Urban ecological environment evaluation model construction method and system

Abstract

The invention relates to an urban ecological environment evaluation model construction method. The method comprises the following steps: inputting a remote sensing image, night light annual synthesis data and auxiliary space data; calculating a normalized differential vegetation index and an enhanced vegetation index; calculating the surface temperature; calculating urban heat island intensity as a pressure index; producing land cover data; calculating primary indexes of each criterion layer in the model, and performing gridding and normalization processing; designing primary indexes of the model; carrying out improved normal cloud combination empowerment; comprehensive evaluation based on the closeness of cloud similarity is carried out; and outputting a final comprehensive evaluation product. Performing space-time dynamic analysis and driving mechanism analysis; future scene simulation based on the improved LSTM model is carried out; and outputting a final comprehensive prediction product. The invention also relates to an urban ecological environment evaluation model construction system. According to the method, a new generation of urban ecological environment quantitative evaluation model with cognitive robustness, spatial insight and dynamic predictability can be constructed.

Description

Technical Field

[0001] This invention relates to a method and system for constructing an urban ecological environment assessment model. Background Technology

[0002] With the acceleration of urbanization, urban ecosystems have become complex mega-systems composed of a strong interaction between natural foundations and human and social elements. Their health directly impacts sustainable development and residents' well-being. Remote sensing technology, with its macroscopic, dynamic, and objective advantages, has become an indispensable tool for urban ecological monitoring. However, how to accurately quantify the entire process of "driving forces-stress-state-services" in urban ecosystems from massive, multi-source remote sensing data, and effectively characterize its spatial heterogeneity, temporal dynamics, and the uncertainty of evaluation and cognition, is a cutting-edge challenge facing current urban ecology research and practice. Shenzhen, as a typical representative of rapid urbanization in China, with its unique "mountain, sea, city" pattern and high-intensity development and construction activities, makes the construction of a quantitative evaluation model that can accurately reflect its ecological background, respond to human disturbances, and support future scenario projections of great theoretical and practical significance.

[0003] Currently, research and practice in the field of urban ecological environment assessment, both domestically and internationally, are mostly based on the classic "Stress-State-Response" (PSR) model or its extended model framework, constructing a static comprehensive evaluation index system. Among existing publicly available solutions, the implementations most similar to this invention can be mainly divided into two categories: The first category is a comprehensive index evaluation model based on deterministic weights. This type of solution typically uses the Analytic Hierarchy Process (AHP) or entropy weight method to determine the index weights, and synthesizes a comprehensive index through linear weighting to characterize the ecological environment quality. Although this method has a clear structure and is computationally simple, its weight determination process often separates subjective and objective information and completely ignores the inherent fuzziness and randomness (i.e., cognitive uncertainty) in expert judgment, resulting in evaluation results that are overly sensitive to subjective settings and lack robustness. The second category is an evaluation model that introduces simple spatiotemporal analysis. This type of solution, based on the comprehensive index, supplements it with tools such as trend analysis or geographic detectors to reveal the spatiotemporal evolution patterns of the ecological environment and its driving factors. However, most of these models fail to deeply reveal the spatial non-stationarity of driving relationships, meaning that the intensity and direction of the impact of the same driving factor on the ecological environment may dynamically change in different geographical locations. More importantly, these models are essentially "post-hoc" assessments, lacking the ability to dynamically simulate and predict the future evolution trajectory of the system, making it difficult to support forward-looking ecological risk early warning and planning decisions. Therefore, while existing approximate solutions can, to some extent, achieve a "snapshot" assessment of the current state of the urban ecological environment, they have systemic shortcomings in three key dimensions: handling cognitive uncertainty, analyzing the driving mechanisms of spatial heterogeneity, and achieving dynamic scenario prediction. They cannot meet the modern demand for accurate, robust, and forward-looking evaluation of highly complex and dynamically evolving urban ecosystems.

[0004] It is evident that the core flaws of existing urban ecological environment assessment models lie in the static nature of their assessment framework, the rigidity of their weight determination, and the averaging of their driving force analysis. These limitations make it difficult for them to accurately depict the complex dynamics of the ecological environment in rapidly urbanizing areas. Specifically: (1) Weight determination ignores cognitive uncertainty: Existing solutions mostly adopt deterministic weight allocation methods (such as AHP or entropy weight method), or simply combine subjective and objective weights in a linear manner. These methods completely ignore the inherent fuzziness and randomness of domain experts in the judgment process, simplify the weights to fixed values, resulting in evaluation results that are abnormally sensitive to subjective judgments, lack robustness, and cannot reflect the real cognitive process of multi-factor trade-offs in complex ecosystems.

[0005] (2) Lack of spatial nonstationarity perspective in the analysis of driving mechanisms: Although some advanced models have introduced driving factor analysis, they are mostly based on the global regression assumption (such as OLS), which implicitly assumes that the impact of a certain driving factor on the ecological environment is homogeneous throughout the study area. This is seriously inconsistent with the reality of highly heterogeneous urban ecosystems, making it impossible to reveal the spatial differentiation law of "different effects of the same factor", which greatly restricts the formulation of differentiated and precise environmental governance strategies.

[0006] (3) The model architecture is limited to current situation assessment and lacks dynamic foresight: Most existing models are essentially "static snapshots" of the past or present situation. Although they can assess the past and present situation, they are completely unable to simulate and predict future development trends. This makes it impossible for managers to predict the potential ecological consequences under different development strategies, and it is difficult to support forward-looking ecological risk prevention and spatial planning decisions.

[0007] In summary, the aforementioned systemic flaws collectively pose serious challenges to the reliability of existing evaluation models, the depth of their mechanistic explanations, and the timeliness of their decision support when dealing with highly complex and rapidly evolving mega-city ecosystems. Summary of the Invention

[0008] In view of this, it is necessary to provide a method and system for constructing an urban ecological environment assessment model, which can construct a new generation of quantitative assessment models for urban ecological environment that combine cognitive robustness, spatial insight and dynamic predictability.

[0009] This invention provides a method for constructing an urban ecological environment assessment model, comprising the following steps: S1, inputting time-series Sentinel-2 remote sensing images, time-series Landsat-8 / 9 remote sensing images, annual composite data of nighttime light, and auxiliary spatial data; the auxiliary spatial data includes: the city's administrative divisions, road network, points of interest (POIs), and meteorological data; S2, calculating the Normalized Differential Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) using the input time-series Sentinel-2 remote sensing images; S3, calculating the Land Surface Temperature (LST) using the input time-series Landsat-8 / 9 remote sensing images; S4, calculating the Urban Heat Island Intensity (SUHI) as a pressure index using the LST; S5, generating land cover number data using the input time-series Sentinel-2 remote sensing images. According to LUCC; S6, using land cover data LUCC and POI, calculate the primary indicators of each criterion layer in the DPSPE-RS model, and perform gridding and normalization processing; then design the primary indicators of the model; S7, perform improved normal cloud combination weighting on all primary indicators to determine the final weight of each indicator in the model; S8, perform a comprehensive evaluation based on cloud similarity and proximity on all primary indicators and the determined final weight of each indicator in the model; S9, output the final comprehensive evaluation product of the above comprehensive evaluation; S10, perform spatiotemporal dynamic analysis and driving mechanism analysis on the above comprehensive evaluation; S11, perform future scenario simulation based on the improved LSTM model on the above comprehensive evaluation, spatiotemporal dynamic analysis and driving mechanism analysis; S12, output the final comprehensive prediction product.

[0010] Preferably, step S2 includes: The calculation formula is as follows:

[0011]

[0012] in, This represents the brightness value of the nth band in the Sentinel-2 data.

[0013] Preferably, step S3 includes: The calculation formula is as follows:

[0014] in, This indicates the brightness value of the thermal infrared band in Landsat-8 / 9 data; and The calibration constants are preset before launch and can be obtained from the metadata of the satellite data.

[0015] Preferably, step S4 includes: The calculation formula is as follows:

[0016] in, Indicates the first The surface temperature value of each pixel; This represents the average forest surface temperature during the same period.

[0017] Preferably, step S6 includes: (1) Driving force index: Based on NPP / VIIRS nighttime light data and the density of POIs of high-tech enterprises, the economic density index is calculated. The calculation formula is as follows:

[0018] in, This represents normalized NPP / VIIRS nighttime light intensity data; This represents the normalized kernel density data of POIs for high-tech enterprises. (2) Stress indicators: Ecological land fragmentation index is calculated based on land cover data LUCC; (3) Status indicators: Mangrove health index was calculated based on Sentinel-2 data. The calculation formula is as follows:

[0019] in, This represents the brightness value of the nth band in Sentinel-2 data; (4) Service indicators: vegetation cover is calculated based on NDVI and pixel bisection model. The calculation formula is as follows:

[0020] in, Indicates a pixel completely covered by bare soil. Mean; Representing pixels with complete vegetation cover Mean; Based on vegetation coverage Based on the empirical equation for vegetation biomass, the carbon sequestration and oxygen release are estimated using the following formula:

[0021]

[0022] in, Indicates the amount of carbon sequestration; This represents the amount of oxygen released; a and b are the coefficients of the empirical equation for vegetation biomass.

[0023] Preferably, step S7 includes: (1) Subjective weight Calculation: Using the analytic hierarchy process (AHP), experts in ecology, remote sensing, and urban planning were invited to conduct pairwise comparisons of the indicators, constructing a judgment matrix and calculating the eigenvectors as... If the consistency ratio CR < 0.1, then adjust the judgment matrix until consistency is satisfied. (2) Objective weight Calculation: The entropy is calculated using the entropy weight method, based on the dispersion of the data for each indicator. Coefficient of difference Finally obtained ; (3) Combination weights Calculation: Based on game theory, seek the optimal compromise between subjective and objective weights by minimizing... and , Based on the differences, construct the objective function:

[0024] The objective function is to find a combined weight. To make it consistent with subjective weight and objective weight The objective function is to minimize the sum of squared Euclidean distances, and then solve for the objective function to obtain the optimal linear combination coefficients. (4) Weight cloudification: Using a reverse cloud generator to transform deterministic combined weights Convert to a normal cloud model:

[0025] Among them, expected value ;entropy Calculated based on the degree of disagreement among experts regarding the importance of the indicators. , This represents the subjective weight vector given by the k-th expert; hyperentropy. according to The uncertainty setting.

[0026] Preferably, step S8 includes: (1) Generate a standard evaluation cloud: map the evaluation level to The interval is defined, and a standard normal cloud model is generated for each level. (2) Calculate the overall closeness: for each evaluation unit Its comprehensive index Instead of a direct weighted sum, it calculates the standardized values ​​of all its indicators. With corresponding weight cloud After the combined effect, it is used in conjunction with various standard evaluation cloud platforms. similarity The calculation formula is:

[0027] in, As an evaluation unit A comprehensive index, based on the weight of each indicator. A large number of instantaneous weights are randomly generated, and the standardized index values ​​are summed using these instantaneous weights each time. Indicates the first The expected value of the standard cloud model corresponding to each evaluation level is usually determined by expert experience, defining the evaluation domain. Division; Indicates the first The entropy of a standard cloud model for each evaluation level reflects the fuzziness of that level; Indicates the first Hyperentropy of a standard cloud model with a rating level; (3) Determine the final grade: Based on the principle of maximum similarity, determine the final evaluation grade of the unit:

[0028] Among them, the maximum value Corresponding level This is the final evaluation level for that unit.

[0029] Preferably, step S10 includes: (1) Spatiotemporal trend analysis: Using the MK nonparametric test and Sen slope estimation, the trend and significance of the SZ-EQI comprehensive index and the scores of each criterion layer in the time series were analyzed. (2) Driving mechanism analysis: The geographically weighted regression model was adopted, with the SZ-EQI score as the dependent variable and the indicators of the driving force layer and pressure layer as independent variables, to explore the influence of each driving factor on the spatial differentiation of ecological environment quality and its local effect; finally, the local t-test p-value diagram of each independent variable was output.

[0030] Preferably, step S11 includes: (1) Model construction: A long short-term memory network LSTM with an attention mechanism is constructed. The input sequence is the time series data of the SZ-EQI index and its key driving factors over the past N years. ; (2) Model training and prediction: 80% of the data is used for training and 20% for testing; by adjusting the number of hidden layer nodes and learning rate of LSTM, the training aims to minimize the root mean square error (RMSE) between the predicted value and the true value. (3) Scenario setting and simulation: Set different future development scenarios, change the values ​​of the input driving factors in future years, run the trained LSTM model, and predict the spatial pattern of the city's ecological environment quality.

[0031] This invention provides a system for constructing an urban ecological environment assessment model. The system includes an input module, a calculation module, a production module, a design module, a weighting module, an evaluation module, an output module, an analysis module, and a simulation module. The input module is used to input time-series Sentinel-2 remote sensing images, time-series Landsat-8 / 9 remote sensing images, annual composite data of nighttime light, and auxiliary spatial data. The auxiliary spatial data includes the city's administrative divisions, road network, points of interest (POIs), and meteorological data. The calculation module is used to calculate the Normalized Differential Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) from the input time-series Sentinel-2 remote sensing images. The calculation module is also used to calculate the Land Surface Temperature (LST) from the input time-series Landsat-8 / 9 remote sensing images. Furthermore, the calculation module is used to calculate the Urban Heat Island Intensity (SUHI) as a pressure indicator from the LST. The production module is used to input time-series Sentinel-2 remote sensing images, road network, points of interest (POIs), and meteorological data. The system uses INIEL-2 remote sensing imagery to produce Land Cover Data (LUCC). The design module calculates primary indices for each criterion layer in the DPSPE-RS model using LUCC and POI data, and performs gridding and normalization. Then, it designs the primary indices for the model. The weighting module assigns improved normal cloud combinations to all primary indices to determine their final weights in the model. The evaluation module performs a comprehensive evaluation based on cloud similarity and proximity, considering all primary indices and their final weights. The output module outputs the final comprehensive evaluation product. The analysis module performs spatiotemporal dynamic analysis and driving mechanism analysis on the comprehensive evaluation. The simulation module simulates future scenarios based on an improved LSTM model, considering the comprehensive evaluation, spatiotemporal dynamic analysis, and driving mechanism analysis. The output module also outputs the final comprehensive prediction product.

[0032] This application enables the construction of a DPSPE-RS quantitative dynamic evaluation model for urban ecological environment that integrates "driving force—pressure—state—products and services" and deeply combines remote sensing technology and cloud model theory. By introducing an improved normal cloud combination weighting method to handle cognitive uncertainty, combining geographically weighted regression to mine spatial non-stationary driving relationships, and integrating attention mechanism long short-term memory networks to achieve multi-scenario simulation, it achieves a systematic, precise, and forward-looking assessment of the urban ecological environment, providing quantifiable, traceable, and predictable decision-making basis for urban ecological governance and sustainable development. This invention has the following beneficial effects: (1) This invention achieves a paradigm shift from "rigid evaluation" to "flexible cognition," significantly improving the scientific rigor and robustness of the evaluation results. Existing technical solutions often employ rigid assignment or simple linear combinations in determining weights, completely ignoring the inherent ambiguity and randomness in expert cognition, leading to evaluation results that are exceptionally sensitive to subjective settings. This invention innovatively introduces an improved normal cloud combination weighting method, seeking an equilibrium solution between subjective and objective weights through game theory, and using a reverse cloud generator to convert it into a weight cloud model that considers cognitive disagreements. This expands the weights from a fixed value to a distribution representing uncertainty, mathematically quantifying and transmitting cognitive ambiguity. The final evaluation result is no longer a single, definite value, but a comprehensive judgment containing uncertain information, greatly enhancing the scientific rigor and decision robustness of the model in the evaluation of complex urban ecosystems.

[0033] (2) This invention breaks through the limitations of the "homogenization assumption" and achieves precise spatial diagnosis of the driving mechanisms of the ecological environment. Traditional driving analysis is mostly based on global regression models, which implicitly assume that the driving factors have a homogeneous influence throughout the study area, which is seriously inconsistent with the reality of highly heterogeneous urban ecosystems. This invention creatively introduces the geographically weighted regression (GWR) model, and by constructing local regression equations, it accurately reveals the unique influence intensity and direction (i.e., spatial non-stationarity) of each driving factor (such as economic density and heat island intensity) in different geographical locations. The spatial effect map of driving factors generated by this method can intuitively show "where, which factor, and how much influence," providing an unprecedented spatial explicit decision-making basis for implementing differentiated and precise environmental governance with "one policy for one area," and overcoming the macro-extensive drawbacks of traditional schemes that "only see the forest but not the trees."

[0034] (3) This invention represents a leap from "static snapshots" to "dynamic extrapolation," providing forward-looking decision support for urban ecological governance. Most existing models are essentially static assessment tools focused on the past or present, unable to predict the future, severely limiting their application value in strategic planning. This invention constructs a dynamic simulation system with temporal learning and prediction capabilities through a Long Short-Term Memory (LSTM) network integrating an attention mechanism. This system can not only capture the evolutionary patterns of the ecological environment from historical data, but also dynamically extrapolate the spatial pattern and uncertainty range of future ecological environment quality by setting different development scenarios (such as ecological priority and high-speed development). This enables decision-makers to assess the potential ecological consequences of different development paths in advance, achieving a revolutionary shift from "post-event assessment" to "pre-event warning" and "in-event simulation," providing a powerful "decision laboratory" for optimizing the path of sustainable urban development.

[0035] In summary, this invention is not a partial improvement on existing technologies, but rather a novel quantitative and dynamic evaluation paradigm for the urban ecological environment constructed through systematic and original innovations across multiple dimensions, including multi-source data fusion, quantification of cognitive uncertainty, spatial heterogeneity analysis, and dynamic scenario prediction. It represents a qualitative leap in the completeness of the evaluation theory, the advancement of the model architecture, and the effectiveness of decision support, demonstrating significant technological advantages. Attached Figure Description

[0036] Figure 1 This is a flowchart of the method for constructing an urban ecological environment assessment model according to the present invention; Figure 2 This is a hardware architecture diagram of the urban ecological environment assessment model construction system of the present invention. Detailed Implementation

[0037] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0038] See Figure 1 The diagram shown is a flowchart of a preferred embodiment of the urban ecological environment assessment model construction method of the present invention.

[0039] Step S1: Input time-series Sentinel-2 remote sensing imagery, time-series Landsat-8 / 9 remote sensing imagery, annual composite data of nighttime light, and auxiliary spatial data. The auxiliary spatial data includes: the city's administrative divisions, road network, POIs (Points of Interest), and meteorological data (temperature, precipitation). Specifically: In this embodiment, time-series Sentinel-2 remote sensing imagery, covering the entire Shenzhen area, is input to extract vegetation index data and land cover data; time-series Landsat-8 / 9 remote sensing imagery, covering the entire Shenzhen area, is input to extract surface temperature data; and time-series NPP / VIIRS annual composite nighttime light data, covering the entire Shenzhen area, is input to quantify the intensity of human activities. Users can selectively input auxiliary spatial data such as Shenzhen's administrative divisions, road network, POIs (points of interest), and meteorological data (temperature, precipitation).

[0040] Step S2: Using the input time-series Sentinel-2 remote sensing imagery, calculate the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI). Specifically: In this embodiment, the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) of Shenzhen are calculated using the input time-series Sentinel-2 data. The calculation formula is as follows:

[0041]

[0042] in, This represents the brightness value of the nth band in the Sentinel-2 data.

[0043] Step S3: Calculate the land surface temperature (LST) using the input time-series Landsat-8 / 9 remote sensing imagery. Specifically: In this embodiment, the land surface temperature (LST) of Shenzhen is calculated using the input Landsat-8 / 9 time-series data. The calculation formula is as follows:

[0044] in, This indicates the brightness value of the thermal infrared band in Landsat-8 / 9 data; and The calibration constants are preset before launch and can be obtained from the metadata of the satellite data.

[0045] Step S4: Calculate the urban heat island intensity (SUHI) as a pressure index using the land surface temperature (LST). Specifically: In this embodiment, the urban heat island intensity (SUHI) is calculated as a pressure index using land surface temperature (LST) data. The calculation formula is as follows:

[0046] in, Indicates the first The surface temperature value of each pixel; This represents the average forest surface temperature during the same period.

[0047] Step S5: Using the input temporal Sentinel-2 remote sensing imagery, generate land cover data (LUCC). Specifically: In this embodiment, land cover data (LUCC) for Shenzhen is generated from the input temporal Sentinel-2 data using visual interpretation.

[0048] Step S6: Using land cover data LUCC and POI, calculate the primary indices for each criterion layer in the DPSPE-RS model, and perform gridding and normalization processing; then design the primary indices for the model. Specifically: In this embodiment, primary indices for each criterion layer in the DPSPE-RS model are calculated using land cover data (LUCC) and points of interest (POI), and then... Spatial gridding and normalization were performed. Then, the primary indices of the model were designed: (1) Driving force index: Based on NPP / VIIRS nighttime light data and the density of POIs of high-tech enterprises, the economic density index is calculated ( The calculation formula is as follows:

[0049] in, This represents normalized NPP / VIIRS nighttime light intensity data; This represents the normalized kernel density data of POIs for high-tech enterprises.

[0050] (2) Stress index: Ecological land fragmentation index was calculated based on land cover data (LUCC) (Patch density PD was calculated using landscape pattern software Fragstats). (3) Status indicators: The mangrove health index was calculated based on Sentinel-2 data. The calculation formula is as follows:

[0051] in, This represents the brightness value of the nth band in the Sentinel-2 data.

[0052] (4) Service indicators: Vegetation coverage was calculated based on NDVI and a pixel-based bipartite model. The calculation formula is as follows:

[0053] in, Indicates a pixel completely covered by bare soil. Mean; Representing pixels with complete vegetation cover Mean.

[0054] Based on vegetation coverage ( Based on the empirical equation for vegetation biomass in Shenzhen, the carbon sequestration and oxygen release are estimated using the following formula:

[0055]

[0056] in, Indicates the amount of carbon sequestration; This represents the amount of oxygen released; a and b are coefficients of the empirical equation for vegetation biomass, which can be found in local ecological observation datasets.

[0057] Step S7 involves weighting the improved normal cloud combination for all primary indicators to determine the final weights of each indicator in the model. Specifically: An improved normal cloud portfolio weighting method is used to determine the final weight of each indicator in the model, based on all primary indicator data. The specific method is as follows: (1) Subjective weight ( Calculation: The Analytic Hierarchy Process (AHP) was used, and experts in ecology, remote sensing, and urban planning were invited to conduct pairwise comparisons of the indicators to construct a judgment matrix. The eigenvectors were then calculated as... If the consistency ratio CR < 0.1, then adjust the judgment matrix until consistency is satisfied.

[0058] (2) Objective weight ( Calculation: The entropy is calculated using the entropy weight method, based on the dispersion of each indicator data. Coefficient of difference Finally obtained Assuming there are m samples and n indicators, and a standardized matrix P is obtained, then the information entropy is... Coefficient of difference The calculation formula is as follows:

[0059]

[0060] in, Indicates the first Under this indicator, the first The proportion of each sample value ; The matrix element represents the normalized matrix P.

[0061] (3) Combination weights ( Calculation: Based on game theory, seek the optimal compromise between subjective and objective weights by minimizing... and , Based on the differences, construct the objective function:

[0062] The objective function is to find a combined weight. To make it consistent with subjective weight and objective weight The objective function minimizes the sum of squared Euclidean distances. Solving this objective function yields the optimal linear combination coefficients.

[0063] (4) Weighted cloudification: using a reverse cloud generator (a method that converts precise numerical data into three digital features of a cloud model). The algorithm will use deterministic combined weights. Convert to a normal cloud model:

[0064] Among them, expected value ;entropy Calculated based on the degree of disagreement (standard deviation) among experts regarding the importance of the indicator. , This represents the subjective weight vector given by the k-th expert; hyperentropy. according to The uncertainty setting is usually This move expands the weights from a fixed value to a distribution that takes into account cognitive uncertainty.

[0065] Step S8 involves performing a comprehensive evaluation based on cloud similarity, considering all primary indicators and their final weights in the model. Specifically: In this embodiment, a comprehensive evaluation based on cloud similarity is performed on all primary indicator data and the generated weighted cloud model. The specific algorithm is as follows: (1) Generate a standard evaluation cloud: Map the evaluation level (e.g., excellent, good, average, poor) to The interval is defined, and a standard normal cloud model is generated for each level. For example, the "Excellent" level can be represented as... ; (2) Calculate the overall closeness: for each evaluation unit Its comprehensive index Instead of a direct weighted sum, it calculates the standardized values ​​of all its indicators. With corresponding weight cloud After the combined effect, it is used in conjunction with various standard evaluation cloud platforms. similarity The calculation formula is:

[0066] in, As an evaluation unit The composite index is not a fixed value, but is obtained through Monte Carlo simulation, that is, using a forward cloud generator (the inverse algorithm of the reverse cloud generator), based on the weight of each indicator. A large number of instantaneous weights are randomly generated, and each time the standardized index values ​​are summed using these instantaneous weights (e.g., 1000). (sample mean) Indicates the first The expected value of the standard cloud model corresponding to each evaluation level (such as "Excellent") is usually determined by expert experience, defining the evaluation domain. Division, such as ,but ; Indicates the first The entropy of a standard cloud model with a rating level can be obtained through... The estimation reflects the ambiguity of this level; Indicates the first The hyperentropy of a standard cloud model with a rating level is usually set to a small value, such as 0.01, based on experience.

[0067] (3) Determine the final grade: Based on the principle of maximum proximity (i.e., select the grade with the highest membership degree as the final evaluation result of the unit), determine the final evaluation grade of the unit.

[0068]

[0069] Among them, the maximum value Corresponding level This is the final evaluation level for that unit.

[0070] Step S9: Output the final comprehensive evaluation product based on the above comprehensive evaluation. Specifically: In this embodiment, the final comprehensive evaluation product is output, including the comprehensive evaluation map of the Ecological Environment Quality Index (SZ-EQI), the evaluation maps of each criterion layer, and the grade classification results.

[0071] Step S10 involves conducting spatiotemporal dynamic analysis and analyzing the driving mechanism based on the above comprehensive evaluation. Specifically: In this embodiment, spatiotemporal dynamic analysis and driving mechanism analysis are performed on the evaluation results. The specific methods are as follows: (1) Spatiotemporal trend analysis: Using the MK (Mann-Kendall) nonparametric test and Sen's slope estimation, the trend and significance of the SZ-EQI comprehensive index and the scores of each criterion layer in the time series were analyzed.

[0072] The Mann-Kendall nonparametric test is a mature and readily available algorithm that can be used directly. Its result is a significance determination, i.e., at a given significance level... (Usually taken as 0.05) if If the result is positive, the null hypothesis of no trend is rejected, and a significant trend is considered to exist. Sen's Slope estimation is also a mature existing algorithm that can be used directly. Its result, Q, represents the interannual rate of change of the time series. Indicates an upward trend This indicates a downward trend. Combining Q and Z generates a trend classification chart. and For "significant improvement", and The value is marked as "significant degradation" and the rest as "not significant".

[0073] (2) Driving mechanism analysis: The geographically weighted regression (GWR) model was used, with the SZ-EQI score as the dependent variable and the indicators of the driving force layer and pressure layer as independent variables, to explore the influence of each driving factor on the spatial differentiation of ecological environment quality and its local effects.

[0074] The GWR model is a mature existing algorithm that can be used directly. The algorithm implementation path is as follows:

[0075] in, Indicates the first Spatial coordinates of a spatial unit; Indicates the first The first spatial unit The values ​​of the independent variables (indicators from the driving force and pressure layer, such as economic density and heat island intensity); Indicates the first The dependent variable value of each spatial unit (i.e., the SZ-EQI score of that unit). Indicates the first Local intercept term of each spatial unit; Indicates the first The first spatial unit Local regression coefficients of the independent variables; Indicates the first The random error term of each spatial unit.

[0076] For location Its coefficient This can be obtained by minimizing the following objective function:

[0077] in, Indicates position The observed values ​​at the estimated location The spatial weights of the coefficients are determined by the kernel function.

[0078] in, Indicates position With position The Euclidean distance between them; Bandwidth is the most important hyperparameter of the model, which determines the rate at which weights decay with distance. The optimal bandwidth can be automatically selected using the Akaike Information Criterion (AICc).

[0079] Finally, output the local t-test p-value plot for each independent variable.

[0080] Step S11 involves conducting a future scenario simulation based on the improved LSTM model, taking into account the aforementioned comprehensive evaluation, spatiotemporal dynamic analysis, and driving mechanism analysis. Specifically: In this embodiment, based on the evaluation results and driving analysis, a future scenario simulation based on an improved LSTM model is performed. The specific method is as follows: (1) Model Construction: Construct a Long Short-Term Memory (LSTM) network incorporating an attention mechanism. The input sequence is the sequence of the past N years (suggested). The application uses time-series data of the SZ-EQI index and its key driving factors (from GWR analysis results). An attention layer is used to automatically learn and weight the importance of states at different historical moments for prediction. The LSTM model is a mature existing algorithm that can be used directly and will not be described in detail here.

[0081] (2) Model Training and Prediction: 80% of the data was used for training and 20% for testing. The LSTM was trained by adjusting hyperparameters such as the number of hidden layer nodes and the learning rate, with the goal of minimizing the root mean square error (RMSE) between the predicted and true values. The objective was to reduce the RMSE on the test set by 20%-25% compared to the traditional ARIMA model. (3) Scenario setting and simulation: Set different future development scenarios (such as natural development, ecological priority, and high-speed development), and by changing the values ​​of the input driving factors in future years (such as setting the construction land expansion rate to 0 under the ecological priority scenario), run the trained LSTM model to predict the spatial pattern of Shenzhen's ecological environment quality in 2030 and 2035.

[0082] Step S12: Output the final comprehensive forecast product. Specifically: In this embodiment, the final comprehensive prediction product is output, including comprehensive prediction products of ecological environment quality prediction maps, change trajectories and uncertainty analysis under different future scenarios.

[0083] See Figure 2 The diagram shown is a hardware architecture diagram of the urban ecological environment evaluation model construction system 10 of the present invention. The system includes: an input module 101, a calculation module 102, a production module 103, a design module 104, a weighting module 105, an evaluation module 106, an output module 107, a parsing module 108, and a simulation module 109.

[0084] The input module 101 is used to input time-series Sentinel-2 remote sensing imagery, time-series Landsat-8 / 9 remote sensing imagery, annual composite data of nighttime light, and auxiliary spatial data. The auxiliary spatial data includes: the city's administrative divisions, road network, POIs (Points of Interest), and meteorological data (temperature, precipitation). Specifically: In this embodiment, the input module 101 inputs time-series Sentinel-2 remote sensing imagery, spatially covering the entire Shenzhen area, for extracting vegetation index data and land cover data; inputs time-series Landsat-8 / 9 remote sensing imagery, spatially covering the entire Shenzhen area, for extracting surface temperature data; inputs time-series NPP / VIIRS annual composite nighttime light data, spatially covering the entire Shenzhen area, for quantifying the intensity of human activities; users can selectively input auxiliary spatial data such as Shenzhen's administrative divisions, road network, POIs (points of interest), and meteorological data (temperature, precipitation).

[0085] The calculation module 102 is used to calculate the Normalized Differential Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI) from the input temporal Sentinel-2 remote sensing imagery. Specifically: In this embodiment, the calculation module 102 calculates the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) of Shenzhen City using the input time-series Sentinel-2 data. The calculation formula is as follows:

[0086]

[0087] in, This represents the brightness value of the nth band in the Sentinel-2 data.

[0088] The calculation module 102 is also used to calculate the land surface temperature (LST) using the input time-series Landsat-8 / 9 remote sensing imagery. Specifically: In this embodiment, the calculation module 102 calculates the land surface temperature (LST) of Shenzhen City using the input time-series Landsat-8 / 9 data. The calculation formula is as follows:

[0089] in, This indicates the brightness value of the thermal infrared band in Landsat-8 / 9 data; and The calibration constants are preset before launch and can be obtained from the metadata of the satellite data.

[0090] The calculation module 102 is also used to calculate the urban heat island intensity (SUHI) as a pressure index using the land surface temperature (LST). Specifically: In this embodiment, the calculation module 102 calculates the urban heat island intensity (SUHI) as a pressure index using land surface temperature (LST) data. The calculation formula is as follows:

[0091] in, Indicates the first The surface temperature value of each pixel; This represents the average forest surface temperature during the same period.

[0092] The production module 103 is used to produce land cover data (LUCC) from input temporal Sentinel-2 remote sensing imagery. Specifically: In this embodiment, the production module 103 produces land cover data (LUCC) for Shenzhen City using the input time-series Sentinel-2 data, and the production method is visual interpretation.

[0093] The design module 104 is used to calculate the primary indices of each criterion layer in the DPSPE-RS model using land cover data LUCC and POI, and then perform gridding and normalization processing; subsequently, it designs the primary indices of the model. Specifically: In this embodiment, the design module 104 calculates the primary indices of each criterion layer in the DPSPE-RS model using land cover data (LUCC) and POIs, and then performs... Spatial gridding and normalization were performed. Then, the primary indices of the model were designed: (1) Driving force index: Based on NPP / VIIRS nighttime light data and the density of POIs of high-tech enterprises, the economic density index is calculated ( The calculation formula is as follows:

[0094] in, This represents normalized NPP / VIIRS nighttime light intensity data; This represents the normalized kernel density data of POIs for high-tech enterprises.

[0095] (2) Stress index: Ecological land fragmentation index was calculated based on land cover data (LUCC) (Patch density PD was calculated using landscape pattern software Fragstats). (3) Status indicators: The mangrove health index was calculated based on Sentinel-2 data. The calculation formula is as follows:

[0096] in, This represents the brightness value of the nth band in the Sentinel-2 data.

[0097] (4) Service indicators: Vegetation coverage was calculated based on NDVI and a pixel-based bipartite model. The calculation formula is as follows:

[0098] in, Indicates a pixel completely covered by bare soil. Mean; Representing pixels with complete vegetation cover Mean.

[0099] Based on vegetation coverage ( Based on the empirical equation for vegetation biomass in Shenzhen, the carbon sequestration and oxygen release are estimated using the following formula:

[0100]

[0101] in, Indicates the amount of carbon sequestration; This represents the amount of oxygen released; a and b are coefficients of the empirical equation for vegetation biomass, which can be found in local ecological observation datasets.

[0102] The weighting module 105 is used to perform improved normal cloud combination weighting on all primary indicators to determine the final weight of each indicator in the model. Specifically: The weighting module 105 performs improved normal cloud combination weighting on all primary indicator data to determine the final weight of each indicator in the model. The specific method is as follows: (1) Subjective weight ( Calculation: The Analytic Hierarchy Process (AHP) was used, and experts in ecology, remote sensing, and urban planning were invited to conduct pairwise comparisons of the indicators to construct a judgment matrix. The eigenvectors were then calculated as... If the consistency ratio CR < 0.1, then adjust the judgment matrix until consistency is satisfied.

[0103] (2) Objective weight ( Calculation: The entropy is calculated using the entropy weight method, based on the dispersion of each indicator data. Coefficient of difference Finally obtained Assuming there are m samples and n indicators, and a standardized matrix P is obtained, then the information entropy is... Coefficient of difference The calculation formula is as follows:

[0104]

[0105] in, Indicates the first Under this indicator, the first The proportion of each sample value ; The matrix element represents the normalized matrix P.

[0106] (3) Combination weights ( Calculation: Based on game theory, seek the optimal compromise between subjective and objective weights by minimizing... and , Based on the differences, construct the objective function:

[0107] The objective function is to find a combined weight. To make it consistent with subjective weight and objective weight The objective function minimizes the sum of squared Euclidean distances. Solving this objective function yields the optimal linear combination coefficients.

[0108] (4) Weighted cloudification: using a reverse cloud generator (a method that converts precise numerical data into three digital features of a cloud model). The algorithm will use deterministic combined weights. Convert to a normal cloud model:

[0109] Among them, expected value ;entropy Calculated based on the degree of disagreement (standard deviation) among experts regarding the importance of the indicator. , This represents the subjective weight vector given by the k-th expert; hyperentropy. according to The uncertainty setting is usually This move expands the weights from a fixed value to a distribution that takes into account cognitive uncertainty.

[0110] The evaluation module 106 is used to perform a comprehensive evaluation based on cloud similarity proximity, considering all primary indicators and their final weights in the model. Specifically: In this embodiment, the evaluation module 106 performs a comprehensive evaluation based on cloud similarity closeness for all primary indicator data and the generated weighted cloud model. The specific algorithm is as follows: (1) Generate a standard evaluation cloud: Map the evaluation level (e.g., excellent, good, average, poor) to The interval is defined, and a standard normal cloud model is generated for each level. For example, the "Excellent" level can be represented as... ; (2) Calculate the overall closeness: for each evaluation unit Its comprehensive index Instead of a direct weighted sum, it calculates the standardized values ​​of all its indicators. With corresponding weight cloud After the combined effect, it is used in conjunction with various standard evaluation cloud platforms. similarity The calculation formula is:

[0111] in, As an evaluation unit The composite index is not a fixed value, but is obtained through Monte Carlo simulation, that is, using a forward cloud generator (the inverse algorithm of the reverse cloud generator), based on the weight of each indicator. A large number of instantaneous weights are randomly generated, and each time the standardized index values ​​are summed using these instantaneous weights (e.g., 1000). (sample mean) Indicates the first The expected value of the standard cloud model corresponding to each evaluation level (such as "Excellent") is usually determined by expert experience, defining the evaluation domain. Division, such as ,but ; Indicates the first The entropy of a standard cloud model with a rating level can be obtained through... The estimation reflects the ambiguity of this level; Indicates the first The hyperentropy of a standard cloud model with a rating level is usually set to a small value, such as 0.01, based on experience.

[0112] (3) Determine the final grade: Based on the principle of maximum proximity (i.e., select the grade with the highest membership degree as the final evaluation result of the unit), determine the final evaluation grade of the unit.

[0113]

[0114] Among them, the maximum value Corresponding level This is the final evaluation level for that unit.

[0115] The output module 107 is used to output the final comprehensive evaluation product of the above comprehensive evaluation. Specifically: In this embodiment, the output module 107 outputs the final comprehensive evaluation product, including the comprehensive evaluation map of the ecological environment quality index (SZ-EQI), the evaluation maps of each criterion layer, and the grade classification results.

[0116] The analysis module 108 is used to perform spatiotemporal dynamic analysis and driving mechanism analysis on the above comprehensive evaluation. Specifically: In this embodiment, the analysis module 108 performs spatiotemporal dynamic analysis and driving mechanism analysis on the evaluation results, and the specific method is as follows: (1) Spatiotemporal trend analysis: Using the Mann-Kendall nonparametric test and Sen's Slope estimation, the trend and significance of the SZ-EQI composite index and the scores of each criterion layer in the time series were analyzed.

[0117] The Mann-Kendall nonparametric test is a mature and readily available algorithm that can be used directly. Its result is a significance determination, i.e., at a given significance level... (Usually taken as 0.05) if If the result is positive, the null hypothesis of no trend is rejected, and a significant trend is considered to exist. Sen's Slope estimation is also a mature existing algorithm that can be used directly. Its result, Q, represents the interannual rate of change of the time series. Indicates an upward trend This indicates a downward trend. Combining Q and Z generates a trend classification chart. and For "significant improvement", and The value is marked as "significant degradation" and the rest as "not significant".

[0118] (2) Driving mechanism analysis: The geographically weighted regression (GWR) model was used, with the SZ-EQI score as the dependent variable and the indicators of the driving force layer and pressure layer as independent variables, to explore the influence of each driving factor on the spatial differentiation of ecological environment quality and its local effects.

[0119] The GWR model is a mature existing algorithm that can be used directly. The algorithm implementation path is as follows:

[0120] in, Indicates the first Spatial coordinates of a spatial unit; Indicates the first The first spatial unit The values ​​of the independent variables (indicators from the driving force and pressure layer, such as economic density and heat island intensity); Indicates the first The dependent variable value of each spatial unit (i.e., the SZ-EQI score of that unit). Indicates the first Local intercept term of each spatial unit; Indicates the first The first spatial unit Local regression coefficients of the independent variables; Indicates the first The random error term of each spatial unit.

[0121] For location Its coefficient This can be obtained by minimizing the following objective function:

[0122] in, Indicates position The observed values ​​at the estimated location The spatial weights of the coefficients are determined by the kernel function.

[0123] in, Indicates position With position The Euclidean distance between them; Bandwidth is the most important hyperparameter of the model, which determines the rate at which weights decay with distance. The optimal bandwidth can be automatically selected using the Akaike Information Criterion (AICc).

[0124] Finally, output the local t-test p-value plot for each independent variable.

[0125] The simulation module 109 is used to simulate future scenarios based on an improved LSTM model, taking into account the aforementioned comprehensive evaluation, spatiotemporal dynamic analysis, and driving mechanism analysis. Specifically: In this embodiment, the simulation module 109 performs future scenario simulation based on an improved LSTM model, taking into account the evaluation results and driving analysis. The specific method is as follows: (1) Model Construction: Construct a Long Short-Term Memory (LSTM) network incorporating an attention mechanism. The input sequence is the sequence of the past N years (suggested). The application uses time-series data of the SZ-EQI index and its key driving factors (from GWR analysis results). An attention layer is used to automatically learn and weight the importance of states at different historical moments for prediction. The LSTM model is a mature existing algorithm that can be used directly and will not be described in detail here.

[0126] (2) Model Training and Prediction: 80% of the data was used for training and 20% for testing. The LSTM was trained by adjusting hyperparameters such as the number of hidden layer nodes and the learning rate, with the goal of minimizing the root mean square error (RMSE) between the predicted and true values. The objective was to reduce the RMSE on the test set by 20%-25% compared to the traditional ARIMA model. (3) Scenario setting and simulation: Set different future development scenarios (such as natural development, ecological priority, and high-speed development), and by changing the values ​​of the input driving factors in future years (such as setting the construction land expansion rate to 0 under the ecological priority scenario), run the trained LSTM model to predict the spatial pattern of Shenzhen's ecological environment quality in 2030 and 2035.

[0127] The output module 107 is also used to output the final comprehensive prediction product. Specifically: In this embodiment, the output module 107 outputs the final comprehensive prediction product, including comprehensive prediction products of ecological environment quality prediction maps, change trajectories and uncertainty analysis under different future scenarios.

[0128] This invention overturns the traditional static evaluation paradigm and constructs a new generation of quantitative evaluation model for urban ecological environment that combines cognitive robustness, spatial insight, and dynamic predictability: (1) Introducing a cognitive uncertainty quantification mechanism: By integrating game theory combination weighting and the normal cloud model, an improved weight cloudification method is created. The purpose is to convert the determined weight values ​​into a weight distribution (cloud model) that takes into account the cognitive differences of experts, thereby mathematically representing and quantifying the fuzziness and randomness in the evaluation process, and significantly improving the scientificity and robustness of the evaluation results.

[0129] (2) Achieve explicit spatial analysis of driving relationships: By introducing the geographically weighted regression (GWR) model, the aim is to reveal the unique influence intensity and direction of each driving factor (such as economic density and heat island intensity) in different geographical locations (i.e., spatial non-stationarity), generate a spatial action map of driving forces, and provide direct and intuitive scientific basis for precise environmental management of "one policy for one region".

[0130] (3) Building dynamic simulation and scenario extrapolation capabilities: By integrating attention mechanism Long Short-Term Memory (LSTM) networks, the aim is to empower the model with the ability to learn evolution patterns from historical data and predict the future. By setting different development scenarios, the spatial pattern of future ecological environment quality is dynamically simulated, ultimately achieving a leap from "current status evaluation" to "future prediction", providing a powerful simulation experiment platform for the path selection and policy evaluation of urban sustainable development.

[0131] This invention not only realizes the paradigm shift from "static description" to "dynamic simulation" in evaluation, but also solves key scientific problems such as the quantification of cognitive uncertainty, the analysis of spatial heterogeneity-driven processes, and the prediction of future scenarios through a complete technology chain, providing a brand-new systematic solution for the refined governance and intelligent decision-making of urban ecological environment.

[0132] Although the present invention has been described with reference to the present preferred embodiments, those skilled in the art should understand that the above preferred embodiments are only used to illustrate the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing an urban ecological environment assessment model, characterized in that, The method includes the following steps: S1, Input time-series Sentinel-2 remote sensing imagery, time-series Landsat-8 / 9 remote sensing imagery, annual composite data of nighttime lights, and auxiliary spatial data; the auxiliary spatial data includes: the city's administrative divisions, road network, points of interest (POIs), and weather. S2, using the input time-series Sentinel-2 remote sensing images, calculates the Normalized Differential Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI). S3 calculates the land surface temperature (LST) using the input time-series Landsat-8 / 9 remote sensing images; S4, using the surface temperature LST to calculate the urban heat island intensity SUHI as a pressure index; S5 generates land cover data LUCC from the input temporal Sentinel-2 remote sensing imagery; S6. Using land cover data LUCC and POI, calculate the primary indices of each criterion layer in the DPSPE-RS model and perform gridding and normalization processing; then design the primary indices of the model. S7, Improved Normal Cloud Combination Weighting for all primary metrics to determine the final weights of each metric in the model; S8, based on cloud similarity and proximity, performs a comprehensive evaluation of all primary indicators and the final weights of each indicator in the model. S9 outputs the final comprehensive evaluation product based on the above comprehensive evaluation. S10, based on the above comprehensive evaluation, conduct spatiotemporal dynamic analysis and analyze the driving mechanism; S11, based on the above comprehensive evaluation, spatiotemporal dynamic analysis and driving mechanism analysis, conduct future scenario simulation based on the improved LSTM model; S12 outputs the final comprehensive forecast product.

2. The method as described in claim 1, characterized in that, Step S2 includes: The calculation formula is as follows: in, This represents the brightness value of the nth band in the Sentinel-2 data.

3. The method as described in claim 2, characterized in that, Step S3 includes: The calculation formula is as follows: in, This indicates the brightness value of the thermal infrared band in Landsat-8 / 9 data; and The calibration constants are preset before launch and can be obtained from the metadata of the satellite data.

4. The method as described in claim 3, characterized in that, Step S4 includes: The calculation formula is as follows: in, Indicates the first The surface temperature value of each pixel; This represents the average forest surface temperature during the same period.

5. The method as described in claim 4, characterized in that, Step S6 includes: (1) Driving force index: Based on NPP / VIIRS nighttime light data and the density of POIs of high-tech enterprises, the economic density index is calculated. The calculation formula is as follows: in, This represents normalized NPP / VIIRS nighttime light intensity data; This represents the normalized kernel density data of POIs for high-tech enterprises. (2) Stress indicators: Ecological land fragmentation index is calculated based on land cover data LUCC; (3) Status indicators: Mangrove health index was calculated based on Sentinel-2 data. The calculation formula is as follows: in, This represents the brightness value of the nth band in Sentinel-2 data; (4) Service indicators: vegetation cover is calculated based on NDVI and pixel bisection model. The calculation formula is as follows: in, Indicates a pixel completely covered by bare soil. Mean; Representing pixels with complete vegetation cover Mean; Based on vegetation coverage Based on the empirical equation for vegetation biomass, the carbon sequestration and oxygen release are estimated using the following formula: in, Indicates the amount of carbon sequestration; This represents the amount of oxygen released; a and b are the coefficients of the empirical equation for vegetation biomass.

6. The method as described in claim 5, characterized in that, Step S7 includes: (1) Subjective weight Calculation: Using the analytic hierarchy process (AHP), experts in ecology, remote sensing, and urban planning were invited to conduct pairwise comparisons of the indicators, constructing a judgment matrix and calculating the eigenvectors as... And satisfy the consistency ratio CR < 0.1, otherwise adjust the judgment matrix until consistency is satisfied; (2) Objective weight Calculation: The entropy is calculated using the entropy weight method, based on the dispersion of the data for each indicator. Coefficient of difference Finally obtained ; (3) Combination weights Calculation: Based on game theory, seek the optimal compromise between subjective and objective weights by minimizing... and , Based on the differences, construct the objective function: The objective function is to find a combined weight. To make it consistent with subjective weight and objective weight The objective function is to minimize the sum of squared Euclidean distances, and then solve for the objective function to obtain the optimal linear combination coefficients. (4) Weight cloudification: Using a reverse cloud generator to transform deterministic combined weights Convert to a normal cloud model: Among them, expected value ;entropy Calculated based on the degree of disagreement among experts regarding the importance of the indicators. , This represents the subjective weight vector given by the k-th expert; hyperentropy. according to The uncertainty setting.

7. The method as described in claim 6, characterized in that, Step S8 includes: (1) Generate a standard evaluation cloud: map the evaluation level to The interval is defined, and a standard normal cloud model is generated for each level. (2) Calculate the overall closeness: for each evaluation unit Its comprehensive index Instead of a direct weighted sum, it calculates the standardized values ​​of all its indicators. With corresponding weight cloud After the combined effect, it is used in conjunction with various standard evaluation cloud platforms. similarity The calculation formula is: in, As an evaluation unit A comprehensive index, based on the weight of each indicator. A large number of instantaneous weights are randomly generated, and the standardized index values ​​are summed using these instantaneous weights each time. Indicates the first The expected value of the standard cloud model corresponding to each evaluation level is usually determined by expert experience, defining the evaluation domain. Division; Indicates the first The entropy of a standard cloud model for each evaluation level reflects the fuzziness of that level; Indicates the first Hyperentropy of a standard cloud model with a rating level; (3) Determine the final grade: Based on the principle of maximum similarity, determine the final evaluation grade of the unit: Among them, the maximum value Corresponding level This is the final evaluation level for that unit.

8. The method as described in claim 7, characterized in that, Step S10 includes: (1) Spatiotemporal trend analysis: Using the MK nonparametric test and Sen slope estimation, the trend and significance of the SZ-EQI comprehensive index and the scores of each criterion layer in the time series were analyzed. (2) Driving mechanism analysis: The geographically weighted regression model was adopted, with the SZ-EQI score as the dependent variable and the indicators of the driving force layer and pressure layer as independent variables, to explore the influence of each driving factor on the spatial differentiation of ecological environment quality and its local effect; finally, the local t-test p-value diagram of each independent variable was output.

9. The method as described in claim 8, characterized in that, Step S11 includes: (1) Model construction: A long short-term memory network LSTM with an attention mechanism is constructed. The input sequence is the time series data of the SZ-EQI index and its key driving factors over the past N years. ; (2) Model training and prediction: 80% of the data is used for training and 20% for testing; by adjusting the number of hidden layer nodes and learning rate of LSTM, the training aims to minimize the root mean square error (RMSE) between the predicted value and the true value. (3) Scenario setting and simulation: Set different future development scenarios, change the values ​​of the input driving factors in future years, run the trained LSTM model, and predict the spatial pattern of the city's ecological environment quality.

10. A system for constructing an urban ecological environment assessment model, characterized in that, The system includes an input module, a calculation module, a production module, a design module, a weighting module, an evaluation module, an output module, a parsing module, and a simulation module, among which: The input module is used to input time-series Sentinel-2 remote sensing images, time-series Landsat-8 / 9 remote sensing images, annual composite data of nighttime lights, and auxiliary spatial data; the auxiliary spatial data includes: the city's administrative divisions, road network, points of interest (POIs), and weather. The calculation module is used to calculate the Normalized Differential Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI) from the input temporal Sentinel-2 remote sensing images. The calculation module is also used to calculate the land surface temperature (LST) using the input time-series Landsat-8 / 9 remote sensing images; The calculation module is also used to calculate the urban heat island intensity (SUHI) as a pressure index using the surface temperature (LST). The production module is used to produce land cover data LUCC from the input time-series Sentinel-2 remote sensing images; The design module is used to calculate the primary indices of each criterion layer in the DPSPE-RS model using land cover data LUCC and POI, and to perform gridding and normalization processing; then, the primary indices of the model are designed. The weighting module is used to assign weights to the improved normal cloud combination for all primary indicators in order to determine the final weight of each indicator in the model. The evaluation module is used to conduct a comprehensive evaluation based on cloud similarity and proximity for all primary indicators and the final weights of each indicator in the model. The output module is used to output the final comprehensive evaluation product of the above comprehensive evaluation; The analysis module is used to perform spatiotemporal dynamic analysis and driving mechanism analysis on the above comprehensive evaluation. The simulation module is used to simulate future scenarios based on an improved LSTM model, taking into account the above comprehensive evaluation, spatiotemporal dynamic analysis and driving mechanism analysis. The output module is also used to output the final comprehensive forecast product.

Images