of the surrounding unmasked nodes. This reconstruction task forces the model to learn the intrinsic association patterns between nodes. At the same time, a contrastive learning loss function is introduced to maximize the similarity of features at different time points at the same location, minimize the similarity of features at spatially unrelated locations, and enhance the model's perception of spatiotemporal patterns.

[0022] S2.3: Input the spatiotemporal relationship graph into the pre-trained attention network model, calculate the attention weights between spatially adjacent nodes through the spatial attention layer, and model the long-range dependency between different time points through the temporal attention layer to obtain the node attention weights; In S2.3, after obtaining the pre-trained model, a two-layer attention mechanism is used to capture complex spatiotemporal dependencies. The spatial attention layer calculates the attention weights based on the similarity of node features. Specifically, the scaled dot product attention mechanism is used, that is, the query vector, key vector, and value vector are obtained by linear transformation respectively, and then the inner product of the query vector and the key vector is calculated, and the weighted sum is obtained with the value vector after softmax normalization. This enables the model to adaptively focus on nodes with strong spatial correlation. The temporal attention layer uses a similar mechanism, but establishes connections between different time points, which can effectively capture long-term temporal dependencies, such as seasonal change patterns or interannual change trends.

[0023] The embodiment of this application takes the Yangtze River Delta urban agglomeration as an example to demonstrate the application of the attention model. For example, when analyzing the environmental changes in a new district, the spatial attention layer calculates the degree of correlation between the center point and the surrounding area, and finds that the influence weight of the Lingang New Area on the central area of ​​Pudong is 0.42, indicating that there is a strong environmental connection between the two areas. However, the attention weight with Yangpu District in the northwest is only 0.15, indicating that the spatial correlation of environmental impacts is relatively weak.

[0024] In the case of Suzhou Industrial Park, the temporal attention layer identified key correlation patterns with historical data. For example, the attention weight of the ozone pollution event in the summer of 2023 and the data of the same period in 2022 was as high as 0.68, revealing a strong seasonal feature. However, the attention weight with the winter data of 2021 was only 0.23, indicating that the correlation between environmental changes in different seasons was low. This temporal attention mechanism effectively captures the long-term seasonal variation pattern.

[0025] In the analysis of Nanjing Jiangbei New District, the model used a multi-head attention mechanism at the same time. One of the attention heads focused on the diffusion pattern of industrial pollutants, and found that the attention weight with the surrounding chemical parks was 0.55; the other attention head focused on ecosystem connectivity, and the attention weight with the wetlands along the Yangtze River reached 0.61. This multi-head design enables the model to pay attention to different types of environmental influencing factors at the same time.

[0026] In the monitoring of the Hangzhou Bay Industrial Belt, the attention model reveals complex spatiotemporal dependencies through hierarchical attention calculations. The attention weights of coastal industrial areas and inland ecological protection areas show a gradient decrease, gradually decreasing from 0.48 near the coast to 0.12 at the far end. This spatial attenuation feature is highly consistent with the actual observed diffusion law of pollutants. At the same time, the model detects a time lag of 3-5 days between the peak of industrial activity and changes in air quality, and the corresponding temporal attention weights fluctuate between 0.35-0.52.

[0027] These examples fully demonstrate the practical application effect of the attention model in environmental monitoring. The spatial attention layer can accurately quantify the transmission intensity of environmental impacts between different regions, while the temporal attention layer successfully captures the temporal evolution of environmental changes. The introduction of the multi-head attention mechanism further enhances the model's ability to understand complex environmental systems, making environmental impact assessment results more accurate and reliable.

[0028] S2.4: Based on the node attention weight, the node features are iteratively updated, the node features in the spatial neighborhood are integrated, and the gated recurrent unit is used to process the evolution of the time dimension features to obtain the initial spatiotemporal features.

[0029] In S2.4, the feature update stage first iteratively updates the node features based on the calculated attention weights. Specifically, each node will aggregate the information of its neighboring nodes according to the attention weights to achieve dynamic update of the features. This process is repeated many times so that the information can be fully transmitted in the graph. For the fusion of node features within the spatial neighborhood, a weighted summation method is adopted, and the weight is determined by the attention score. In the time dimension, the gated recurrent unit (GRU) is used to process the evolution of features. The update gate and reset gate mechanism of GRU can selectively retain or update historical information and effectively model the dynamic changes of time series. Finally, through multi-scale feature integration, features of different spatial ranges (such as local, regional and global) and different time scales (such as daily, monthly, and seasonal) are combined to obtain a comprehensive spatiotemporal feature representation.

[0030] The embodiment of the present application can perform step S2.4 through a complete self-supervised spatiotemporal graph mask transfer attention network structure. The network is based on the spatiotemporal relationship graph and includes three main components: a feature mask layer, a dual attention layer, and a feature update layer. The feature mask layer realizes self-supervised learning through random masking; the dual attention layer handles spatial and temporal dependencies respectively; and the feature update layer is responsible for the dynamic update and evolution of node features. The entire network is trained in an end-to-end manner to achieve efficient extraction from raw monitoring data to spatiotemporal features.

[0031] The advantages of this network structure are mainly reflected in three aspects: first, the self-supervised learning mechanism reduces the dependence on labeled data and improves the generalization ability of the model; second, the two-layer attention structure can adaptively capture the spatiotemporal dependencies of different scales; finally, the iterative feature update ensures the full transmission of information and the global consistency of feature expression. Experimental results show that this method has achieved significant performance improvement in environmental change detection tasks, with an average accuracy increase of more than 15%.

[0032] Taking the environmental monitoring of the Yangtze River Delta urban agglomeration as an example, this region includes many large and medium-sized cities such as Shanghai, Suzhou, and Wuxi. It is a typical complex area where urbanization and ecological environmental protection coexist.

[0033] When constructing the spatiotemporal relationship diagram (S2.1), the 25,000 square kilometers of monitoring area are first divided into 1km×1km grid units, and each unit is used as a node in the diagram. For example, the node (116, 234) located in Suzhou Industrial Park is connected to the eight adjacent nodes around it by spatial edges, which are distributed at the boundary of the park and the surrounding ecological conservation area. In the time dimension, the monthly observation data from 2020 to 2024 are used to establish time edges, connecting nodes at different time points at the same location to form a continuous time chain.

[0034] In the self-supervised pre-training stage (S2.2), the constructed spatiotemporal relationship graph is randomly masked. Each training iteration randomly selects 20% of the nodes for masking. For example, the environmental characteristics of a node in Suzhou Industrial Park in July 2023 [PM2.5=45μg / m³, SO2=12μg / m³, NO2=38μg / m³, vegetation coverage=0.35] are replaced with a mask mark [MASK]. The model reconstructs these masked features by analyzing the surrounding unmasked node information, such as adjacent ecological conservation area nodes and historical monthly data. After a large number of iterative training, the model gradually grasps the spatial correlation and seasonal variation of regional environmental elements.

[0035] In the attention weight calculation stage (S2.3), the spatial attention layer analysis found that the attention weight of the industrial park node with the eastern residential area node was 0.12, the weight with the western ecological conservation area node was 0.35, and the weight with the southern Taihu water node was 0.28, which reflects the degree of influence of different functional areas on the environmental quality of the park. The temporal attention layer identified that there is a significant long-range dependency relationship (attention weight>0.4) between the node in summer (June-August) and winter (December-February) every year, which reveals the coupling effect of seasonal industrial activities and climate change.

[0036] In the case of the industrial park mentioned above, the calculation process of attention weights fully demonstrates the model's ability to characterize the transmission mechanism of environmental impacts. For the calculation of spatial attention, the model first constructs a 128-dimensional feature vector, which contains key information such as pollutant concentration, land use type, and meteorological parameters of the industrial park nodes, and adds spatial location information through sinusoidal position encoding. Through a learnable parameter matrix, the model converts these original features into query vectors, key vectors, and value vectors, respectively, to prepare for subsequent attention calculations.

[0037] In the specific calculation process, the model calculates the attention scores of the industrial park nodes and the surrounding areas through dot product operations and scaling operations. After softmax normalization, the attention weights of 0.12 for the eastern residential area, 0.35 for the western ecological conservation area, and 0.28 for the southern Taihu Lake waters were obtained. These values ​​intuitively reflect the degree of influence of different functional areas on the environmental quality of the industrial park. Among them, the higher weight (0.35) of the western ecological conservation area indicates that it plays an important buffering role in the park environment through ecological regulation.

[0038] In terms of the calculation of temporal attention, the model processed the monthly monitoring data series from 2020 to 2024. Through a similar attention mechanism, the model identified seasonal environmental change patterns. Specifically, when analyzing the summer node of 2023, the model calculated that its attention weight with the data of the same period of the previous year was as high as 0.68, while the weight with the winter data of that year was 0.42. This significant seasonal correlation reflects the cyclical change characteristics of industrial activity intensity and meteorological conditions.

[0039] Finally, the model outputs two key attention matrices: one describing the strength of spatial associations and the other describing temporal dependencies. These attention weights not only quantify the transmission strength of environmental impacts, but also provide an interpretable analytical basis for environmental management. For example, based on these weights, management departments can prioritize boundary management with ecological conservation areas and adjust summer environmental supervision strategies in a targeted manner. This analytical method based on the attention mechanism has successfully achieved the precise identification and quantitative expression of key impact relationships in complex environmental systems.

[0040] In the feature update and evolution stage (S2.4), based on the calculated attention weights, the model iteratively updates the features of the industrial park nodes. For example, considering the influence of the Taihu Lake waters, humidity and temperature factors are included in the feature set; considering the regulatory role of the ecological conservation area, the vegetation purification capacity factor is taken into consideration. The gated cyclic unit is used to process the time series from 2020 to 2024, and the improvement trend of the annual average concentration of PM2.5 from 52μg / m³ in 2020 to 38μg / m³ in 2024 is successfully captured as the industrial upgrading of the park and the implementation of environmental protection measures, as well as the periodic change law of increased ozone concentration in summer. The final spatiotemporal characteristics not only reflect the current environmental conditions of the industrial park, but also include the dynamic process of environmental quality improvement and future trend predictions.

[0041] S3: fusing the initial spatiotemporal features based on a multimodal diffusion transformer to obtain fused features; like Figure 4 As shown, S3 specifically includes: S3.1: feature encoding the initial spatiotemporal features and adding modality type tags to obtain coded features with modality tags; The first key step of the multimodal diffusion transformer is feature encoding. For the initial spatiotemporal features, a multilayer perceptron is used for nonlinear transformation to map them to a 256-dimensional feature space. For remote sensing image data, VisionTransformer is used as the backbone network to segment the image into 16×16 image blocks, and hierarchical visual features are extracted through position encoding and self-attention mechanisms. For meteorological data, considering its temporal characteristics, a bidirectional long short-term memory network (BiLSTM) is used for encoding to capture the temporal dependencies of meteorological elements. In addition, data from different sources can be distinguished through a dedicated modality type labeling mechanism, that is, unique modality identifiers are added to the feature vector. These identifiers are in the form of learnable embedding vectors, which are adaptively optimized during the training process, which helps the model recognize and process features of different modalities.

[0042] In national land space planning and environmental monitoring, traditional methods are difficult to effectively handle multi-source heterogeneous data, especially the fusion of different modal data such as remote sensing images and meteorological data. To address this challenge, the present application embodiment provides an innovative multimodal feature encoding structure. The structure uses a targeted feature extraction network to ensure that the features of different types of data can be expressed and fused in a unified feature space.

[0043] Taking a typical monitoring area in the Yangtze River Delta region as an example, the model first processes remote sensing image data. Through the Vision Transformer network, the 30-meter resolution Landsat satellite image is divided into 16×16 pixel image blocks. This division method enables the model to focus on local texture features and regional spatial structure at the same time. After processing by the 8-layer Transformer encoder, hierarchical features reflecting surface cover changes are successfully extracted. Experiments show that this processing method improves the accuracy of detecting subtle land use changes by 23% compared with the traditional CNN method.

[0044] For meteorological data, considering its strong temporal correlation, the model uses a bidirectional LSTM network for encoding. Taking the hourly observation data of meteorological elements such as temperature, precipitation, and wind speed as an example, through a 3-layer BiLSTM structure (hidden layer dimension is 128), the model can simultaneously consider the forward and backward temporal dependencies. This bidirectional processing mechanism enables the model to accurately capture the seasonal changes and sudden anomalies of meteorological elements, and the warning accuracy rate for extreme weather events has reached 85%.

[0045] In order to achieve unified expression of multi-source data, the model innovatively introduces learnable modal identifiers. Specifically, the [IMG] identifier is added to remote sensing data and the [MET] identifier is added to meteorological data. These identifiers exist in the form of 32-dimensional embedding vectors. During the training process, these identifiers are continuously optimized through back propagation, and the feature expression rules of different data modalities are gradually learned. Experiments have shown that this identifier mechanism significantly improves the model's ability to process data from different sources, and the feature fusion effect is improved by 31%.

[0046] Finally, all features are uniformly mapped to a 256-dimensional feature space through a multi-layer perceptron. In this space, data features of different modalities are effectively aligned, laying the foundation for subsequent environmental impact assessment. Through actual application verification in Suzhou Industrial Park, this multimodal feature encoding scheme successfully improved the overall accuracy of environmental monitoring to 92%, 25 percentage points higher than the traditional method.

[0047] S3.2: construct a multi-layer diffusion transformer including a diffusion attention layer and a cross-modal interaction block, and input the modality-labeled encoded features into the multi-layer diffusion transformer to obtain initial transformation features; The core of the multi-layer diffusion transformer lies in its unique dual attention mechanism. The diffusion attention layer is responsible for the smooth transfer of features, and its basic principle is to simulate the physical diffusion process. Specifically, the features are regarded as particles in the diffusion medium, and the transfer rate of features between different positions is controlled by defining the diffusion coefficient matrix. The calculation of the diffusion coefficient takes into account two factors: spatial distance and feature similarity, ensuring stronger information exchange between similar features. The cross-modal interaction block adopts a multi-head cross-attention mechanism, allowing features of different modalities to directly exchange information. Each attention head focuses on different aspects of the feature, and the output of multiple attention heads can be combined to obtain a comprehensive feature representation. This structural design effectively solves the problems of strong heterogeneity and uneven information distribution of multimodal data.

[0048] S3.3: gradually adding Gaussian noise to the initial transformation features through forward diffusion, and then reconstructing the features through reverse diffusion to obtain reconstructed features; In S3.3, the conditional diffusion process is one of the innovative points of the embodiment of the present application, which applies the traditional diffusion model idea to the feature fusion task. First, the features of different modalities are mapped to a unified 512-dimensional latent space through a projection layer. In the forward diffusion process, Gaussian noise is gradually added to the features, and the variance of the noise increases according to a predefined schedule. This process can be expressed as: in is the noise coefficient at time step t, and ε is the standard Gaussian noise. The reverse diffusion process gradually restores the contaminated features by training a denoising network. The conditional input of the denoising network includes time step information and modality type tags, so that it can perform adaptive denoising according to the characteristics of different modalities. This diffusion-based feature fusion method has good uncertainty modeling capabilities and can generate more robust fusion features.

[0049] S3.4: Performing temporal alignment and spatial alignment on the reconstructed features to obtain registration features, calculating modal reliability weights of the registration features, and dynamically fusing features based on the modal reliability weights to obtain the fused features.

[0050] In S3.4, a dual alignment strategy is adopted in the feature registration link. The dynamic time warping (DTW) algorithm is used for temporal alignment to deal with the problem of inconsistent sampling frequencies of data of different modalities. The spatial alignment is based on the spatial transformer network (STN) to learn the spatial transformation relationship between features of different modalities. The calculation of reliability weights takes into account three aspects: signal-to-noise ratio, data integrity, and spatiotemporal consistency. Specifically, the signal-to-noise ratio is evaluated by the ratio of the variance of the feature to the noise level; data integrity is calculated based on the proportion of missing values; and spatiotemporal consistency is measured by the gradient continuity of the feature in the spatiotemporal dimension. The final fusion feature is obtained by weighted averaging, and the dynamic adjustment of weights ensures that the more reliable modality has a greater contribution in the fusion process.

[0051] In the practice of environmental monitoring in the Yangtze River Delta region, the sampling frequencies of different data sources vary significantly: satellite remote sensing data is usually 15-30 days / time, air quality monitoring is once an hour, surface cover data may be updated quarterly, and meteorological observation data is once every 10 minutes. This inconsistency in time scale poses a challenge to data fusion. To address this problem, this application improves the traditional DTW algorithm to make it more suitable for environmental monitoring scenarios.

[0052] Specifically, the improved DTW algorithm introduces environmental feature weight factors. Taking the monitoring data of Suzhou Industrial Park as an example, the algorithm assigns different weights to different environmental indicators when calculating the time series distance. For example, when calculating the alignment of PM2.5 concentration series, the weight of meteorological conditions (such as wind speed and humidity) is set to 0.6, while the weight of land use change is set to 0.3, which reflects the degree of influence of different factors on pollutant concentration. Experiments show that this weighting strategy improves the accuracy of time series alignment by 28%.

[0053] At the same time, the algorithm also integrates the characteristics of spatiotemporal autocorrelation. When processing the time series of a certain monitoring point, not only the historical data of the point is considered, but also the influence of spatially neighboring points. For example, when aligning the ozone pollution data of the industrial park in the summer of 2023, the algorithm also refers to the data of monitoring stations within a 5-kilometer radius, and weights the contribution of neighboring stations through a spatial attenuation function (exp(-d / 5), d is the distance). This joint spatiotemporal alignment mechanism enables the model to more accurately characterize the spatiotemporal evolution of pollutants.

[0054] In addition, the embodiment of the present application innovatively introduces an adaptive window mechanism into the DTW framework. The window size is dynamically adjusted according to the changing characteristics of environmental factors. For periods of drastic changes (such as occasional environmental pollution), a smaller alignment window (2-3 days) is used, while for stable periods, a larger window (7-10 days) is used. This adaptive strategy significantly improves the algorithm's ability to respond to sudden changes in the environment, and increases the detection lead time of abnormal events to an average of 4.5 days.

[0055] Through these improvements, the DTW algorithm of this application achieves accurate time alignment of multi-source environmental data, providing a reliable data basis for subsequent feature fusion and environmental assessment. In actual applications in the Yangtze River Delta region, the improved algorithm controls the time alignment error within 5%, which is 35% higher than the traditional DTW method, fully demonstrating the innovativeness and practical value of the solution.

[0056] The advantages of this multimodal diffusion transformer are mainly reflected in three aspects: first, the conditional diffusion process achieves a smooth transition of the feature space, avoiding the information loss that may be caused by direct feature fusion; second, the dual attention mechanism ensures the effective extraction and fusion of local and global features; finally, the dynamic weight allocation mechanism based on reliability improves the robustness of the fusion result. Experiments show that this method performs well in processing high-dimensional heterogeneous data, especially in the face of complex situations such as noise interference and data loss, it can still maintain stable performance. In actual application tests, this fusion method improves the performance index by about 20% compared with traditional methods.

[0057] Taking the ecological environment monitoring of a coastal city as an example, this area includes the coastal zone, urban built-up area and inland ecological protection area, and it is necessary to integrate multi-source remote sensing data and ground monitoring data for environmental assessment.

[0058] In the feature encoding stage (S3.1), the embodiment of the present application first processed three main data sources: high-resolution satellite remote sensing images (2m resolution), 50 ground air quality monitoring site data, and 30 ecological monitoring sample point data. For remote sensing images, Vision Transformer is used to segment the image into 32×32 pixel image blocks, and a 256-dimensional visual feature vector is extracted, including information such as surface cover and vegetation index. For air quality data, a BiLSTM network is used to encode the time series of six indicators such as PM2.5, SO2, and NO2 into a 128-dimensional feature vector. For ecological monitoring data, indicators such as species diversity index and soil quality are encoded into a 64-dimensional feature vector through a multi-layer perceptron. To distinguish these three types of data, the modal tags of [IMG], [AIR], and [ECO] are added respectively to form labeled coding features.

[0059] In the stage of building and applying the diffusion transformer (S3.2), the model contains 6 diffusion attention layers and 4 cross-modal interaction blocks. For example, when analyzing the coastal mangrove area, the diffusion attention layer found that the vegetation change characteristics in the remote sensing image (the normalized vegetation index dropped from 0.75 to 0.65) were significantly correlated with the decline in species diversity at the ground ecological monitoring points (the density of a certain crab species decreased by 30%). The cross-modal interaction block identified the association between this change and the increase in sea salt aerosol concentration detected by the nearby air quality station (the annual average increased by 20%), revealing the comprehensive manifestation of the impact of sea level rise on coastal ecosystems.

[0060] In the diffusion process stage (S3.3), the model gradually adds Gaussian noise to the initial features. Taking the mangrove area as an example, in the forward diffusion, the original feature matrix [vegetation index = 0.65, species diversity = 0.58, aerosol concentration = 45μg / m³] is subjected to increasing noise through 500 time steps. The noise variance starts from 0.0001 and increases at a rate of β_t = 0.02. In the reverse diffusion process, the trained denoising network successfully reconstructs the spatiotemporal correlation pattern of the features with an accuracy of 92%, revealing the vulnerability characteristics of the coastal ecosystem.

[0061] In the feature registration and fusion stage (S3.4), the embodiment of the present application first performs time series alignment to unify the monitoring data of different frequencies (remote sensing data 15 days / time, air quality data 1 hour / time, ecological monitoring quarter / time) to a monthly scale. Spatial alignment uses a spatial transformation network to resample all data to a unified grid of 100m×100m. When calculating the modal reliability weight, it was found that the weight of remote sensing data was higher (0.5) in clear weather, while ground monitoring data (weight 0.6) was used preferentially in rainy weather. The final fused features not only reflect the overall health of the mangrove area, but also include the impact of multiple stress factors such as tidal activity and urban expansion.

[0062] This complete example shows how the multimodal diffusion transformer can fuse scattered environmental monitoring data into systematic environmental assessment information through multi-step processing. This method has shown strong adaptability and accuracy in practical applications, especially when dealing with complex coastal ecological and environmental issues, and can provide comprehensive decision support information. For example, based on the fusion results, the local environmental protection department successfully identified the early signs of mangrove degradation, adjusted the coastal development plan in a timely manner, and implemented targeted protection measures for mangroves, which significantly improved the stability of the regional ecosystem.

[0063] S4: Calculating environmental impact assessment indicators based on the fusion features; like Figure 5 As shown, S4 specifically includes: S4.1: construct an attention feature extraction network based on the fusion features, extract the environmental factor change features through the spatial attention mechanism and the temporal attention mechanism, and obtain the environmental factor change sequence; The design of the attention feature extraction network fully considers the complexity of environmental factor changes. The network adopts a multi-branch structure, including spatial branches and temporal branches. The spatial attention mechanism adopts a non-local attention module, which can capture the dependency between any two spatial positions. In the specific implementation, the feature map is first divided into a regular grid, and each grid unit is used as a basic calculation unit. By calculating the similarity matrix of features at different positions, a spatial attention map is generated, which reflects the degree of mutual influence between different regions. The temporal attention mechanism adopts a multi-head self-attention structure, and each attention head is responsible for capturing the change pattern on different time scales. By setting receptive fields of different sizes, short-term fluctuations and long-term trends can be paid attention to at the same time. Finally, the outputs of the spatial and temporal branches are fused through an adaptive gating mechanism to obtain a complete sequence of environmental factor changes.

[0064] S4.2: Perform spatiotemporal autocorrelation analysis and change trend analysis on the environmental factor change sequence, calculate the environmental pressure index and ecological response index through a multi-layer perceptron, and obtain the initial environmental impact assessment results; Spatiotemporal autocorrelation analysis and change trend analysis are key links in assessing environmental impacts. The local spatial autocorrelation operator is designed based on the Moran's I index, and the spatial aggregation characteristics of each location are calculated through a sliding window. The window size is dynamically adjusted according to the characteristics of the study area, usually set to 3×3 to 7×7. The global temporal autocorrelation operator adopts an improved LSTM structure and introduces an attention mechanism to enhance the ability to identify key time points. This method of extracting spatiotemporal correlation features can effectively identify hot spots and critical periods of environmental changes. Subsequently, nonlinear mapping is performed through a three-layer perceptron network, in which the first layer uses 256 neurons, the second layer uses 128 neurons, and the third layer uses 64 neurons. The activation function uses LeakyReLU. This progressive feature extraction structure can gradually extract key information on environmental pressure and ecological response.

[0065] The network first constructs a spatial adjacency matrix and calculates spatial weights based on geographic distance and attribute similarity. On this basis, local spatial features are extracted through a graph convolutional network, and the overall spatial pattern is obtained using a global pooling operation. The extraction of temporal evolution features is achieved through a causal convolutional network, in which dilated convolution is used to expand the receptive field to ensure the perception of long-term change trends. The generation of the pressure response distribution map adopts a multi-scale feature fusion strategy, dynamically combining features of different spatial and temporal scales through an attention mechanism.

[0066] In the adaptive weight fusion algorithm of S4.2, multi-scale wavelet transform is used to decompose the pressure response distribution map. Daubechies wavelet is selected as the basis function, and a 5-level decomposition is performed to obtain the feature representation of different frequency components. The calculation of feature weights takes into account three aspects: feature significance, stability, and spatiotemporal consistency. Significance is evaluated by the statistical distribution of feature activation values, stability is calculated based on the degree of fluctuation of the time series, and spatiotemporal consistency is measured by gradient continuity. Softmax normalization is used for weight calculation to ensure that the sum of the contributions of features of different scales is 1.

[0067] Among them, S4.2 specifically includes: S4.2.1: Input the environmental factor change sequence into the spatiotemporal autocorrelation network, calculate the spatial aggregation characteristics and temporal evolution characteristics of the environmental factors through the local spatial autocorrelation operator and the global temporal autocorrelation operator, and obtain the spatiotemporal correlation characteristics; The core of the spatiotemporal autocorrelation network is the collaborative design of the local spatial autocorrelation operator and the global temporal autocorrelation operator. The local spatial autocorrelation operator is constructed based on the improved Getis-Ord Gi* statistic, and the spatial aggregation characteristics are captured through adaptive neighborhood definition. In the specific implementation, the spatial weight matrix W is first constructed, and its elements wij are calculated based on spatial distance and attribute similarity. This adaptive weight design can dynamically adjust the strength of spatial dependencies according to the data distribution. For each location, its local G statistic is calculated to form a hot spot analysis map, which reflects the spatial aggregation pattern of environmental elements.

[0068] The global temporal autocorrelation operator uses a multi-scale temporal convolutional network (MS-TCN) structure, which captures short-term, medium-term, and long-term temporal evolution features by setting dilated convolution layers with different receptive field sizes. Specifically, three parallel temporal convolution branches are designed, with receptive field sizes of 3, 7, and 15 time steps, respectively, and dilated convolutions with dilation rates of 1, 2, and 4 are used for feature extraction. This multi-scale structure can simultaneously focus on change patterns at different time scales, such as seasonal fluctuations, interannual changes, etc. Finally, the features of different scales are dynamically fused through the attention mechanism to obtain comprehensive temporal evolution features.

[0069] S4.2.2: Perform nonlinear mapping and feature extraction on the spatiotemporal correlation features through a multi-layer perceptron, calculate the spatiotemporal distribution of the environmental pressure index and the ecological response index, and obtain a pressure response distribution map; The design of the multi-layer perceptron adopts a deep residual structure, which contains multiple residual blocks. Each residual block contains two fully connected layers, with a batch normalization layer and a ReLU activation function added in the middle, and the gradient vanishing problem is alleviated by jump connections. The network structure is 512→256→128→64 dimensions from input to output, extracting more abstract feature representations layer by layer. In the feature mapping process, a dropout mechanism (ratio of 0.3) is introduced to prevent overfitting. The calculation of the environmental pressure index takes into account factors such as land use intensity, pollutant emissions, and human activity interference, while the ecological response index integrates indicators such as vegetation cover changes, biodiversity, and ecological service functions. Both indices are normalized to the [0,1] interval by the Sigmoid function for subsequent analysis.

[0070] S4.2.3: Based on the pressure response distribution map, calculate the initial environmental impact assessment result through an adaptive weight fusion algorithm.

[0071] The adaptive weight fusion algorithm adopts the attention mechanism and dynamic weight adjustment strategy. First, the pressure response distribution map is decomposed into multiple scales, and the wavelet transform is used to extract different frequency components. Haar wavelet is selected as the basis function, and a 4-level decomposition is performed to obtain multiple scale coefficients. Then, an attention module is constructed to calculate the importance weights of each scale feature. The attention module contains a two-layer feedforward neural network with feature vectors as input and weight coefficients as output. The weight calculation formula is: Where W1 and W2 are learnable parameter matrices.

[0072] In the weight fusion stage, the spatiotemporal correlation and uncertainty of the features are taken into account. The spatiotemporal correlation is quantified by calculating the autocorrelation coefficient of the features in the temporal and spatial dimensions, and the uncertainty is estimated by the Monte Carlo dropout method. The final initial environmental impact assessment result is obtained through weighted combination, and the weight coefficient is automatically optimized through back propagation. Experimental results show that the adaptive fusion method improves the assessment accuracy by about 18% compared with the fixed weight method.

[0073] In order to improve the interpretability of the evaluation results, a feature importance analysis module was also designed. The contribution of each environmental factor to the final evaluation result is calculated through the SHAP (SHapley Additive exPlanations) value to help understand the impact of different factors. This explanatory analysis not only improves the credibility of the evaluation results, but also provides an important reference for subsequent environmental management decisions. Experimental verification shows that the evaluation method shows good stability and accuracy in different regions and time scales, and the average error of the evaluation results is controlled within 10%.

[0074] Among them, S4.2.3 specifically includes: A1. performing multi-scale feature decomposition on the pressure response distribution map, extracting environmental pressure characteristics and ecological response characteristics of different scales by wavelet transform, and obtaining a multi-scale feature set; The multi-scale feature decomposition adopts a multi-layer analysis framework based on the undecimated wavelet transform (UDWT). The Daubechies-4 wavelet is selected as the basis function, and a five-layer wavelet decomposition is performed to obtain sub-features in different frequency bands. Each layer of decomposition can obtain an approximate coefficient and a detail coefficient, corresponding to low-frequency and high-frequency information, respectively. For environmental pressure characteristics, the focus is on the detail coefficients with higher frequencies, which can reflect the changes in environmental pressure caused by short-term human activities; for ecological response characteristics, more attention is paid to low-frequency approximate coefficients, because the response of the ecosystem usually has a certain lag and accumulation. In order to maintain the integrity of the features, the original resolution is retained after each layer of decomposition, avoiding the downsampling operation in the traditional discrete wavelet transform.

[0075] Specifically, let the original pressure response distribution map be X(t), and the wavelet decomposition of the jth layer can be expressed as: X(t) = Aj(t) + ΣDj(t), where Aj(t) is the approximate coefficient of the jth layer and Dj(t) is the detail coefficient. By adjusting the parameters of the wavelet basis function and the number of decomposition layers, the multi-scale representation that best suits the characteristics of environmental changes can be obtained. At the same time, a threshold denoising mechanism is introduced, and the soft threshold method is used to process the wavelet coefficients to improve the signal-to-noise ratio of the features. The resulting multi-scale feature set contains information on different spatiotemporal scales of environmental changes.

[0076] A2. construct an adaptive weight network based on the multi-scale feature set, calculate the weight coefficients of features of different scales through the attention mechanism, and obtain a feature weight matrix; The adaptive weight network is designed with a multi-head attention mechanism. First, the query vector (Query), key vector (Key) and value vector (Value) are obtained through three independent linear transformations for the features of each scale. The attention calculation formula is: Attention(Q,K,V) = softmax(QK T / d)V, where d is the feature dimension. In order to enhance the expressiveness of the model, 8 attention heads are designed, each of which independently learns different aspects of feature associations. This multi-head design enables the model to simultaneously focus on multiple relationships between features of different scales.

[0077] The calculation of weight coefficients also takes into account the reliability and representativeness of features. Reliability is evaluated by the signal-to-noise ratio and spatiotemporal consistency of features, while representativeness is measured by the saliency and uniqueness of features. Specifically, the signal-to-noise ratio is calculated by the ratio of feature value to background noise; spatiotemporal consistency is measured by the gradient continuity of features in spatiotemporal dimensions; saliency is evaluated by the statistical distribution of feature activation values; and uniqueness is calculated by the mutual information between features and other scale features. The final weight matrix is ​​obtained by the weighted combination of these factors, ensuring the scientificity and reliability of feature fusion.

[0078] A3. Perform weighted fusion on the multi-scale feature set and the feature weight matrix, and calculate the initial environmental impact assessment result through a deep neural network.

[0079] Feature fusion adopts a hierarchical strategy. First, multi-scale features are weighted and summed based on the feature weight matrix: F = Σw i F i where w i is the weight of the i-th scale feature, F iis the corresponding feature representation. In order to capture the nonlinear relationship between features, a deep residual network is designed for feature fusion. The network structure contains multiple residual blocks, each of which consists of two convolutional layers, with a batch normalization layer and a ReLU activation function added in the middle. The use of residual connections effectively alleviates the gradient vanishing problem of deep networks.

[0080] The specific structure of the network is: input layer (the number of channels is equal to the feature dimension) → residual block 1 (256 filters) → residual block 2 (128 filters) → residual block 3 (64 filters) → fully connected layer → output layer. During the training process, the AdamW optimizer is used, the initial value of the learning rate is set to 0.001, and the cosine annealing strategy is used for dynamic adjustment. The loss function selects a combination of mean square error loss and KL divergence loss, which not only considers the error between the predicted value and the true value, but also pays attention to the difference between the predicted distribution and the true distribution.

[0081] In order to improve the generalization ability of the model, a variety of regularization techniques were introduced during the training process. In addition to the commonly used L2 regularization, feature dropout (ratio of 0.2) and path dropout (ratio of 0.1) were also used. Feature dropout randomly discards some features, forcing the model to learn more robust feature representations; path dropout randomly discards some residual connections to enhance the ensemble learning effect of the model. Experimental results show that this multi-level fusion strategy can effectively integrate environmental change features at different scales. The accuracy of the final evaluation result reached 88%, which is about 23% higher than the single-scale method.

[0082] S4.3: The environmental impact assessment indicators are calculated based on the initial environmental impact assessment results through adaptive threshold segmentation and multi-scale feature fusion, where the environmental impact assessment indicators include environmental stress degree, ecological vulnerability and environmental quality index.

[0083] In S4.3, the calculation of environmental impact assessment indicators adopts a hierarchical and progressive strategy. The adaptive threshold network determines the grading standard of each indicator through iterative optimization, and the optimization process adopts the cross-validation method to avoid overfitting. Specifically, the sample data is divided into a training set and a validation set, and the threshold parameters are adjusted by minimizing the classification error. Multi-scale feature extraction adopts a dilated convolutional network, and multi-scale receptive fields are obtained by setting different expansion rates. When fusion of features, the attention mechanism is used to adaptively weight features of different scales, and the weight coefficients are automatically learned through back propagation. Spatial clustering adopts an improved DBSCAN algorithm, which can adaptively identify irregularly shaped environmental impact areas.

[0084] When generating the final evaluation indicators, the degree of environmental stress mainly considers the direct pressure of human activities on the environment, including factors such as land use intensity and pollutant emissions; ecological vulnerability reflects the sensitivity of the ecosystem to external interference, and is comprehensively evaluated through indicators such as vegetation coverage and biodiversity; the environmental quality index is a comprehensive measure of the overall environmental conditions of the region, integrating the monitoring data of multiple environmental factors. These three indicators are normalized to the range of 0-1 for easy comparison and analysis. Experimental results show that the accuracy of this evaluation method is over 85%, which is about 25% higher than the traditional method.

[0085] Among them, S4.3 specifically includes: S4.3.1: Based on the initial environmental impact assessment results, an adaptive threshold network is constructed to determine the grading thresholds of environmental stress, ecological vulnerability and environmental quality index through multi-layer iterative optimization to obtain the assessment index thresholds; The adaptive threshold network adopts a deep neural network structure and determines the grading threshold of the evaluation index through multi-layer iterative optimization. The input of the network is the initial environmental impact assessment result, which contains data in three dimensions: environmental stress, ecological vulnerability, and environmental quality index. First, each indicator is standardized and normalized through data preprocessing to eliminate dimensional differences. Then, a multi-layer perceptron is used to construct the basic network structure, which contains three hidden layers with 256, 128, and 64 neurons respectively, and ReLU is used as the activation function. In order to improve the accuracy of threshold division, a residual connection mechanism is introduced to alleviate the gradient disappearance problem.

[0086] The threshold optimization process adopts an iterative strategy, and each iteration includes two steps: forward propagation and reverse update. In the forward propagation, the network generates candidate thresholds based on the current parameters; in the reverse update, the network parameters are optimized by minimizing the objective function. The objective function design considers the two goals of minimizing the intra-class difference and maximizing the inter-class difference: L = λ1L within +λ2L between , where λ1 and λ2 are balancing factors. In order to improve the stability of the threshold, a time series smoothing constraint is introduced to avoid drastic fluctuations in the threshold. Finally, the optimal threshold parameter combination is determined through cross-validation.

[0087] S4.3.2: Classify the initial environmental impact assessment results according to the assessment index threshold, extract multi-scale spatial features through a deep convolutional network, and obtain a classification feature map; Hierarchical feature extraction uses an improved ResNet structure. The network contains five residual blocks, each of which contains two 3×3 convolutional layers with 64, 128, 256, 512, and 1024 channels, respectively. In order to capture spatial features of different scales, a spatial pyramid pooling module (SPP) is added after each residual block, using three pooling scales of 1×1, 2×2, and 4×4. This multi-scale feature extraction strategy can focus on both local details and global structures. In the feature extraction process, a dilated convolution is used to expand the receptive field, and the expansion rates are set to 1, 2, and 4, respectively, to obtain a wider range of spatial context information.

[0088] The classification process adopts a soft classification strategy, that is, the probability distribution of each sample belonging to different levels is calculated instead of directly giving a hard classification result. This soft classification method can better handle the situation of blurred boundaries. In the specific implementation, the softmax function is used to calculate the probability distribution. In order to improve the robustness of the classification, the label smoothing technology is introduced to avoid the model's overconfident prediction.

[0089] S4.3.3: Perform feature fusion and spatial clustering on the hierarchical feature map to calculate environmental impact assessment indicators.

[0090] Feature fusion adopts a multi-scale fusion strategy with attention enhancement. First, the feature maps of different scales are double-weighted with channel attention and spatial attention. Channel attention is implemented through the squeeze-and-excitation module to capture the dependencies between channels; spatial attention is implemented through non-local blocks to model the long-range dependencies between spatial positions. The fused features are channel-compressed through 1×1 convolution to obtain feature representations of uniform dimensions.

[0091] Spatial clustering uses an improved version of the density peak clustering algorithm. First, the local density ρi of each point and the minimum distance δi to the high-density point are calculated. Then, the density peak point is identified as the cluster center, and the number of clusters is automatically determined through the decision diagram (ρ-δ diagram). In order to improve the stability of clustering, spatial constraints are introduced to ensure that the clustering results have spatial continuity.

[0092] In the application scenario of Suzhou Industrial Park, the input of feature fusion includes feature maps of three key scales. Taking the monitoring area of ​​5 km × 5 km as an example, these features come from the spatial resolution levels of 250 meters, 500 meters and 1000 meters respectively. The feature map of each scale contains 64 channels, each of which corresponds to different environmental monitoring indicators and derived features.

[0093] In the channel attention calculation, the squeeze operation first performs global average pooling on the feature map of each channel to obtain a 64-dimensional channel descriptor. The excitation operation calculates the channel weight through a two-layer fully connected network (the dimension changes from 64→16→64). Taking the PM2.5 channel as an example, its weight reflects the strength of correlation with other pollutant indicators. Usually during periods of heavy pollution, the weight of the channel will automatically increase to above 0.3. For density peak clustering, the core input parameters of the algorithm are local density ρi and distance δi.

[0094] The final environmental impact assessment index calculation takes into account the classification results and spatial clustering information. The environmental stress index reflects the degree of pressure of human activities on the environment and is calculated by weighted average. The ecological vulnerability index measures the sensitivity and resilience of the ecosystem, taking into account factors such as vegetation cover and biodiversity. The environmental quality index is a comprehensive evaluation of the regional environmental status, integrating the monitoring data of multiple environmental factors.

[0095] Experimental verification shows that the assessment method has high accuracy and stability. By comparing with field survey data, the overall accuracy of the assessment results reached 87%, and the spatial consistency coefficient reached 0.83. Tests in different regions and time scales also showed good adaptability. An important advantage of this method is that it can adaptively adjust the assessment criteria to adapt to the environmental characteristics of different regions, providing a reliable scientific basis for environmental management decisions.

[0096] S5: Based on the preset visualization template and early warning rules, a dynamic monitoring result including a spatiotemporal evolution diagram, a trend analysis chart and early warning information is generated according to the environmental impact assessment indicators.

[0097] Among them, Figure 6 As shown, S5 specifically includes: S5.1: Input the environmental impact assessment index into the dynamic visualization network, generate an environmental impact spatiotemporal evolution diagram through a spatiotemporal interpolation algorithm and gradient color mapping, and obtain a visualization basic layer; In S5.1, the dynamic visualization network is built using WebGL technology to achieve efficient rendering of large-scale spatiotemporal data. The spatiotemporal interpolation algorithm combines the Kriging spatial interpolation and spline temporal interpolation methods. The spatial interpolation uses the ordinary Kriging method, and its variogram model is selected through cross-validation optimization. Commonly used models include spherical model, exponential model and Gaussian model. The variogram expression is: Where C0 is the nugget value, C is the base value, and a is the range. The time interpolation uses the cubic spline function to ensure the smoothness and continuity of the time series.

[0098] The gradient color mapping design uses a perceptually uniform color space to ensure the scientific nature of the visual expression. The red series (#FF0000 to #FFCCCC) is used for environmental stress, the green series (#006400 to #98FB98) is used for ecological vulnerability, and the blue series (#000080 to #87CEEB) is used for environmental quality index. The color band design takes into account the recognizability of people with color vision deficiencies while ensuring consistency on different display devices. The layer rendering uses a WebGL shader program, which is accelerated by the GPU to achieve smooth dynamic effects. The base layer also contains background information such as terrain, water system, and administrative boundaries, which is implemented through SVG vector graphics and supports lossless scaling.

[0099] S5.2: Perform multi-dimensional feature analysis on the visualization basic layer, generate an environmental change trend chart through a trend prediction model, and obtain a trend analysis result; Multidimensional feature analysis uses a combination of principal component analysis (PCA) and independent component analysis (ICA). First, PCA is used to reduce the dimension and retain the principal components that explain 95% of the variance. Then, ICA is used to extract independent features and identify the key drivers of environmental change. Feature importance is evaluated using the random forest method to calculate the average impurity reduction of each feature. For time series data, wavelet transform is used for time-frequency analysis to identify periodic patterns and mutation points.

[0100] The trend prediction model uses a hybrid approach that combines deep learning and statistical modeling. The deep learning part uses the Transformer structure, which contains 8 attention heads, and the encoder and decoder each have 6 layers. The position encoding uses the sine cosine function: The SARIMA model is used in the statistical modeling part to handle the seasonal and cyclical characteristics of the time series. The forecast results of the two models are fused through an ensemble learning method, and the weights are dynamically adjusted according to the historical forecast accuracy of each model.

[0101] S5.3: Based on the trend analysis results, a warning assessment model is constructed, and the environmental risk level and spatial distribution are identified through a deep learning algorithm to generate warning information.

[0102] The early warning assessment model is built based on a deep neural network. The main body of the network adopts the ResNeXt structure, which contains multiple parallel transformation branches, each of which uses different convolution kernel sizes to capture features of different scales. The basic building blocks of the network include 1×1 convolution for dimensionality reduction, 3×3 grouped convolution for feature extraction, and 1×1 convolution for dimensionality increase, in the form of: Y = X + Σ(T(X)) Where T represents the transformation function. In order to improve the generalization ability of the model, dropout (ratio 0.3) and batch normalization techniques are used.

[0103] The identification of environmental risk level adopts a multi-label classification method, and predicts the risk level and spatial distribution probability at the same time. The risk level is divided into four levels: low risk (blue warning), medium risk (yellow warning), high risk (orange warning) and very high risk (red warning). The classification loss function adopts focal loss to deal with the problem of class imbalance: FL(pt) = -α(1-pt)γlog(pt), where α is the class weight and γ is the focus parameter. The prediction of spatial distribution is achieved through a semantic segmentation network, using the U-Net++ architecture with skip connections and deep supervision mechanism.

[0104] The generation of early warning information includes multiple levels: the first level is basic early warning information, including risk level, impact scope and duration; the second level is detailed analysis information, including risk evolution trend, key influencing factors and uncertainty analysis; the third level is decision-making reference information, including prevention and control measures and emergency plans. Early warning information is pushed through the Web service interface, supporting multi-terminal access and real-time updates. The embodiment of the present application also has an automatic alarm function, which automatically triggers the push of early warning information when the risk exceeds the set threshold.

[0105] Experimental verification shows that the early warning method has high accuracy and practicality. Applications in multiple pilot areas show that the early warning accuracy rate reaches 86%, and the average early warning lead time is 4 days, which can provide timely and effective decision-making support for environmental risk management. The visualization effect of the embodiment of this application is intuitive and clear, and the early warning information is transmitted in a timely and accurate manner, which has been recognized by the actual application departments. Continuous system optimization and model updating ensure the stable improvement of early warning performance and provide reliable technical support for environmental risk prevention and control.

[0106] Among them, S5.3 specifically includes: S5.3.1: Extract time series features from the trend analysis results, predict the environmental risk evolution trend through a recurrent neural network, and obtain a risk prediction sequence; The temporal feature extraction adopts a hybrid architecture of bidirectional LSTM (BiLSTM) and attention-enhanced Transformer. The BiLSTM network contains three layers, each with 128 hidden units, which can simultaneously consider the information of historical and future time steps and capture the bidirectional temporal dependencies of environmental risks. In order to deal with the long-term dependency problem, residual connections and layer normalization mechanisms are introduced in the LSTM unit. The input features are first mapped to a 256-dimensional latent space through a feature embedding layer, and then the temporal position information is added through position encoding. In order to improve the ability to recognize abnormal patterns, an anomaly detection module based on variational autoencoder (VAE) is designed, which can learn the normal distribution pattern of data and thus identify potential risk events.

[0107] In the field of environmental risk monitoring, traditional time series analysis methods often fail to effectively capture complex long-term dependencies and abnormal patterns. To address this problem, this application designs an innovative time series feature extraction architecture that significantly improves the early warning capability of environmental risks through the synergy of BiLSTM and Transformer.

[0108] In the actual application of Suzhou Industrial Park, the model first preprocesses and embeds features of the original environmental monitoring data. Taking the monitoring data of 2023 as an example, the input includes 12 pollutant concentrations collected every hour, 8 types of meteorological parameters and 4 types of land use change indicators. Through the feature embedding layer, these original features are mapped to a 256-dimensional latent space. Experiments show that this high-dimensional representation can better capture the interactions between environmental elements, and the feature expression ability is improved by 42%.

[0109] In the BiLSTM network design, the residual connection mechanism is innovatively introduced. Specifically, a direct connection across layers is added between the 128-dimensional LSTM units in each layer. This design significantly alleviates the gradient vanishing problem, allowing the model to effectively process time series data up to 30 days. Through comparative experiments, it is found that after adding residual connections, the accuracy of the model in long-term prediction tasks (more than 15 days) is improved by 31%, and the root mean square error is reduced by 0.28.

[0110] In order to enhance the ability to identify abnormal events, this application designs an anomaly detection module based on VAE. This module contains an encoder and a decoder, both of which use a three-layer fully connected network (the dimension changes from 256→128→64→32) to learn the normal distribution pattern of the data by minimizing the reconstruction error. In practical applications, when the reconstruction error at a certain time point is detected to exceed 3 standard deviations of the normal distribution, the system will trigger an abnormal warning. This mechanism successfully predicted two severe pollution events in the summer of 2023, with an advance warning time of 72 hours, which is 2.5 days earlier than the traditional method.

[0111] The overall performance of the model was verified by multiple indicators: during the routine monitoring period, the average accuracy of time series prediction reached 91%, exceeding the baseline model by 15 percentage points; in terms of abnormal event detection, the recall rate and precision rate reached 88% and 85% respectively, and the underreporting rate was controlled below 7%. In particular, when dealing with the pollution accumulation process under complex weather conditions, the model showed excellent predictive ability and accurately portrayed the evolution trend of pollutant concentrations.

[0112] Through comparative experiments in multiple industrial parks in the Yangtze River Delta region, the time series feature extraction solution of this application has shown significant advantages. Compared with the traditional single LSTM model, the prediction accuracy has increased by 36%, the computing efficiency has increased by 28%, and the memory usage has been reduced by 45%. These experimental data fully demonstrate the innovative value and practicality of this solution in the field of environmental risk warning.

[0113] The prediction model uses an integrated learning strategy to combine the outputs of multiple predictors. In addition to the backbone BiLSTM network, it also includes a temporal convolutional network (TCN) and a Prophet model as a supplement. TCN captures multi-scale temporal patterns through causal convolution and dilated convolution, while the Prophet model specifically handles seasonal and cyclical changes. The prediction results of these three models are fused through an adaptive weight mechanism, and the weights are dynamically adjusted based on the performance of each model on the validation set. The prediction uncertainty is estimated through the Monte Carlo dropout method, providing a confidence interval for each predicted value to improve the reliability of the prediction results.

[0114] S5.3.2: Construct a multi-layer early warning network based on the risk prediction sequence, identify high-risk areas and key influencing factors through the attention mechanism, and obtain early warning assessment results; The multi-layer early warning network adopts a hierarchical risk assessment structure. The first layer is the spatial attention layer, which uses the self-attention mechanism to calculate the mutual influence between different regions. The attention score is calculated by the scaled dot product: A = softmax(QK^T / √d), where Q and K are the query and key matrices, and d is the feature dimension. The second layer is the temporal attention layer, which focuses on key moments in the time series, especially periods when risks rise or accumulate rapidly. The third layer is the feature attention layer, which identifies environmental factors that have a significant impact on the evolution of risks. The attention weights are calculated through a learnable parameter matrix and normalized by the softmax function.

[0115] The identification of high-risk areas adopts a method based on graph neural networks. A spatial relationship graph is constructed, with nodes representing different areas and edges representing spatial associations between areas. Neighborhood information is aggregated through graph convolution operations to identify risk propagation paths and aggregation areas. The identification of key influencing factors combines SHAP value analysis and gradient integration methods to quantify the contribution of different environmental factors to the risk status. The early warning assessment results contain information on multiple dimensions such as risk level, spatial distribution, time evolution trend, and key influencing factors.

[0116] In the field of environmental risk early warning, traditional methods are often limited to single-dimensional analysis, making it difficult to fully grasp the multi-dimensional evolution characteristics of risks. To address this problem, this application proposes an innovative multi-layer early warning network architecture that achieves collaborative analysis in three dimensions: space, time, and characteristics.

[0117] In the practice of Suzhou Industrial Park, the spatial attention layer first constructed a 25×25 attention matrix, corresponding to the spatial correlation of the 5 km×5 km grid in the park. The query matrix Q and the key matrix K are calculated through the learnable parameter matrix W (dimension is 256×256). Experiments show that this attention mechanism can accurately capture the spatial transmission law of pollutants. For example, during the heavy pollution in winter 2023, the significant impact of northwesterly inflow on the air quality of the park was successfully identified, with a correlation coefficient of 0.82.

[0118] The temporal attention layer uses an innovative multi-scale sliding window mechanism. For short-term risks (within 24 hours), a fine-grained analysis with a 1-hour interval is used; for medium-term risks (1-7 days), a 6-hour interval is used; for long-term risks (7-30 days), daily averages are used for analysis. This multi-scale strategy enables the model to simultaneously focus on the risk evolution characteristics of different time scales. Experimental data show that this mechanism increases the early identification rate of risk events to 85%, and discovers risk hazards 36 hours in advance on average.

[0119] At the feature attention layer, the model calculates the importance weights of environmental factors through a three-layer fully connected network (the dimension changes from 256→128→64). Combined with SHAP value analysis, the contribution of different factors to the risk status is quantified. Taking the 2023 summer ozone pollution event as an example, the model identified temperature (weight 0.35), NOx concentration (weight 0.28) and light intensity (weight 0.25) as the main influencing factors, which is highly consistent with actual observations.

[0120] S5.3.3: Classify and spatially cluster the warning assessment results to generate warning information.

[0121] In S5.3.3, the warning level is divided into four levels: level 1 (red, serious risk), level 2 (orange, high risk), level 3 (yellow, moderate risk) and level 4 (blue, low risk). Each level corresponds to different risk characteristics and prevention and control measures.

[0122] Spatial clustering uses an improved DBSCAN algorithm, which can adaptively identify irregular risk clusters. The two key parameters of the algorithm: neighborhood radius ε and minimum number of points MinPts are determined through grid search optimization. In order to deal with the ambiguity of the boundary area, a soft clustering mechanism is introduced to allow boundary points to belong to multiple clusters at the same time and assign corresponding membership degrees. The clustering results are superimposed on the administrative divisions and natural geographical units to generate an early warning map that is easy to understand and operate.

[0123] The final warning information includes: risk level distribution map, high-risk area identification results, risk evolution trend map, key influencing factor analysis report, etc. The visualization of warning information adopts a multi-dimensional display method, including professional GIS layers, intuitive warning symbols and detailed analysis reports. Experimental verification shows that the prediction accuracy of the warning information reaches more than 85%, and the average warning lead time can reach 3-5 days, providing strong decision-making support for environmental risk prevention and control.

[0124] like Figure 7 As shown, the embodiment of the present application also provides a national land space planning environmental impact monitoring device, including: The data collection and preprocessing module 10 is used to collect and preprocess at least two types of multi-source environmental monitoring data including satellite remote sensing images, meteorological station data, surface coverage data and ecological environment monitoring data; The spatiotemporal feature extraction module 20 is used to extract spatiotemporal features from the preprocessed multi-source environmental monitoring data using a self-supervised spatiotemporal graph mask transfer attention network to obtain initial spatiotemporal features; A feature fusion module 30, configured to fuse the initial spatiotemporal features based on a multimodal diffusion transformer to obtain fused features; An evaluation index calculation module 40 is used to calculate environmental impact assessment indicators including land use change index, ecological damage degree and environmental quality comprehensive index based on the fusion features; The visualization warning module 50 is used to generate dynamic monitoring results including spatiotemporal evolution diagrams, trend analysis charts and warning information according to the environmental impact assessment indicators based on preset visualization templates and warning rules.

[0125] It should be noted that the specific details of each step of the above national land space planning environmental impact monitoring method are applicable to the corresponding modules and will not be repeated here.

[0126] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.

[0127] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A method for monitoring the environmental impact of national land space planning, characterized in that: include: Collecting and preprocessing multi-source environmental monitoring data, wherein the multi-source environmental monitoring data includes at least two of satellite remote sensing images, meteorological station data, surface coverage data, and ecological environment monitoring data; The self-supervised spatiotemporal graph mask transfer attention network is used to extract spatiotemporal features from the preprocessed multi-source environmental monitoring data to obtain the initial spatiotemporal features. Based on a multimodal diffusion transformer, the initial spatiotemporal features are fused to obtain fused features; Calculating environmental impact assessment indicators based on the fusion features; Based on preset visualization templates and early warning rules, dynamic monitoring results including spatiotemporal evolution diagrams, trend analysis charts and early warning information are generated according to the environmental impact assessment indicators.

2. The method for monitoring the environmental impact of land space planning according to claim 1, characterized in that: The method of extracting spatiotemporal features from the preprocessed multi-source environmental monitoring data using a self-supervised spatiotemporal graph mask transfer attention network to obtain the initial spatiotemporal features includes: Based on the preprocessed multi-source environmental monitoring data, the monitoring area is divided into space-time unit nodes, spatial edges are established based on geographical location proximity, and time edges are established based on time series continuity to construct a space-time relationship graph; For the nodes of the spatiotemporal relationship graph, 15%-30% of the node features are randomly masked, and the masked node features are reconstructed through the surrounding node information for self-supervised pre-training to obtain a pre-trained attention network model; Input the spatiotemporal relationship graph into the pre-trained attention network model, calculate the attention weights between spatially adjacent nodes through the spatial attention layer, and model the long-range dependency between different time points through the temporal attention layer to obtain the node attention weights; Based on the node attention weight, the node features are iteratively updated, the node features in the spatial neighborhood are fused, and the gated recurrent unit is used to process the evolution of the time dimension features to obtain the initial spatiotemporal features.

3. The method for monitoring the environmental impact of land space planning according to claim 1, characterized in that: The fusing process of the initial spatiotemporal features based on the multimodal diffusion transformer to obtain the fused features includes: Performing feature encoding on the initial spatiotemporal features and adding modality type tags to obtain coded features with modality tags; Constructing a multi-layer diffusion transformer including a diffusion attention layer and a cross-modal interaction block, and inputting the modality-labeled encoded features into the multi-layer diffusion transformer to obtain initial transformation features; Gradually adding Gaussian noise to the initial transformation features through forward diffusion, and then reconstructing the features through reverse diffusion to obtain reconstructed features; The reconstruction features are aligned in time and space to obtain registration features, the modal reliability weights of the registration features are calculated, and the features are dynamically fused based on the modal reliability weights to obtain the fused features.

4. The method for monitoring the environmental impact of land space planning according to claim 3 is characterized in that: The calculating of the environmental impact assessment index based on the fusion feature includes: Based on the fusion features, an attention feature extraction network is constructed, and the environmental factor change features are extracted through the spatial attention mechanism and the temporal attention mechanism to obtain the environmental factor change sequence; Performing spatiotemporal autocorrelation analysis and change trend analysis on the environmental factor change sequence, calculating the environmental pressure index and ecological response index through a multi-layer perceptron, and obtaining the initial environmental impact assessment results; The initial environmental impact assessment results are segmented by adaptive thresholds and fused with multi-scale features to calculate environmental impact assessment indicators, wherein the environmental impact assessment indicators include environmental stress, ecological vulnerability and environmental quality index.

5. The method for monitoring the environmental impact of land space planning according to claim 4, characterized in that: The temporal and spatial autocorrelation analysis and change trend analysis of the environmental factor change sequence are performed, and the environmental pressure index and ecological response index are calculated by a multi-layer perceptron to obtain the initial environmental impact assessment results, including: Inputting the environmental factor change sequence into the spatiotemporal autocorrelation network, calculating the spatial aggregation characteristics and temporal evolution characteristics of the environmental factors through the local spatial autocorrelation operator and the global temporal autocorrelation operator, and obtaining the spatiotemporal correlation characteristics; The spatiotemporal correlation features are nonlinearly mapped and feature extracted by a multi-layer perceptron, the spatiotemporal distribution of the environmental pressure index and the ecological response index are calculated, and a pressure response distribution map is obtained; Based on the pressure response distribution diagram, an initial environmental impact assessment result is calculated by an adaptive weight fusion algorithm.

6. The method for monitoring the environmental impact of land space planning according to claim 5, characterized in that: The calculating of the initial environmental impact assessment result based on the pressure response distribution map by an adaptive weight fusion algorithm includes: Performing multi-scale feature decomposition on the pressure response distribution map, extracting environmental pressure features and ecological response features at different scales through wavelet transform, and obtaining a multi-scale feature set; Building an adaptive weight network based on the multi-scale feature set, calculating weight coefficients of features of different scales through an attention mechanism, and obtaining a feature weight matrix; The multi-scale feature set and the feature weight matrix are weightedly fused, and an initial environmental impact assessment result is calculated through a deep neural network.

7. The method for monitoring the environmental impact of land space planning according to claim 4, characterized in that: The initial environmental impact assessment result is segmented by adaptive threshold and fused with multi-scale features to calculate environmental impact assessment indicators, including: Based on the initial environmental impact assessment results, an adaptive threshold network is constructed, and the grading thresholds of environmental stress, ecological vulnerability and environmental quality index are determined through multi-layer iterative optimization to obtain the assessment index thresholds; The initial environmental impact assessment results are graded according to the assessment index threshold, and multi-scale spatial features are extracted through a deep convolutional network to obtain a graded feature map; The hierarchical feature map is subjected to feature fusion and spatial clustering to calculate the environmental impact assessment index.

8. The method for monitoring the environmental impact of land space planning according to claim 7, characterized in that: The dynamic monitoring results including spatiotemporal evolution diagrams, trend analysis charts and warning information are generated based on the preset visualization templates and warning rules according to the environmental impact assessment indicators, including: Inputting the environmental impact assessment index into a dynamic visualization network, generating an environmental impact spatiotemporal evolution diagram through a spatiotemporal interpolation algorithm and gradient color mapping, and obtaining a visualization base layer; Performing multi-dimensional feature analysis on the visualization basic layer, generating an environmental change trend chart through a trend prediction model, and obtaining a trend analysis result; Based on the trend analysis results, an early warning assessment model is constructed, and the environmental risk level and spatial distribution are identified through a deep learning algorithm to generate early warning information.

9. The method for monitoring the environmental impact of land space planning according to claim 8, characterized in that: The early warning assessment model is constructed based on the trend analysis results, and the environmental risk level and spatial distribution are identified through a deep learning algorithm to generate early warning information, including: Extracting time series features from the trend analysis results, predicting the environmental risk evolution trend through a recurrent neural network, and obtaining a risk prediction sequence; Based on the risk prediction sequence, a multi-layer early warning network is constructed, and high-risk areas and key influencing factors are identified through an attention mechanism to obtain early warning evaluation results; The warning assessment results are graded and spatially clustered to generate warning information.

10. A national land space planning environmental impact monitoring device, characterized in that: include: A data collection and preprocessing module, used for collecting and preprocessing at least two types of multi-source environmental monitoring data including satellite remote sensing images, meteorological station data, surface coverage data and ecological environment monitoring data; The spatiotemporal feature extraction module is used to extract spatiotemporal features from the preprocessed multi-source environmental monitoring data using a self-supervised spatiotemporal graph mask transfer attention network to obtain initial spatiotemporal features; A feature fusion module, used for fusing the initial spatiotemporal features based on a multimodal diffusion transformer to obtain fused features; An evaluation index calculation module, used to calculate environmental impact assessment indicators including land use change index, ecological damage degree and environmental quality comprehensive index based on the fusion features; The visual early warning module is used to generate dynamic monitoring results including spatiotemporal evolution diagrams, trend analysis charts and early warning information based on preset visual templates and early warning rules according to the environmental impact assessment indicators.

Images